Research On Generative Conversational Recommendation System Based On Knowledge Aggregation

In today’s life,with the rapid development of the Internet,people’s information needs are growing day by day.Traditional search engines gradually expose their limitations in meeting users’ personalized needs.Therefore,providing users with more accurate recommendations has increasingly attracted the attention and research of both industry and academia.As a new type of recommendation method,conversation-based recommendation systems have also received considerable attention.Conversationbased recommendation systems aim to mine users’ interests,needs,and preferences from their conversation history and generate smooth and user-oriented reply conversations.This thesis mainly studies generative conversational recommendation systems based on knowledge aggregation and discusses their principles,model design,and system design.The specific content is as follows:(1)The focus is on designing a model for building a conversation-based recommendation system.In this thesis,the model for building a conversation-based recommendation system is divided into two parts,each responsible for different tasks.In the dialogue model,the Transformer model is used to generate smooth conversations,and axial position encoding is used to optimize the position encoding method.In the recommendation model,an improved relational graph convolutional neural network model is used in conjunction with a knowledge graph and attention mechanism to generate recommended items for users.On the Redial dataset,the Recall metric is used to evaluate and compare the new model and the baseline model,finding that the new model with four layers performs better than the baseline model in terms of Recall.In the strategy mechanism,a conversion mechanism based on Pointer Softmax is used to incorporate the generated recommended items into the generated conversation,providing users with more personalized and accurate recommendations while maintaining smooth and understandable sentences.(2)A conversation-based recommendation system is designed based on the proposed new model,and explanations are made from three aspects: system requirements analysis,architectural design,and workflow.The feasibility of the conversation-based recommendation system designed based on the proposed model is further demonstrated through example testing.(3)The dialogue model based on Transformer and improved relational graph convolutional neural network is verified through experiments,and the effectiveness of the method is compared and analyzed.On the TG-Re Dial,Du Rec Dial,Open Dial KG,and INSPIRED datasets,KGSF,REDIAL,GRU4 rec,Text CNN,and SASRec are chosen as baseline models for the recommendation model,and REDIAL,KGSF,and TG-Re Dial are chosen as baseline models for the dialogue model.The models are evaluated using HIT,NDCG,BLEU,and DIST metrics.Experiments show that the proposed model performs better than other baseline models in both dialogue and recommendation models,overall achieving the expected research objectives.