Extraction Of Subject Features And Sentiment Analysis Of Commodities Online Reviews

Due to the rapid development of Internet technology and E-commerce,more and more consumers take online shopping as their main way of shopping.E-commerce platform not only brings convenience to consumers and enterprises,but also produces a lot of online evaluation information.Online evaluation collects the personal opinions of each product's history consumer,changes the previous situation that consumers can only understand the product through product description and acquaintances' word of mouth,and provides support for consumers' purchase decision from various angles.In addition,the online evaluation contains a large amount of market information,such as the direction of product improvement,product and service quality,etc.,which can help enterprises realize customer-centric product and service concepts,so as to expand market share and improve profitability.It mainly extracts,organizes and analyzes the viewpoint information in the text through data mining,machine learning and other algorithms,and judges the emotion expressed in the text.Through the sentiment analysis of the online reviews of e-commerce,consumers can know the information such as product quality and function in advance,e-commerce platforms can obtain the preferences of different users,and manufacturers can grasp the timely changes in consumer demand.Traditional sentiment analysis research,however,is mainly around the chapters,paragraphs,sentences,text coarse-grained sentiment analysis,to analyze the consumer attitude is good for the product or derogatory,but was not able to analyze consumer emotional attitude for a specific product features,so around the product attributes of fine-grained sentiment analysis research has important practical significance and theory significance.On the basis of domestic and foreign research status,text emotion classification,text vector-quantization,multi-task learning model and other theories,this paper USES feature engineering and machine learning algorithm to conduct in-depth research on key task feature extraction and emotion classification in online fine-grained emotion analysis.Firstly,according to online evaluation of text feature extraction problem,with the introduction of external Chinese language corpus and evaluation of online text corpus to carry on the comprehensive training,reduce error,semantic understanding and using the clustering algorithm of machine learning to consumers focus on product attributes synthetically classified,then proposed a fusion of clustering algorithm based on word vector feature extraction method,and the method based on word frequency and analyzed the feature extraction method based on the subject.Then,in order to deep mining online consumer fine-grained emotional information in the comment text,through the introduction of SIF,therefore,embedding method dimension vector layer,in as much as possible under the premise of reducing the loss of semantic document feature vector,and the neural network algorithm into multitasking learning popular MMo E framework,design contains the word vector classification,pooling layer,layer three layer system based on the analysis of MMo E fine-grained emotions Shared neural network model.Finally,through Meituan review data set analysis of the random forest,support vector machine(SVM),logistic regression,analysis of three-layer neural network model of emotional effect,and the effectiveness of the proposed model,the Shared neural network based on MMo E Framework for fine-grained sentiment analysis,is verified.The experimental results show that the fine-grained emotion analysis model based on MMo E is superior to other models in the evaluation indexes such as accuracy and loss value.Therefore,it is suggested that future manufacturers choose online evaluation text as information input in future customer word-of-mouth research and use multitask learning neural network to mine fine-grained sentiment information of online commodity evaluation.