Research And Implementation Of Novelty Recommendation System Based On BER

In recent years,with the rapid development of artificial intelligence,big data and other high technologies in the Internet era,the amount of information data in the Internet has grown exponentially,and it has become very difficult for people to find target information from the huge amount of information data.Service providers provide people with high-quality recommendation services,which can effectively reduce the time cost of obtaining target information as well as improve people’s viscosity to service providers and increase the revenue of the platform.In high quality recommendation service,the recommendation model will data mine the user’s behavioral data and extract the user’s interest preferences,so as to provide personalized recommendation service.Currently,the research of recommendation models has made some progress,but there is still room for improvement in the performance of models and satisfaction of recommendation results.This thesis is based on deep learning technology to study users’ behavioral patterns,to explore users’ long-and short-term interest preferences,to design models and methods in multi-stage recommendation schemes,and to develop recommendation systems to provide users with high-quality recommendation services,as follows.(1)A study of BERT-based novelty recommendation.To address the problem that existing recommendation models and methods ignore the user interest preferences are dynamic and change,as well as the problem that homogeneous movies will be provided within a certain time,the thesis designs several models and methods in a multi-stage recommendation scheme.Firstly,a recall model based on user interest modeling is proposed in the recall layer,which makes full use of the BERT network to mine the feature information in user behavior sequences and narrow down the search scope in the ranking layer.Secondly,a recommendation model integrating time perception and interest preference is proposed in the ranking layer,which designs a time function to obtain temporal location information and captures the fine-grained relationships in the sequence using temporal convolutional networks and BERT networks,so as to obtain a high-quality Top-N recommendation list.Finally,a novelty recommendation method based on user interest clustering is proposed in the reordering layer,using Mean-Shift clustering algorithm to obtain user interest clusters,designing novel activation function to add nonlinear features,realizing Top-N recommendation list reordering,and completing novelty recommendation.Experiments on publicly available datasets show significant improvements in all evaluation metrics compared to existing models and methods.(2)Design and implement a novelty recommendation system based on BERT.Based on the models and methods proposed in the multi-stage recommendation scheme,the requirements analysis and design of the recommendation system are carried out from the perspective of software development.The functional and nonfunctional requirements of the recommendation system are clarified,and the overall architecture of the system is scientifically planned.Based on the B/S architecture,the Django back-end framework and the HTML + Vue.js front-end framework are used to implement the novelty recommendation system based on BERT.Through system testing,the system has good compatibility and response speed,and correctly provides high-quality,novel recommendation services.