Research And Application Of Target-Dependent Fine-grained Sentiment Analysis

The traditional text sentiment analysis aims at judging the sentiment polarity or sentiment intensity of a given text from the whole sentence.However,this often overlooks the bias of sentiment due to different targets.This paper focuses on the target-dependent fine-grained sentiment analysis,which aims to analyze the sentiment of the text based on a particular target.For example,in product reviews,the target may be the different attributes of the laptop,such as price,battery,performance,etc.In the first part of this paper,the traditional machine learning method is used to conduct the fine-grained sentiment analysis for different types of targets in different areas.Specifically,our first work is to build the target-dependent fine-grained sentiment classification models for product reviews in laptop and restaurant domain using nature language processing and machine learning methods,where the targets are the entities and attributes of the product.This work has been applied to the task 5,i.e.,Aspect-Based Sentiment Analysis,in International Workshop on Semantic Evaluation in 2016(SemEval-2016)and achieved the third place.The corresponding research has been published in the SemEval conference in 2016.Our second work is to build the stock-dependent fine-grained sentiment regression models for Financial Twitter and News using well-designed features combined with machine learning methods,aiming at identifying the sentiment polarity(bullish or bearish)and the sentiment intensity associated with the target companies and stocks.This model has been applied to the task 5,i.e.,Fine-Grained Sentiment Analysis on Financial Microblogs and News,in SemEval-2017 and achieved the first place.The corresponding research has been published in the SemEval conference in 2017.However,the traditional nature language processing and machine learning methods have a strong domain dependency when it is used to deal with target-dependent sentiment analysis.It relies on complicated hand-crafted features engineering,experts'domain-specific knowledge and has poor generalization performance.Therefore,in the third part of this paper,we propose a novel Gated and Attention-based Bidirectional Long-Short Term Memory neural network model(GABi-LSTM)for target-dependent fine-grained sentiment analysis,and apply this model to product reviews and financial domains,where the performance in both domains is superior to traditional machine learning methods.This part of work has been published in International Conference on Knowledge Science,Engineering and Management 2017(KSEM 2017).In this paper,we carried out a large number of experiments using a variety of methods(machine learning and deep learning)at different granularity(coarse-grained sentiment polarity,fine-grained sentiment intensity)from different data sources(product reviews,Twitter,news)in different domains(laptop,restaurant,financial stock).The experimental results show the effectiveness of our proposed machine learning and deep learning models for target-dependent fine-grained sentiment analysis.