O2O Store Food Safety Public Opinion Analysis System

In recent years,with the rapid development of the Online To Offline(O2O)catering industry and the continuous expansion of the takeaway market,the issue of online food safety cannot be ignored.With the rapid development of artificial intelligence and big data processing technologies,using big data to promote online food safety governance has become an effective means for supervisors.The use of artificial intelligence technologies to assist online food safety governance can improve the efficiency of the platform and regulatory authorities,and achieve early warning of food safety issues.Among these technologies,the public opinion analysis method is the key means.However,the current food safety public opinion analysis technology focuses on social media public opinion analysis,which cannot meet the current O2 O store food safety supervision requirements.Therefore,this thesis conducts related research on the food safety public opinion analysis technology of O2 O catering platforms.The main content as follows:1.In order to mine the rich information in user reviews of O2 O stores and improve the accuracy of prediction,data processing and feature engineering are carried out.In this thesis,the active learning algorithm is used to label the original comment data,which reduces the cost of sample labeling and improves the accuracy of the classification.In the word vector training process,a combination of pre-training word vectors and word vectors trained with the classic algorithm is adopted to increase the amount of information of the word vector.2.In order to improve the effect of model prediction and increase the richness of analysis results,three different solutions are proposed for experiments.In the machine learning method,various models are trained and compared.In the deep learning method,the Attention-based LSTM(AT-LSTM)model is used to increase the model interpretability,giving users more fine-grained results.In the pre-trained model method,a variety of pre-trained models are tested for fine-tuning,and a variety of model construction methods are compared to find the most suitable model construction plan in the current scenario.The pre-trained model absorbs the language knowledge in the large-scale corpus and further improves the accuracy of prediction.3.Based on the above-mentioned feature engineering and model construction schemes that are most suitable for the current task,an O2 O store food safety public opinion analysis system was designed and implemented to help supervisors realize efficient and automatic food safety public opinion analysis.The data used in this thesis comes from user reviews of shops in Meituan and Dianping,and the evaluation index is the F1 score of the prediction by food safety hazards.This thesis explores three solutions,compares various model construction methods through experiments,and finally increases the F1 score to 0.9412.The core algorithm of this thesis ranked third among the 3048 teams in the F1 score ranking of the O2 O store food safety-related reviews discovery competition hosted by China Computer Federation in 2019.Based on the above algorithm,this thesis implements an O2 O store food safety public opinion analysis system,and provides efficient and accurate public opinion monitoring and analysis services for O2 O platform supervisors and health supervision departments.