Using Genetic Algorithms for Personalized Recommendation 109 3.2 Neighborhood Selection Module( Nsm) In NSM, firstly, the GAs is used to fine-tune the feature weights for each active cus tomer. Then, the collaborative filtering approach is applied to form the neighborhood of the active customer The chromosome in the ga process is represented as a weight vector with 17 genes. Each gene is encoded with 8 bits. The Ga begins with random genotypes and an initial population of 100 chromosomes. For each active customer, a randomly selected chromosome is assigned and tested by the fitness function. The fitness func- tion measures the prediction accuracy of products based on the current chromosome ∑|P(k,)-A(, Accuracy(k)=1-2- Each active customer k is tested by a random selection of l products. P(k, j) and A(,j) are the predicted and the actual ratings of customer k to product j, respec- tively. The predicted ratings are calculated by the collaborative filtering algorithm with different neighborhood size The algorithm continues to evolve until the termination criteria are met. In experiment, we set the maximum generation number to 100. For each generation evolution, chromosomes for the next generation are selected using the roulette wheel selection scheme to implement proportionate random selection. All of the chromo somes are then paired up using the single-point crossover strategy with a probability of 0.9. After the crossover, for each of the genes of the chromosomes, the gene is obability of 0.05 After obtaining customer's best feature weights, we can now select the most similar n neighbors(denoted as NB, ) by computing the similarity value ∑Wx(CP-CPp)2 where Wr=(Wi, W2, . W)is the feature weights of customer k obtained from GAs 3.3 Recommendation Module ro RC module recommends products for active customer k by collecting the information from his/her neighbors For each product j we compute its recommendation score as Score(k,j)=>Similarity(k, i)x purchase(i, j) (7)Using Genetic Algorithms for Personalized Recommendation 109 3.2 Neighborhood Selection Module ( NSM ) In NSM, firstly, the GAs is used to fine-tune the feature weights for each active cus￾tomer. Then, the collaborative filtering approach is applied to form the neighborhood of the active customer. The chromosome in the GA process is represented as a weight vector with 17 genes. Each gene is encoded with 8 bits. The GA begins with random genotypes and an initial population of 100 chromosomes. For each active customer, a randomly selected chromosome is assigned and tested by the fitness function. The fitness func￾tion measures the prediction accuracy of products based on the current chromosome. l P k j A k j Accuracy k l j ∑= − = − 1 1 ( , ) ( , ) ( ) (5) Each active customer k is tested by a random selection of l products. P(k, j) and A(k, j) are the predicted and the actual ratings of customer k to product j, respec￾tively. The predicted ratings are calculated by the collaborative filtering algorithm with different neighborhood size. The algorithm continues to evolve until the termination criteria are met. In our experiment, we set the maximum generation number to 100. For each generation evolution, chromosomes for the next generation are selected using the roulette wheel selection scheme to implement proportionate random selection. All of the chromo￾somes are then paired up using the single-point crossover strategy with a probability of 0.9. After the crossover, for each of the genes of the chromosomes, the gene is mutated with a probability of 0.05. After obtaining customer's best feature weights, we can now select the most similar n neighbors (denoted as NBk ) by computing the similarity value. ∑ ∑ = = × − = − 17 1 17 1 2 1 i i k i i a i k i k W W CP CP Similarity k a ( ) ( , ) (6) where ( , , , ) 1 2 17 Wk = Wk Wk " Wk is the feature weights of customer k obtained from GAs. 3.3 Recommendation Module (RC) RC module recommends products for active customer k by collecting the information from his/her neighbors. For each product j we compute its recommendation score as ∑∈ = × NBk i Score(k, j) Similarity(k,i) purchase(i, j) (7)