1. 2. Research motivation fittest individuals. a collection of individuals are represented by chromosomes which are coded in The main objective of this research work is numeric real-value commander that exploits advantages of hybrid CF for high quality prediction and 2.1. Chromosome Encoding recommendation. at the same time ove limitations inherited from hybrid CF technique In order to implement a recommendation Genetic algorithm-based approach is proposed here a case study of Movie recommender, we apply Genetic for it is one of the most powerful search techniques for Algorithm to find the fine-tuned feature weights so that finding an optimized solution of a matching item to be each feature used in C can be characterized; the recommended to the user. In order to support an chromosome structure is presented as in Figure 2 extensive range of input data required for Hybrid CF This Ga chromosome structure is designed in mind (c f. Table 1), the GA approach allows the data to be that it could embrace the full range of input data from features into relativel Table 1 for supporting chromosome structure can be downsized to leave out With the chromosomes ready, GA searches for a some data components such as user demographic data solution that has the highest value of fitness function or item contents, in cases of Item-based CF or User (to be defined later). The feature weights which are based CF were in use respectivel relevant remain, noises are eliminated. During the early Feature I is the movie rating. Features 2-4 represent period of usage, there wasnt enough rating data which the user profile from MovieLens(source is known as the data sparsity problem [9]. We define a movielens umn. edu): age, gender and occupation are ay of solving this problem in our GA recommender the most influential demographic attributes that by coding an additional preference component in the describe the background characteristic of auser chromosome. So during the cold boot-up stage, the Features 5-12 represent the additional profile features preference component will dominate the search calculated in our work; they are the extra component function in the GA operation. dded into the chromosome. So they can be used in the d of 2. GA-based Collaborative Filtering Features 13-30 represent from MovieLens; features 3-37 represent the movie ga is a heuristic search method based on the mechanics of natural selection and genetic evolution, attributes from IMDb(source: imdb. com).Theya.the component of item content that describes about introduced by John Holland in 1975[10]. It maintains a movie. Both data from MovieLens and IMDb are used population of computing feature transformation in order to prevent any bias in any set of the data, for a matrices. By using selection, crossover and mutation fairer evaluation in our experiments ethods of GA, it finds the fitness value for picking the '量 w咧呦wws-Jwda2- Figure 2. Chromosome of full features Figure 6. Chromosome of 7 features Figure 3. GA-based Recommendation System1.2. Research Motivation The main objective of this research work is to design a recommender that exploits advantages of hybrid CF for high quality prediction and recommendation, at the same time overcoming the limitations inherited from hybrid CF techniques. Genetic algorithm-based approach is proposed here for it is one of the most powerful search techniques for finding an optimized solution of a matching item to be recommended to the user. In order to support an extensive range of input data required for Hybrid CF (c.f. Table 1), the GA approach allows the data to be encoded as ‘features’ into relatively elongated chromosomes. With the chromosomes ready, GA searches for a solution that has the highest value of fitness function (to be defined later). The feature weights which are relevant remain, noises are eliminated. During the early period of usage, there wasn’t enough rating data which is known as the data sparsity problem [9]. We define a way of solving this problem in our GA recommender by coding an additional ‘preference’ component in the chromosome. So during the cold boot-up stage, the preference component will dominate the search function in the GA operation. 2. GA-based Collaborative Filtering GA is a heuristic search method, based on the mechanics of natural selection and genetic evolution, introduced by John Holland in 1975 [10]. It maintains a population of computing feature transformation matrices. By using selection, crossover and mutation methods of GA, it finds the fitness value for picking the fittest individuals. A collection of individuals are represented by chromosomes which are coded in numeric real-value. 2.1. Chromosome Encoding In order to implement a recommendation system as a case study of Movie recommender, we apply Genetic Algorithm to find the fine-tuned feature weights so that each feature used in CF can be characterized; the chromosome structure is presented as in Figure 2. This GA chromosome structure is designed in mind that it could embrace the full range of input data from Table 1 for supporting hybrid CF. Optionally the chromosome structure can be downsized to leave out some data components such as user demographic data or item contents, in cases of Item-based CF or User￾based CF were in use respectively. Feature 1 is the movie rating. Features 2-4 represent the user profile from MovieLens (source: movielens.umn.edu); age, gender and occupation are the most influential demographic attributes that describe the background characteristic of a user. Features 5-12 represent the additional profile features calculated in our work; they are the extra component added into the chromosome. So they can be used in the initial period of usage when the movie ratings are scarce. Features 13-30 represent the movie genres from MovieLens; features 3-37 represent the movie attributes from IMDb (source: imdb.com). They are the component of item content that describes about the movie. Both data from MovieLens and IMDb are used in order to prevent any bias in any set of the data, for a fairer evaluation in our experiments. w1 w2 w3 w4 w5 … w12 Rating Age Gender Occupation type Preferred film 8 user profile features w13 … w30 w31 … w37 Action 18 genres language Preferred Western Director 7 characters Language w1 w2 w3 w4 w5 … w12 Rating Age Gender Occupation type Preferred film 8 user profile features w13 … w30 w31 … w37 Action 18 genres language Preferred Western Director 7 characters Language Figure 2. Chromosome of full features w1 w2 w3 w4 w5 Rating Prefer Director w6 w7 Prefer Actress Prefer Actor Prefer Producer Prefer Writer Prefer Editor w1 w2 w3 w4 w5 Rating Prefer Director w6 w7 Prefer Actress Prefer Actor Prefer Producer Prefer Writer Prefer Editor Figure 6. Chromosome of 7 features Profile Capture: User registered to provide the demographics New user Movie Database Collect Movie Info Recommendations Data Collect selected items from neighbors Search for neighbors from user database User Database Neighbor Set GA Recommender Get Top N neighbors Collect Ratings: Get ratings of movies Create New User Profile Features Phase I Phase II Phase III: Recommender Profile Capture: User registered to provide the demographics New user Movie Database Collect Movie Info Recommendations Data Collect selected items from neighbors Search for neighbors from user database User Database Neighbor Set GA Recommender Get Top N neighbors Collect Ratings: Get ratings of movies Create New User Profile Features Phase I Phase II Phase III: Recommender Figure 3. GA-based Recommendation System