Feature-Based Opinion Summarization for Arabic Reviews

Opinion mining applications work with a large number of opinion holders. This means a summary of opinions is important so we can easily interpret holders' opinions. The aim of this paper is to provide a feature-based summarization for Arabic reviews. In our work, a system is proposed using Natural Language Processing (NLP) techniques, information extraction and sentiment lexicons. This provides users to access the opinions expressed in hundreds of reviews in a concise and useful manner. We start with extracting feature for a specific domain, assigned sentiment classification to each feature, and then summarized the reviews. We conducted a set of experiments to evaluate our system using data corpus from the hotel domain. The accuracy for opinion mining we calculated using objective evaluation was 71.22%. We, also, applied subjective evaluation for the summary generation and it indicated that our system achieved a relevant measure of 73.23% accuracy for positive summary and 72.46% accuracy for a negative summary.