Building a Lifestyle Recommender System Supiya Ujin and Peter J. Bentley University College London Department of Computer Science Gower Street. London WC1E 6BT (s ujin, p. bentley l@cs. ucl. ac uk ABSTRACT research movie recommendation website, makes use of Recommender systems are new types of internet-based software collaborative filtering technology to make its suggestions. This tools, designed to help users find their way through today technology captures user preferences to build a profile by complex on-line shops and entertainment websites. Here we asking the user to rate movies. It searches for similar profil rovide an overview of current recommender systems, and then (i.e. users that share the same or similar taste)and uses them to outline a new Lifestyle Recommender System, which employs generate new suggestions. One shortcoming that most websit techniques such as evolutionary search and a 3D avatar to using collaborative filtering suffer from is that they do not have provide tailored and friendly suggestions for users any facility to provide explanations of how recommendations are derived. This is addressed in [6 which proposes Kevwords explanation facilities for recommender systems in order to Recommender system, avatar, collaborative filtering ncrease users' faith in the suggestions BycontrastLibrA(Http://www.cs.utexas.edw/users/ibra 1. INTRODUCTION combines a content-based approach with machine learning to The rapid expansion of the Internet has brought about a new make book recommendations. The content-based approach market for trading. Electronic commerce or e-commerce has differs from collaborative filtering that it carries out analysis enabled businesses to open up their products and services to a n the contents of the items being recommended Furthermore massive client base that was once available only to the largest each user is treated individually-there is no sense of community" which forms the basis of collaborative filtering. It businesses becomes increasingly fierce, consumers are faced with a myriad of choices. Although this might seem to be techniques to build a model of each user's preferences relative nothing but beneficial to the consumer, the sheer wealth to the content of the items. The key advantage is explanations lating the various choices can can be very easily produced. However a content-base overwhelming. One would normally rely on the opinions and Is inappropriate when the items being considered are in a advice of friends or family members but unfortunately even they textual form such as images, and video or music clips. have limited knowledge. 3. BUILDING A LIFESTYLE Recommender systems provide one way of circumventing this RECOMMENDER SYSTEM problem. As the name suggests, their task is to recommend or suggest items or products to the customer based on his/her 3.1 System Overview preferences. These systems are often used by E-commerce This work focuses on the development of a Recommender System. It will gather preferences from products that the customer is likely to buy. An intermet site a broad range of topics allowing recommendations to ing a recommender syste exploit knowledge of for general lifestyle activities such as shopping for customers' likes and dislikes n understanding of their eating out and going to the cinema customer loyalty [1, 2 The system is based around a collaborative filtering approach, From the literature. it seems that the definition of the term building up profiles of users and then using an algorithm to find recommender system"varies depending on the author. Some profiles similar to the current user. Selected data from those profiles are then used to build recommendations. Because " collaborative filtering"and"social filtering"interchangeably profiles contain many attributes, many of which have sparse or [3, 4]. Conversely, others regard "recommender system' as a Incomplete data[6], the task of finding appropriate similarities generic descriptor that represe is often difficult. To overcome these problems, current systems recommendation/prediction techniques including collaborative, (such as MovieLens) use stochastic and heuristic-based models association rules [5]. In this paper, we adopt the latter definition research takes such ideas one step further, by applying a when referring to recommender systems random-based search algorithm known as an evolutionar algorithm to the problem of profile matching. Most current 2. CURRENT RECOMMENDER systems use standard nearest neighbour algorithms that consider SYSTEMS only "voting information"as the feature on which the comparison between two profiles is made [3]. This system There are a number of websites that incorporate recommender differs in that it takes into account multiple features such as system technologies. Here we will describe two systems that users age, gender and movie genres and employs evolution to exemplify the main approaches in this area. select and weight features to be used in Movielens(http://www.movielens.umn.edu),awell-know feature selection approach adapts to each users preferences and
