| CPC G06F 16/2423 (2019.01) [G06N 20/00 (2019.01)] | 18 Claims |
|
1. A method comprising:
with a computerized search engine, retrieving a plurality of electronic documents relevant to a query;
obtaining, via computerized term embedding, from said retrieved documents, a plurality of most similar terms with respect to said query;
for each of said most similar terms, determining a pervasiveness score and a relevance score;
filtering out, from said most similar terms, those of said terms that are pervasive, based on said pervasiveness score, the pervasiveness score being a probability of a given term appearing in a document related to the given term in the plurality of electronic documents relevant to the query, those of said terms that are irrelevant, based on said relevance score, the relevance score for the given term being a measure of popularity of the given term computed using a rank of documents where the given term is observed in the plurality of electronic documents relevant to the query, and those of said terms that are redundant to improve a relevance of query terms and reduce a quantity of the query terms;
outputting a top number of terms remaining in said most similar terms after said filtering, based on similarity to said query, as discriminative facets; and
with said computerized search engine, retrieving an updated plurality of electronic documents relevant to said query by resubmitting said query with at least one of said discriminative facets.
|