Guess what i want: Inferring the semantics of keyword queries using evidence theory

Abstract

The tagged and nested structure of an XML document provides quite detailed information about its structure and semantic, which is neglected by traditional keyword search model like TF-IDF and BM25 etc. Popular XML search models such as SLCA and XRANK tend to return the "deepest" node containing all given keywords, which usually leads to semantic loss. In this paper, we introduce the concept of belief in D-S evidential theory to evaluate primary search results, and propose a novel ranking model XSRET to rank them. In XSRET, We utilize XML's rich tag system to predict the semantics of keyword queries. For evaluating our SLCA-E model, we compare it with some state-of-the-art models, such as XSeek and XReal, and experimental result shows that XSRET outperforms these models. In addition, XSRET won the championship in the contest of data-centric track of INEX 2010. © 2012 Springer-Verlag Berlin Heidelberg.