Measuring Semantic Relatedness Between Words

Semantic relatedness between words is a concept which represents degree of correlation for two words, it reflects the degree of association of the words, that is, once you see one word, whether or not think of another word, we can use the possibility of two words co-occur in the same context to measure the semantic relatedness of two words. Semantic similarity and semantic relatedness are two concepts which are very easy to confuse, Semantic similarity refers to the similarity between the words. There is correlation between semantic relatedness and semantic similarity, if two words semantic similarity, then they must be semantically related, but in turn, if two words are semantically related, they are not necessarily semantic similarity, so we can use semantic similarity degree as an integral part of the semantic relatedness.Measuring semantic relatedness for machine translation, information retrieval, text analysis, natural language processing research task is of great significance, is also a fundamental research work. This paper studies the existing semantic relatedness method, then present a search engine based method measuring semantic relatedness, specific work are as follows:Firstly, the method of the existing semantic relatedness between words can be roughly divided into traditional method and method based on encyclopedia; while the traditional method can be further divided into two categories:based on semantic dictionary (WordNet and HowNet) and based on the corpus. In this paper, these methods need to use the semantics of resources to do a detailed introduction, followed by elaborate representative in each category are several semantic relatedness, analyze their theoretical basis and characteristic.Secondly, we propose a new method measuring semantic relatedness between words which combine kernel function and Page Counts, Page Counts is the number of pages returns when using search engine query. This semantic relatedness study provides a new direction for us, and tightly rapid development of network technology for our research. At the same time, we also have the following three aspects to verify the validity of the method:first, the analysis of its theoretical basis; second, doing experiments on a standard test set, and then compare with artificial analyzing the results do; third, estimating under certain circumstances. Verified by experiment, comparing to using kernel function or Page Counts alone, the result obtained from our propose method more close to the result by human judgment. So our method is effective.Thirdly, this paper describes text clustering which using semantic relatedness. On the basis of the result of measuring semantic relatedness between words, calculate semantic relatedness of text, we can improve the accuracy of text clustering.