Calculation of Relatedness by Using Search Results

Semantic relatedness is a very important research field and it has been used in many applications. Most of current works depend on the hierarchy of WordNet, which results in restricted calculation objects and language, although context vector method and Google distance do not have such drawbacks, but the accuracy is not satisfied. This paper presents a new semantic relatedness calculation method by using search results returned by search engines, which is called search-based relatedness measure. It takes statistical information and content of search results into consideration. Search-based relatedness measure is based on assumption that words appear in a same result have some relatedness and some top ranked search results can represent meaning of the searched term. Search-based relatedness measure does not have the limitation on part of speech and language, and experiment results show its accuracy is better than all current measures.