Using Genetic Algorithms for Personalized Recommendation 105 feature may be associated with different importance. Most of the systems either ig- nored or used a fixed weight for each feature, which often caused a poor recommen dation performance Genetic algorithms are adaptive algorithms based on the Darwinian principle of natural selection and are often used to solve optimization problems. In this paper, we propose a hybrid recommender system which uses genetic algorithms for feature weighting. The proposed system consists of three modules In PGM(Profile Genera- tion Module), the customers transaction data are analyzed to establish the customers preference profile candidate table. In NSM(Neighborhood Selection Module), a clus tering method is first adopted to segment customers into groups using the profile candidate table. The genetic algorithm is then used to fine-tune profile matching for each active customer. Finally, in the RC Model(Recommendation Module), a list of recommendation is derived and presented. This will enable the recommender system to make more accurate predictions of users' likes and dislikes, and hence will provide better recommendations to users 2 Research Background Recommender systems have been successfully applied in a number of difference applications such as recommending movies, books, music and products. There are two major techniques used in recommender systems [2-4], content-based approach and collaborative filtering approach 2.1 Collaborative Filtering The term collaborative filtering was coined by the Goldberg et al. [5], the developers of the first recommender system-Tapestry. Tapestry allows users to annotate docu ments that they read. Users can then retrieve a document based on the content of the document or other users opinions in terms of annotation on that document. However, the recommendations are not automated and require users to explicitly define their collaborative relationships. GroupLens[6][7] provides an automated recommendations using a neighborhood-based algorithm. The system uses the ratings of items to find people who are most similar to you and use their opinions for recommendations GroundLens provides personalized predictions for Usenet news articles, while other systems use this approach for recommending movies, music, jokes and web pages. The original collaborative filtering algorithm contains two main steps: neighbor hood formation and generation of recommendation. Neighborhood formation finds a set of users known as neighbors that have similar preference ratings. Common simi- ity metrics used include Pe correlation, mean squared difference, and vector similarity. In the generation of recommendation step, the system then computes the predicted ratings on items the active user has not yet seen based on his neighbors ratings for those items. Finally, the system derives and sorts a set of recommendations by the predicted ratinUsing Genetic Algorithms for Personalized Recommendation 105 feature may be associated with different importance. Most of the systems either ig￾nored or used a fixed weight for each feature, which often caused a poor recommen￾dation performance. Genetic algorithms are adaptive algorithms based on the Darwinian principle of natural selection and are often used to solve optimization problems. In this paper, we propose a hybrid recommender system which uses genetic algorithms for feature weighting. The proposed system consists of three modules. In PGM (Profile Genera￾tion Module), the customer’s transaction data are analyzed to establish the customers’ preference profile candidate table. In NSM (Neighborhood Selection Module), a clus￾tering method is first adopted to segment customers into groups using the profile candidate table. The genetic algorithm is then used to fine-tune profile matching for each active customer. Finally, in the RC Model (Recommendation Module), a list of recommendation is derived and presented. This will enable the recommender system to make more accurate predictions of users' likes and dislikes, and hence will provide better recommendations to users. 2 Research Background Recommender systems have been successfully applied in a number of difference applications such as recommending movies, books, music and products. There are two major techniques used in recommender systems [2-4], content-based approach and collaborative filtering approach. 2.1 Collaborative Filtering The term collaborative filtering was coined by the Goldberg et al. [5] , the developers of the first recommender system – Tapestry. Tapestry allows users to annotate docu￾ments that they read. Users can then retrieve a document based on the content of the document or other users’ opinions in terms of annotation on that document. However, the recommendations are not automated and require users to explicitly define their collaborative relationships. GroupLens[6][7] provides an automated recommendations using a neighborhood-based algorithm. The system uses the ratings of items to find people who are most similar to you and use their opinions for recommendations. GroundLens provides personalized predictions for Usenet news articles, while other systems use this approach for recommending movies, music, jokes and web pages. The original collaborative filtering algorithm contains two main steps: neighbor￾hood formation and generation of recommendation. Neighborhood formation finds a set of users known as neighbors that have similar preference ratings. Common simi￾larity metrics used include Pearson correlation, mean squared difference, and vector similarity. In the generation of recommendation step, the system then computes the predicted ratings on items the active user has not yet seen based on his neighbors’ ratings for those items. Finally, the system derives and sorts a set of recommendations by the predicted ratings