Multi-dimension Label Classification And Extraction Based Restaurant Review Visualization System

The explosive growth of online shopping has led to a large number of e-commerce reviews.On the one hand,analyzing the commodity reviews can help consumers make decisions on shopping,on the other hand,the mining of shop reviews can make the merchants accurately locate the problems in business operation.The task of multidimensional label classification and extraction focuses on the value of the comments.Two tasks are: task one,emotional analysis and key information extraction algorithm for catering comments.Facing the catering scene review data set in AI Challenger 2018 global AI challenge,on the one hand,emotional classification is carried out for 20 evaluation dimensions,such as location,service and price;On the other hand,on the premise of knowing the evaluation dimension and emotional polarity,the users' evaluation words,phrases and sentences are located in the corresponding evaluation dimension,and the diversity of user comment text is explored in different evaluation dimensions.Task 2: review the research of visual product design.Mining the pain points of users when using the visual function of comments of comment products,sorting out and verifying requirements,integrating the algorithm results in task 1,and completing product design,and evaluating product availability.This research belongs to the field of natural language processing and product design research under the artificial intelligence discipline,and it is a social computing research.Methodologically,after a solid investigation of papers,inspired by domain dependence of sentiment analysis,this study innovatively uses millions of “dianping.com”food review data to unsupervised post training on the pre-trained model,which is named as “Ro BERTa-wwm-ext-large-PT” model,and verifies its effectiveness on multidimensional label sentiment classification and label extraction task.In the research of product design,the study conducts user interview to see user needs Also,the study implements heuristic evaluation,and analyzes competitive products,designs the product prototype,optimizes interaction and vision.Based on the results of task 1,the technical implementation of the product is carried out,also carry out the heuristic evaluation and user experience of the product is measured to verify the satisfaction of the requirements.As a result,in the multi-dimensional label classification and extraction task for restaurant reviews,the method proposed in this study achieves the state-of-the-art hrough multiple groups of experiments,and demonstrates the effectiveness of domain enhanced learning by pre-trained language model under the restaurant corpus,in which the classification task improves by 4.84%(F1 value),and the multi-extraction task increased by 7.18%(F1 value).In the research of product design,after need analysis and verification,analysis of competitive product,product prototype design,interaction and visual optimization,web visualization technology is used to complete the product launch.After heuristic evaluation of online products,the second optimization of products is carried out,and the positive gain of user experience before and after product improvement is proved by statistical means.This research comes into conclusion,in the catering corpus,domain enhanced learning of the pre-trained language model can effectively improve the effect of the model.The review product proposed in this study has a positive gain from the perspective of user experience.