The Research Of Information Retrieval Optimization Based On User Behavior Mining

This paper is written in the background of e-book web store information retrieval system. Firstly, through latent Dirichlet analysis process, we turn e-Book document which consists of tags into Book-topic distribution. Secondly, by user behavior such as purchasing、collecting、sharing、commenting、browsing a specific e-book, according to the book-topic distribution, we propose UPMA algorithm to calculate users behavior on each topic. Since the number of user behavior is not limited, each behavior will increase the interest weight on some topics; the UPMA algorithm can ensure convergence of the final result. The effect of search optimization is achieved by re-sorting of search results. The harmonic mean score of each result combines query and document similarity score with user interest matching score, we call it CIRA algorithm.This paper completed the following tasks:1. Research personalized search, user modeling, topic modeling techniques inside and outside China.2. Make use of LDA method to mine the latent topic in every e-book documents (consist of label text), establish e-books and topic distribution model.3. Based on user behavior in bookstore logs, combined with book-topic distribution, we proposed UPMA algorithm and prove its convergence; through this algorithm, we can get quantified vector of user interest preferences (weighed preference vector, WPV). In the reordering algorithm design, we make use of harmonic mean of query-document similarity and interest-topic similarity score to optimize the final search result.4. In terms of system design and implementation, we use Solr as a framework, analyze the structure of Solr, illustrate key configuration of the system. Through remote procedure calling and http, our system can be utilized by other applications.