Search prediction using context modeling

摘要

A method and apparatus for predicting search queries for a user on a search-based system is described. A search engine tracks search queries from a user received at the search engine. A prediction engine compares a search query history from the user with search query histories from other users on the search engine. The search query histories are stored in a storage device of the search-based system. A prediction of a search query for the user is generated based on the comparison. Search query histories from other users having greater similarities with the search query history from the user have a greater computation weight than search query histories from other users with less similarities with the search query history from the user.

主权项

1. A method comprising:
tracking search queries of a user of a search engine to generate a search query history; comparing the search query history of the user with search query histories of other users of the search engine; determining a number of common search queries between the search query history of the user and the search query history of each other user in view of the comparing; associating the search query history of the user with the search query history of at least one other user having the most number of common search queries; determining where a test sequence of the search query history of the user is most similar to a reference sequence of at least one other user having the most number of common search queries; and generating, by a processing device, a predicted search query for the user that the user is predicted to use to perform a next search in relation to other possible searches from the search query history of at least one other user having the most number of common search queries in view of the test sequence and the reference sequence, wherein the predicted search query is the next sequential predicted search relative to the reference sequence, and wherein the search query histories of the other users with greater similarities to the search query history of the user have a greater computation weight in generating the predicted search query than the search query histories of the other users with fewer similarities to the search query history of the user.