Research And System Implementation Of Personalized Public Opinion Recommendation In Colleges And Universities

With the explosive growth of Internet information resources,people are exposed to more and more information.While enjoying the convenience brought by rich Internet resources,people are inevitably affected by various rumors and negative information.Especially in colleges and universities,because college students are not mature enough,they are more likely to be affected by bad information and make irrational behaviors.In order to solve the problem of public opinion in colleges and universities,and allow managers to respond quickly to various events,a system of public opinion in colleges and universities came into being.However,in the public opinion system of colleges and universities,users need to face massive amounts of information,and a serious information overload problem occurs.In order to solve this problem,on the basis of researching commercial public opinion system and personalized recommendation algorithm,this paper has completed multiple work to assist users in information retrieval,and developed a university public opinion system.First,this paper analyzes the demand of the university public opinion system and makes a detailed design.In addition to the basic functions of the public opinion system,the system functions also include sentiment analysis and personalized search keyword recommendation.The system is mainly developed in Java,the data processing and algorithms are implemented in Python,and the search function is implemented in Elasticsearch.In the system,the related functions realized by Python are encapsulated into functional scripts,and the system implements the corresponding functions by running scripts through scheduled tasks.The two interact through the database,which effectively reduces system coupling.Then,this paper implements a variety of sentiment analysis models.This paper built four neural network models,fastText,BiLSTM,BiGRU and CNN,and uses the word vector generated by word2 vec and the word vector of the BERT pre-training model as embedded features.After conducting experiments on public data sets,this paper selected the best combination of Bi LSTM and word2 vec word vectors to integrate into the university public opinion system to help users filter public opinion information and alleviate the problem of information overload.Then,this paper proposes two personalized search keyword recommendation methods,recommending search keywords to users,helping users to search,alleviating the problem of information overload,and improving the usability of public opinion system.The first method is to calculate the similarity of keywords through an item-based collaborative filtering algorithm,and recommend the keywords that are most similar to the user's latest search keywords to the user.The second method is a personalized keyword recommendation model combining word vector clustering and popularity ranking.First,cluster according to word vectors,and divide all keywords into 100 word classes.Then use the user's historical behavior data to calculate the user's rating for some word classes,and predict the user's rating for all word classes through the SVD algorithm to get the 10 word classes that are most interesting to each user.Finally,select the most popular keywords in the most interesting word classes of the user to recommend to the user.In addition,for the situation after the user searches,this paper provides word vectorbased synonyms recommendation as a supplement to personalized keywords recommendation.While expanding the user's search range,real-time keywords recommendation is also achieved.Finally,this paper tests the function and performance of the system,and the test results prove that the system availability meets the demand.