Elasticsearch-based Research On Public Opinion Retrieval And Analysis Of Domestic Outbound Tour

Since the beginning of the new crown epidemic,China's outbound travel industry has suffered a cold winter,and countries have carried out prevention and control of the epidemic by restricting entry and exit,etc.As the epidemic gradually stabilizes,outbound travel still has prospects for development in the future,so it is important to pay attention to outbound travel public opinion information for travel industry personnel to make business decisions or relevant departments to adjust travel policies in a timely manner.The popularity of mobile Internet makes it easier for public opinion to break out and spread,so it is especially important to locate public opinion information quickly.This paper conducts experimental and applied research on the theoretical techniques related to public opinion retrieval and analysis.The main research contents :(1)Designing and implementing an opinion retrieval system based on Elasticsearch framework with the news data set of domestic outbound travel field,which can manage and apply the opinion data;(2)Proposing a re-scoring embedding scheme based on ranking learning to improve the relevance of the retrieval results in response to the problem of insufficient refinement of the returned results in the opinion retrieval system;(3)To address the problem of high cost and time consuming to annotate the outbound travel news corpus with sentiment,propose a sentiment classification model based on pre-trained language models,and use Mixture-Of-Experts to implement a model fusion scheme to achieve automatic classification of text sentiment;(4)To design and implement an analysis module for opinion news data with emotion labels combined with the main steps of opinion analysis to satisfy further mining and display of opinion data.Specific studies of the above are as follows:(1)Collection and pre-processing of opinion news data,using a web crawler framework to collect travel news of popular outbound places and adopting regular expressions and removing deactivated words for data pre-processing to ensure the feasibility of subsequent experiments.(2)Reordering of opinion retrieval,by comparing and analyzing three machine learning-based ranking learning models,embedding custom query statements in Elasticsearch to obtain features and feature value sets,and inserting the trained model into the scoring mechanism again to reorder the retrieved results.(3)Sentiment classification of opinion data,using three pre-trained language models to predict the sentiment categories of the collected news data and determining the labeling results with a post-fusion scheme based on Mixture-Of-Experts models.(4)Analysis of public opinion data,using the aggregation module in the search engine to do descriptive statistical analysis on the sentiment category fields of the data,visualize them in the form of graphs and tables,and provide users with public opinion suggestions.To sum up,outbound travel public opinion data contains a large amount of potential knowledge,and analyzing public opinion data has important practical significance and value in the application of online public opinion.Using the public opinion retrieval and analysis module,it can provide public opinion-related subjects with the collection and mining of public opinion information,judge the development trend of public opinion in the field of outbound tourism,and help to better understand and use public opinion data.