Abstract: For their greater contribution to the semantics of the paragraph, the first and the last paragraphs and the first and the last sentences of the paragraph should be taken as the main factors in computing the similarity of the paragraphs. By using them in SiteQ with appropriate weight, we propose Topic-SiteQ calculation algorithm. It uses a multi-feature fusion algorithm to compute the semantic similarity of the first and the last sentences that contribute to the paragraph similarity by weight. At the same time, we improve the score of the first and the last paragraphs, recommend and sort the paragraphs by the final score. Experiments show that, with Topic-SiteQ, the MRR value of relevance ranking of paragraph increased about 0.032, and the F-measure increased about 1.4%. The experimental results show that the optimized algorithm is effective.

Key words: automatic question answering system, SiteQ, semantic similarity, multi-feature fusion