When the media curation tool is launched, there may be no default definition already in place. In a sense, no description values may be allowed in the default state—all tracks may be filtered out because all descriptor values are effectively being filtered. The user may intervene to add allowed descriptor values to a definition, and thus see any tracks listed in the station library view. Descriptors may be selected at any level in any hierarchy.

A user may select any one of the various categories or “Artist Profile” from the category selector. To add any of the descriptor values displayed, the user may select a checkbox immediately to the left of the descriptor value name. A red check may appear in the check box and the descriptor name text color may change to red, indicating that that descriptor value is now added to the station definition; therefore tracks annotated with this descriptor value as the primary value of that type may be allowed in the station library. The primary value may be the descriptor value of a given type (e.g., genre) that has the greatest weight associated with it for the selected track (e.g., 75%) in cases where there is more than one value of the same type assigned to a single track.

There may be no limit to the number of descriptor values that can be selected for a single station definition, and the values may be selected from any combination of hierarchy levels from any combination of descriptor types.

Selecting a name or checkbox of any previously selected descriptor value may have the effect of removing it from the station definition, which may be indicated by the check not appearing in the check box and the text reverting to black from red. De-selecting a previously selected descriptor category may also de-select any and all child descriptor values. When any one or more child descriptor values, at any level in the hierarchy, are de-selected, the color of any parent descriptor values may revert to black, though the count of selected child descriptor values may still appear in the red oval of any descriptor value which may have one or more child values selected in the immediately adjacent child level.

To completely remove all values from the entire tree under a particular L1 root of a particular descriptor type, a user may de-select the root (e.g., “Peaceful” Mood or “Blues” Genre). To add back all values under the same L1 parent, the user may select the parent L1 node again.

For each tab in the category selector view, there may be a single “All” node representing all values of the descriptor type above the L1 descriptor values. Selecting the “All” node may select all values of the selected descriptor type, and all checkboxes may be selected for every node. Stated another way, there may be no filters on any content based on the primary descriptor value of the selected type.

De-selecting or “un-checking” the “All” node may de-select all values of the descriptor type. If one or more values in any other descriptor type are selected for the definition, this may have the effect of allowing all values of the selected descriptor type.

If a user wishes to include most values of a particular descriptor type in a station definition, and only filter out a small set, a user may start out by selecting the highest parent node applicable (including the “All”) root, and then prune away the child nodes that are not desired. If, on the other hand, a user wishes to only include a relatively small number of descriptor values of a particular descriptor type in a station definition, then the user may start out by ensuring the entire tree is de-selected, and only select the desired nodes by exception.

If the station definition specifies an unusual combination of descriptor values, the results displayed in the station library may be limited in number or null.

By default, station definitions may select a mix of content (e.g., a percent mix or mix percent) that is equally distributed across selected descriptor categories. To adjust this default mix (for example, to put a greater emphasis on Two-Tone “Ska Revival” versus “Synth Pop”), a user may first select a bars icon in the upper right of the station definition summary section in order to put the station definition summary into “genre mix” mode. Once in genre mix mode, the user may use slider bars to adjust the relative mix of each genre contained in the station definition. In an example case, the user can slide the “Ska Revival” mix slider bar to the right to increase the relative proportion of “Ska Revival” Tracks versus “Synth Pop” Tracks in the station library. Genre mix sliders for parent nodes may also adjust their child nodes in a proportional manner. The content of the station library may be re-computed after each adjustment of the genre mix. The genre mix is recorded as part of the station definition.

To set a primary descriptor type, a user may select a gear wheel icon 624 at the right of the station library and playlist results display header 630 to open up the station settings view dialog. In the station setting view dialog a user may change the primary descriptor type by using the drop-down menu to the right of “Primary Descriptor Type”. There may be various options (e.g., genre, mood, origin, era, artist type, none, etc.). Genre may be the default primary descriptor type.

Each time a change is made in the station definition (e.g., the user makes an update), the content of the station library may be recomputed. A user may review the content of the station library.

The station library and playlist results display 600 may provide various fields of information to the user in order to fully identify and understand the nature of each recording in the library. Some examples of fields includes rank, title (e.g., track or song title), artist, album, genre, mood, era, origin, artist type, popularity (POP), similarity (SIM), score (e.g., total score), and XID (e.g., external identifier). The fields described below are example fields, and embodiments described herein are not limited to only these fields. Some embodiments may include some, all, and/or other fields.

A rank field may be an integer-ordered rank of the track, starting at 1, based on the score of its underlying the recording. The track with the highest score may receive a rank of 1 and the track with the lowest score may have the highest rank value, equal to the number of recordings in the particular station library.

A title field may be the title of the composition performed in the recording contained in the selected release track. An artist field may be the recording artist who created the recording contained in the selected track. The album field may be the title of the album contained in the selected track.

The genre field may be the primary master genre code label associated with the recording contained in the selected track. A recording may be associated with more than one genre. The primary genre may be the one that has the greatest weight associated with it for the selected recording, in cases where more than one genre is assigned.

A mood field may be the primary master mood code label associated with the recording contained in the selected track. A recording may be associated with more than one mood. The primary mood may be the one that has the greatest weight associated with it for the selected recording, in cases where more than one mood is assigned.

An era field may be the primary master era code label associated with the recording contained in the selected track. A recording may be associated with more than one era. The primary era may be the one that has the greatest weight associated with it for the selected recording, in cases where more than one era is assigned. When a valid release year is available for the recording, the release year may be displayed; if no release year is available, the era(s) associated with the artist may be displayed.

An origin field may be a primary master origin code label associated with the recording contained in the selected track. A recording may be associated with more than one origin. The primary origin may be the one that has the greatest weight associated with it for the selected recording, in cases where more than one origin is assigned.

An artist type field may be a primary master artist type code label associated with the recording contained in the selected track. A recording may be associated with more than one artist type. The primary origin may be the one that has the greatest weight associated with it for the selected recording, in cases where more than one artist type is assigned.

A popularity (POP) field may be a popularity score associated with the recordings contained in the selected track. Popularity may be based on the combination of several different indicators of popularity. For example, popularity may be based on how many times an album with the track is searched for by users, how many times users recognize a song using fingerprint technology, music charts data for popular songs, and so forth.

A similarity (SIM) field may be a similarity score based on how strongly the recording's descriptor vectors align with the specific station definition. For example, for a station definition comprised of a single mood of “Gentle Bittersweet”, a recording which has a primary mood of Genre Bittersweet with a weight of 82 may have a higher similarity score than another one with a primary mood of Genre Bittersweet, but with a weight of only 64. Likewise, the former recording may also have a higher similarity score versus another recording that also has a primary Gentle Bittersweet weight of 82 but whose other mood vector values may have a lower correlation to “Gentle Bittersweet” than those of the former recording.

A score field may be an overall relevance score (e.g., the total score for a track), created via a weighted combination of the popularity and similarity scores (as described in further detail below). The relevance score may be what is used to provide the default ordering of tracks displayed in the station library results.

An external identifier field (XID) indicates whether one or more XIDs are available for the selected track. XIDs may be external identifiers for the tracks which can be used to link the track to third party catalogs.

The standard columnar display described above may only display the primary value assigned for each descriptor type. If there are two or more values assigned to the track for the given descriptor type, an indicator of +[count of additional values] may appear to the immediate right of the primary value name. For example if the track has three genres associated with it and the primary genre is Dream Pop, the genre display may appear as “Dream Pop (+2)”.

To view the other values assigned, a user may hover over (e.g., mouse-over or other method) the primary value and the other values may appear in an overlay window. Note that this display may not show the weights associated with each descriptor value, just the names of the values themselves.

A user may adjust the station library view by increasing the number of tracks displayed, adjusting the column width, changing the sort order, changing the view from full vector to brief view and vice versa, adjusting the station library size, and so forth.

