Research Recommendation Algorithm Based On Forgetting Curve

With the rapid development of information technology and Internet technology, people enter the age of "information explosion", facing massive information, users do not know how their interested information is extracted from the mass of information, the search engine has already can’t satisfy people’s needs, recommender system has become the important way to solve the problem. Collaborative filtering because its algorithm is simply, can deal with complex issues and has good effect is widely used by people, also became the most successful recommender system technotogy. However, the user’s interest is always changing, and for new users system cannot predict the user’s preference, the original of the recommendation technology can’t meet these demands, this paper is devoted to solve these problems.Firstly, based on the Ebbinghaus forgetting curve proposed our recommendation algorithm, because people’s interest are constantly changing, and this kind of change is the process of natural forgetting,that is to say, it is keeping with the curve, so we applied the forgetting function to simulate the change of user’s interest. Taking into account the time playing an important role to score, when using similarity algorithm introduced time factor in it, made a attenuation for the original score of the users. Then the algorithm proposed in this paper designed two groups of experiments to verify the effectiveness of the algorithm. Through two groups of experimental results demonstrated that, generally speaking, the proposed similarity computing method based on the forgetting curve was better than the traditional algorithm. So in the recommendation system, combined with the laws of nature, through using Ebbinghaus forgetting curve as the forgetting rules could reflect user’s interest, made a attenuation for users score, could improve the system accuracy rate obviously. It also shows that, the human cognitive rules can play a very important role in recommendation system.next, based on the above content, we researched on the cold start problem in recommendation system in detail, a analyzed and compared various algorithms of predecessors to solve the cold start problem, made the advantages and shortcomings of each algorithm clearly; at the same time introduced the thinking caused by cross recommended, because people living in society, different people have different social circles, people’s preferences in target user’s social circle can reflect the target user’s preferences in some ways, based on this idea, combined with the algorithm Sahebi had proposed and the forgetting curve thought, put forward a scheme that is in a multidimensional network based on user-community division to solve the cold start. Through clearing the users’book reviews in dataset to simulate the cold start problem in the system, in the condition of overcoming the cold start problem with interest drift of community-groups to predict the users’score on books. At the part of experiment, compared traditional collaborative filtering recommendation algorithm with our scheme in paper. experiment proved that with the number of neighbors growing, The prediction accuracy was better than that of the latter, that is to say, the algorithm could overcome the cold start problem and improve the recommendation quality.