Research And Implementation Of News Recommendation System Based On Text Features

In response to the needs of managing massive amounts of Internet news data and building an efficient information retrieval platform,this paper designs and implements a recommendation algorithm of this platform focuses on the entity tag and content feature of news and the user's interest transfer feature,proposes an interest transfer model based on user tag weight,and improves the extraction model of news entity tag.In the stage of news text feature modeling,this paper constructs an ERNIE-BiLSTM-CRF model through model fusion to improve the accuracy of entity recognition.And for the issue of multiple ambiguities in military and current affairs news entities,the ERNIE-RE relationship extraction model is used to extract alias relationships,build an entity database,and perform entity disambiguation.Based on all the extracted entities,the entity space is constructed,and the entity tag vector of the news is calculated through the TF-IDF algorithm.For the content vector of news,it is obtained through the prediction of the doc2vec model.In the stage of user feature modeling,this paper refers to the ItemCF recommendation algorithm and the recommendation algorithm based on tag weights,and proposes a user vector extraction method based on tag weights for interest transfer.then uses the LSTM neural network model to finally Obtain a user interest transfer model with tag weights.And generate passing results by mixing hot news,news based on user vectors,and news based on tagsFinally,this article evaluates the accuracy of the named entity recognition and entity relationship extraction process through manually annotated military and current affairs news data sets;verifies the accuracy of the recommendation effect through multiple rounds of recommendation effect evaluation tests for system users;The access time and concurrent performance measurement of the system verify the usability of the recommended system.