Evaluation method of automatic summarization calculating the similarity of text based on HowNet

Abstract

In order to do automatic evaluation for summarization much more accurately and efficiently, this paper analyzed the present evaluation methods of automatic summarization concretely, and pointed out the shortcoming of these evaluation methods. Based on the method of vector space model, it presents an evaluation method of automatic summarization calculating the similarity of text based on HowNet. It analyzes the meaning of words concretely using HowNet in the vector space model, considering the effect of part of speech serving as role in the sentences when calculating the weight of feature item and improving the formula of weight of feature item. The experiment shows that evaluation result of this method is better than that of P/R and the method based on the similarity of text.