Research On Recommendation Algorithm For Distributed Recommendation Engine Based On Cloud Computing

With the evolvement of information technology and the growing scale of theInternet, the mount of information received by users will grow geometrically, and theera of information overload is coming. In this era, we received mass of informationfrom external everyday, with inability to select. In this case, both the informationconsumers and the information providers, have had to face unprecedented challenges.How to find the useful mass of information, that customers need, and how to getinformation pushed to the consumers who need it, both sides want to solve the problem.Recommendation system is generated in order to solve such difficulties: associate userswith information to help users to filter information by analyzing the user’s personalpreferences, and select the useful information to present to the users.Recommendation algorithm is the crucial part of the recommendation system,almost all researchers and recommended service providers attempt to achieve abreakthrough in the field of study of the recommendation algorithm. In this thesis thecurrent popular recommendation algorithm is discussed, including content-basedrecommendation algorithm neighborhood-based recommendation algorithm,dimensionality reduction methods, graph-based recommendation algorithm, etc, andthese algorithms have their own advantages and disadvantages and differentenvironment of application. Item-based collaborative filtering algorithm is one of themost popular recommendation algorithms. It has higher prediction accuracy and has theability to explain the recommendation result. As a simple and efficient algorithms, Slopeone is proposed to study in recent years. In this thesis, drawing the idea of the similaritymeasure in item-based recommendation algorithm, an improvement of the Slope Onealgorithm is proposed, which is using an adjusted cosine similarity measure to weightthe algorithm. The results show that the improved algorithm has higher predictionaccuracy than traditional Slope One algorithm.Fusion technology has been concerned by the researchers in recent years, It canfuse some single recommendation algorithm as a complex algorithms according to somestrategy. Single recommendation algorithm may compensate for each other’s deficiencies, so that the new fusion algorithm has better universality and predictionaccuracy. The algorithm of the winner team of the Netflix Prize uses the fusiontechnology. In this thesis, adjusted cosine similarity is used to weight the Slope Onealgorithm, and the item-based collaborative filtering algorithm is fused with it. Byfinding the optimal parameters of the new fusion algorithm for prediction, we comparethe prediction accuracy with the algorithms’ before. We get the conclusion that, the newfusion algorithm contains the advantages of the two algorithms. It has a niceperformance in the prediction with two measures. This is the achievement which thetwo algorithms could not achieve.