An automatic Persian text summarization system based on linguistic features and regression

Considering the vast amount of existing written information and the shortage of time, optimal summarization of books, articles, news reports, etc. on the Web is a major concern of researchers. In this paper, we propose a new approach for Persian single-document Summarization based on several linguistic features of text. In our approach after extracting the linguistic features for each sentence, the weight of features is learned by a linear regression method. We select one sentence with maximum score at each step of algorithm. The score of each sentence is calculated based on two factors: first, sum of the weighted features and second, the amount of its similarity to the sentences that are selected for final summary previously. We use an automatic evaluation tool to compare our approach with other existing approaches. The result indicates that our method improves the performance of summarization.