Research On Recognition Of Product Innovation Features Based On LSTM Model

Product innovation is the guarantee for an enterprise to maintain a sustainable competitive advantage.Through product innovation,an enterprise can more effectively meet the ever-changing needs of consumers.Traditional user needs acquisition methods based on questionnaires and focus interviews have disadvantages such as long time,high cost,and small sample size.In addition,users pay more and more attention to subjective experience elements in the process of consuming products,and traditional methods are difficult to help enterprises to discover the real potential needs of users.The massive online product reviews of online shopping platforms contain a wealth of product demand information about product needs,user experience,improvement suggestions.Based on this,this paper designs an attention mechanism based on long and short-term memory networks(At-LSTM).The model framework is used to mine user demand information in online comment data,identify product features that users pay attention to,and provide companies with product innovation and improvement directions.Existing research on online review product feature recognition based on sentiment dictionaries,machine learning,neural networks.This has effectively provided improvement directions for product innovation,but there are still problems such as lack of contextual semantics and low classification accuracy.LSTM can effectively improve these problems,but there are still shortcomings in the fine-grained product feature emotional polarity evaluation,the capture of important components in sentences,and the long-distance dependent feature capture.Therefore,a feature recognition model of product innovation based on LSTM is constructed in this paper.First of all,on the basis of summarizing relevant research results and analyzing the language characteristics of online product reviews,a process framework for LSTM online review product innovation feature identification integrated with the attention mechanism is constructed,and the data collection and preprocessing processes are carried out to analysis.Secondly,for the problem of fine-grained sentiment analysis in LSTM,the usefulness of reviews is screened through the construction of product feature vocabulary and sentiment vocabulary,which effectively ensures that each review contains product features and corresponding sentiment words.Thirdly,through the integration of the attention mechanism to capture the internal structure of the sentence,and through the learning and training of the word dependence in the sentence,the attention weight of different aspects of product characteristics is given,and the accuracy of emotion classification is improved.Finally,the identified product feature frequency,coverage,and emotional coefficient are combined with the Kano model to more effectively identify product features that help improve consumer satisfaction.In order to test the feasibility of the model method constructed in this paper,the online reviews of four smartphones,oppo,vivo,Huawei and Xiaomi,crawled on JD and Taobao platforms are used as experimental data to test the relevant models.The experimental results show that:the At-LSTM accuracy,precision,and recall of the LSTM model are 91.52%,91.73%,and 91.53%,respectively,which are improved compared to text sentiment classification models such as KNN,NB,and SVM.The identified product features are similar to the Kano model The combined experimental analysis shows that the combination of LSTM network and Kano model can more accurately identify the innovation and improvement direction of smart phone products.This paper uses the method of combining the LSTM network and the Kano model to construct a product innovation feature identification method for online reviews.The application of oppo,Huawei and other smart phone products has verified the effectiveness of the method,which can be used as a reference for enterprise product innovation and improvement.At the same time,this research also addresses the shortcomings of the emotional dictionary semantic analysis method that does not consider the context and the shortcomings of the LSTM network model to measure the importance of text data features.By introducing the attention mechanism algorithm into the LSTM network,different attention weights are given to improve.It is true that this research still needs improvement in terms of identifying false comments,optimizing sentiment classification algorithms,and setting thresholds in the Kano model.