To expose more tracks for review, a user may drag and drop a rounded tab at the top center of a station library display header. To reduce the number of tracks shown (and thus expose more of the station definition workspace), the user may drag and drop the tab back down.

The width of each station library display column may be adjusted by dragging the right border of the column header.

The sort order of the station library display may be changed by clicking on the column headers. The tracks in the station library may be sorted in the order of the column selected. To reverse the order of the selected sort, a user may click on the column header again.

The default view for both station library and playlist content may be “Full Vectors” where the primary value of each descriptor type may be shown in the appropriate column, and any other values contained in the same vector may be viewed by hovering over the value name as described above.

Alternatively, to simply viewing the track, album, and artist name metadata along with the popularity, similarity, and score values (e.g., without any of the descriptor values shown), a user may click on a gear wheel icon 624 at the right of the station library and playlist results display header 630 to open up a station settings view dialog. In the station settings view dialog the user may change the output format from “Full Vectors” to “Brief”. Using the drop-down menu. To return to the full vectors view, a user may select “Full Vectors”.

Station Libraries may have a predetermined maximum track count (e.g., 2,000, 500, 5,000, etc.). Stations are populated with as many unique tracks available in the selected catalog that meet the criteria of the station definition.

The curation tool application may provide the ability to export station libraries. In one example, to export a listing of a current station library appearing in a station library and playlist results display (e.g., 600), a user may first confirm that the station library (and not the sequenced playlist) that he wishes to export is displayed in the results display 600. Then the user may select an export icon 626 in the station library and playlist results display header. A dialog interface may appear that allows the user to enter a name for the station library export (e.g., Station Definition Name+Version #), then select “Export”. The listing may be exported as a .TSV file (as an example) and saved to the browser's default, or currently selected, download location, usually the user's “Downloads” folder.

Once a station library has been generated based on the station definition (as described in further detail below), the user may also view example playlists produced from the station library. This provides an opportunity for evaluation and further tuning of the listening experience by adjusting the station definition. The example playlist may also contain up to a predetermined maximum amount of tracks (e.g., 2,000), however, in the playlist the more relevant tracks may appear more than once in rotation, with the default rotation potentially containing only a subset (e.g., 300-400) of the most relevant of the total (e.g., 2,000) tracks available in the station library. For example, there may be specific rules (e.g., DCMA rules, rules specified by a user, etc.) for how often a particular track and/or album or artist may be played (e.g., more popular tracks played more often, a particular track may only be played once every 4 hours, a track by the same artist or the same album only once every 2 hours, etc.). It is possible for any of the tracks to appear on the play list, but the actual playlist may not have all of the tracks based on the various playlist rules. In one example, the system may choose the five best tracks that are eligible to be played (e.g., that may be played based on the rules) and choose randomly among them.

To review an example playlist generated from the currently selected station library, a user may first click on the gear wheel icon 624 in the station library results header to open up the station settings view dialog. In the station setting view dialog, the user may change the station view radio button to, for example, “Radio Sequence” from the default “Station Library (Rank)”.

The station library and playlist results display 600 may now show an example playlist comprised of tracks selected from the station library, ordered in a playout sequence with various rules enforced, such as minimum separation between tracks by the same artist and between the same track. In this view, the leftmost column may change from the default “Rank” 602 value for the underlying station library to the playlist sequence number.

The curation tool application may provide a user the ability to view various pie and bar charts showing the composition of the current playlist. For example, a user may select an option 628 in the station library and playlist results display header 630. By default, charts showing the genre mix may be displayed. The user may select alternate primary descriptor views of the playlist content (e.g., mood, origin, era, artist type, etc.), via a drop-down menu or other method. The categories shown may be at the most granular, or “leaf node” level, regardless of how the user defined the station. For example, if the user defined a station by selecting a single L1 genre of “Blues”, the genre chart may actually say “21 Genres” because this is the total count of all “leaf nodes” that are children of “Blues” in the Blues hierarchy. This may cause the distribution shown to be very dispersed, depending on the nature of the station definition.

To export a listing of the current sequenced playlist appearing in the station library and playlist results display 600, a user may first confirm that the example playlist (and not station library) that he wishes to export is displayed in the results display 600. Then the user may select an export icon 626 in the station library and playlist results display header 630. An “Export Tracklist” dialog may appear that allows the user to enter a name for the playlist and a version identifier (ID), and then select “Export”. The listing may be exported as a .TSV file (in one example) and saved to the browser's default, or currently selected download location, usually the user's “Downloads” folder.

The curation tool application allows a user to create, save, open, and delete station definitions via a menu items 704 as shown in example display 700. When a user initially launches or opens the curation tool application, it may already be in a new station mode, with no definition elements created. If a user has already set some elements and wishes to start over from scratch, the user may select an option 706 that allows the user to create a new station definition. Selecting the new option 706 may cause all existing definition elements to be cleared.

Once a user has defined a station definition to his satisfaction, the user may click on the option “Save As” 708. The curation tool application may then display an interface 800 as shown in FIG. 8 allowing a user to enter information about the station definition (e.g., station name) into the interface 800 such as via text box 802. Once entered, the user may select “Save” 804 in the interface 800. The station name may now appear in the station library results header.

Referring again to FIG. 7, after making edits to an existing station definition, the user may select “Save” 710 to save the updated definition to the same station definition name already established.

A user may select “Open/Delete” 712 to open or delete an existing station definition. The curation tool application may then display an interface such as the example interface 900 shown in FIG. 9. The user may select the station definition (e.g., 902) he wishes to open or delete and then select “open” 904 or “delete” 906 depending on which operation he desires. The interface 900 may also include an option to “Cancel” 908 for the user to cancel the operation to open or delete a station library.

Example embodiments show the user interface of the media curation tool in English and based on a United States regional view. Other regional views may also be provided (e.g., China, Japan, Australia, etc.). A user may choose a different region. For example, if a user chooses Japan as the region, a different hierarchical tree structure for the presentation of the genre categories may be utilized that more closely aligns with the perception of the relative importance and affinities of individual genres to a Japanese music listener. For example, names of categories or greater relevance in Japan (e.g. J-Pop, Enka, Anime Music) may be exposed at a higher level in the hierarchy, whereas categories of lesser relevance in Japan (e.g. Bro Country, Gospel Choir) may be aggregated with similar genres at higher levels and only exposed if the user drills down to lower levels in the hierarchy. Additionally, the genre name labeling terminology, character set and language may also be localized (e.g., to be presented in Japanese or another relevant language depending on the region selected) in addition to the structural localization of the hierarchy.

To generate a station library based on criteria selected by the user, the server system 102 may have access to a master database 126 with a comprehensive collection of media such as various recordings for audio tracks. The user of the curation tool application may be associated with a third party that may only have access to a subset of the master database. For example, the third party may have negotiated rights to access particular media. The third party may have one or more catalogs of media which it may access, based on the licenses or other agreements with record labels, music publishers, content authors, and so forth. The third party may communicate the one or more catalogs to the server system 102 so that the server system 102 may ingest the one or more catalogs associated with the third party. In one example, the server system 102 may create a mapping whereby the tracks in the one or more catalogs of media from the third party may be mapped to the master database 126 so that the server system 102 may determine which audio tracks, for example, may be part of which catalog(s). Each track in the third party catalogs may have an external identifier (XID) that is used by the third party to identify the track. The XID may be stored in the mapping of the third party catalogs in the master database 126. Each third party catalog may have a catalog identifier, which may also be stored in the master database 126.

FIG. 10 is a flow chart illustrating aspects of a method 1000, according to some example embodiments, for generating a station library. For illustrative purposes, method 1000 is described with respect to the networked system 100 of FIG. 1. It is to be understood that method 1000 may be practiced with other system configurations in other embodiments.

