The Study Of Time And Space Context Based Unifying Collaborative Filtering Algorithm

With the development of share economic recommendation systems help Internet users find valuable information from a huge volume of information timely and correctly which greatly improve the utilization efficiency of information. At the same time with the vigorous development of e-commerce, users' requirement for products and services are toward the direction of diversification and individuation. Recommendation system is facing problems of lacking diversification and individuation.The thesis proposes KNN Time Unified Nearest Neighbors algorithm(KTUNN) and KNN Place Unified Nearest Neighbors algorithm (KPUNN)which is aimed at solving the serious problem that the recommendation system is lacking personalized. At the same time the thesis proposes KNN Time and Place Unified Nearest Neighbors algorithm (KTPUNN) based on the two algorithms above which considered relieving the cold-start problem. As for the time characteristic, the algorithm uses the attenuation characteristic of Logistic equation in the similarity calculation. By considering the time decay factor, the algorithm set the user's recent interests with a higher weight, while the long-term interests with a lower weight in the recommendation process. As for the space characteristics, the algorithm considers the distance factor into the calculation of the location similarity. At the same time by using the character of TF-IDF for the screening of personalized sites, the algorithm set a higher weight to the sites which received little attention and a lower weight to popular sites. As for the cold start problem the thesis combine the new users and the new items with the exiting users and exiting items into similarity calculation which improves accuracy of the recommendation results.The algorithms is put forward by the experiment and compared with other traditional collaborative filtering algorithms. Experiments show that the unifying collaborative filtering algorithm based on space and time algorithms not only enhance the time and space characteristics of the recommendation system, but also improve the cold start problem which upgrade the quality of the recommendation.