Analyzing large patent portfolios has been challenging. This is one of the major factors that has made patents an illiquid asset class. Discovering infringements is often random. No easy or scalable process yet exists that could uniformly be applied to patents in all technology areas.

Patents were created to encourage innovation by giving exclusivity to the inventors to expand and protect their market share for a limited time. However, the challenge in discovering infringement has made this task difficult.

We believe that artificial intelligence (AI) will change this situation in the coming years. This week, we use an AI-based solution to identify strengths and weaknesses in the Apple patent portfolio.

To show an infringement, one has to map all the elements of a patent claim to various functions, components, and/or features of a product (claim chart). But most large companies still face two main challenges.

Large Patent Portfolios: With thousands of patents in their portfolios, large companies often struggle to track all of the patents they own. This means properly protecting their market share is a challenge (a difficulty in out-licensing)
Feature-Rich Products: With large product portfolios and features that are constantly evolving, even if large companies know all the details of their patent portfolios, they often do not know what others have. This unknown factor may cause them headaches down the road (a difficulty in in-licensing).

Even when they understand their portfolios (both patents and products), the challenge goes back to the mapping process. Often, features of the products may have been described with different keywords than that of patents. In this case, a simple keyword search is not sufficient to identify infringements.

All of these factors have made the in-licensing, out-licensing, and monetization very slow, complex, and expensive.

In order to measure the strength of a portfolio, companies need to examine their patent portfolios in the context of other companies’ portfolios. In other words, regardless of the size of one’s portfolio, patents may exist that should be licensed and are necessary for proper operation of the company. Unfortunately, this is not a simple problem.

Measuring Portfolio Quality
Patexia Data Science team has been working on this problem for quite some time now. We first developed a portfolio ranking methodology to measure the quality of patents for each company. We measure 30 to 40 factors (e.g., claim size, forward and backward citations, etc.) to develop the Quality Metric for a company’s portfolio.

In March, we published the portfolio Quality Metric for the 1,000 most active companies in our 2019 Patent Intelligence Report. We calculate these quality metrics for each patent. We then average them for each portfolio.

While quality is important, it does not tell us anything about the strengths and weaknesses of a portfolio in relation to other companies’ portfolios. So our Data Science Team continued working on this problem by developing an AI-based solution that examines the portfolio in relation to all other patents.

Measuring Portfolio Strengths and Weaknesses (Apple Patent Portfolio)
Our solution has been implemented at multiple levels using International Patent Classification (IPC), U.S. Patent Classification (USPC), and patent/product word clouds. (Patexia has developed its own word cloud to model the key subject matters and their importance over time for all patents.)

Here, we will show you a use case for this powerful technology, identifying the strengths and weaknesses of Apple’s portfolio.

Since January 2000, Apple has filed more than 21,000 patent applications and has been issued approximately 20,000 patents by the U.S. patent office. We clustered these patents by their IPC codes and identified 729 unique IPC codes. (Only the top 3 layers of IPC codes were considered for this study). Then, using neural networks, we developed a model and compared the Apple portfolio with all other companies that had at least 10 or more patents in the last 20 years.

This analysis, which is quite time consuming and requires multi-processing computation power, will automatically bring up the key IPC codes that are important with respect to the Apple product portfolio and predicts the right portfolio size for each IPC code for a company like Apple.

For the remainder of the study, we will only focus on the top 10 IPC codes for Apple.

The above table shows the 10 most popular IPC codes for Apple, which cover more than 15,000 patents (roughly 75% of the Apple patent portfolio). The prediction column indicates the result of our neural network analysis with the predicted number of patents that Apple should have for each category, based on all other companies with patents in the overlapping areas (e.g., smartphone and smartwatch makers, chip makers, etc.).

In the last column of the table, we calculated the strength as the ratio of Apple’s patents to this predicted number. We then generated the colored heat map to easily illustrate whether Apple has strength or weakness in a particular area.

Verification of Our Findings (Comparison with Apple’s Lawsuits)
To examine and verify our findings, we looked at Apple’s patent lawsuits over the last decade to find any possible correlation.

Based on our records, Apple has been involved in 1,057 patent lawsuits over this period (district court, PTAB, ITC, etc.). The following table, groups the cases by IPC codes of patents involved in the lawsuits. It shows the relevant counts for the 10 most popular IPC codes of Apple.

As our last step, we compare our findings from Apple’s patent portfolio with Apple’s patent lawsuits in the following table.

As you can see, Apple shows relative strength (zero or few lawsuits) in areas where the portfolio is stronger. For example, H05K is an area where, based on our calculations, Apple has relatively a healthy patent portfolio compared to its competitors. The current count is above our predicted number (e.g., 116%). Additionally, in this area, Apple has had no lawsuits in recent years.

About one-third of all of Apple’s litigation is in G06F, which is the most popular IPC code in the Apple patent portfolio. This area is generally related to digital data processing. While our prediction shows Apple has 92% strength, we still see 340 (or 32%) of lawsuits in this area in recent years.

To better understand this IPC code, we decided to perform our AI-based analysis again but this time, we clustered the patents not only by IPC codes but also by the year of filing. The outcome was very interesting and should explain the reasons for the relative weakness in this area (see below).

Once we divided the portfolio by age into four groups of 0-5, 5-10, 10-15, and 15-20 years and performed our AI-based analysis, we noticed that, despite the large number of 5,817 patents in G06F code, Apple’s patent portfolio in this area is relatively young (less than 10 years old). Most of the filings were done in recent years. In fact, 5,338 out of 5,817 patents or 92% of all filings were done in the last decade.

This means that Apple’s portfolio may not cover the fundamental technologies related to digital data processing (G06F) that the company is using in its products. As a result, it is an area where the company is facing more lawsuits.

Other Applications of AI-Based Analysis
We have only covered one application of our AI-based analysis engine at a very high level, using it to identify the strengths and weaknesses of a company’s portfolio. In some cases, we can combine the same engine with our Quality Metric tool and our Mining Engine to discover potential portfolios for acquisition or divestiture for our clients.

Depending on the needs, these portfolios can be used for the purpose of in-licensing, cross-licensing, out-licensing, monetization, etc. For example, a company (defendant) that is involved in a patent lawsuit can look at the weaknesses of the plaintiff and acquire patents for cross-licensing and settlement purposes.

In the more advanced model, we use our own homegrown word cloud, which is more useful and relevant than IPC codes. The IPC codes are sometimes inaccurate and often don’t correctly show the latest technologies.

While humans cannot be completely removed from the process, meaning the use of technical experts is still required at later stages, our goal is to shorten the first round of analysis and discovery to a matter of days rather than months. This is done by filtering a portfolio of thousands of patents to something more manageable by humans. As a result, it will allow companies to expedite the negotiation phase and move one step closer to a more liquid patent market.

In the coming weeks, we plan to expand our earlier analysis on ITC Section 337 (Patexia Insight 65), as we gradually prepare our first ITC Intelligence Report. The report is now available for pre-order at a discount. Stay tuned.