At operation 1002, the server system 102 receives a request for a station library. For example, a user may select criteria for a station library via a curation tool application (e.g., client application 114), such as via a user interface of the curation tool application as described above. Each time the user selects or changes criteria for the station library, the client device 110 sends a request to generate a station library to the server system 102 (e.g., via API server 120). In this way the station library is generated and updated in real-time or near real-time as the user is selecting and changing criteria. The request may comprise criteria for the station library. For example, the request may comprise criteria such as one or more of genres, moods, eras, origins, artist types, artist profiles, catalogue identifiers, percent mix, and so forth. In an alternative embodiment, instead of sending a request each time a user selects or changes criteria for a station library, a request may be submitted once the user has completed putting together all of the criteria for the station library.

The server system 102 receives the request and filters one or more media content catalogs based on criteria included in the request, as shown in operation 1004. For example, the server system 102 via API server 120 may receive the request and determine one or more third party catalogs which are associated with the request. The one or more third party catalogs may be determined based on one or more catalog identifiers included in the request, based on a third party identifier from which the API server 120 could determine the one or more catalogs associated with the third party, or other means. The criteria in the request may include at least one genre. The server system 102 (e.g., via curation server 124) may filter the one or more catalogs based on the genre included in the request to generate a first list of candidate tracks for the station library that are associated with that genre (and/or sub-genre, mood, era, origin, artist type, artist profile, etc.). For example, the criteria may include a genre (punk) and an era (1970s). The server system 102 would filter the one or more catalogs to only tracks that are associated with 1970s punk.

At operation 1006, the server system 102 (e.g., via curation server 124) determines a total number of tracks to be selected for the first genre based on the criteria. In one example, there may be a predetermined maximum number of tracks for a station library. For illustration purposes we will use a predetermined maximum number of 2000 tracks. If there is only one genre included in the criteria, then the total number of tracks to be selected for the genre will be 2000. If there is more than one genre, the predetermined total number of tracks will be equally divided between genres if there is no percent mix specified in the request. For example, if there are two genres specified in the criteria (e.g., punk and indie) the total number of tracks to be selected for each genre would be 1000 (e.g., 50% for punk and 50% for indie). In another example, the user may specify a percent mix for each genre. For example, the user may specify 75% for a first genre (e.g., punk) and 25% for a second genre (e.g., pop). The criteria in the request may include the percent mix. Thus, the total number of tracks to be selected for the first genre would be 1500 and the total number of tracks to be selected for the second genre would be 500.

In operation 1008, the server system 102 generates a similarity score for each track of the first list of candidate tracks. The similarity score may be determined by comparing the criteria included in the request with qualities of each track in the first list of candidate tracks to come up with a similarity score for each track. Example algorithms for generating a similarity score are described in further detail below.

In operation 1010, the server system 102 (e.g., via curation server 124) determines a popularity score for each track of the first list of candidate tracks. The popularity score may be based on how many times an album with the track is searched for by users, how many times users recognize a song using fingerprint technology, music charts data for popular songs, and so forth. The popularity score may be determined in advance and associated with the track. For example, it may be stored with details for the track in one or more databases 126, and the curation server 124 may determine the popularity score by accessing the one or more database 126 to get the popularity score associated with each track in the first list of candidate tracks.

In operation 1012, the server system 102 (e.g., via curation server 124) generates a total score for each track by combining the similarity score and popularity score for each track of the first list of candidate tracks. In one example, the total score may be a weighted combination/sum based on a predetermined weight for the similarity score and a predetermined weight for the popularity score. For example, the predetermined weight for similarity score may be 25% and the predetermined weight for popularity score may be 75%, and thus the calculation of the total score would be 0.25*similarity score+0.75*popularity score. Using this example, if the similarity score is 904 and the popularity score is 387, the total score would be 516.25 (e.g., 0.25*904+0.75*387=516.25). The total score may be rounded to the nearest whole number. Thus, for this example the total score would be 516.

In operation 1014, the server system 102 (e.g., via curation server 124) selects the top ranked tracks for the first list of candidate tracks based on the total number of tracks to be selected for the first genre to generate a list of top ranked tracks for the genre. For example, the server system 102 may rank the first list of candidate tracks based on the total score for each track, and then take the top ranked tracks in the ranked list. For example, if the total number of tracks to be selected is 500, the server system 102 would take the top ranked 500 tracks (e.g., the 500 tracks with the highest total scores). In operation 1016, the server system 102 returns the list of top ranked tracks of the first genre as part of the station library. If there is only one genre, then the server system 102 returns the station library containing the top ranked tracks for the genre.

If there is more than one genre included in the criteria of the request, then the server system 102 generates a list of top ranked tracks for each genre. For example, the request may comprise criteria including a second genre. As described above, the server system 102 may filter at least one media catalog based on the criteria for the station library to generate a second list of candidate tracks for the station library. The server system 102 may determine the total number of tracks to be selected for the second genre based on the criteria. The server system 102 may generate a similarity score for each track of the second list of candidate tracks and determine a popularity score for each track of the second list of candidate tracks. The server system 102 may combine the similarity score and popularity score for each track of the second list of candidate tracks to generate a total score for each track of the second list of candidate tracks.

The server system 102 may select the top ranked tracks of the second list of candidate tracks based on the total number of tracks to be selected for the second genre to generate a list of top ranked tracks for the second genre. For example, the server system 102 may rank the second list of candidate tracks based on the total score for each track.

If there are other genres, similar steps will be taken for each genre. In some cases, there may not be enough tracks in the current catalog to meet the number of tracks to be selected for a particular genre. For example, the user may specify 50% old school punk, but the current catalog may only contains 810 old school punk tracks (instead of the 1000 to be selected). In one example embodiment, this may be addressed by taking the leftover amount to be selected (e.g., 290) and re-distributing to other selections according to the ratio selected by the user. For example, if the user chose 50% old school punk, 25% indie, and 25% alternative, the 810 would be redistributed to indie and alternative (e.g., 145 to indie and 145 to alternative since the percent mix for each is the same).

The server system 102 may combine the list of top ranked tracks for the first genre with the list of top ranked tracks for the second genre (and any other genre), and then return the station library that includes the list of top ranked tracks for the first genre and the list of top ranked tracks for the second genre (and the list of top ranked tracks for each of any further genres). The station library list may be displayed in the user interface sorted in order of total score, regardless of the genre. The station library list may comprise a list of tracks and at least one descriptor such as rank, title, artist, artist type, album, genre, mood, era, origin, popularity score, similarity score, total score, one or more catalog identifiers, one or more external identifiers, and so forth.

As explained above, the server system 102 may have access to a master database 126 with a comprehensive collection of media such as various recordings for audio tracks. Data related to the media may include descriptive data (also referred to herein as descriptors or qualities) that may be assigned to both artists and tracks. Descriptor values may have a unique identifier and label. There may be thousands of unique music descriptor values, for example, in the server system 102 (e.g., stored in one or more databases 126). For example, artists may have one or more genres, eras, artist types, origins, and so forth. In one example, an artist may have an example identifier for a “Retro Soul” genre of 2963, an example identifier for “Late 2000's” era of 2653, an example identifier “Female Solo-Solo Vocal” artist type of 2656, an example identifier for “Wales” origin of 4285, and so forth. In another example, tracks may have all of the above (e.g., genre, era, artist type, origin, etc.) and also moods (e.g., 34205 for “Dark Romantic Lively Rhythm”), tempos (e.g., 102130 for “130s”), and so forth. In one example, out of all of the descriptor types, only genre may be mandatory for a similarity calculation. The similarity calculation may handle missing descriptions of other types (e.g., eras, artist types, origins, moods, tempos, etc.).

