A Context-Aware Computing Method of Sentence Similarity Based on Frame Semantics.

Sentence similarity computing is a typical technology used in natural language processing, which aims at finding valuable information from documents. By adapting advanced technologies, such as machine learning and deep learning, current sentence similarity computing methods mainly deal with key words and structures of sentences. The main drawback of current methods is taking no consideration of the influence of sentences context. In this paper, we propose a frame semantics theory based computing method that is built upon FrameNet. By quantitatively analyzing the semantic relations among frames, sentences similarity can be calculated based on the frames that are evoked by the sentences. We also carry out experiments to evaluate the performance of this method with the help of a prototype software tool we developed.