A new syntax-based aspect detection approach for sentiment analysis in Arabic reviews

With the explosive growth of different form of social media on the Web, individuals and organizations are increasingly trying to use the content in these media for decision making. Thanks to the automatic and accurate measurement of customer opinions, the organization can integrate, understand and analyze the needs of its targets and thus adapt its marketing strategy. In this paper, we describe a new syntax-based aspect detection approach for sentiment analysis in Arabic reviews. Many characteristics of the Arabic subjective text were studied and used in order to elaborate syntactic rules and disambiguation techniques that have been integrated in our approach. The proposed approach is based on a five-step process, namely: pre-processing, lexical entities disambiguation, separation between aspects, extraction of aspects and grouping of aspects. To demonstrate the effectiveness of our proposal, we used two datasets of Arabic social content related to hotel and product reviews. The former is based on content extracted from TripAdvisor while the latter is from the Souq website. The Experimental results showed that the syntax-based approach obtains an F-measure of 69,56% (resp. 68.29%) for the hotels dataset (resp. the products dataset).