Artists and tracks may have one or more of certain types of descriptors assigned. This is referred to herein as a vector. Vectors may consist of an array of one or more descriptor-weight pairs. For example, an era vector for the “Beastie Boys” may comprise:

Type
ID
Weight
Label

ERA
2645
40
1990's

ERA
2641
32
1980's

ERA
2650
14
2000's

ERA
34814
14
Early 2010's

The data in this vector tells us that the Beastie Boys were active from the 80's until the 2010's, with their most significant period of activity being in the 90's. The vector element with the highest weight may be referred to as the primary value. In this example, the Beastie Boys' primary era is the 90s. Weights in a vector may usually, but not always, add up to 100.

As another example, a vector with all of the descriptors for Echo & the Bunnymen may comprise:

Type
ID
Weight
Label

ATYPE
23710
100
Male Group-Solo Vocal

ERA
2641
32
1980's

ERA
34814
19
Early 2010's

ERA
2650
19
2000's

ERA
2645
19
1990's

ERA
2640
11
Late 70's

GENRE
2740
48
Original Post-Punk

GENRE
2748
26
Neo-Psychedelic

GENRE
19042
26
Pre-Grunge Alternative Rock

ORIGIN
4583
100
Liverpool

Vectors may be sorted by weight, as shown in this example. From this table we can see that Echo & The Bunnymen are a male group with solo vocals, formed in Great Britain in the late 70's, and active up until now. Their primary genre is “Original Post-Punk”, and their primary era is the 80's.

Below is a table showing the descriptors for the track “The District Sleeps Alone Tonight” by The Postal Service.

Type
ID
Weight
Label

ATYPE
23719
100
Mixed Group-Male Solo

Vocal

ERA
2650
100
2000's

GENRE
2755
75
Techno Pop

GENRE
2763
25
Indie Pop

MOOD
34182
21
Deep Dreamy Relaxing Beat

MOOD
34188
9
Lite Melancholy

MOOD
34179
9
Light Soft Soulful

MOOD
34181
7
Soft Sensual/Intimate

MOOD
34180
5
Lite Glossy Sensual Groove

MOOD
34184
5
Enigmatic/Brooding/

Mysterious

MOOD
34371
5
Energetic Emo-Yearning

MOOD
34205
4
Dark Romantic Lively Rhythm

MOOD
34400
4
Idealistic/Stirring

MOOD
34191
3
Determined/Bitter/

Serious/Jaded

ORIGIN
4440
100
Los Angeles

TEMPO MAIN
102154
99
150's

TEMPO MAIN
102077
79
70's

TEMPO MAIN
102303
40
300+

Within a given type of descriptor, correlate values may exist for each unique pair or descriptors. The correlate value is a measure of “closeness” between the two descriptors, and range within a theoretical maximum of −1,000 (opposite) to 1,000 (identical). Some example correlates may include:

Type
Descriptor 1
Correlation
Descriptor 2

GENRE
2942-Neo Soul
750
2963-Retro Soul

GENRE
2717-Classic New
300
2793-Psychobilly

York Punk

ERA
2601-1960's
−50
34807-2010's

ORIGIN
4440 Los Angeles
750
4446-San Francisco

ORIGIN
4440 Los Angeles
563
4508-New York City

ORIGIN
4440 Los Angeles
350
4583-Liverpool

For description pairs that do not have a correlate value defined in the data, a correlation value of 0 may be assumed.

An example similarity algorithm will now be described that takes all of the descriptors of two items (e.g. the station definition and a candidate track) and derives a single similarity value (e.g., between −1000 to 1000) between them. In one example the similarity score may be determined by comparing criteria included in a request for a station library (as a first item) with qualities of each track (as a second item) in a list of candidate tracks to come up with a similarity score for each track, as explained above.

The algorithm is a multi-step process consisting of a number of different value lookups and computations. The algorithm utilizes correlates and vector similarity to get the most out of the rich and fine-grained nature of descriptive data. For example, two artists most listeners would consider similar might have different (but highly correlating) primary genres. If one were to look for artists that are similar to the Pet Shop Boys (Primary Genre: 2756—Synth Pop) just on the primary genre alone, then one would miss both OMD and Tears For Fears (Primary Genre for both: 2752—New Romantic).

FIG. 11 is a flow chart illustrating aspects of method 1100 according to some example embodiments, for generating a similarity score. For illustrative purposes, method 1100 is described with respect to the networked system 100 of FIG. 1. It is to be understood that method 1100 may be practiced with other system configurations in other embodiments.

In one example embodiment, the similarity algorithm may be a two-step process. The first step is a quick pass/fail type of check based on how strongly the two items' primary genres (and if available, primary moods) correlate. If the correlation value for either of these is below a specified threshold, then the result of the first step is fail, the similarity algorithm ends, and the similarity score between the two items is considered to be 0. In one example, a specified threshold for a primary genre may be 350. In another example, (e.g., for a track) a specified threshold for a primary mood comparisons may be −160.

Accordingly, the server system 102 (e.g., via curation server 124) determines a correlation value for two items, as shown in operation 1102. The two items may be a station library and a track, two artists, two tracks, and so forth. In one example, the first item may be criteria for a station library and the second item may be a track.

For example, the server system 102 may access a correlation table (as explained above) in one or more databases 126 to look up the correlation value between the primary genres of the two items (e.g. a genre specified in criteria for a station library and a genre for a track). If the correlation value is below a specified threshold, the similarity algorithm ends and the similarity score between the two items is considered 0. If both of the items also have a primary mood, the server system 102 may access the correlation table to look up the correlation between the primary moods of the two items. If the correlation value is below a specified threshold, the similarity algorithm ends and the similarity score between the two items is considered 0.

At operation 1104, the server system 102 determines that the correlation value is not below the specified threshold if the correlation value for genre (and mood, if applicable) is not below the specified threshold. Thus, if the first step was passed, a weighted similarity score is computed at step two. A pseudo-code example of the two step similarity calculation process may comprise:

Step 1:

Look up correlation between primary genres of the two items.

If below genre correlation threshold:

Similarity = 0

Exit

If both items have primary mood, look up correlation between these

If below mood correlation threshold:

Similarity is 0

exit

Step 1 passed, proceed to Step 2

Step 2:

Compute weighted similarity score

Similarity = weighted similarity score

Exit

At operation 1106, the server system 102 generates a weighted similarity score. For the weighted similarity score, the similarity of each of the different descriptor types is computed, and then combined based on weights assigned to types.

A pseudo-code example for computing a weighted track similarity score (e.g., criteria for a station library to track similarity) may comprise:

station_library
⁢
_to
⁢
_track
⁢
_similarity
⁢
_score

=

(

50
*
mood_similarity

+
 

40
*
genre_similarity

+

5
*
era_similarity

+

3
*
origin_similarity

+
 

2
*

⁢
artisttype_similarity

)

/
100

;

The weighted algorithm may be tolerant of one or more items not containing a particular type of descriptor. In this case, the division in the computation may be adjusted to account for the missing type's weight. For example if a station library and track are compared and the mood is missing for either one of them, then the sum of genre, era, origin, and artist type similarity should be divided by 50 instead of 100.

Next, how to compute these individual similarities is described. In this example, the station definition and the descriptors assigned to tracks come in the form of vectors (as described above), and thus, the similarity between these are computed using a vector similarity algorithm.

The vector similarity is an iterative process to derive a single similarity score between two descriptor vectors of the same type. In sum, the server system 102 determines the pair of elements in the two vectors (e.g., one element from each) that have the highest correlation to each other. Once that is determined, the server system 102 allocates the weights from this correlation to the overall score. For example, the server system 102 may add the product of the correlation value and the common weight of the correlation to the running score sum kept by the server system 102, and the common weight is added to the allocated weight sum. This process is repeated, until all of the weights in either vector are allocated. The correlation sum is then divided by the allocated weight sum to arrive at the vector similarity value.

