Will similarity scores be available in the list view?
Yes, the similarity score is available in the list view and the results can be sorted ascending or descending by that.
Similarity searching starts with vectorising every patent family in the universe (think of this like giving each patent family a unique fingerprint). Each patent family can then have its vector (fingerprint) compared against others to identify vectors (patent families) that is closest to it, returning the closest results based on the chosen sample size (50, 100, 1000 etc.). The deep learning model (“algorithm”) is specifically designed for patent linguistic tasks and uses the patent title, abstract and claims to generate a vector for each individual patent family. TechDiscovery uses generative AI to assist the vectorisation (or fingerprinting) of information when you have very little information e.g. a technology name or brief description.
What do these show me?
The score is an indication of how similar the result is to the input in the query. Every patent family in the results list is assigned a similarity score. This score is between 100 and 0, with 100 being the highest similarity score, and zero the lowest. Patent Families are sorted by the highest score for you to review. The similarity score is an indication of how similar the patent family is to your input query.
You can select patent families in the 'similar patent families' list that you decide are most relevant to the search criteria you entered, by clicking the 'plus' icon. This helps to refine the search by adding these examples to your input query.
You can also use the minus icon to simply remove non-relevant examples from your results, or if you're unsure on certain examples, you can use the ‘Review Later’ button to store these away.
All of these actions help to refine the search that you have input to generate the most relevant patent family results in your list.