| CPC G06F 16/338 (2019.01) [G06F 16/313 (2019.01); G06F 16/3347 (2019.01); G06F 16/382 (2019.01); G06F 40/20 (2020.01); G06N 20/00 (2019.01)] | 17 Claims |
|
1. A method comprising:
receiving, at a computing device, training data comprising a plurality of paragraphs from a plurality of legal briefs, wherein each paragraph of the training data is labeled as to whether the paragraph is an argument paragraph containing one or more legal arguments;
using the training data to train a machine learning model to determine whether an input paragraph is an argument paragraph by using supervised learning techniques;
receiving, at the computing device, an electronic document, wherein the electronic document is a legal brief comprising a plurality of paragraphs;
inputting the plurality of paragraphs of the electronic document into the trained machine learning model to identify one or more argument paragraphs from among the plurality of paragraphs, wherein each of the one or more argument paragraphs includes one or more legal arguments;
performing a textual search of a corpus of legal documents by comparing text of the identified argument paragraphs to text of the legal documents in the corpus;
performing a semantic search of the corpus of legal documents by comparing semantic meanings of the identified argument paragraphs to semantic meanings of the legal documents in the corpus, wherein performing the semantic search comprises:
determining a vector representation of one or more words of an argument paragraph;
comparing the vector representation of the one or more words of the argument paragraph to a vector representation of an index of the corpus by:
identifying a plurality of documents of the corpus using an approximate nearest neighbor search; and
determining a cosine similarity between the vector representation of the one or more words of the argument paragraph and a vector representation of the documents identified using the approximate nearest neighbor search; and
returning a score for one or more documents in the corpus based on the comparison;
combining results of the textual search and the score returned from the semantic search; and
presenting the combined results to a user by displaying the combined results associated with each identified argument paragraph adjacent to the associated identified argument paragraph in a word processing program.
|