To begin with a simple example, the genre similarity between a station definition comprised of 2 genres and a track by Elliott Smith will be computed. The following table shows the station definition genre vector:

Type
ID
Weight
Label

GENRE
2780
64
Alternative Singer-

Songwriter

GENRE
2781
36
Experimental Rock

The following tables shows the Elliot Smith track's genre vector:

Type
ID
Weight
Label

GENRE
2780
56
Alternative Singer-

Songwriter

GENRE
71063
44
Indie Folk

The server system 102 determines the correlation values between the genres of the two items. For example, the server system 102 may access a correlation table (as explained above) in on or more databases 126 to look up the correlation value between the genres of the two items. The server system 102 may sort the correlation values between the genres of the two items from highest to lowest:

- 2780~2780=950
- 2780~71063=600
- 2781~2780=0
- 2781~71063=0

The server system 102 determines the pair of genres with the highest correlation. In this example, the highest correlation is genre 2780 and genre 2780, as both the station definition and the track have the same primary genre. The “common weight” between these is the lower of the two weights (e.g., 56). The server system 102 allocates the weight (e.g., adds the correlation and the common weight to the sums), and removes the allocated weights from the vectors. The result may comprise:

Type
ID
Weight
Label

GENRE
2780
8
Alternative Singer-

Songwriter

GENRE
2781
36
Experimental Rock

Station Definition genre vector

Type
ID
Weight
Label

GENRE
2780
0
Alternative Singer-

Songwriter

GENRE
71063
44
Indie Folk

Elliot Smith Track genre vector

The sum after the first step is 53200 (e.g., 950*56=53200). The common weight of 56 is allocated, and so the weightsum is 56. Accordingly, the Station Definition has a weight of 8 left for genre 2780, whereas the Elliott Smith track's entire weight for genre 2780 was consumed.

The server system 102 repeats this process for each genre pair. For example, the server system 102 determines the next pair with the highest correlation. In this example, this is 2780 and 71063, with a correlation of 600, and the common weight is 8. Once allocated and added up, we end up with:

Type
ID
Weight
Label

GENRE
2780
0
Alternative Singer-

Songwriter

GENRE
2781
36
Experimental Rock

Station Definition genre vector

Type
ID
Weight
Label

GENRE
2780
0
Alternative Singer-

Songwriter

GENRE
71063
36
Indie Folk

Elliot Smith track genre vector

The server system 102 adds 4800 (e.g., 600*8=4800) to the sum to get 58000, and 8 to the weightsum to get 64. The process is repeated once again for the next highest correlation, which is the only one left now with unallocated weights (2781, 71063), with a correlation of 0, and a common weight of 36. Once allocated, the result comprises:

Type
ID
Weight
Label

GENRE
2780
0
Alternative Singer-

Songwriter

GENRE
2781
0
Experimental Rock

Station Definition genre vector

Type
ID
Weight
Label

GENRE
2780
0
Alternative Singer-

Songwriter

GENRE
71063
0
Indie Folk

Elliot Smith Track genre vector

The sum remains unchanged at 58000 and the weightsum is 100 (64+36).

Now, there are no more unallocated weights left in either vector. (Though in most cases both vectors will have no unallocated weights left at the end of the iterations, technically the iterations are complete when either vector has no more unallocated weights.) The server system 102 calculates the vector similarity by dividing the sum by the weightsum. Accordingly, the genre similarity between the Station Definition and the Elliott Smith track is 58000/100=580.

The algorithm may be the same, no matter how many elements the compared vectors have, or what types of descriptor vectors are being compared, with one notable exception of mood. Mood adds a slight modification to the algorithm.

Mood vectors may be a bit different from other vectors, in that their weights frequently may not add up to 100, but may add up to less than 100. This is because the mood vector may actually only contain, for example, the top ten most strongly weighted available mood values for each track, out of a total number (e.g., 101) available mood values for each track. If the vector similarity algorithm is applied as described above, it may lead to a mood similarity score that is disproportionately far in either the “similar” or “different” direction. Because of this, for mood vector similarity, the unallocated (or unknown) weights are taken into consideration, with an assumed correlation sum of half of the allocated weights' correlation sum. This effectively pulls the similarity score towards zero, particularly if there is a large portion of the weights unallocated.

The following is an example mood vector similarity calculation and output between two mood vectors:

Computing MOOD similarity:

0. 34212/39 (950) 34212/3

Sum=2850, Weightsum=3

1. 34373/28 (950) 34373/5

Sum=7600, Weightsum=8

2. 34201/6 (950) 34201/3

Sum=10450, Weightsum=11

3. 34387/3 (950) 34387/17

Sum=13300, Weightsum=14

4. 34203/1 (750) 34411/3

Sum=14050, Weightsum=15

5. 34201/3 (700) 34411/2

Sum=15450, Weightsum=17

6. 34388/3 (600) 34391/5

Sum=17250, Weightsum=20

7. 34212/36 (400) 34389/20

Sum=25250, Weightsum=40

8. 34180/2 (300) 34170/7

Sum=25850, Weightsum=42

9. 34212/16 (200) 34387/14

Sum=28650, Weightsum=56

10. 34373/23 (200) 34399/8

Sum=30250, Weightsum=64

11. 34373/15 (200) 34391/2

Sum=30650, Weightsum=66

12. 34373/13 (100) 34170/5

Sum=31150, Weightsum=71

13. 34212/2 (−600) 34384/3

Sum=29950, Weightsum=73

14. 34201/1 (−700) 34384/1

Sum=29250, Weightsum=74

--------------------

Known = 395 / 74

Unknown = 197 / 26

====================

Result = 343

As can be seen above, after 14 iterations of the allocations, we ended up with an allocated weightsum of only 74. To account for the missing weightsum of 26, we assume a correlation sum of 197 (half of 395, in integer arithmetic), and incorporate both sets (known and unknown) into the final score, according to their weights. The result is ((395*74)+(197*26))/100=343.

In sum, to compute the music similarity between two items is a two-step process. Step one is a quick comparison of primary genre (and for tracks, both primary genre and primary mood). If step one fails, the similarity score is 0. If step one is passed, then a more in-depth weighted similarity score is computed. For the components of the weighted score, which are based on vectors (e.g., genre, era, mood), the vector similarity algorithm is used, and for mood vector similarity only, we take the “unknown weights” into consideration.

Some example output showing the entire process will next be described.

The following is an example similarity calculation comparing an Indie-Pop Station Definition to a track by The Ladybug Transistor:

Name=”Indie Pop Focus” Station Definition

GOETs (6):

1. Type=ATYPE, ID=2662, Weight=100, Name=Mixed Group - Mixed

Duo/Group Vocal

2. Type=ERA, ID=2650, Weight=64, Name=2000's

3. Type=ERA, ID=34814, Weight=36, Name=Early 2010's

4. Type=GENRE, ID=2763, Weight=64, Name=Indie Pop

5. Type=GENRE, ID=2765, Weight=36, Name=Chamber Pop

6. Type=ORIGIN, ID=4401, Weight=100, Name=Montreal

ID=149069

Name=Track by The Ladybug Transistor

GOET's (6):

1. Type=ATYPE, ID=2662, Weight=100, Name=Mixed Group -

Mixed Duo/Group Vocal

2. Type=ERA, ID=2650, Weight=75, Name=2000's

3. Type=ERA, ID=34814, Weight=36, Name=Early 2010's

4. Type=GENRE, ID=2764, Weight=75, Name=Baroque Pop

5. Type=GENRE, ID=2763, Weight=25, Name=Indie Pop

6. Type=ORIGIN, ID=4509, Weight=100, Name=Brooklyn

STEP 1: 2763 (775) 2764 >= 350 −> PASS

STEP 2: (Genre=60 Era=20 Origin=10 Artist Type=10)

Computing GENRE similarity:

0. 2763/64 (900) 2763/25

Sum=22500, Weightsum=25

1. 2765/36 (900) 2764/75

Sum=54900, Weightsum=61

2. 2763/39 (775) 2764/39

Sum=85125, Weightsum=100

--------------------

Known = 851 / 100

====================

Result = 851

Computing ERA similarity:

0. 2650/64 (935) 2650/75

Sum=59840, Weightsum=64

1. 34814/36 (935) 2650/11

Sum=70125, Weightsum=75

2. 34814/25 (680) 2648/25

Sum=87125, Weightsum=100

--------------------

Known = 871 / 100

====================

Result = 871

Computing ORIGIN similarity:

SINGLE: 4401 (765) 4509

Computing ARTIST TYPE similarity:

SINGLE: 2662 (975) 2662

RESULT:

Step1 = PASS

Step2:

* Genre = 851

* Origin = 765

* Era = 871

* Artist Type = 975

* Weighted = 858

====================

SIMILARITY = 858

Step one is passed, since the correlation between the primary genres is high enough (e.g., 775). The similarity score then is the result of the weighted similarity computation, which includes the genre and era vectors, and the simple origin and artist type correlations.

