Relevance Analysis

Semantic search using AI delivers impressive results, but sometimes there can be a bit of a "black box" in understanding why specific results are deemed relevant. Our Relevance Analysis feature eliminates this phenomenon, providing users with a transparent view of the factors contributing to result relevance. It also offers users and idea if they want to spend time to further review a document or not.

Note: Relevance Analysis is only available for Natural Language and Patent Number Search

Generating Relevance Analysis

Relevance analysis can be done after conducting a search. Click a result and then go to the “Relevance Analysis” tab in the pop-up. 

Once you click the button, it will take a few moments to analyze and then start typing out the relevance analysis.

Sample of relevance analysis

The analysis is bespoke to each patent, delving into details and specific features on why it is similar or dissimilar. Generally speaking, it will point out what is similar and dissimilar about the patent to the query, followed by a conclusion.


You can now also export a generated relevance analysis by click "Relevance_Analysis" in the report field. Please note that the relevance analysis field will only be populated if

Understanding Relevance Drop Off 

Beyond explaining why a patent is relevant, this feature also helps identify where relevancy begins to taper off. In traditional Boolean searches, users receive a fixed number of results (e.g., 64), with no meaningful order—forcing users to manually assess each one. In contrast, NLPatent ranks results by semantic similarity, bringing the most relevant documents to the top..


By default, NLPatent shows the top 100 results (up to 10,000 can be selected). Since these are ordered by similarity—not a hard cutoff—it can be difficult to tell where relevance starts to fade. That’s where relevance analysis helps: it offers insight into whether it’s worth digging deeper into lower-ranked results. For instance, if the analysis shows “limited relevance,” that often means the patent isn’t particularly aligned with your query.

An example of where the relevance may start to taper off.

Disclaimer

While Relevance Analysis undergoes rigorous monitoring, occasional false results may occur. We emphasize the importance of user due diligence and cross-verification of generated results. Relevance Analysis serves as a powerful tool for expediting searches, but it is not a replacement for professional human opinions.

Did this answer your question? Thanks for the feedback There was a problem submitting your feedback. Please try again later.