What is APIX?

Advanced Patent IndeX (APIX) is a new tool developed exclusively by Unified Patents to examine the probable validity of any granted patent in the US.


HOW DOES APIX WORK?

APIX uses a machine learning algorithm to rate the quality of a patent from AA to D. The algorithm relies on 30 variables derived from a patent's specification, claims, and prosecution history. Unified Patents members can input a list of patents or search by assignee to rate up to 5,000 patents at a time.  Results can also be further parsed by date of issuance and other attributes.


WHAT DO THE RATINGS MEAN?

The rating signifies the chances that a patent would be found invalid by the PTAB if challenged in inter partes review.  A “D” means the patent has the lowest rating on the scale and has the highest likelihood of being found invalid if challenged.  Conversely, an “AA” has the highest likelihood of surviving a validity challenge.


HOW ARE PATENT RATINGS DETERMINED?

Unified uses a “raw score” based on the probability of survival for each patent.  We then convert the raw score into a rating by comparing it against the 6M+ patents issued by the PTO.  Patents are then rated based on how they compare to the overall average for all patents.  The diagram below shows this in greater detail.  The ratings are separated by how many standard deviations they are from the mean.

Rating

Perc. of
Total Patents

From

To

D

15.8%

-

-1 σ

C

34.1%

-1 σ

0

B

34.1%

0

+1 σ

A

13.6%

+1 σ

+2 σ

AA

2.2%

+2 σ

+


HOW DOES UNIFIED CREATE ITS “RAW SCORE”?

We do 3 things which provide better patent quality assessments that other rating tools cannot duplicate:

 

1. Use our PTAB portal data as a unique training set to teach the tool what is valid.  Unified's PTAB portal is based on the thousands of institution decisions made to date by the PTAB. We instructed the tool to learn that a “good” patent was one that successfully survived a validity challenge while a “bad” patent was one that did not. Institution decisions are clear determinations of validity from domain area experts, and the high volume of these decisions therefore provides an excellent data set with which to train the tool. Additionally, almost all the patents in the sample were in litigation at the time and therefore there is a strong interest in successfully challenging them.  We also believe there is a strong incentive for the challenger to only challenge patents they believe they can successfully invalidate due to the high costs of filing a petition and the prospect that an unsuccessful challenge may make the patent stronger.

 

2. Use all available patent data to help the tool determine which variables have the greatest influence on determining the likelihood a patent is “good."  The roughly 30 variables include not only attributes of a patent after it is granted such as citations, family size, and words, but also data from a patent’s prosecution history to the extent it was important in surviving a challenge.

 

3. Utilize a machine learning algorithm (AI) to provide results with much higher statistical significance than is possible with linear regressions.  Because an AI uses thousands of computations which can vary based on key initial variables, the accuracy can be much higher, but specific determinants can be much harder to measure since they may change significantly for each result.  This is why similar AI is now used by many of the leading companies to help predict very complex scenarios more accurately than ever before.


WHAT IS A KEY TAKEAWAY?

The most interesting result that we have found so far is that several variables derived from a patent’s prosecution history are among those most strongly correlated with a patent’s ability to survive a challenge.  For example, a patent examiner's allowance rate and its variance from their art unit's overall allowance rate weighs particularly strongly in determining a “good” or “bad” patent.  There could be many reasons for this result.  One possibility is that examiners with higher allowance rates permit broader patents to be granted, which are more likely to be used in litigation since they are easier to allege infringement against.  Broader patents are also more likely to be found invalid if challenged because there are often more prior inventions which predate parts of the patent’s claimed invention.