Personalized information retrieval method and device

Abstract

The invention discloses a personalized information retrieval method and device. The method comprises generating retrieval results according to an assigned user's retrieval task; subjecting each query term in the user's retrieval task and each query word in each query term to Bayesian probability matrix decomposition to obtain query-document relevancy; acquiring the user's long-term behaviors and short-term behaviors, and acquiring user-document attention according to the user's long-term behaviors, short-term behaviors and behavior weighted parameters; generating the user's sorting weighted parameters, and determining relevancy of each document in the retrieval results according to the query-document relevancy, user-document attention and sorting weighted parameters; sorting the documents in the retrieval results according to the relevancies of the documents in the retrieval results, and returning to the user. The personalized information retrieval method and device have the advantages that the accuracy of a personalized information retrieval sorting model can be improved and retrieving may be independent of specific contents of query and documents.

Description

technical field [0001] The present invention relates to the field of information retrieval, in particular, to a personalized information retrieval method and device. Background technique [0002] As the information retrieval system can record information such as each user's query and browse the records of retrieval results, the personalized information retrieval model has been developed rapidly. In these personalized information retrieval ranking models, user behavior information can be used to predict users' long-term attention. When the user enters the query again to perform information retrieval, the results related to his long-term focus will be returned to the user first, so as to improve the accuracy of information retrieval, meet the user's retrieval needs, and then improve the user's satisfaction with the information retrieval system. This method of improving the quality of information retrieval and sorting by mining information such as historical queries and behavi...

AI Technical Summary

Problems solved by technology

[0003] In the prior art personalized information retrieval ranking model, the user's long-term behavior records or short-term behavior records under the current information retrieval task are used to predict the user's retrieval intention and improve the accuracy of information retrieval, but both are not considered at the same time , which makes the retrieval performance of the ranking model relatively low when the user's current query intent differs from his long-term attention

At the same time, the existing information retrieval ranking model is mainly based on the specific content of the query and the document to estimate the correlation between the two. This model will not work if the specific content of the query and the document cannot be obtained.

[0004] For the problems of low accuracy and high content dependence of the personalized information retrieval and ranking model in the existing technology, there is currently no effective solution

Images