王文杰-LLM4Rec.pdf

1、Large Language Models for Recommendation:Progresses and Future DirectionSpeaker:Wenjie WangNational University of SingaporeWWW24 Tutorial Organizers:Jizhi Zhang,Keqin Bao,Yang Zhang,Wenjie Wang,Fuli Feng,Xiangnan He1OutlineIntroduction Progress of LLM4Rec LLM4Rec Trustworthy LLM4RecOpen Problems Fut

2、ure Direction&Conclusions2Background of RecSys3RecommendationsInteractionsUser feedbackRecommenderTrainingInferenceSystem sideUser sideItemdatabaseUserq Workflow of Recommender System(1)Train recommender on collected interaction data to capture user preferences.(2)Recommender genrates recommendation

3、s based on estimated preferences.(3)User engage with the recommended tiems,forming new data,affected by open world.(4)Train recommender with new data again,either refining user interests or capturing new ones.Open worldBackground of RecSys4q Core idea of personalized recommendationCollaborative filt

4、ering(CF):Making automatic predictions(filtering)about the interests of a user by collecting preferences from many users(collaborating).Memory-based CFUser CFItem CFModel-based CFMFFISM Neural CFGCN-based CFXiangnan He et al.LightGCN:Simplifying and Powering Graph Convolution Network for Recommendat

5、ionBackground of RecSys5q Core idea of personalized recommendationCollaborative filtering(CF):collaborative informationContent/context-aware models(CTR models):side information+context informationClick-Through Rate(CTR)predictionImages from:Deep Interest Network for Click-Through Rate PredictionFact

6、orization machines:FM,NFM,DeepFMNeural network:DIN,AutoIntBenefit of LMs6q How can recommender systems benefit from LMs Learning paradigm:Pretrain-finetune,Prompt learning Model architecture:Transformer、Self-attention Representation:Textual feature,item representation,knowledge representation Task f