The following is an example artist similarity calculation comparing a “Serious Synth Pop” Station Definition to a track by Nine Inch Nails:

GOETs (11):

1. Type=ATYPE, ID=2660, Weight=100, Name=Male Group -

Duo/Group Vocal

2. Type=ERA, ID=2644, Weight=49, Name=Late 80's

3. Type=ERA, ID=2645, Weight=17, Name=1990's

4. Type=ERA, ID=2650, Weight=17, Name=2000's

5. Type=ERA, ID=2641, Weight=17, Name=1980's

6. Type=GENRE, ID=2756, Weight=44, Name=Synth Pop

7. Type=GENRE, ID=2700, Weight=14, Name=Alternative Dance

8. Type=GENRE, ID=23672, Weight=14, Name=Darkwave

9. Type=GENRE, ID=2752, Weight=14, Name=New Romantic

10. Type=GENRE, ID=2741, Weight=14, Name=Brit Rock

11. Type=ORIGIN, ID=4578, Weight=100, Name=England

Name=Track by Nine Inch Nails

GOET's (7):

1. Type=ATYPE, ID=2655, Weight=100, Name=Male Solo - Solo

Vocal

2. Type=ERA, ID=2645, Weight=60, Name=1990's

3. Type=ERA, ID=2650, Weight=20, Name=2000's

4. Type=ERA, ID=2644, Weight=20, Name=Late 80's

5. Type=GENRE, ID=2694, Weight=75, Name=Industrial Dance

6. Type=GENRE, ID=2698, Weight=25, Name=Alternative Rock

7. Type=ORIGIN, ID=4521, Weight=100, Name=Cleveland

STEP 1: 2756 (0) 2694 < 350 −> FAIL

STEP 2: (Genre=60 Era=20 Origin=10 Artist Type=10)

Computing GENRE similarity:

0. 23672/14 (850) 2694/75

Sum=11900, Weightsum=14

1. 2741/14 (650) 2698/25

Sum=21000, Weightsum=28

2. 2700/14 (600) 2698/11

Sum=27600, Weightsum=39

3. 2756/44 (0) 2694/61

Sum=27600, Weightsum=83

4. 2700/3 (0) 2694/17

Sum=27600, Weightsum=86

5. 2752/14 (0) 2694/14

Sum=27600, Weightsum=100

--------------------

Known = 276 / 100

====================

Result = 276

Computing ERA similarity:

0. 2644/49 (970) 2644/20

Sum=19400, Weightsum=20

1. 2645/17 (935) 2645/60

Sum=35295, Weightsum=37

2. 2650/17 (935) 2650/20

Sum=51190, Weightsum=54

3. 2644/29 (798) 2645/43

Sum=74332, Weightsum=83

4. 2641/17 (680) 2645/14

Sum=83852, Weightsum=97

5. 2641/3 (630) 2650/3

Sum=85742, Weightsum=100

--------------------

Known = 857 / 100

====================

Result = 857

Computing ORIGIN similarity:

SINGLE: 4578 (356) 4521

Computing ARTIST TYPE similarity:

SINGLE: 2660 (475) 2655

RESULT:

Step1 = FAIL

Step2:

* Genre = 276

* Origin = 356

* Era = 857

* Artist Type = 475

* Weighted = 420

====================

SIMILARITY = 0

As can be seen above, step one failed, since the primary genres are too dissimilar. The weighted similarity score computation is shown just for completeness; it is not necessary to do it when step one fails. In one example embodiment, no matter what the weighted similarity score is, the final similarity score will always be 0, if step one fails.

The following is an example track similarity calculation comparing an “Edgy Urban Crossover” Station Definition and the Track “What You Waiting For” by Gwen Stefani. This example incorporates the mood attribute into the station definition.

Name=”Edgy Urban Crossover” Station Definition

GOETs (14):

1. Type=ATYPE, ID=2656, Weight=100, Name=Female Solo -

Solo Vocal

2. Type=ERA, ID=2652, Weight=60, Name=Mid 2000's

3. Type=ERA, ID=2653, Weight=20, Name=Late 2000's

4. Type=ERA, ID=34814, Weight=20, Name=Early 2010's

5. Type=GENRE, ID=19002, Weight=53, Name=Urban Crossover:

R&B

6. Type=GENRE, ID=2853, Weight=29, Name=Dance Pop

7. Type=GENRE, ID=19041, Weight=18, Name=Dancehall:

Singer

8. Type=MOOD, ID=34213, Weight=91, Name=Dark Intense Pop

Determination

9. Type=MOOD, ID=34182, Weight=1, Name=Deep Dreamy

Relaxing Beat

10. Type=MOOD, ID=34398, Weight=1, Name=Edgy / Sexy

11. Type=ORIGIN, ID=4004, Weight=100, Name=Barbados

Name= Track “What You Waiting For?” by Gwen Stefani

GOET's (9):

1. Type=ATYPE, ID=2656, Weight=100, Name-Female Solo -

Solo Vocal

2. Type=ERA, ID=2652, Weight=100, Name=Mid 2000's

3. Type=GENRE, ID=2853, Weight=100, Name=Dance Pop

4. Type=MOOD, ID=34213, Weight=96, Name=Dark Intense Pop

Determination

5. Type=MOOD, ID=34393, Weight=1, Name=Arousing /

Energizing Groove

6. Type=ORIGIN, ID=4447, Weight=100, Name=Orange County

STEP 1: 19002 (700) 2853 >= 350 −> Compare MOOD

STEP 1: 34213 (950) 34213 >= −160 −> PASS

STEP 2: (Genre=40 Mood=50 Era=5 Origin=3 Artist Type=2)

Computing GENRE similarity:

0. 2853/29 (925) 2853/100

Sum=26825, Weightsum=29

1. 19002/53 (700) 2853/71

Sum=63925, Weightsum=82

2. 19041/18 (0) 2853/18

Sum=63925, Weightsum=100

--------------------

Known = 639 / 100

====================

Result = 639

Computing MOOD similarity:

0. 34213/91 (950) 34213/96

Sum=86450, Weightsum=91

1. 34398/1 (300) 34393/1

Sum=86750, Weightsum=92

2. 34182/1 (−700) 34213/5

Sum=86050, Weightsum=93

--------------------

Known = 925 / 93

Unknown = 462 / 7

====================

Result = 892

Computing ERA similarity:

0. 2652/60 (970) 2652/100

Sum=58200, Weightsum=60

1. 34814/20 (935) 2652/40

Sum=76900, Weightsum=80

2. 2653/20 (890) 2652/20

Sum=94700, Weightsum=100

--------------------

Known = 947 / 100

====================

Result = 947

Computing ORIGIN similarity:

SINGLE: 4004 (225) 4447

Computing ARTIST TYPE similarity:

SINGLE: 2656 (975) 2656

RESULT:

Step1 = PASS

Step2:

* Genre = 639

* Origin = 225

* Era = 947

* Artist Type = 975

* Mood = 892

* Weighted = 775

====================

SIMILARITY = 775

Step one was passed since the primary genre correlation is 700 (e.g., greater than example specified threshold 350), and both the station definition and the candidate track have the same primary mood, with a correlation of 970 (e.g., greater than the example specified threshold −160), so we compute the weighted score, which will become the final similarity score.

