Fine-grained Sentiment Analysis For Case-related Weibo Comments

With the rapid development of the Internet,the influence of Internet public opinion on court trials has become increasingly prominent.Weibo is a gathering place for hot case discussions.Case Weibo refers to Internet Weibo focusing on case-related incidents.The sentiment analysis technology for case-related microblogs can quickly obtain emotional information from a large number of comments,so as to understand people's attitudes towards case holders in different aspects and guide the prediction of case public opinion trends.Significance.The sentiment analysis methods are different.Based on the characteristics of the case Weibo field,the role of case element information in the sentiment analysis process is analyzed.The case micro-study is studied under the framework of deep learning networks such as convolutional neural networks and bidirectional LSTM The identification of blog opinion sentences,the removal of evaluation objects and the analysis of sentiment orientation.Here,the research on the fine-grained sentiment analysis of case-related Weibo comments is mainly completed with the following characteristics:Propose a case microblog opinion sentence recognition model based on feature extended convolutional neural network,analyze the relationship between the case microblog original text and comments,after extracting the keywords of the case original text,use the case microblog The keyword information of the original blog post expands the feature of the input layer.The convolutional layer extracts the viewpoint features associated with the keywords of the case,and finally identifies the required viewpoint sentence.The experimental results show that the method has the accuracy of recognition.effective.(1)Case Weibo data acquisition and data set construction.Data acquisition and labeled training corpora are the basis of supervised learning.This article requests through the requests toolkit,and then parses out the text with html tags.Through keyword search,it crawls from Sina Weibo to extract the current popular case Weibo original and comment data.According to the different task requirements of object level sentiment analysis of opinion sentence recognition and evaluation,a total of 39 case Weibo original texts and 4711 case Weibo related case comments were constructed on #BUS Case #,#Benz Case #.Sentence recognition data set;Based on this,a case-level sentiment analysis data set consisting of 2003 evaluation objects and2379 reviews was constructed for case evaluation objects and polarity labeling.The work in this chapter has an important supporting role for later research points.(2)A method of case microblog opinion sentence recognition based on feature extended convolutional neural network is proposed.Compared with the general domain opinion sentence recognition task,case microblog opinion sentence recognition usually involves specific case element information.This information is generally present in the original case microblog and plays an important role in guiding the opinion sentence recognition process.Therefore,combined with the deep learning framework based on convolutional neural network,a case microblog opinion sentence recognition method combining case information feature expansion is proposed.First,use the text Rank technology to extract the keywords in the original text of the case,use the keyword information with case elements to expand the feature of the input layer,and use the convolution layer to extract the viewpoint features associated with the case keywords,and finally identify the required Opinion sentence.In order to verify the effectiveness of the method,experiments were performed on the case Weibo opinion sentence recognition data set constructed.The experimental results show that the method has better accuracy than the benchmark model in recognition accuracy.(3)An evaluation object extraction method based on dual-embedded convolutional neural network is proposed.Case Weibo evaluation objects have different meanings in the general field and the case field.They cannot accurately characterize the relevance to the case in the general field,but have case-specific meaning in the case field.Therefore,an evaluation object extraction method based on the double-embedded convolutional neural network is proposed.First,pre-train the case Weibo review text to obtain an embedded layer with the characteristics of the case field.At the same time,through the case Weibo field and the general field embedding layer,obtain the characterization results of the evaluation objects in different fields and perform stitching operations.The layer extracts the features related to the case,and finally uses the classifier to mark the sequence to extract the case's Weibo evaluation object.In order to verify the validity of the experiment,experiments are performed on the constructed case Weibo evaluation object-level sentiment analysis data set.The experimental results show that it has better results than the traditional neural network model.(4)A method for extracting evaluation objects based on double-embedded multi-layer convolutional neural network is proposed.Case Weibo evaluation objects have different meanings in the general field and the case field.They cannot accurately characterize the relevance of the case in the general field,but have a case-specific meaning in the case field.Therefore,a method for extracting evaluation objects with double embedded multi-layer convolutional neural network is proposed.First,pre-train the case microblog comment text to obtain the embedding layer with case domain characteristics.At the same time,through the embedding layer of the case microblog domain and the general domain,the representation results of the evaluation objects in different domains are obtained and the stitching operation is performed.The layer extracts the features related to the case,and finally uses the classifier to mark the sequence to extract the evaluation object of the case Weibo.In order to verify the validity of the experiment,the experiment was conducted on the sentiment analysis data set of the evaluation object level of the case microblog constructed,and the experimental results showed that it had a better effect than the traditional neural network model.(5)Prototype system of case Weibo sentiment analysis.Based on the above research results,a case-based micro-blog sentiment analysis prototype system was designed and implemented.It integrates data acquisition,recognition model of opinion sentence corresponding to review text,extraction model of evaluation objects,and sentiment analysis model,and provides users with a visual information acquisition platform.