Measuring Sentence Similarity from Both the Perspectives of Commonalities and Differences

摘要

Similarity between two sentences can be determined by either comparing their commonalities or their differences. Commonalities, which reflect similarity judgment, connect the two sentences while differences, which reflect dissimilarity judgment, represent the unique way of self-identification. Although both of them are essential in determining sentence similarity, however, the existing methods only focus on single perspective, mostly the perspective of commonalities. This paper presents a method which calculates the sentence similarity from multiple perspectives by taking both the commonalities and differences into consideration. The experimental result on a standard data set shows that the proposed method outperforms the baseline, which is the existing most outstanding single perspective measure, with statistically significant improvement.