A short text modeling method combining semantic and statistical information

Abstract

A novel modeling method for a collection of short text snippets is presented in this paper to measure the similarity between pairs of snippets. The method takes account of both the semantic and statistical information within the short text snippets, and consists of three steps. Given a set of raw short text snippets, it first establishes the initial similarity between words by using a lexical database. The method then iteratively calculates both word similarity and short text similarity. Finally, a proximity matrix is constructed based on word similarity and used to convert the raw text snippets into vectors. Word similarity and text clustering experiments show that the proposed short text modeling method improves the performance of existing text-related information retrieval (IR) techniques.