Building a Lifestyle Recommender System Supiya Ujjin and Peter J. Bentley University College London Department of Computer Science Gower Street, London WC1E 6BT {s.ujjin, p.bentley}@cs.ucl.ac.uk ABSTRACT Recommender systems are new types of internet-based software tools, designed to help users find their way through today’s complex on-line shops and entertainment websites. Here we provide an overview of current recommender systems, and then outline a new Lifestyle Recommender System, which employs techniques such as evolutionary search and a 3D avatar to provide tailored and friendly suggestions for users. Keywords Recommender system, avatar, collaborative filtering 1. INTRODUCTION The rapid expansion of the Internet has brought about a new market for trading. Electronic commerce or e-commerce has enabled businesses to open up their products and services to a massive client base that was once available only to the largest multinational companies. As the competition between businesses becomes increasingly fierce, consumers are faced with a myriad of choices. Although this might seem to be nothing but beneficial to the consumer, the sheer wealth of information relating to the various choices can be overwhelming. One would normally rely on the opinions and advice of friends or family members but unfortunately even they have limited knowledge. Recommender systems provide one way of circumventing this problem. As the name suggests, their task is to recommend or suggest items or products to the customer based on his/her preferences. These systems are often used by E-commerce websites as marketing tools to increase revenue by presenting products that the customer is likely to buy. An internet site using a recommender system can exploit knowledge of customers' likes and dislikes to build an understanding of their individual needs and thereby increase customer loyalty [1, 2]. From the literature, it seems that the definition of the term “recommender system” varies depending on the author. Some researchers use the concepts, “recommender system”, “collaborative filtering” and “social filtering” interchangeably [3,4]. Conversely, others regard “recommender system” as a generic descriptor that represents various recommendation/prediction techniques including collaborative, social, and content based filtering, Bayesian networks and association rules [5]. In this paper, we adopt the latter definition when referring to recommender systems. 2. CURRENT RECOMMENDER SYSTEMS There are a number of websites that incorporate recommender system technologies. Here we will describe two systems that exemplify the main approaches in this area. MovieLens (http://www.movielens.umn.edu), a well-known research movie recommendation website, makes use of collaborative filtering technology to make its suggestions. This technology captures user preferences to build a profile by asking the user to rate movies. It searches for similar profiles (i.e. users that share the same or similar taste) and uses them to generate new suggestions. One shortcoming that most websites using collaborative filtering suffer from is that they do not have any facility to provide explanations of how recommendations are derived. This is addressed in [6] which proposes explanation facilities for recommender systems in order to increase users' faith in the suggestions. By contrast, LIBRA (http://www.cs.utexas.edu/users/libra) combines a content-based approach with machine learning to make book recommendations. The content-based approach differs from collaborative filtering in that it carries out analysis on the contents of the items being recommended. Furthermore, each user is treated individually - there is no sense of “community” which forms the basis of collaborative filtering. It also uses Bayesian text-categorisation machine learning techniques to build a model of each user's preferences relative to the content of the items. The key advantage is explanations can be very easily produced. However a content-based approach is inappropriate when the items being considered are in a nontextual form such as images, and video or music clips. 3. BUILDING A LIFESTYLE RECOMMENDER SYSTEM 3.1 System Overview This work focuses on the development of a Lifestyle Recommender System. It will gather preferences from users on a broad range of topics allowing recommendations to be made for general lifestyle activities such as shopping for clothes, eating out and going to the cinema. The system is based around a collaborative filtering approach, building up profiles of users and then using an algorithm to find profiles similar to the current user. Selected data from those profiles are then used to build recommendations. Because profiles contain many attributes, many of which have sparse or incomplete data [6], the task of finding appropriate similarities is often difficult. To overcome these problems, current systems (such as MovieLens) use stochastic and heuristic-based models to speed up and improve the quality of profile matching. This research takes such ideas one step further, by applying a random-based search algorithm known as an evolutionary algorithm to the problem of profile matching. Most current systems use standard nearest neighbour algorithms that consider only “voting information” as the feature on which the comparison between two profiles is made [3]. This system differs in that it takes into account multiple features such as user’s age, gender and movie genres and employs evolution to select and weight features to be used in the comparison. This feature selection approach adapts to each user's preferences and
