| With the advent of Internet and e-commerce,sentiment analysis has become one of hot topics in the field of natural language processing.Sentiment analysis has a wide range of applications,such as information retrieval,recommendation system,etc.Recently,the ma-jor e-commerce platforms have launched Question-Answering(QA)board one after anoth-er,which gives birth to a new type of QA comment.In this way,users can ask and answer questions about the product information they are interested in.Compared with traditional way of comment,the advantage of this approach is that it avoids fraud and generates more reliable data.However,there is no related research on fine-grained sentiment analysis of the QA text.This dissertation focus on attribute-level sentiment analysis towards QA text.Specifically,it is divided into opinion target extraction task,attribute classification task and attribute-level sentiment classification task.Our details of research content are as follow:Firstly,this dissertation studies the method of opinion target extraction towards QA text.Opinion target extraction aims at extracting opinion target from QA text.Aiming at this task,this dissertation proposes a method of opinion target extraction based on attention mechanism to extract the opinion target of QA text.In particular,Firstly,the question and answer text are encoded by a bi-directional LSTM model respectively.Secondly,we use attention mechanism to obtain question text representation that integrates the answer in-formation.Finally,Conditional Random Field layer is used to obtain global optimal tag sequence of the question text.Experimental results show that our proposed method has better recognition performance than traditional opinion target extraction methods.Secondly,this dissertation studies the attribute classification method towards QA text.Attribute classification aims at identifying attribute label information of QA text.Aiming at this task,this dissertation proposes an attribute classification method based on mul-ti-dimensional text representation.In particular,firstly,we segment the question text into sentences and a long short-term memory network(LSTM)is used to encode each question and answer obtained by segmentation.Secondly,the acquired codes are merged to form a multi-dimensional text representation.Finally,feature extraction is performed on the mul-ti-dimensional text representation using Convolutional Neural Networks layer and classi-fication the text.Experimental results show that the performance of our proposed method is obviously superior to traditional attribute classification methods.Finally,this dissertation studies the attribute-level sentiment classification method to-wards QA text.Attribute-level sentiment classification aims at identifying sentiment in-formation of the specified object attributes in QA text.Aiming at this task,this dissertation proposes an attribute-level sentiment classification method based on attention mechanism.In particular,firstly,the answer text with specific attribute information and the question text are encoded by LSTM respectively.Secondly,we capture the sentiment information of specific attributes between the question text and answer text through attention mechanism.Finally,the sentiment polarity of specific attribute are output by using classifier.Experi-mental results show that the performance of our proposed attribute-level sentiment classi-fication method is obviously superior to traditional attribute-level sentiment classification methods. |
