Research And Application Of Text Sentiment Analysis For Catering Review

Due to the rapid development of the Internet,people's life has gradually become intelligent,and their eating habits have changed from offline to online.After daily consumption,users are used to leave comments on the online platform to express the feeling of this consumption.With the accumulation of time,a huge data set of consumer reviews has been formed.Using big data means to analyze and mine these emotional data reasonably can not only effectively and quickly understand the advantages and disadvantages of the business,but also grasp the user's preferences and consumption needs,and improve the product.Traditional emotion analysis is mainly based on machine learning model and rule matching method.The rule-based approach is to construct an emotion dictionary for each kind of emotion in a certain field.The effect of the final prediction largely depends on whether the emotion words contained in each kind of emotion are perfect and accurate.It is difficult to build a general emotional dictionary for different fields.On the other hand,based on the traditional machine learning method,we need to extract the shallow syntax and semantic features of text,such as part of speech information and entity information,which can not extract the semantic information of text context,so the effect of the model is general.Based on this,this thesis mainly uses deep learning algorithm to represent the text as word vector,considering the context information,semantic,grammar and other information of the text,and uses deep learning algorithm to analyze emotion.The main contents of this thesis are as follows:(1)This thesis proposes an emotion classification model based on seq2 seq.Firstly,a language model of Elmo,which can be used in food and beverage comments,is pre trained on the input.It can generate word vectors including context semantics and grammar,effectively represent the text comments,and improve the accuracy of the model.The seq2 seq model mainly uses that the input of this thesis is a sequence,and the output is a sequence.There is a complex relationship between the 20 different granularity emotional granularity of the output.Through the seq2 seq model,we can learn this relationship.This thesis improves the attention mechanism,the characteristics of sharing parameters,allocates different weights in different directions,and uses Ge decoding for sequence generation module It can capture the correlation between labelsand predict that different labels can focus on different parts of the input.Finally,the accuracy value of the improved seq2 seq model in the meituan review data set is 89.32%,and the average F1 value is 0.7190.Through the comparative study with other baseline models,it is found that the model proposed in this thesis is the best in the evaluation index.(2)In this thesis,we propose a second fine-grained emotion classification deep learning model based on self-attention mechanism,which uses two models based on self attention mechanism to encode text,gradually obtains 20 specific aspects of information,and finally outputs 20 aspects of emotion labels.The model has achieved good results in the data set of meituan reviews,with an average accuracy of 88.64% and an average F1 of 0.7079.(3)In order to speed up the application of fine-grained sentiment analysis method in the actual production environment,this thesis develops a text application of fine-grained sentiment analysis of food and beverage comments,which integrates data crawling,data preprocessing,model calculation,statistical information display and other modules,and provides the function of fine-grained sentiment analysis of food and beverage comment data,which verifies the usability and effectiveness of the model proposed in this thesis Effectiveness.