Collaborative Filtering Alejandro Bellogn 1 , Jun Wang 2 , and Pablo - PowerPoint PPT Presentation

Text Retrieval Methods for Item Ranking in Collaborative Filtering Alejandro Bellogín 1 , Jun Wang 2 , and Pablo Castells 1 1 Escuela Politécnica Superior, Universidad Autónoma de Madrid 2 Department of Computer Science, University College London @abellogin alejandro.bellogin@uam.es European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Collaborative Filtering – in a glimpse  Input: a rating matrix European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Collaborative Filtering – in a glimpse  The users European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Collaborative Filtering – in a glimpse  The items European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Collaborative Filtering – in a glimpse  A rating: from a user towards an item European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Collaborative Filtering – in a glimpse  User profile European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Collaborative Filtering – in a glimpse  Item profile European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Collaborative Filtering – in a glimpse  Unknown rating European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Collaborative Filtering – in a glimpse  Goal If user Boris watched Love Actually , how would he rate it? European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Collaborative Filtering – in a glimpse  Prediction: how Boris rated similar items European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Text Retrieval Query Process Query Text Term Retrieval Output occurrences Inverted Engine (term-doc Index matrix) European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Collaborative Filtering? User Profile Process User profile (as query) Text User Item Retrieval Output Profiles Inverted Similarity Engine (User-item Index matrix) European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

In this work  A first attempt • Item ranking • Item-based CF  Good results  Improvements – current work • More models • Rating prediction  Now… algorithmic details European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Text Retrieval  In (Metzler & Zaragoza, 2009)       s q d , s q d t , ,    t g q • In particular: factored form        s q d t , , w q t w , d t , 1 2 European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Text Retrieval  Examples      w q t , qf t • TF: 1      w d t , tf t d , 2     • TF-IDF:  , qf w q t t 1   N        w d t , tf t d , log df     2 t        • BM25: k 1 qf t    3 w q t ,    1 k qf t 3            N df t 0.5 k 1 tf t d ,      1 w d t , log                  2 df 0.5  t  k 1 b b dl d / dl tf t d , 1 European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Collaborative Filtering  Standard item-based formulation (Adomavicius & Tuzhilin 2005)      sim , i j     rat , rat , u i u j   sim , i j  j I u  j I u European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Collaborative Filtering  Standard item-based formulation (Adomavicius & Tuzhilin 2005)      sim , i j     rat u i , rat u j ,   sim , i j  j I u  j I u European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Collaborative Filtering  Standard item-based formulation      sim , i j     rat u i , rat u j ,   sim , i j  j I u  j I u • More general             rat u i , f u i j , , f u j f , i j , 1 2       j g u j g u European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Text Retrieval for Collaborative Filtering  In item-based Collaborative Filtering  t j      tf t d , sim , i j  d i      qf rat , t u j  q u  Apply different models • With different normalizations and norms: s qd , L 1 and L 2 Document s qd No norm Norm ( /|D|) s 00 s 01 No norm Query s 10 s 11 Norm ( /|Q|) European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Text Retrieval for Collaborative Filtering  TF L1 s01 equivalent to item-based CF   tf t d ,             s q d , w q t w , d t , qf t    1 2 tf t d ,       t g q t g q    t g q     sim , i j     rat u i , rat u j ,    sim , i j  j I u  j I u      tf t d , sim , i j      qf t rat u j , European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Results  Movielens 1M • Movielens100k: equivalent results 0.40 nDCG 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00 TF L1 TF-IDF TF-IDF BM25 TF L1 BM25 TF-IDF BM25 TF-IDF TF-IDF TF L2 BM25 TF L1 BM25 BM25 BM25 TF L2 TF-IDF TF L2 s01 L1 s01 L2 s11 L2 s11 s10 L1 s01 L1 s10 L1 s00 L2 s10 L1 s00 s10 L2 s10 s00 L1 s11 L1 s10 L2 s01 s11 L2 s01 s01  TF L1 s01 equivalent to item-based CF (baseline) European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

Conclusions  It is possible to use Text Retrieval methods in rating-based Collaborative Filtering  Our methods outperform classic Collaborative Filtering methods  …Questions? European Conference on Information Retrieval 2011 April 20, Dublin, Ireland

References  Adomavicius, G., Tuzhilin, A.: Toward the next generation of recommender systems: a survey of the state-of-the-art and possible extensions. IEEE TKDE 17(6), 734-749 (2005)  Metlzer, D., Zaragoza, H.: Semi-parametric and non-parametric term weighting for information retrieval. LNCS, vol. 5766, pp. 42-53. Springer (2009) European Conference on Information Retrieval 2011 April 20, Dublin, Ireland