Research On Recommendation Models Based On User’s Instant And Inherent Interests

Information overload has become one of the main problems in the information age.As one of the effective solutions to this problem,recommender system has been in depth research and extensive development.This thesis is dedicated to the research of recommendation models in recommender systems.Based on our understanding of realworld recommendation scenarios and data,we have proposed a recommendation model based on user’s instant and inherent interests.Comparing with the state-of-the-art recommendation models,our proposed model is more in line with the thinking logic of recommendation scenarios in the real world.The main work of this thesis includes the following points:1.We proposed a novel recommendation model based on user’s instant and inherent interests named Ⅲ-DNN.The Ⅲ means Instant and Inherent Interests.This model extracts user’s instant interest and inherent interest respectively and uses deep neural network for feature fusion.It can alleviate the dependence of the recommendation model on the user’s single interest and adaptively judge the user’s interest in the target item from the perspective of multiple interests.2.We proposed an improved model with ancillary neural networks named ⅢDNN-Anc based on Ⅲ-DNN.Though the analysis of the testing loss,we found there existed the problem of overfitting.Then we proposed the ancillary neural networks which is a way to alleviate overfitting without redundant computation,and the generalization ability of our model is improved through this method.3.We designed experiments to compare and analyze the effectiveness of these two models we proposed in this thesis.We conducted experiments on MovieLens,Amazon and XiaMenRenCai datasets to compare the recommending efficiency among our proposed model and other state-of-the-art models.We have also verified the effectiveness of each module of our proposed model and visualized the results.We have compared and analyzed the recommendation effectiveness of multiple models before and after integrating the ancillary neural networks.And the results have verified the effectiveness of the ancillary neural networks.4.We have designed and implemented a job recommender system.We designed the system and the database based on the XiaMenRenCai dataset.The job recommender system is built using Django,a Python-based web framework.The recommendation module is formed by our proposed model Ⅲ-DNN-Anc.We also developed other functions which contains job searching,resume editing,job applying,data visualization and so on.