ommendations. A more detailed love this fii Chek it an description of the system can be found in[8] If you we already seen It tell ne what you think Evolutionary algorithms are inspired by natural evolution. Populations of solutions are maintained, these are evaluated to find out how well each solution solves the problem. Fitness ★★★★ scores are given to each solution(where lower fitness values THE KOWARENAI dwcd by Kyok mean the solution is better )."Parents"are then selected from the population based on fitness- the fitter the solution, the ore likely it is to become a parent. Child solutions are enerated from the parents, employing crossover and mutation operators to ensure children resemble their parents with minor variations. The children replace the current population, the solutions are evaluated, parents are picked, and so on, for a YOUR OPINION number of generations. Evolution causes the solutions to mprove until they satisfy the current problem [7] O Love it O Not sure O Hate it This system uses an evolutionary algorithm in a slightly unusual way. Firstly, evaluation involves feedback from the users-as Figure 2: Lifestyle Recommender System Avatar interface they judge the quality of recommendations, their input is used help derive fitness scores for corresponding solutions in the 4. SUMMARY population. Secondly, initial solutions are not random as is As electronic commerce and on-line entertainment internet sites typical in such algorithms (a random matching function would multiply, and as the number of novice users of the internet be of little use in a recommender system). Thirdly, open-ended is a evolution is used -the system never trying to evolve software advisors. As described in this paper, recommender newer and better ways of matching systems provide such a solution for users. By employing how the evolutionary profile matcher is used within the system techniques such as evolutionary search and a 3D avatar, the Lifestyle Recommender System outlined here increases the 5. ACKNOWLEDGMENTS Thanks to Philip Treleaven for his guidance and Akira Sato for Matched profiles implementing the Avatar web interface. This work is funded by a scholarship provided by the Thai go 6. REFERENCES 1] Schafer, J B, Konstan, J. A and Riedl, J. January 2001.E- Commerce Recommendation Applications. Journal of Data Figure I Block diagram of the Lifestyle Recommender System [2] Schafer, J B, Konstan, J and Riedl, J Recommender Systems in E-Commerce 3.2 Consumer interaction the acm 1999 Conference on electron As we have seen mender systems can bring added value [3] Breese, I.S., Heckerman, D and Kadie, C.1998.Empirical to e-commerce websites by being able to understand the users analysis of predic references and needs. One possible way to take this further is to introduce a"to enhance the way in which both In Proceedings of the algorithms for collaborative filtering 14th Conference on Uncertainty in items can be recommended to, and feedback received from the Artificial Intelligence, pp 43-52 customer. This character or avatar would be a visual [4] Goldberg, K, Roeder, T, Gupta, D and Perkins, C representative of the website with which customers can interact. August 2000. Eigentaste: A Constant Time Collaborative The rationale behind this is that a user would find it easier to Filtering Algorithm. UCB ERL Technical Report M00/41 relate to another person or character than a standard web interface. There is then the possibility of a relationship being 5 Terveen, L and Hill, W. 2001. Beyond Recommender Systems: Helping People Help Each Other. In HCl In The established between user and avatar. From the point of view New Millenium, Carroll, J. ed. Addison-Wesley increased use of the website and allow more preference data to [6] Herlocker, J.L., Konstan, J. A. and riedl, J. 2000 be captured which would consequently lead to more accurate Explaining Collaborative Filtering Recommendations Proceedings of the ACM 2000 Conference on Computer upported Cooperative Work. The lifestyle recommender system will employ the use of a 3D avatar to interact with the user. Experiments will be conducted [7 Bentley, P J. 2001. Creative Evolutionary Systems to validate the assumption that users would participate more Morgan Kaufman Pub with the avatar. Figure 2 shows an early version of the avatar to [8] Ujin, S. 2001. An Adaptive Lifestyle Recommender be used for the lifestyle recommender system. System Using a Genetic Algorithm. Submitted to the Computation Conference 2001(GECCO 2001)
