The Patent Score
One of the most popular new features in AcclaimIP is our Patent Score, which we call the P-Score. P-Score ranks patents by various statistical metrics that contribute to their statistical quality.
Patents that rank high on the P-Score scale tend to be better patents, with a higher likelihood of being infringed, being used in a company’s products, or having current or future monetization potential.
Depending on how you define a “data point”, AcclaimIP uses from 30 to 30,000 data points to develop the P-Score for each patent. To simplify the scoring concept, the P-Score is a weighted average of three contributing component scores:
- Citation Score (C-Score)
- Technical Score (T-Score)
- Legal Score (L-Score)
Each sub-score is calculated from a number of contributing factors, and each is included in the AcclaimIP interface so you can make better “at-a-glance” judgments on the statistical metrics that contributed to a patent’s score.
Try sorting on a search result by the P-Score, and simultaneously view the other component scores to identify in which areas the patents scored high and which scored low, where they are statistically strong, and where they are statistically weak.
We experimented with several methods to normalize the scores, but I think the one we picked is the most intuitive—and most useful.
Scores are normalized from 1 to 100 corresponding to precise percentages. So if you have a patent that scores 100, you’ll know that it is a top 1%er. If your portfolio averages a 55, you’ll know that you are 5 points above the statistical average of exactly 50. This way there is little guesswork in how your patents and portfolios stack up compared to the general population of patents.
Each of the three component scores is also normalized to 100 using the same linear percentage method.
What Data Points Contribute to Scores?
Let’s consider the Citation Score. You can imagine that two patents with the same number of forward citations could be ranked quite differently. There are lots of ways to tease out what makes a quality citation profile, which includes far more than just raw citation counts. Some of these include:
- Cited by Highly Cited patents
- Examiner Citations
- Co-Pending Citations
- Citations that are a result of a forward 102 argument (that is directly blocking someone else’s claims.)
- Citations that are a result of a forward 103 argument (that is used in an obviousness argument)
- Age weighted citation rate
AcclaimIP’s Citation Score includes citations to the parent application as well. Application citations are often the most important. Until your patent is granted, your patent cannot receive any direct citations. But the parent application may have 20 or more. In fact, most co-pending citations are to the application, and they are cited before your patent is even granted, and can lead to unintentional infringement because the potentially infringing party could not have known about your un-published prior art when they made the decision to use the technology in their products.
The technology score, or T-Score, does not score a patent directly, but rather it scores the CPC classifications (technology classes) in which your patent is classified. We use statistics to determine which classes are strong and which are relatively weak. A strong classification is one that scores highly in each of the areas below:
- Growth of class
- Maintenance rate of class
- Transaction rate of class
- Allowance rate of class
The T-Score algorithm accounts for the hierarchical nature of the CPC class system, otherwise it would not work, since a parent class may have little activity when all the “action” is pushed down to its children classes.
If a patent is classified in a class that is growing, with a high maintenance rate, with lots of recorded transactions or company acquisitions, AcclaimIP assumes that it covers a strong technology, and it then impacts the technical component of the patent’s P-Score.
Legal Score (L-Score)
The legal component is composed of several metrics that are indicative of a patent’s strength including the following measurements:
- Pendency (time between file date and grant date)
- Length of independent claims
- Number of claims
- Degree to which claims were modified during prosecution
- Number of office actions
- Family size
- Remaining life
What is a patent metric?
A patent metric is any quantitative data point (a number) associated with a patent. Metrics can relate to anything as long as it can be quantitatively measured.
I like to break up patent metrics into two categories:
1. Those that apply to a individual patent
2. Those that apply to a set of patents
For example, on an individual patent, the number of inventors can be explicitly measured, the pendency (in days) of the prosecution, the number of family members, the number of reverse citations, the number of office actions, number of examiner citations, the number of figures, the length of claims, the number of forward rejections… and the list goes on and on.
Further metrics from a set of patents in the patent corpus at large can be measured and applied to a patent sharing characteristics of the set.
For example, allowance rates, maintenance rates, growth rates and transaction rates can all be applied to patent classifications. In turn, each patent can inherit some quantitative measure based on the classifications in which it was assigned.
How to Use and Interpret Patent Scoring Systems
Clearly the P-Score is not something that is pulled out of thin air, but rather a result of a careful study of the academic literature, and years of experimenting with regression testing to confirm the metrics are indicative of a patent’s quality.
Of course, there are outliers. A patent with characteristics of a quality patent can actually be useless, and conversely, a patent with poor quality metrics can be outstanding, but these are the exceptions and not the rule.
If you want 100% confidence that your scoring is accurate, you have to read every patent and file wrapper, and make judgments to the quality in context of the problem that the patent may help solve. There is no way around that.
Nonetheless, patent scoring systems are valuable, because they focus the researcher on those patents that are most likely to be useful or important to the searcher.
Scores also clue you in on what happened during prosecution. Let’s say you have a patent where the first independent claim was narrowed by 100 words. Immediately you should know that what the applicant filed and what was granted are quite different, and surely some investigation is required. Another example might be a patent that received a blind citation to your patent by an examiner making a 102-novelty argument. Clearly cases like these are great indicators of potential infringement. These important cases are reflected in the P-Score.
Personally, I like to expose all 4 scores in my search results grids. Let’s say I have a patent whose score profile looks like this:
If you investigate the citations further, you’ll notice that most of the citations are from Semiconductor Energy Lab via IDS citations. SEL also cites the application and not the patent even though the patent has been published well before most of the citing documents were filed. You have to wonder what is going on here. Even so, this patent is highly cited by other patents from examiners.
Querying Patent Scores
Each of the four scoring metrics can be queried using advanced syntax. Each score has its own easy-to-remember field code:
- PSCORE –> queries the Patent Score field
- CSCORE –> queries the Citation Score field
- LSCORE –> queries the Legal Score field
- TSCORE –> queries the Technology Score field
Each field code supports a number from 1 to 100 or range queries. For example:
…AND PSCORE:[80 to 100] –> Finds patents with a P-Score between 80 and 100, inclusive.
Evolution of the Patent Scores
I’d like to conclude this article by letting you know that the scoring system is designed to be agile and improved as we learn more and get feedback from our users. Please weigh in, point out outliers, or suggest a new visualization, so we can continue to improve the scoring metrics and your reporting capabilities.