The following is an example similarity calculation comparing the station definition “Neo-Lounge Nitecap” to the candidate track “High Noon” by Kruder & Dorfmeister.

Name=“Neo-Lounge Nitecap” Station Definition

GOETs (25):

1. Type=ATYPE, ID=2662, Weight=100, Name=Mixed Group -

Mixed Duo/Group Vocal

2. Type=ERA, ID=2650, Weight=60, Name=2000's

3. Type=ERA, ID=34814, Weight=20, Name=Early 2010's

4. Type=ERA, ID=2648, Weight=20, Name=Late 90's

5. Type=GENRE, ID=2792, Weight=22, Name=Neo-Lounge

6. Type=GENRE, ID=18981, Weight=14, Name=Chanson Jazz

7. Type=GENRE, ID=3164, Weight=14, Name=Cabaret

8. Type=GENRE, ID=18811, Weight=14, Name=Indie Cabaret

Pop-Rock

9. Type=GENRE, ID=3337, Weight=14, Name=Bossa Nova

10. Type=GENRE, ID=3175, Weight=14, Name=Cocktail Music

11. Type=GENRE, ID=18833, Weight=8, Name=Soft Jazz Vocals

12. Type=MOOD, ID=34163, Weight=27, Name=Sophisticated /

Lush / Romantic

13. Type=MOOD, ID=34409, Weight=20, Name=Lively / Playful

/ Swingin’

14. Type=MOOD, ID=34207, Weight=7, Name=Sultry / Swank

15. Type=MOOD, ID=34415, Weight=6, Name=Charming / Easy-

Going

16. Type=MOOD, ID=34190, Weight=5, Name=Blue / Depressed

/ Lonely

17. Type=MOOD, ID=34408, Weight=5, Name=Showy / Dramatic

/ Rousing / Lively

18. Type=MOOD, ID=34192, Weight=3, Name-Gritty / Earthy /

Soulful

19. Type=MOOD, ID=34168, Weight=2, Name=Sweet Ballad

20. Type=MOOD, ID=34172, Weight=2, Name=Smoky Romantic

21. Type=MOOD, ID=34189, Weight=2, Name=Tender Sad /

Soulful

22. Type=ORIGIN, ID=4525, Weight=100, Name=Portland, OR

Name=Track “High Noon” by Kruder & Dorfmeister

GOETs (13):

1. Type=ATYPE, ID=2657, Weight=100, Name=Male Duo - Duo

Vocal

2. Type=ERA, ID=2645, Weight=75, Name=1990's

3. Type=ERA, ID=2650, Weight=25, Name=2000's

4. Type=GENRE, ID=2914, Weight=49, Name=Trip Hop

5. Type=GENRE, ID=2912, Weight=17, Name=Chill Out

6. Type=GENRE, ID=2887, Weight=17, Name=Electronica

Lounge Fusion

7. Type=GENRE, ID=2865, Weight=17, Name=Electronica

8. Type=MOOD, ID=34182, Weight=94, Name=Deep Dreamy

Relaxing Beat

9. Type=MOOD, ID=34191, Weight=1, Name=Determined /

Bitter / Serious / Jaded

10. Type=ORIGIN, ID=4267, Weight=100, Name=Germany

STEP 1: 2792 (550) 2914 >= 350 −> Compare MOOD

STEP 1: 34163 (−200) 34182 < − 160 −> FAIL

STEP 2: (Genre=40 Mood=50 Era=5 Origin=3 Artist Type=2)

Computing GENRE similarity:

0. 2792/22 (700) 2912/17

Sum=11900, Weightsum=17

1. 2792/5 (600) 2887/17

Sum=14900, Weightsum=22

2. 18981/14 (0) 2914/49

Sum=14900, Weightsum=36

3. 3164/14 (0) 2914/35

Sum=14900, Weightsum=50

4. 18811/14 (0) 2914/21

Sum=14900, Weightsum=64

5. 3337/14 (0) 2914/7

Sum=14900, Weightsum=71

6. 3337/7 (0) 2887/12

Sum=14900, Weightsum=78

7. 3175/14 (0) 2887/5

Sum=14900, Weightsum=83

8. 3175/9 (0) 2865/17

Sum=14900, Weightsum=92

9. 18833/8 (0) 2865/8

Sum=14900, Weightsum=100

--------------------

Known = 149 / 100

====================

Result = 149

Computing MOOD similarity:

0. 34190/5 (650) 34191/1

Sum=650, Weightsum=1

1. 34207/7 (200) 34182/94

Sum=2050, Weightsum=8

2. 34189/2 (100) 34182/87

Sum=2250, Weightsum=10

3. 34192/3 (−100) 34182/85

Sum=1950, Weightsum=13

4. 34163/27 (−200) 34182/82

Sum = −3450, Weightsum=40

5. 34172/2 (−200) 34182/55

Sum = −3850, Weightsum=42

6. 34409/20 (−300) 34182/53

Sum = −9850, Weightsum=62

7. 34190/4 (−300) 34182/33

Sum = −11050, Weightsum=66

8. 34168/2 (−400) 34182/29

Sum = −11850, Weightsum=68

9. 34415/6 (−500) 34182/27

Sum = −14850, Weightsum=74

10. 34408/5 (−600) 34182/21

Sum = −17850, Weightsum=79

--------------------

Known = −225 / 79

Unknown = −112 / 21

====================

Result = −201

Computing ERA similarity:

0. 2650/60 (935) 2650/25

Sum=23375, Weightsum=25

1. 2648/20 (873) 2645/75

Sum=40835, Weightsum=45

2. 2650/35 (680) 2645/55

Sum=64635, Weightsum=80

3. 34814/20 (680) 2645/20

Sum=78235, Weightsum=100

--------------------

Known = 782 / 100

====================

Result = 782

Computing ORIGIN similarity:

SINGLE: 4525 (281) 4267

Computing ARTIST TYPE similarity:

SINGLE: 2662 (585) 2657

RESULT:

Step1 = FAIL

Step2:

* Genre = 149

* Origin = 281

* Era = 782

* Artist Type = 585

* Mood = −201

* Weighted = 18

====================

SIMILARITY = 0

Step one failed because, while the primary genres correlate by 550 (which is above the example specified threshold of 350), the primary moods correlate only by −200, which is below the example specified threshold of −160. As before, going through step two is not necessary if step one fails; it is shown here for the sake of completeness. Regardless of the weighted score, the overall similarity score will be 0, since step one failed.

In addition to the capability to generate a station library based on direct attribute filters (e.g., utilizing station library criteria to track similarity comparisons), a particular artist or track may be used as an indirect way to define a complex descriptive vector as a component of a station definition or as all of the station definitions, in another example embodiment.

In another example embodiment, the system may allow a user to select and filter out specific tracks, albums, and artists as part of a station definition. For example, the user may select specific tracks or identify particular albums and artists from a station library result set displayed in a user interface. The results generated going forward would exclude the selected content.

