Availableonlineatwww.sciencedirectcom ° Science Direct Expert Sy stems with Applications ELSEVIER Expert Systems with Applications 35(2008)1386-1399 www.elsevier.com/locate/eswa Fuzzy-genetic approach to recommender systems based on a novel hybrid user model Mohammad Yahya H. Al-Shamri, Kamal K. Bharadwaj School of Computer and Systems Sciences, Jawaharlal Nehru Unicersity, New Delhi 110 067, India Abstract The main strengths of collaborative filtering(CF), the most successful and widely used filtering technique for recommender systems e its cross-genre or outside the box'recommendation ability and that it is completely independent of any machine-readable represen tation of the items being recommended. However, CF suffers from sparsity, scalability, and loss of neighbor transitivity. CF techniques are either memory-based or model-based. While the former is more accurate, its scalability compared to model-based is poor. An impor tant contribution of this paper is a hybrid fuzzy-genetic approach to recommender systems that retains the accuracy of memory-based CF and the scalability of model-based CF. Using hybrid features, a novel user model is built that helped in achieving significant reduc- tion in system complexity, sparsity, and made the neighbor transitivity relationship hold. The user model is employed to find a set of like linded users within which a memory-based search is carried out. This set is much smaller than the entire set, thus improving systems scalability. Besides our proposed approaches are scalable and compact in size, computational results reveal that they outperform the classical approach. e 2007 Elsevier Ltd. All rights reserved Keywords: Recommender systems; Collaborative filtering: Web personalization; User model; Fuzzy sets 1. Introduction down the information overload, and efficiently guide the user in a personalized manner to interesting items within The popular use of web as a global information system a very large space of possible options(Burke, 2002). Typi- has flooded us with a tremendous amount of data and cally rs recommend information (URLS, netnews articles ), information. The end users are overloaded with a huge entertainment(books, movies, restaurants), or individuals amount of information from myriad resources. This explo- (experts). Amazon. com (Linden, Smith, York, 2003) sive growth of data has generated an urgent need for pow- and Movie Lens. org(Miller, Albert, Lam, Konstan, erful automated web personalization tools that can Riedl, 2003)are two well-known examples of Rs on the intelligently assist us in transforming the vast amount of web data into useful information and knowledge. In other Recommender systems employ four information filter words, these tools ensure that the right information is ing techniques, demographic filtering(DMF)(Krulwich, delivered to the right people at the right time( Adomavicius 1997), content-based filtering(CBF)(Lang, 1995), collabo- Tuzhilin, 2005; Eirinaki vazirgiannis, 2003). Web rec- rative filtering(Resnick, Iakovou, Sushak, Bergstrom, ommender systems(RS), the most successful example of Riedl, 1994; Shardanand Maes, 1995), and hybrid filter- Web personalization tools, tailor information access, trim ing techniques(Balabanovic Shoham, 1997; Pazzani, 1999: Shahabi, Banaei-Kashani, Chen, McLeod, 2001) DMF categorizes the user based on the user personal attr Corresponding author. Mobile: \9\W9810196636: fax: +91(11) butes and makes recommendations based on demographic 26717528 mohamad. alshamriagmaiLcom(M Y.H. Al-Shamri), classes while the CBF suggests items similar to the ones the (KK. Bharadwan user preferred in the past 0957-4174/S- see front matter 2007 Elsevier Ltd. All rights reserved doi:10.1016/eswa2007.08.016Fuzzy-genetic approach to recommender systems based on a novel hybrid user model Mohammad Yahya H. Al-Shamri , Kamal K. Bharadwaj School of Computer and Systems Sciences, Jawaharlal Nehru University, New Delhi 110 067, India Abstract The main strengths of collaborative filtering (CF), the most successful and widely used filtering technique for recommender systems, are its cross-genre or ‘outside the box’ recommendation ability and that it is completely independent of any machine-readable represen￾tation of the items being recommended. However, CF suffers from sparsity, scalability, and loss of neighbor transitivity. CF techniques are either memory-based or model-based. While the former is more accurate, its scalability compared to model-based is poor. An impor￾tant contribution of this paper is a hybrid fuzzy-genetic approach to recommender systems that retains the accuracy of memory-based CF and the scalability of model-based CF. Using hybrid features, a novel user model is built that helped in achieving significant reduc￾tion in system complexity, sparsity, and made the neighbor transitivity relationship hold. The user model is employed to find a set of like￾minded users within which a memory-based search is carried out. This set is much smaller than the entire set, thus improving system’s scalability. Besides our proposed approaches are scalable and compact in size, computational results reveal that they outperform the classical approach. 2007 Elsevier Ltd. All rights reserved. Keywords: Recommender systems; Collaborative filtering; Web personalization; User model; Fuzzy sets 1. Introduction The popular use of web as a global information system has flooded us with a tremendous amount of data and information. The end users are overloaded with a huge amount of information from myriad resources. This explo￾sive growth of data has generated an urgent need for pow￾erful automated web personalization tools that can intelligently assist us in transforming the vast amount of data into useful information and knowledge. In other words, these tools ensure that the right information is delivered to the right people at the right time (Adomavicius & Tuzhilin, 2005; Eirinaki & Vazirgiannis, 2003). Web rec￾ommender systems (RS), the most successful example of Web personalization tools, tailor information access, trim down the information overload, and efficiently guide the user in a personalized manner to interesting items within a very large space of possible options (Burke, 2002). Typi￾cally RS recommend information (URLs, netnews articles), entertainment (books, movies, restaurants), or individuals (experts). Amazon.com (Linden, Smith, & York, 2003) and MovieLens.org (Miller, Albert, Lam, Konstan, & Riedl, 2003) are two well-known examples of RS on the web. Recommender systems employ four information filter￾ing techniques, demographic filtering (DMF) (Krulwich, 1997), content-based filtering (CBF) (Lang, 1995), collabo￾rative filtering (Resnick, Iakovou, Sushak, Bergstrom, & Riedl, 1994; Shardanand & Maes, 1995), and hybrid filter￾ing techniques (Balabanovic & Shoham, 1997; Pazzani, 1999; Shahabi, Banaei-Kashani, Chen, & McLeod, 2001). DMF categorizes the user based on the user personal attri￾butes and makes recommendations based on demographic classes while the CBF suggests items similar to the ones the user preferred in the past. 0957-4174/$ - see front matter 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2007.08.016 * Corresponding author. Mobile: +91 (11) 9810196636; fax: +91 (11) 26717528. E-mail addresses: mohamad.alshamri@gmail.com (M.Y.H. Al-Shamri), kbharadwaj@gmail.com (K.K. Bharadwaj). www.elsevier.com/locate/eswa Available online at www.sciencedirect.com Expert Systems with Applications 35 (2008) 1386–1399 Expert Systems with Applications