Research On Recommendation Diversity And Recommendation Cold Starting

With the continuous expansion of the scale of the Internet, the Internet has entered the era of information overload. It has become a serious problem that how to quickly extract information on the Internet which users may want or interest. With these opportunities and challenges, the recommendation system was born. Recommendation system is aiming to find the items which users and customers may interest according to users' hobbies and interests, thus to achieve the win-win situation which the users and the businesses are both satisfied.Recommendation system development is impressive, but the attendant issues are also increasingly prominent, for example, how to improve the diversity of the recommendation system, how to relieve cold starting issue and so on. In the field of the Internet, especially e-commerce,solving these problems has become an urgent need.On the problem of how to improve the diversity of recommendation system, thesis proposes a new algorithm to improve the recommendation diversity. This algorithm combines both of the collaborative filtering algorithms, which are user-based collaborative filtering and item-based collaborative filtering. Since both of the methods have advantages and disadvantages, thesis combines the two methods together, and divides the two methods into two different weights. A large number of experimental data show that the proposed method is effective and stable.On the recommendation system cold starting problem, thesis dealt with both of the item cold starting and user cold starting. Based on the decision tree classification to ease cold starting,we propose a new method to improve the original algorithm. We add the trusted network into this algorithm and classify the users whose similarity is more than 0.6 into one group. We assume that if the similarity is more than 0.6,both of the users are more similar than others. The MAE of the improved model decreases to some extent. Aiming to relieve the user cold starting, the new model combined both of the third application and statistics, and uses the proposed scheme, thus the precision, individual diversity and aggregate diversity are all improved, then the user cold starting problem is relieved.