In another example embodiment, the system may provide the ability to filter, boost, and/or penalize candidate tracks based on other flags or attributes of the candidate tracks to further focus the nature of the station library content. Some examples of filters may include: duration, live, explicit, occasion appropriateness (e.g., Christmas), lyric topic, demo, alternative take, re-recording, explicit, clean, record label, language, energy, and acoustic/electric. The system may further provide the ability to use lists of previously flagged or scored “key artists,” “key albums,” and/or “key tracks” in order to boost the likelihood of certain tracks appearing in a station library based on editorial preference. In addition, the system may provide the ability to use artists or tracks as “seeds” as a component (along with criteria based filters) of a station definition, or as all of a station definition. Furthermore, the system may provide the ability to utilize “related artist” associative relationship data to boost the likelihood of certain tracks appearing in the station library.

Embodiments described herein generally refer to media curation of audio content. It is understood that embodiments described herein may be used for other media content (e.g., video, mixed media, etc.) and other media curation. In one example, embodiments described herein may be used for curation of music videos or other videos.

FIG. 12 is a block diagram 1200 illustrating a software architecture 1202, which can be installed on any one or more of the devices used to implement any of the embodiments described herein. For example, in various embodiments, client devices 110 and server machines or systems 130, 102, 120, 122, and 124 may be implemented using some or all of the elements of software architecture 1202. FIG. 12 is merely a non-limiting example of a software architecture, and it will be appreciated that many other architectures can be implemented to facilitate the functionality described herein. In various embodiments, the software architecture 1202 is implemented by hardware such as machine 1300 of FIG. 13 that includes processors 1310, memory 1330, and I/O components 1350. In this example, the software architecture 1202 can be conceptualized as a stack of layers where each layer may provide a particular functionality. For example, the software architecture 1202 includes layers such as an operating system 1204, libraries 1206, frameworks 1208, and applications 1210. Operationally, the applications 1210 invoke application programming interface (API) calls 1212 through the software stack and receive messages 1214 in response to the API calls 1212, consistent with some embodiments.

In various implementations, the operating system 1204 manages hardware resources and provides common services. The operating system 1204 includes, for example, a kernel 1220, services 1222, and drivers 1224. The kernel 1220 acts as an abstraction layer between the hardware and the other software layers, consistent with some embodiments. For example, the kernel 1220 provides memory management, processor management (e.g., scheduling), component management, networking, and security settings, among other functionality. The services 1222 can provide other common services for the other software layers. The drivers 1224 are responsible for controlling or interfacing with the underlying hardware, according to some embodiments. For instance, the drivers 1224 can include display drivers, camera drivers, BLUETOOTH® or BLUETOOTH® Low Energy drivers, flash memory drivers, serial communication drivers (e.g., Universal Serial Bus (USB) drivers), WI-FI® drivers, audio drivers, power management drivers, and so forth.

In some embodiments, the libraries 1206 provide a low-level common infrastructure utilized by the applications 1210. The libraries 1206 can include system libraries 1230 (e.g., C standard library) that can provide functions such as memory allocation functions, string manipulation functions, mathematic functions, and the like. In addition, the libraries 1206 can include API libraries 1232 such as media libraries (e.g., libraries to support presentation and manipulation of various media formats such as Moving Picture Experts Group-4 (MPEG4), Advanced Video Coding (H.264 or AVC), Moving Picture Experts Group Layer-3 (MP3), Advanced Audio Coding (AAC), Adaptive Multi-Rate (AMR) audio codec, Joint Photographic Experts Group (JPEG or JPG), or Portable Network Graphics (PNG)), graphics libraries (e.g., an OpenGL framework used to render in two dimensions (2D) and three dimensions (3D) in graphic content on a display), database libraries (e.g., SQLite to provide various relational database functions), web libraries (e.g., WebKit to provide web browsing functionality), and the like. The libraries 1206 can also include a wide variety of other libraries 1234 to provide many other APIs to the applications 1210.

The frameworks 1208 provide a high-level common infrastructure that can be utilized by the applications 1210, according to some embodiments. For example, the frameworks 1208 provide various graphic user interface (GUI) functions, high-level resource management, high-level location services, and so forth. The frameworks 1208 can provide a broad spectrum of other APIs that can be utilized by the applications 1210, some of which may be specific to a particular operating system 1204 or platform.

In an example embodiment, the applications 1210 include a home application 1250, a contacts application 1252, a browser application 1254, a book reader application 1256, a location application 1258, a media application 1260, a messaging application 1262, a game application 1264, and a broad assortment of other applications such as a third party application 1266 and curation application 1267. According to some embodiments, the applications 1210 are programs that execute functions defined in the programs. Various programming languages can be employed to create one or more of the applications 1210, structured in a variety of manners, such as object-oriented programming languages (e.g., Objective-C, Java, or C++) or procedural programming languages (e.g., C or assembly language). In a specific example, the third party application 1266 (e.g., an application developed using the ANDROID™ or IOS™ software development kit (SDK) by an entity other than the vendor of the particular platform) may be mobile software running on a mobile operating system such as IOS™, ANDROID™, WINDOWS® Phone, or another mobile operating system. In this example, the third party application 1266 can invoke the API calls 1212 provided by the operating system 1204 to facilitate functionality described herein.

FIG. 13 is a block diagram illustrating components of a machine 1300, according to some embodiments, able to read instructions from a machine-readable medium (e.g., a machine-readable storage medium) and perform any one or more of the methodologies discussed herein. Specifically, FIG. 13 shows a diagrammatic representation of the machine 1300 in the example form of a computer system, within which instructions 1316 (e.g., software, a program, an application 1210, an applet, an app, or other executable code) for causing the machine 1300 to perform any one or more of the methodologies discussed herein can be executed. In alternative embodiments, the machine 1300 operates as a standalone device or can be coupled (e.g., networked) to other machines. In a networked deployment, the machine 1300 may operate in the capacity of a server machine or system 130, 102, 120, 122, 124, etc., or a client device 110 in a server-client network environment, or as a peer machine in a peer-to-peer (or distributed) network environment. The machine 1300 can comprise, but not be limited to, a server computer, a client computer, a personal computer (PC), a tablet computer, a laptop computer, a netbook, a personal digital assistant (PDA), an entertainment media system, a cellular telephone, a smart phone, a mobile device, a wearable device (e.g., a smart watch), a smart home device (e.g., a smart appliance), other smart devices, a web appliance, a network router, a network switch, a network bridge, or any machine capable of executing the instructions 1316, sequentially or otherwise, that specify actions to be taken by the machine 1300. Further, while only a single machine 1300 is illustrated, the term “machine” shall also be taken to include a collection of machines 1300 that individually or jointly execute the instructions 1316 to perform any one or more of the methodologies discussed herein.

In various embodiments, the machine 1300 comprises processors 1310, memory 1330, and I/O components 1350, which can be configured to communicate with each other via a bus 1302. In an example embodiment, the processors 1310 (e.g., a central processing unit (CPU), a reduced instruction set computing (RISC) processor, a complex instruction set computing (CISC) processor, a graphics processing unit (GPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC), a radio-frequency integrated circuit (RFIC), another processor, or any suitable combination thereof) include, for example, a processor 1312 and a processor 1314 that may execute the instructions 1316. The term “processor” is intended to include multi-core processors 1310 that may comprise two or more independent processors 1312, 1314 (also referred to as “cores”) that can execute instructions 1316 contemporaneously. Although FIG. 13 shows multiple processors 1310, the machine 1300 may include a single processor 1310 with a single core, a single processor 1310 with multiple cores (e.g., a multi-core processor 1310), multiple processors 1312, 1314 with a single core, multiple processors 1310, 1312 with multiples cores, or any combination thereof.

The memory 1330 comprises a main memory 1332, a static memory 1334, and a storage unit 1336 accessible to the processors 1310 via the bus 1302, according to some embodiments. The storage unit 1336 can include a machine-readable medium 1338 on which are stored the instructions 1316 embodying any one or more of the methodologies or functions described herein. The instructions 1316 can also reside, completely or at least partially, within the main memory