M. YH. Al-Shamri, KK Bharadwaj Expert Systems with Applications 35(2008)1386-1399 FGRs performed worse than Prs but better than extensions. IEEE Transaction on Knowledge and Data engineering, FRS for the active users 45(split-1), and 39, 41(split 17(6),734-749 3). This is because the evolved weights obtained till Balabanovic, M.,&Shoham,Y(1997).Fab:Content-based,collaborative now in our experiments did not describe those users recommendation. Communications of the ACM, 403). 66-72 asu, C, Hirsh, H,& Cohen, W.(1998). Recommendation as realistically cation: Using social and content-based information in recom tion. In Proceedings of the 15th national conferen From the above observations, we can say that demo- intelligence, Madison, Wisconsin to the recommendations made by their friends who are Burke,R(2002 ) Hybrid recommender systems mam. pp g>is or graphical data is more than or at least as important as rat- Breese, J. S, Heckerman, D,& Kadie, C(1998) predictive algorithms for collaborative filtering. ings for some active users. The logic for FRS agrees with 14th conference on uncertainty in artifcial intel the real life everyday experience where most people listen Madison, WI/San Francisco, CA: Morgan Kar most likely to be in their own group. Table 9 illustrates User Modeling and User-Adapted Interaction, 12, 331-370 one such case where the active user is a female aged 32 umitrescu,D, Lazzerini, B,& Jain, L C(2000). Fuzzy sets and their application to clustering and training. CRC Press LLC Out of 30 recommenders for this female active user, 23 Eirinaki. M.& vazirgiannis. M. (2003). Web mining for web personal- are males with age ranging from 19 to 70 using the classical ization. ACM Transactions on Internet Technology. 3(1). 1-27 PRS. whereas for FRS. 29 recommenders are females in the Froschl, C.(2005). User Modeling and User Profiling in Adaptive E- age group 22-51 learning Systems. Master Thesis, Graz University of Technology Gadi, T, Daoudi, R. B. M.,& Matusiak, S(1999). Fuzzy similarity 6. Conclusion measure for shape retrieval. In Vision Interface 99, Trois-RiDier Canada(pp 19-21) This work has shown a considerable reduction in the Goldberg, D.(1989). Genetic algorithms in search, optimization, and complexity of recommender system(RS). Building concise machine learning. Pearson Education and representative user model reduces the effect of user Goldberg, D, Nichols, D, Oki, B. M,& Terry, D.(1992). Using collaborative g to weave an information tapestry. Communica- m matrix sparsity, and the computational time complex tions of the ACM, 35(12), 61-70 Ity that is reduced by approximately a factor of six for our Han, J.,& Kamber, M.(2001). Data mining, concepts and techniques approaches as compared to Pearson recommender system Klir, G,& Yuan, B(1995). Fuzzy sets and fuzzy logic Academic Press (PRS). Human decisions employ a wide range of fuzzy ry and terms therefore a substantial improvement is gained by Koch, N.(2000). Software ng for Adaptive Hyper fuzzifying the user model. Amongst the various distance lunch/ functions, the proposed fuzzy distance function reflects more appropriately the fuzziness of each fuzzy feature. Krulwich, B.(1997). Lifestyle Finder: Inte user profiling using Experimental results show that the proposed fuzzy and large-scale demographic data. Artificial Intelligence Magazine, 18(2). / -genetic approaches consid ang,K.(1995) the accuracy of Rs as compared to simple rs like PRS conference on machine learning, Tahoe City, By employing a GA, each user priority for e ea ature is captured. The computational time complexity is the Linden, G, Smith, B,& York, J(2003). Amazon. com recommendations major problem for this approach, however this difficulty Item-to-item collaborative filtering. IEEE Internet Computing, 7(1), gets overcome by performing fine-tuning offline and the Massa, P,& Avesani, P(2004). Trust-aware collaborative filtering for commender systems Coop/S/DOAlODBASE (1), 492-508 machine or in a separate weight matrix along with the user Michalski, R (2000). Learnable evolution model: Evolutionary item matrix. This set of weights that we have evolved for guided by machine learning. Machine