Semantic Textual Similarity (STS) in Arabic using Lexical-Semantic Analysis

Semantic text similarity has a crucial role in natural language understanding (NLU), which is intended to evaluate the semantic and relatedness of two texts and produce a similarity score. Frequently, the Arabic short text consists of very few words, which mean that they usually do not provide adequate data in understanding the sentences. There are many difficulties and challenges that face processing and understanding of the Arabic short texts. The representations of Arabic short text of different lengths are investigated using lexical-semantic technique to eliminate the less informative words. In this paper, a hybrid approach that combines word-based features and text-based features is proposed for MSA Arabic short texts. It is based on lexical and semantic analysis. Semantic relatedness uses the Feature Vector (FV) Model, which is generated by acquiring semantic features of the word and the context. A comprehensive evaluation is done and the accuracy of the links between concepts is enhanced. The results shows that the lexical and Semantic knowledge is important in understanding short text.