Fine-Grained Aspect-based Sentiment Analysis For Online Review

With the development of Electronic Commerce,it becomes important to aggregate opinions on aspects of entities mentioned in online reviews for both producers and consumers.Faced with massive online reviews,it is necessary to automatically discover the above key information.Aspect-based sentiment analysis aims to identify the aspects of entities and corresponding sentiment,which is commonly applied to extract aspects and opinions from online reviews.In order to automatically extract aspects and corresponding opinions from online reviews,it is important to extract aspect,identify aspect words and associated opinion words,classify sentiment polarity.For a fine-grained sentiment analysis,it is also important to identify granularities.Recently,various researches based on topic models are proposed to process some of the above tasks.However,there is little work available to do all simultaneously,so they can not use information from a certain task to improve results of other tasks,as well as process fine-grained sentiment analysis.This paper aims at fine-grained sentiment analysis for online reviews and studies fine-grained aspect-based sentiment analysis approaches.The main contributions of this paper are as follows:First,this paper studies and summarizes key tasks of fine-grained aspect-based sentiment analysis.A method for identifying granularities of aspect and opinion words is proposed to get a fine-grained sentiment analysis result.Second,this paper proposes a fine-grained aspect-based sentiment analysis model called Joint Aspect-Based Sentiment Topic(JABST)model.This model extends topic models and is able to process key tasks of fine-grained aspect-based sentiment analysis simultaneously.JABST model can identify aspect and opinion words,extract aspect,analysis sentiment and is able to identify fine-grained aspects and opinions.Third,this paper proposed MaxEnt-JABST model,in which a maximum entropy classifier is applied to extend JABST model.MaxEnt-JABST model is a semi-supervised version of fine-grained sentiment analysis model.It is able to better separate aspect and opinion words.This paper evaluated the JABST and MaxEnt-JABST models on reviews of electronic devices and restaurants qualitatively and quantitatively.The experimental results show that the proposed models outperform state-of-the-art baselines and are able to identify fine-grained aspects and opinions.