Research And System Implementation Of Sentiment Analysis Model Of Overseas E-commerce Review Data

The sentiment analysis of product review data has always been a research hotspot at home and abroad.Through the sentiment analysis of consumer reviews,the consumer behavior,motivation and preferences of consumers are inferred,so as to formulate product development strategies and improve products for businesses and businesses The quality of goods and services,as well as the conversion rate of new and old users.Although the current research on sentiment analysis of product review data has achieved good results,it has not been better developed due to the variety of review data formats and long training time of neural networks.With the rapid development of text classification and sentiment analysis technology in the field of natural language processing,researchers can better classify the product's Attribute characteristics and design sentiment analysis model.This thesis takes overseas e-commerce review data as the research object and explores the sentiment analysis model of review data.Firstly,the aspect data and sentiment attribute words are classified based on the improved dependency graph structure and the optimized attention mechanism calculation method;then the aspect word communication layer and the improved position attention mechanism are introduced into the text classification process,and combined with LSTM The memory network is used to improve the model's sentiment analysis effect.Finally,a web-based sentiment analysis prototype system for overseas e-commerce reviews data is designed and implemented.The specific work is as follows:(1)Aiming at the problems that the existing sentiment analysis model cannot capture the aspect word attributes better and the computational requirements of sentiment attribute weight assignment are high,a new sentiment analysis model DGBLE-LSTM is proposed.This model improves the original dependency tree structure,introduces dependency graph structure to capture aspect word attributes,and based on the original attention calculation,introduces a learning matrix and balance factors to assign emotional weights.The DGBLE-LSTM sentiment analysis model solves the problem of insufficient capture of the word attributes of the original dependency tree structure on the one hand,and reduces the dimensions and space requirements of the hidden layer representation and weight vector in the attention calculation process,and improves This improves the efficiency of sentiment analysis.The experimental results show that compared with the dependency tree structure classification method,DGBLE-LSTM improves the classification accuracy by 0.71% to 1.175% respectively;compared with the different methods of LSTM sentiment analysis and DASN sentiment analysis model,DGBLE-LSTM has the sentiment analysis accuracy The increase of 6.2% ~ 10.89% shows that DGBLE-LSTM has a better effect on sentiment analysis.(2)Aiming at the problem that the relationship between aspect words and attribute words in DGBLE-LSTM is not fully utilized,the original classification method is further optimized,and an improved sentiment analysis model DGABLE-LSTM is proposed.The model introduces the aspect word communication layer and improved positional attention in the word processing of the text aspect.On the one hand,the emotional attributes of the aspect word are integrated into the model to communicate and interact with the text.The importance degree characteristics of attribute words,the introduction of positional attention mechanism,and the improvement of its expression,fully consider the phenomenon that no aspect words appear in the text,so as to better capture the emotional attribute words corresponding to the aspect words.In addition,the dependency graph structure and improved attention mechanism calculation formula proposed in Job 1 are introduced,and LSTM is used to maintain the integrity of the text and reduce the model training time.The experimental results show that compared with the DGBLE-LSTM sentiment analysis model,DGABLE-LSTM improves the accuracy of aspect word classification and sentiment analysis by 0.64% and 0.69%,which fully demonstrates that the optimized DGABLE-LSTM has a better effect on sentiment analysis..(3)On the basis of the above research,this thesis has completed the design and implementation of a web-based sentiment analysis prototype system for overseas ecommerce review data.