Implicit Rating Methods Based on Interest Preferences of Categories for Micro-Video Recommendation

Collaborative filtering (CF) without explicit information is one of the most challenging research directions in the field of video recommendation, as the effectiveness of traditional CF methods strongly depend on the ratings of videos for users. However, in the actual online video platforms, explicit ratings are very rare or even completely unavailable in most cases. This makes the effects of traditional recommendation algorithms are not satisfactory. In addition, micro-videos have attracted wide attention, while they have not be considered differently by the traditional recommendation algorithms. It is meaningful to study the recommendation methods for micro-videos. Considering that micro-videos have categories, two implicit rating methods based on interest preferences of categories are proposed to improve the performance of recommendation for micro-videos under implicit feedback. Its core idea is to construct a rating matrix based on implicit information by mining users’ implicit interest preference information for different categories of micro-videos, and use it as the basis of recommendation algorithms. The proposed rating methods are validated on a large online video content provider, and they are correct and can effectively mine users’ preferences without explicit ratings according to the experimental results. They can bring better results than some existing algorithms, and can be better applied to the video recommendation system.