Research On Online Commodity Sales Forecast Method Based On Emotional Computing

In recent years,with the rapid development of e-commerce,online shopping reviews have become one of the most important information sources for online consumer evaluation as a digital version of traditional word-of-mouth.On the one hand,a large number of theoretical research and practical applications show that the emotional tendency of consumers in the online shopping commentary information not only has an important guiding role for other consumers' potential purchasing decisions,but also for manufacturers and merchants to improve product quality and continuously optimize business strategies.On the other hand,because different consumers and merchants often pay different attention to the characteristics of goods,the traditional analysis results based on the overall sentiment orientation of online shopping reviews can no longer meet the increasingly diverse real needs of users.Based on the above background,this research summarizes the sentiment analysis theory and its main research results,and uses a certain e-commerce platform's online shopping commentary data on mobile phones as the experimental data for research.In this research,the sentiment analysis is applied to different attributes of commodities,and the online commodity sales volume forecasting model based on the fine-grained sentiment analysis of online shopping commentary information is carried out.The main research work are as follows:(1)An explicit feature extraction method based on BiLSTM-CRF model and part-of-speech features is proposed.The method firstly models the comment sentences based on words and part-of-speech independently,and uses the BiLSTM network to learn the long-term dependence between them.Then,the semantic features of the fusion words and part-of-speech are used to connect the merged semantic features to the single-layer fully connected network,and finally the output is used as the input of the CRF network.The text sequence with feature annotation is obtained through the training of the CRF model.Using this method,this research completed the explicit feature extraction of experimental data and achieved good experimental verification results.(2)An implicit feature extraction method based on context score is proposed.The method firstly extracts feature opinion pairs based on semantic and statistical methods,and constructs a co-occurrence matrix accordingly.On this basis,the correspondence between implicit features and explicit features in the context is combined.Through experimental verification,it is proved that the method can effectively extract the implicit features of commodities.(3)According to the extracted comment sentences containing features,firstly,the emotion dictionary is used as the training set with the high emotional tendency value,and then the remaining comments are judged by the classifier combined with the self-training method in semi-supervised learning,and finally,the emotional tendency of each feature cluster is obtained according to the constructed feature dictionary.This method not only solves the problem that traditional classification algorithms need to perform manual dataset labeling,but also achieves good experimental results in experimental verification.(4)Based on the above research,a predictive model of online merchandise sales volume is constructed.The weighted factors are selected by calculating the influence weight of each influencing factor,and then the BP neural network model is constructed by combining the obtained emotional value of the comment text.The effectiveness of the method is proved by experimental verification of mobile phone sales.