Opinion Mining Based On Implicit Feature Extraction

With the development of Internet technology,the network inevitably becomes a huge collection of information,storing a large amount of valuable information,which contains a large amount of unstructured information.Transforming unstructured information into structured information is a prerequisite for making full use of information resources.Opinion mining is one of them.However,most of the current opinion mining is based on the analysis of explicit features.The analysis of hidden features is not very mature.Therefore,this paper completes opinion mining based on the implicit feature extraction.The task is to obtain tea review data from the e-commerce network platform and explore the valuable information hidden in it.In this paper,firstly,the unsupervised layer-by-layer pre-training and supervised fine-tuning process are carried out on the stack noise reduction self-encoder to construct the dominant feature extraction model,and the product feature set is constructed.Secondly,by establishing the CBOW model of the comment corpus,the implicit feature extraction model is constructed,and the comments that do not contain the dominant features are fully utilized.Thirdly,on the basis of using the existing dictionary,the existing sentiment dictionary is expanded by the double propagation method;then,the PMI is used as the quantitative index to complete the quantification of each feature-emotional word association pair,and the overall evaluation of the product features are completed;finally,the calculated scores are analyzed to show the actual value of product review mining.