Query Aspect Based Term Weighting Regularization in Information Retrieval

摘要: Traditional retrieval models assume that query terms are independent and rank documents primarily based on various term weighting strategies including TF-IDF document length normalization. However, related, groups of semantically related may form aspects. Intuitively, the relations among could be utilized to identify hidden aspects promote ranking covering more Despite its importance, use semantic for regularization has been under-explored in information retrieval. In this paper, we study incorporation into existing focus addressing challenge, i.e., how regularize weights different improve performance. Specifically, first develop a general strategy can systematically integrate function functions, then propose two specific functions guidance provided by constraint analysis. Experiments eight standard TREC data sets show proposed methods effective accuracy.