Web query classification using improved visiting probability algorithm and babelnet semantic graph

Abstract

In this paper, an unsupervised method which is not use log data is offered to solve "the problem of web query classification". The aim of the proposed approach is the mapping of all the problem components to the BabelNet concepts and solving the problem by using these concepts. Therefore, it is considered a three-phase solutions consist of Offline, Online and Classification phases. In offline phase, all categories are mapping to concepts in BabelNet by using a disambiguation system. In the online phase, first a query is enriched then preprocessing on query is required, after that, by using a disambiguation system all components are mapped to BabelNet's concepts. In the last phase, by improving on visiting probability algorithm, classification is done. For testing process, we used KDD2005 test set, which is leading the series have been used. Results indicate that between the approaches which are unsupervised and do not use log data, proposed approach, has acceptable performance in the point of view F 1 measure. In other words, by compare to best approach which is unsupervised and does not use log data, proposed approach improved 2%, but by compare to the best approach which is unsupervised and uses log data the results get worse and shows reduction of 11% in term of F 1 measure.