Research And Implementation Of User Comments Aspect-level Sentiment Classification System

The flourishing development of the Internet has promoted the development of social networks and e-commerce,resulting in a large amount of data on user review information.The text data contains much meaningful and valuable information.The sentiment analysis task is mainly to identify the sentiment polarity of a given text,so as to further extract and analyze the information contained in the text.The aspect-level sentiment classification serves as a fine-grained task for the sentiment classification.Its main purpose is to extract specific aspects of sentiment polarity from the text.This task has made great progress in recent years.This thesis presents the design of an aspect-level user comment sentiment analysis system,which is suitable for users and businesses on e-commerce platforms to obtain product information,enabling them to obtain product information and make decisions intuitively,quickly and accurately.The core part of the system is the aspect-level user comment sentiment classification model.This thesis designs an accurate and efficient aspect-level sentiment classification model by studying the aspect-level sentiment classification method.Considering the strengths and weaknesses of the existing researches,this thesis argues that the key to aspect-level sentiment classification tasks can be summarized as three factors:contextual semantic information towards aspect words,correlations between aspect words and their context words,and location information of context words with regard to aspect words.For aspect-level sentiment classification,one model called AELA-DLSTMs(Attention-Enabled and Location-Aware DLSTMs)is proposed in this thesis.AELA-DLSTMs makes full use of DLSTMs,which can capture the contextual semantic information in both forward and backward directions towards aspect words.At the same time,a novel attention weights generating method that combines aspect words with their contextual semantic information is designed and those weights can make better use of the correlations between aspect words and their context words.This thesis presents the observation that context words with different distances or different directions towards aspect words have different impacts on the sentiment polarities.The location information of context words by assigning different weights is incorporated in AELA-DLSTMs to improve its accuracy.This research shows the experimental results on two English datasets and one Chinese dataset.Experimental results have confirmed that compared with the benchmark models,the proposed model can achieve significant improvement in both Chinese and English datasets,and achieve the best accuracy.At the end of this thesis,the effectiveness of the whole sentiment analysis system is evaluated.