Legal research recommendation system
The present disclosure is directed towards systems and methods for assisting with legal research and for aiding in the discovery of information and documents relevant to a user's research focus or input text. The inventive systems receive identifications of relevant documents explicitly or implicitly from a user's research session or input text and, based on issues relevant to those documents, texts and other connections to those documents and texts, recommends similar or helpful documents to the user for their consideration. Recommendations may be ranked and filtered.
1. A method for improving a computerized search system's research recommendations, embodied as instructions stored in non-transitory computer memory of the computerized search system which, when executed by a computer processor, are configured to:
receive, using an application programming interface (API), a first binary encoded signal associated with an input text provided by a user computing device;
parse the input text to identify one or more conceptual issue topics relevant thereto, including at least identifying any cites to other documents contained in the input text and identifying conceptual issue topics in the input text based on any conceptual issue topics known to be relevant to the identified cited documents;
identify recommendation candidates known to be relevant to the one or more identified conceptual issue topics; and
transmit, using the API, a second binary encoded signal to the user computing device, the second binary encoded signal being configured to render results of the identifying step via a user interface of the user computing device, the results comprising the recommendation candidates ranked according to a respective relevancy score indicating how relevant each recommendation candidate is to the input text, wherein the relevancy score is based on a weighted aggregation of at least (i) a legal jurisdiction score, (ii) an authority score, and (iii) a recency score, and (iv) a recommendation-frequency score, wherein the legal jurisdiction score indicates a closeness of legal jurisdiction scope between the recommendation candidate and the input text based on a comparison between a first legal jurisdiction code associated with the recommendation candidate and a second legal jurisdiction code associated with the input text, wherein the authority score indicates a degree of authoritativeness of the recommendation candidate with respect to one or more other recommendation candidates, wherein the recency score indicates a recentness of the recommendation candidates determined based on document metadata extracted from an electronic record database, and wherein the recommendation-frequency score indicates a degree of similarity between headnotes relevant to the input text and headnotes identified in data included in respective headnote portions of the recommendation candidates.
2. The method of claim 1 , wherein the parsing step includes identifying conceptual issue topics based one or more formatting aspects of the input text.
3. The method of claim 1 , wherein the parsing step includes identifying any pincites to locations within other documents in the input text and identifying conceptual issue topics in the input text based on any issues known to be relevant to the identified pincited locations.
4. The method of claim 3 , further comprising, if more than one conceptual issue topic is known to be relevant to an identified pincited location, determining which known conceptual issue topic is most relevant to the input text based on the context of the pincite in the input text.
5. The method of claim 4 , wherein the context includes a text of any blockquote relevant to the pincite in the input text.
6. The method of claim 4 , wherein the context includes the portion of the input text from the pincite to the previous pincite or the beginning of the input text, whichever is nearer to the pincite.
7. The method of claim 4 , further comprising identifying one or more legal jurisdictions relevant to the input text and the context of a pincite includes the identified one or more legal jurisdictions.
8. The method of claim 1 , wherein the relevancy score is based at least in part on a similarity between each recommendation candidate and the input text.
9. The method of claim 1 , wherein the relevancy score is based at least in part on a similarity between each recommendation candidate and documents cited in the input text.
10. The method of claim 9 , wherein the similarity between each recommendation candidate and documents cited in the input text is determined at least in part by comparing a similarity between conceptual issue topics known to be relevant to each recommendation candidate and conceptual issue topics known to be relevant to the documents cited in the input text.
11. The method of claim 1 , further comprising pre-filtering the results of the identifying step before returning them by only passing results having a higher relevancy score than a document cited in the input texts having a lowest relevancy score of all documents cited in the input text.
12. The method of claim 11 , wherein the pre-filtering step is configured to also pass a predetermined number of results having a lower relevancy score than the document cited in the input texts having the lowest relevancy score of all documents cited in the input text.
13. The method of claim 1 , wherein the weighted aggregation is determined by an SVM model.
14. The method of claim 1 , wherein the parsing step includes converting the input text to HTML format if it is not already in HTML format, executing an HTML parser to extract plain text from the HTML, formatted input text, segmenting the plain text into discrete issues and generating metadata for each of the discrete issues.
15. The method of claim 1 , wherein the determining the degree of similarity between headnotes relevant to the input text and headnotes identified in recommendation candidates on which the recommendation-frequency score is based includes, for each headnote present in a recommendation candidate, determining frequency that a particular headnote is included in the input text and documents cited by the input text relative to all other included headnotes, and summing such frequencies for all headnotes identified in the recommendation candidate.
16. The method of claim 1 , wherein the relevancy score is further based on an issue similarity score that indicates similarity between respective conceptual issue topics of the recommendation candidate and the input text.
17. The method of claim 1 , wherein the relevancy score is further based on an issue similarity scores, wherein the issue similarity score indicates similarity between respective conceptual issue topics of the recommendation candidate and the input text.
18. The method of claim 1 , further comprising determining the authority score based on two or more authority scores associated with the recommendation candidate.
19. The method of claim 1 , further comprising determining the authority score based on one or more metadata values stored in a data structure associated with the recommendation candidate.
20. The method of claim 1 , further comprising determining the authority score based on a graph data structure that represents relationships among the recommendation candidates.