Cascaded Bi-graph Based Dynamic Recommendation Algorithm

In the information explosion age, the recommender system plays a veryimportant role in filtering the data, and its existence and development brings a lot ofconvenience for the internet users.Recommender System analyzes user’s behavior data, mining user’s interest, andthen do the recommendation for the user. Collaborative filtering is one of the mostclassic and common algorithms in recommendation area. It finds out some connectionpatterns between users and items, through analyzing the behaviors that users do on theitems without getting the knowledge of the properties of the users and items, andrecommends the users the specific items that they may be interested in. As therecommender system become more and more popular, there are various collaborativefiltering methods proposed and some of them showed good effect. Unfortunately,most algorithms did the change on how to describe the relationship between users anditems, while few ones focus on other subjective and objective factors, such as time,location, which is also significant. Recent years, researchers begun to study thedynamic properties of the recommender system, took the time, location factor intoconsideration, and made some progress.Compared to the Score predicts, Top-N recommendation is the most importantproblem in the recommendation. It takes full use of the behavior logs of the users. Itsmain task is to recommend N items to each user in the user list. In this paper, thefollowing work is finished based on the Top-N recommendation:1. Introduced the basic information on recommendation system, and explan themost classic algorithm, collaborative filtering.2. The cascaded bi-graph is put forward, which solved the problem ondescribing the relationship between users and items.3. Analyzed the important role that time information plays in users’ interests,took time factor into consideration, finished the dynamic Top-N recommendation.4. Ran the experiment based on CiteULike datasets, and analyzed the results.The performed experiment based on the dataset of CiteULike shows theefficiency of the method the paper proposed, and proves that the time factor is essential to the recommender system.