Phd. Thesis : Temporal Recommendation

  • Published on
    02-Nov-2014

  • View
    7.799

  • Download
    0

Embed Size (px)

DESCRIPTION

 

Transcript

  • 1. Author xlvector, Copyrights belong to CASIA (Recommender System) Netix Top-N 1 2 Top-NTop-N N
  • 2. Author xlvector, Copyrights belong to CASIAii Top-N 3 4
  • 3. Author xlvector, Copyrights belong to CASIA Abstract Recommender System is an important tool for users to discovery informationof their interest, it is also an important tool to overcome information overloadproblem. The main idea of recommender system is to make recommendation byanalyzing users historical behaviors. Early researches on recommender systemalways neglect temporal information, and most of them are focused on the anal-ysis of users static behaviors. In recent years, because of Netix Prize, more andmore data sets including temporal information are released, and more and moreresearchers are studying on temporal recommendation problem. However, thereare many problems left in this research area. This paper investigate the temporal recommendation problem by analyzingmany public released data sets. Following are main contributions of this paper: 1 Temporal recommendation for rating prediction problem: Rating predic- tion problem is the most famous problem in recommender system, its main task is to predict a given users rating on a given item by analyzing her his- torical rating on other items. In this paper, we incorporate temporal infor- mation into this problem, and propose a latent factor model to model four dierent temporal eects. Furthermore, we also proposed a cascade model to model seasonal eects. Experimental results show that our method can achieve higher accuracy in rating prediction problem than non-temporal methods. 2 Temporal recommendation for top-N recommendation problem: Top-N rec- ommendation problem is the most important problem in real recommender system, its main task is to recommend N items to every user which will be of their interests by analyzing users historical behaviors. In this paper, we introduce a new type of node, session node, into user-item bipartite graph to model users long term and short term interests. Furthermore, we also proposed a new graph-based personal ranking method called PathFusion to
  • 4. Author xlvector, Copyrights belong to CASIAiv make recommendation by the new graph model. Experimental results show that our method can make higher accuracy than non-temporal methods and other temporal recommendation methods in top-N recommendation prob- lem. 3 Inuence of system update rate on recommender system: User behavior is inuenced by social factor and personal factor. However, in the websites with dierent update rates, these two factors will have dierent inuence. In fast updating sites, users are more inuenced by social factor while in slow updating sites, users are more inuenced by personal factor. In this way, we need dierent recommender systems in web sites with dierent update rates. We proposed a session graph model which introduce two new types of nodes into user-item bipartite graph to model social factor and personal factor. By controlling the weight of these two new types of nodes, the recommendation algorithm can control the inuence of social factor and personal factor on nal recommendation results. Experimental results show that our method can achieve high accuracy in systems with dierent update rate. 4 Prototype of temporal recommender system: We design a prototype of temporal recommender system. This system can return real-time recom- mendation to users after they have new behavior, and can tune the ranking of results by user feedback. In this way, this system can improve user experience in real time by their feedback.Keywords: recommender systems, personalization, collaborative tering, tem-poral eects, seasonal eects, temporal dynamics in recommender systems
  • 5. Author xlvector, Copyrights belong to CASIA iAbstract iii v 1 1.1 1 1.2 6 1.2.1 6 1.3 7 1.4 7 1.4.1 Netix 8 1.4.2 MovieLens 8 1.4.3 CiteULike 8 1.4.4 Delicious 9 1.5 9 11 2.1 11 2.2 12 2.3 14 2.3.1 15 2.3.2 17 2.3.3 18 2.3.4 20
  • 6. Author xlvector, Copyrights belong to CASIAvi 2.3.5 22 2.4 23 2.5 25 2.5.1 26 2.5.2 27 2.5.3 27 2.5.4 28 2.6 28 31 3.1 31 3.2 32 3.3 33 3.4 36 3.4.1 36 3.4.2 37 3.4.3 38 3.4.4 39 3.4.5 40 3.5 42 3.5.1 43 3.5.2 43 3.5.3 44 3.6 44 3.6.1 44 3.6.2 45 3.6.3 TRSVD 45 3.6.4 47 3.7 48
  • 7. Author xlvector, Copyrights belong to CASIA vii Top-N 49 4.1 49 4.2 50 4.3 51 4.3.1 52 4.3.2 53 4.4 56 4.4.1 56 4.4.2 SGM 58 4.5 60 4.5.1 60 4.5.2 61 4.5.3 61 4.5.4 62 4.5.5 64 4.6 66 69 5.1 69 5.2 70 5.2.1 70 5.2.2 73 5.3 74 5.3.1 75 5.3.2 76 5.3.3 76 5.4 77 5.4.1 77 5.4.2 78
  • 8. Author xlvector, Copyrights belong to CASIAviii 5.4.3 79 5.4.4 80 5.5 83 87 6.1 87 6.2 88 6.2.1 88 6.2.2 89 6.2.3 90 6.2.4 90 6.3 91 6.3.1 92 6.3.2 92 6.3.3 92 6.4 93 95 7.1 95 7.2 96 99 111 113
  • 9. Author xlvector, Copyrights belong to CASIA 1.1 5 1.2 6 3.1 TRSVDRSVDProbeRMSE 46 3.2 TRSVDRSVDQuizRMSE 46 3.3 TRSVDRSVD 47 3.4 Netix Proble 48 4.1 CiteULikeDeliciousSGM 64 4.2 CiteULike 67 4.3 Delicious 67 5.1 5(1), (30) 75 5.2 5 78 5.3 78 5.4 85 80 5.5 Pop5UserCF/ItemCF 81 5.6 5SGMUSGMISGM 82
  • 10. Author xlvector, Copyrights belong to CASIA
  • 11. Author xlvector, Copyrights belong to CASIA 2.1 13 2.2 14 2.3 Amazon 16 2.4 IMDB 18 2.5 Jinni 20 2.6 AmazonFacebook Connector 21 2.7 Facebook 22 2.8 MappyFriends 23 2.9 5 24 2.10 GetGlue 27 3.1 32 3.2 34 3.3 Netix 37 3.4 Netixut 39 r 3.5 Netix 40 3.6 Google Trend2004 42 4.1 - 52 4.2 54 4.3 58 4.4 SGMA2 60 4.5 CiteULikeHitRatio 63 4.6 DeliciousHitRatio 63 4.7 CiteULikeHitRatio 65
  • 12. Author xlvector, Copyrights belong to CASIAxii 4.8 DeliciousHitRatio 66 5.1 71 5.2 72 5.3 5( )(Y) (X) 74 5.4 Wikipedia/ 83 5.5 Sourceforge/ 84 5.6 NYTimes/ 84 5.7 Blogspot/ 85 5.8 Youtube/ 85 6.1 87 6.2 88...