Learning, 38, 9-40 each user can be used as the initial weights during future Miller, B N, Albert, I, Lam, S.K.Konstan,J.A,& Riedl, access to the system MovieLens unplugged: Experiences with an occasionally ecommender system. In Proceedings of the international conference on Because of the computational time cost of GA intelligent user interfaces for FGRS are found based on small parameter values Nasraoui, O,& Petenes, C(2003). An intelligent web recommendation (Table 8). By increasing the parameter values of GA, better engine based on fuzzy approximate reasoning. In Proceedings of the and more representative weights can be evolved. However, IEEE international conference on fuzzy systems(pp. 1116-1121 one of the most promising research directions for evolving Pazzani, M.(1999). A framework for collaborative, content-based, and appropriate weights would be the use of a learnable evolu- Resnick, P, lakovou, N, Sushak, M, Bergstrom, P,& Riedl, J(1994) tion model (LEM)(Michalski, 2000), which has shown GroupLens: An open architecture for collaborative filtering of speed-up of two or more orders of magnitude in terms of Netnews. In Proceedings of ACM conference on computer supported operative work(pp the number of evolutionary steps in its early experiments. Shahabi. c. Banaei-Kashani,F.Chen,Y,&McLeod,D(2001).Yoda An accurate and scalable web-based recommendation systems. In nformation systems(CoopIs 2001), Trento, Italy. avicius, G,& Tuzhilin, A(2005). Toward the next generation of Shardanand, U,& Maes, P (1995). Social information filtering mender systems: A survey of the state-of-the-art and possible rithms for automating Word of Mouth. In Proceedings• FGRS performed worse than PRS but better than FRS for the active users 45 (split-1), and 39, 41 (split- 3). This is because the evolved weights obtained till now in our experiments did not describe those users realistically. From the above observations, we can say that demo￾graphical data is more than or at least as important as rat￾ings for some active users. The logic for FRS agrees with the real life everyday experience where most people listen to the recommendations made by their friends who are most likely to be in their own group. Table 9 illustrates one such case where the active user is a female aged 32. Out of 30 recommenders for this female active user, 23 are males with age ranging from 19 to 70 using the classical PRS, whereas for FRS, 29 recommenders are females in the age group 22–51. 6. Conclusion This work has shown a considerable reduction in the complexity of recommender system (RS). Building concise and representative user model reduces the effect of user item matrix sparsity, and the computational time complex￾ity that is reduced by approximately a factor of six for our approaches as compared to Pearson recommender system (PRS). Human decisions employ a wide range of fuzzy terms therefore a substantial improvement is gained by fuzzifying the user model. Amongst the various distance functions, the proposed fuzzy distance function reflects more appropriately the fuzziness of each fuzzy feature. Experimental results show that the proposed fuzzy and hybrid fuzzy-genetic approaches considerably improved the accuracy of RS as compared to simple RS like PRS. By employing a GA, each user priority for each feature is captured. The computational time complexity is the major problem for this approach, however this difficulty gets overcome by performing fine-tuning offline and the best set of feature weights is stored on the user’s local machine or in a separate weight matrix along with the user item matrix. This set of weights that we have evolved for each user can be used as the initial weights during future access to the system. Because of the computational time cost of GA, weights for FGRS are found based on small parameter values (Table 8). By increasing the parameter values of GA, better and more representative weights can be evolved. However, one of the most promising research directions for evolving appropriate weights would be the use of a learnable evolu￾tion model (LEM) (Michalski, 2000), which has shown speed-up of two or more orders of magnitude