Using Genetic Algorithm for Hybrid Modes of Collaborative Filtering in Online recommenders Simon Fong, Y vonne Ho, Yang Hang Faculty of Science and Technology, University of Macau, Macau SAR ccfong @umac mo Abstract Content-based Filtering selects items based on the orrelation between the content of the items and user Online recommenders are usually referred to those preferences. Current search engines are based on used in e-Commerce websites for suggesting a product automatic analysis of the content of documents and the or service out of many choices. The core technology content of user's query. As an example shown in Figure implemented behind this type of recommenders la, that is a movie recommendation application, in includes content analysis, collaborative filtering and order to recommend movies to a user, the content some hybrid variants. Since they all have certain based recommender system tries to understand the strengths and limitations, combining them may be a commonalities among the movies that the user has rated highly in the past(specific actors, directors overcoming a large amount of input variables genres, subject matter, etc. ) Movie item D is hence especially from combining different techniques. recommended because it is has a certain high degree of Genetic algorithm(GA)is an ideal optimization search similarity to Item a by its content. function, for finding a best recommendation out of a For another example, a Personalized Recommender large population of variables. In this paper we System [l] creates dynamic hyperlinks on a web site presented a GA-based approach for supporting that contains a collection of advises about do it yourself ombined modes of collaborative filtering. In particular, we show that how the input variables can based on the similarity that they have, to the ones the be coded into ga chromosomes in various modes. user has highly rated in the past. In addition to Insights of how Ga can be used in recommenders are hypertext content, the same concept on content analysis derived through our experiments with the input data was extended in [2] for recommending multimedia taken from Movielens and IMDB products such as mp3 music on a recommender website 1 Introduction One common problem with the Content-based recommendation system is that it can only recommend Recommender systems have been widely adopted by items scoring highly against the user profile so the user e-Commerce websites to suggest products or services is restricted to see the items similar to those already to customers. In general, they assist users to narrow rated. new items will never be recommended due to the down their choices for making a purchase decision working on an individual user. from a large pool of items. One simple way of offering Collaborative Filtering(CF) based on the similarity recommendation is based on the top selling products between currently active user and his however does not differentiate customers who new items the active has never seen before but they have different tastes. The other approach is known were guessed to be interested by him because the other as one-to-one marketing, which takes into account the users who have similar interest to his have seen/liked nformation about a particular user and tries to find The similarity can either be measured by the same item personal match of product that is predicted to best sui which known as item-based CF or by the ype his or her flavor. Generally, there are three methods of user, known as user-based CF. This method is to the recommendation systems for one-to-one marketing suggest new items or to predict the utility of a certain Content-based Filtering, Collaborative Filtering and tem for a particular user based on his previous likings Hybrid Model. or the opinions of other like-minded usersUsing Genetic Algorithm for Hybrid Modes of Collaborative Filtering in Online Recommenders Simon Fong, Yvonne Ho, Yang Hang Faculty of Science and Technology, University of Macau, Macau SAR ccfong@umac.mo Abstract Online recommenders are usually referred to those used in e-Commerce websites for suggesting a product or service out of many choices. The core technology implemented behind this type of recommenders includes content analysis, collaborative filtering and some hybrid variants. Since they all have certain strengths and limitations, combining them may be a promising solution provided there is a way of overcoming a large amount of input variables especially from combining different techniques. Genetic algorithm (GA) is an ideal optimization search function, for finding a best recommendation out of a large population of variables. In this paper we presented a GA-based approach for supporting combined modes of collaborative filtering. In particular, we show that how the input variables can be coded into GA chromosomes in various modes. Insights of how GA can be used in recommenders are derived through our experiments with the input data taken from Movielens and IMDB. 1. Introduction Recommender systems have been widely adopted by e-Commerce websites to suggest products or services to customers. In general, they assist users to narrow down their choices for making a purchase decision from a large pool of items. One simple way of offering recommendation is based on the top selling products. This however does not differentiate customers who may have different tastes. The other approach is known as one-to-one marketing, which takes into account the information about a particular user and tries to find a personal match of product that is predicted to best suit his or her flavor. Generally, there are three methods of the recommendation systems for one-to-one marketing: Content-based Filtering, Collaborative Filtering and Hybrid Model. Content-based Filtering selects items based on the correlation between the content of the items and user preferences. Current search engines are based on automatic analysis of the content of documents and the content of user’s query. As an example shown in Figure 1a, that is a movie recommendation application, in order to recommend movies to a user, the content￾based recommender system tries to understand the commonalities among the movies that the user has rated highly in the past (specific actors, directors, genres, subject matter, etc.). Movie item D is hence recommended because it is has a certain high degree of similarity to Item A by its content. For another example, a Personalized Recommender System [1] creates dynamic hyperlinks on a web site that contains a collection of advises about do it yourself home improvement. The advices are recommended based on the similarity that they have, to the ones the user has highly rated in the past. In addition to hypertext content, the same concept on content analysis was extended in [2] for recommending multimedia products such as mp3 music on a recommender website. One common problem with the Content-based recommendation system is that it can only recommend items scoring highly against the user profile; so the user is restricted to see the items similar to those already rated, new items will never be recommended due to the working on an individual user. Collaborative Filtering (CF) based on the similarity between currently active user and other users, finds new items the active has never seen before but they were guessed to be interested by him because the other users who have similar interest to his have seen/liked. The similarity can either be measured by the same item which known as item-based CF or by the same type of user, known as user-based CF. This method is to suggest new items or to predict the utility of a certain item for a particular user based on his previous likings or the opinions of other like-minded users