Research On Diversity Of Personalized Recommendation Technology

With the high-speed development of information technology, the internet resources has been greatly enriched, bringing the convenience of retrieving information for people. However, with the growing number of network resources, it becomes more difficult for users to find the useful information in the mass of complicated information, following which the problem of information overload becomes severe. Personalized recommendation technology is proposed for the problem. It’s appearing gives users more convenient and effective approaches for information acquisition. Basing on the preferences and behaviors of users, personalized recommendation technology recommends users with the information they most probably needed.Most existing recommendation techniques have focused on improving recommendation accuracy. However, the quality of recommendations can be evaluated along a number of dimensions, and relying on the accuracy of recommendations alone may not be enough to find the most relevant items for each user. Recently, diversity of recommendations has also been increasingly recognized in research literature as an important aspect of recommendation quality. More diverse recommendations, presumably leading to more sales of long-tail items, could be beneficial for both individual users and some business models. The inherent trade-off between accuracy and diversity has been observed in previous studies, therefore, indicating that maintaining accuracy while improving diversity constitutes a difficult task. In this paper, this problem has been explored, several recommendation algorithms which can improve the diversity of recommendations are introduced, and one of which named Re-Ranking Approach is emphasized by discussing its advantages and drawbacks. An improved algorithm is proposed based on the Re-Ranking Approach. Furthermore, a new top-N recommendation algorithm based on recommendation expectation is proposed in this paper, which has better performance.The summary of work:1. This paper introduced several recommendation algorithms which can improve the diversity of recommendations are introduced, one of which named Re-Ranking Approach is particularly discussed and analyzed its advantages and drawbacks.2. To solve the problem of rating threshold which containing no individual user information, this paper proposed an improved user-centered re-ranking approach, which added the users’rating preference, to improve the performance of original algorithm and make it more closely to the user-centered purpose of recommendation.3. To globally control the diversity of recommendations, this paper proposes a new top-N recommendations approach based on recommended expectations of items. This approach models the recommendation expectation of the candidate item for recommending, and uses it as a proxy of the recommended times of candidate items. In the process of making top-N recommendations for each user, the expectation of the item for recommending can be adjusted dynamically to control the recommended times of items indirectly and to achieve the purpose of improving diversity of recommendations.