in terms of the number of evolutionary steps in its early experiments. References Adomavicius, G., & Tuzhilin, A. (2005). Toward the next generation of recommender systems: A survey of the state-of-the-art and possible extensions. IEEE Transaction on Knowledge and Data Engineering, 17(6), 734–749. Balabanovic, M., & Shoham, Y. (1997). Fab: Content-based, collaborative recommendation. Communications of the ACM, 40(3), 66–72. Basu, C., Hirsh, H., & Cohen, W. (1998). Recommendation as classifi- cation: Using social and content-based information in recommenda￾tion. In Proceedings of the 15th national conference on artificial intelligence, Madison, Wisconsin. Breese, J. S., Heckerman, D., & Kadie, C. (1998). Empirical analysis of predictive algorithms for collaborative filtering. In Proceedings of the 14th conference on uncertainty in artificial intelligence (pp. 43–52). Madison, WI/San Francisco, CA: Morgan Kaufmann. Burke, R. (2002). Hybrid recommender systems: Survey and experiments. User Modeling and User-Adapted Interaction, 12, 331–370. Dumitrescu, D., Lazzerini, B., & Jain, L. C. (2000). Fuzzy sets and their application to clustering and training. CRC Press LLC. Eirinaki, M., & Vazirgiannis, M. (2003). Web mining for web personal￾ization. ACM Transactions on Internet Technology, 3(1), 1–27. Froschl, C. (2005). User Modeling and User Profiling in Adaptive E￾learning Systems. Master Thesis, Graz University of Technology, Graz, Austria. Gadi, T., Daoudi, R. B. M., & Matusiak, S. (1999). Fuzzy similarity measure for shape retrieval. In Vision Interface ’99, Trois-Rivie`res, Canada (pp. 19–21). Goldberg, D. (1989). Genetic algorithms in search, optimization, and machine learning. Pearson Education. Goldberg, D., Nichols, D., Oki, B. M., & Terry, D. (1992). Using collaborative filtering to weave an information tapestry. Communica￾tions of the ACM, 35(12), 61–70. Han, J., & Kamber, M. (2001). Data mining, concepts and techniques. Academic Press. Klir, G., & Yuan, B. (1995). Fuzzy sets and fuzzy logic, theory and applications. Prentice-Hall. Koch, N. (2000). Software Engineering for Adaptive Hypermedia Systems. PhD thesis, Ludwig-Maximilians-University Munich/ Germany. Krulwich, B. (1997). Lifestyle Finder: Intelligent user profiling using large-scale demographic data. Artificial Intelligence Magazine, 18(2), 37–45. Lang, K. (1995). NewsWeeder: Learning to filter netnews. In Proceedings of the 12th international conference on machine learning, Tahoe City, CA. Linden, G., Smith, B., & York, J. (2003). Amazon.com recommendations: Item-to-item collaborative filtering. IEEE Internet Computing, 7(1), 76–80. Massa, P., & Avesani, P. (2004). Trust-aware collaborative filtering for recommender systems. CoopIS/DOA/ODBASE (1), 492–508. Michalski, R. (2000). Learnable evolution model: Evolutionary processes guided by machine learning. Machine Learning, 38, 9–40. Miller, B. N., Albert, I., Lam, S. K., Konstan, J. A., & Riedl, J. (2003). MovieLens unplugged: Experiences with an occasionally connected recommender system. In Proceedings of the international conference on intelligent user interfaces. Nasraoui, O., & Petenes, C. (2003). An intelligent web recommendation engine based on fuzzy approximate reasoning. In Proceedings of the IEEE international conference on fuzzy systems (pp. 1116–1121). Pazzani, M. (1999). A framework for collaborative, content-based, and demographic filtering. Artificial Intelligence Review, 393–408. Resnick, P., Iakovou, N., Sushak, M., Bergstrom, P., & Riedl, J. (1994). GroupLens: An open architecture for collaborative filtering of Netnews. In Proceedings of ACM conference on computer supported cooperative work (pp. 175–186). NC: Chapel Hill. Shahabi, C., Banaei-Kashani, F., Chen, Y., & McLeod, D. (2001). Yoda: An accurate and scalable web-based recommendation systems. In Proceedings of the sixth international conference on cooperative information systems (CoopIS 2001), Trento, Italy. Shardanand, U., & Maes, P. (1995). Social information filtering: Algo￾rithms for automating ‘Word of Mouth’. In Proceedings of the 1398 M.Y.H. Al-Shamri, K.K. Bharadwaj / Expert Systems with Applications 35 (2008) 1386–1399