Research And Application Of Radiator Products' Review Mining Method Based On Deep Learning

With the development of the Internet and rise of electric business enterprises,shopping online is no longer limited by time or space.As of December 2020,Chinese online shopping users reached 782 million.The rise of online shopping leads to an explosion of online reviews.Reviews contain a large amount of feedback information.The in-depth mining of reviews is not only helpful for the repositioning of existing products,the research and development of new products,and the formulation of marketing strategies,but also can help enterprises to judge customers'emotion towards the products,grasp consumers'concerns in real time,respond to the market demand in time,and learn of the consumption blueprint in customers'minds.At present,most researches focus on sentiment analysis of sentence-level review texts,while attribute-level sentiment analysis is less,and it can only judge the emotional tendency of product attributes.In order to dig deeper into the customer's specific views on product attributes,build a deep learning-based sentiment analysis model and a dictionary-based online review deep mining model,and integrate the two models into a review analysis APP prototype.Firstly,based on the research results of sentence-level text sentiment analysis,construct bidirectional short-termmemory network model based on the attention mechanism,optimize its activation function,correct its loss function,enhance the model's learning ability,and alleviate overfitting in the training process,improving the classification performance of the model.Set up a control group to evaluate the model fromthe three indicators of recall,accuracy and₁.Secondly,on the basis of sentiment classification,use a series of methods such as grammar path template,principal component analysis,K-means clustering algorithmand semantic mutual information algorithmto construct attribute dictionary and viewpoint word set.Build a deep mining model of online reviews to mine the co-occurrence relationship between product attributes and customer views.Finally,design an APP prototype of comment analysis whose functional modules include:preprocessing module,sentiment analysis module,attribute-opinion co-occurrence analysis module.The sentiment analysis module is integrated with the comment sentiment analysis model,and the attribute-opinion co-occurrence analysis module is integrated with the comment deep mining model.Taking the online reviews of a radiator product K of company C as an example,import the reviews into the review analysis APP to mine the reviews,to verify effectiveness of the two models.The model and method proposed and the APP designed in this paper have high feasibility and practicability,replacing inefficient,incomplete,and repetitive manual analysis with one-click automatic comment mining,helping companies improve the efficiency of comment analysis.At the same time,the study has good portability and can be extended to other product fields,promoting the development of text analysis of Chinese comments.