| Personalized information retrieval is one of the most popular research directions nowadays. It provides services like automatic information collection, analysis, delivery and so on. Its services are more specialized and of higher quality compared with those of general information retrieval systems. The service quality is mainly reflected by the sorting results of WebPages, thus the relevance evaluation of WebPages is the key process in the personalized information retrieval system. Probabilistic model has its advantages in user interest modeling. With probabilistic arguments, it can describe user requirements more exactly. Thus it's suitable for the relevance evaluation of WebPages in personalized information retrieval system.This dissertation is mainly about personalized information retrieval of entity WebPages, trying to find out the way to improve the relevance evaluation precision. The author designs and implements the probabilistic model based relevance evaluation algorithm of famous people WebPages. After a thorough discussion of issues like the model training strategies, model improvements, query expansion and so on, the paper makes some conclusions about probabilistic model and the way to improve the relevance evaluation precision and provides detailed experiment results.The innovations of this dissertation are as follows:(1) It puts forward an appropriate training sets choosing method, thus achieves better training results and lower overhead as well.(2) Using improved probabilistic formula, it absorbs more detailed user feedback, thus optimizes the distribution probabilities of terms. It perfects the relevance computing formula, introducing term frequency, webpage length, HTML tags and other webpage factors, and forms the idea of tailoring relevance computing formula based on entity classes.(3) Based on the feature of entity attributes, it proposes method to extract relevant terms on the basis of both relevant WebPages and entity attributes for query expansion. |
PDF Full Text Request