Unsupervised aspect-based Sentiment Analysis in the Persian language: Extracting and clustering aspects

Abstract

Sentiment analysis is the field of natural language processing to analyze user feedback, preferences, and evaluations from the text. Different organizations in most social scopes use this context as an appropriate tool to find their strengths and weaknesses. In this field of research, the objective is to determine the positive or negative orientations of users towards the features of a product or commodity. Solving this problem consists of two main steps: aspect extraction and identifying the positive or negative tendencies of users towards those aspects. Two of the most critical issues of sentiment analysis in the Persian language are the lack of comprehensive labeled data and the significant difference between colloquial and formal sentences in Persian. One of the most important methods to confront the first problem is to apply unsupervised methods. In this research, a system for aspect-based sentiment analysis in the Persian language is proposed using unsupervised methods. The Sentiment words extraction step is done in another article [1]. The aspect extraction and clustering steps are done using topic modeling methods and neural networks and taking benefit of rule-based methods. This system has been evaluated using precision and recall criteria. The F1 criterion for extracting aspect words in the proposed system was 0.766.