結合粒子群分群技術的協同過濾

隨著資訊越來越龐大，推薦系統不斷的創新改良，協同過濾式推薦在近年中已慢慢成為個人化推薦系統的主流，該原理在於相同喜好的群聚中，使用者和其它成員，因為在喜好或想法上接近，其它成員所喜愛之物品，有很大的機會也是該使用者有所興趣。系統將口碑自動化的傳達給使用者A和該使用者A相似的成員，進而推薦使用者可能也會喜愛的產品。在最近幾年中不斷有學者針對協同過濾推薦系統的缺點及限制做改良，由於協同過濾的評分矩陣過大，造成所謂評分稀疏性問題，傳統上常使用K-means來做分群，但由於K-means在初始指定中心點會有困難，陷入局部最優問題，因此本文提出結合粒子群演算法的分配策略來對K-means的缺點做彌補，將用戶評分相似的項目放入同一個分群中，在計算目標項目與每個群聚中心的相似度，以相似度最高的分群作為目標項目的查詢結果進行搜尋。在本研究使用評分資料集來做實驗，並經由資料集的答案來評估結合粒子群改良後的推薦，其結果會以MAE(Mean Absolute Error)來評估分析，實驗結果中，受到資料集的性質影響，基於粒子群K-means對於稀疏性低的資料集有明顯改善，反之沒改善。

關鍵字

推薦系統；個人化推薦；口碑；協同過濾； K-means ；粒子群

並列摘要

Recommender systems keep being innovated and improved as the amount of data increases significantly. Collaborative filtering recommender is now the mainstream of the personalized recommenders in recent years. The working principle is that, in a group of preference, a user and other members gather together for something or idea they love, and therefore there is a great chance that what is loved by the rest of the group is also loved by the user. The system automatically distributes the words of mouth to user A and those similar to user A, thus recommending products that users may be interested in. There have been studies focusing on the improvement of the defects and limits of the collaborative filtering recommender. The rating matrix is huge in the collaborative filtering, causing the sparsity problem of rating. Traditionally K-means is used for grouping. However, it is somewhat difficult to assign the center initially with K-means, resulting in local optimization. In this study, a distribution strategy combining the particle swarm optimization (PSO) is introduced to patch up the defects that come with the use of K-means. Items of similar user rating are placed in the same group. When calculating the target items and the similarity of every clustering center, the group with the highest similarity is selected as the search result of the target items. An experiment is conducted based on the rating data set. The answers obtained from the data set are used to evaluate the recommendations after the combination of particle swarm. The results are analyzed using MAE (Mean Absolute Error). The experiment results are subject to the properties of the data set. The K-means based on the particle is improved significantly for the data set with low sparsity and no improvement is detected for high sparsity. Keywords: Recommender System, Personalized Recommender, Words of Mouth, Collaborative Filtering, K-means, Particle Swarm

並列關鍵字

Recommender System ； Personalized Recommender ； Words of Mouth ； Collaborative Filtering ； K-means ； Particle Swarm

參考文獻

Adomavicius, G., & Tuzhilin, A. (2005). toward the next generation of recommender systems: a survey of the state-of-the-art and possible extensions. IEEE Transactions on Knowledge and Data Engineering, 17(6), 734-749.

Ahn, H. J. (2007). A new similarity measure for collaborative filtering to alleviate the new user cold-starting problem

Balabanovic, M., & Shoham, Y. (1997). Fab: Content-Based, Collaborative Recommendation. Communications of ACM, 40(3), 66-72.

Burke, R. (2002). Hybrid Recommender Systems: Survey and Experiments. User Modeling and User-Adapted Interaction, 12, 331-370.

Cacheda, F., Carneiro, V., Fernández, D., & Formoso, V. (2011). Comparison of collaborative filtering algorithms: Limitations of current techniques and proposals for scalable, high-performance recommender systems. ACM Transactions on the Web (TWEB), 5(1).

國際替代計量

結合粒子群分群技術的協同過濾

未授權

主題瀏覽