generates personalised recommendations. A more detailed description of the system can be found in [8]. Evolutionary algorithms are inspired by natural evolution. Populations of solutions are maintained, these are evaluated to find out how well each solution solves the problem. Fitness scores are given to each solution (where lower fitness values mean the solution is better). “Parents” are then selected from the population based on fitness – the fitter the solution, the more likely it is to become a parent. Child solutions are generated from the parents, employing crossover and mutation operators to ensure children resemble their parents with minor variations. The children replace the current population, the solutions are evaluated, parents are picked, and so on, for a number of generations. Evolution causes the solutions to improve until they satisfy the current problem. [7] This system uses an evolutionary algorithm in a slightly unusual way. Firstly, evaluation involves feedback from the users – as they judge the quality of recommendations, their input is used to help derive fitness scores for corresponding solutions in the population. Secondly, initial solutions are not random as is typical in such algorithms (a random matching function would be of little use in a recommender system). Thirdly, open-ended evolution is used – the system never stops trying to evolve newer and better ways of matching profiles. Figure 1 shows how the evolutionary profile matcher is used within the system. Evolutionary Profile Profile Generator Matcher Best Feature Extractor Matched Profiles Profile Database Person Avatar Web Interface Generate/ Update/ Feedback Figure 1 Block diagram of the Lifestyle Recommender System. 3.2 Consumer Interaction As we have seen, recommender systems can bring added value to e-commerce websites by being able to understand the users' preferences and needs. One possible way to take this further is to introduce a “character” to enhance the way in which both items can be recommended to, and feedback received from the customer. This character or avatar would be a visual representative of the website with which customers can interact. The rationale behind this is that a user would find it easier to relate to another person or character than a standard web interface. There is then the possibility of a relationship being established between user and avatar. From the point of view of the website owner, such a relationship would encourage increased use of the website and allow more preference data to be captured which would consequently lead to more accurate recommendations. The lifestyle recommender system will employ the use of a 3D avatar to interact with the user. Experiments will be conducted to validate the assumption that users would participate more with the avatar. Figure 2 shows an early version of the avatar to be used for the lifestyle recommender system. Figure 2: Lifestyle Recommender System Avatar interface 4. SUMMARY As electronic commerce and on-line entertainment internet sites multiply, and as the number of novice users of the internet increases, there is a greater need for tailored and friendly software advisors. As described in this paper, recommender systems provide such a solution for users. By employing techniques such as evolutionary search and a 3D avatar, the Lifestyle Recommender System outlined here increases the capabilities of such systems further. 5. ACKNOWLEDGMENTS Thanks to Philip Treleaven for his guidance and Akira Sato for implementing the Avatar web interface. This work is funded by a scholarship provided by the Thai Government. 6. REFERENCES [1] Schafer, J.B., Konstan, J. A. and Riedl, J. January 2001. ECommerce Recommendation Applications. Journal of Data Mining and Knowledge Discovery. [2] Schafer, J.B., Konstan, J. and Riedl, J. 1999. Recommender Systems in E-Commerce. Proceedings of the ACM 1999 Conference on Electronic Commerce. [3] Breese, J.S., Heckerman, D. and 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. [4] Goldberg, K., Roeder, T., Gupta, D. and Perkins, C. August 2000. Eigentaste: A Constant Time Collaborative Filtering Algorithm. UCB ERL Technical Report M00/41 [5] Terveen, L. and Hill, W. 2001. Beyond Recommender Systems: Helping People Help Each Other. In HCI In The New Millenium, Carroll, J. ed. Addison-Wesley. [6] Herlocker, J.L., Konstan, J. A. and Riedl, J. 2000. Explaining Collaborative Filtering Recommendations. Proceedings of the ACM 2000 Conference on Computer Supported Cooperative Work. [7] Bentley, P. J. 2001. Creative Evolutionary Systems. Morgan Kaufman Pub. [8] Ujjin, S. 2001. An Adaptive Lifestyle Recommender System Using a Genetic Algorithm. Submitted to the Graduate Student Workshop, Genetic and Evolutionary Computation Conference 2001 (GECCO 2001)
