Leveraging recommendation systems for improving caching emerging short video in content delivery network

Abstract

Short video applications have rapidly become one of the most popular online entertainments. It is also one of the most important ways of social networking. However, the request pattern of short videos shows the characteristics of highly dynamic compared to traditional video. First, the access of short video is changing quickly, that is, user shows less preferences. Second, most short videos have no tag descriptions. These result in a significant reduction in caching efficiency of content delivery network (CDN) server, which is used to cache short videos. To address these challenges, this article proposes a recommendation system to optimize short video caching. For lack of tag description, this article proposes event detection method to recognize tags based on improved dense trajectory and concept dictionary video feature extraction. Then the collaborative filtering method is used to generate caching recommendation list for CDN to guide short video caching. Simulation results show that the proposed method can significantly improve the efficiency of short video caching for CDN.

Network-aware Recommendations in the Wild: Methodology, Realistic Evaluations, Experiments

“…Then, CABaRet searches for video IDs in L that are also included in the list of cached videos C and adds them to the list of video IDs to be recommended R, until all IDs in L are explored or the list R contains N video IDs, whichever comes first (lines 4-9). If after this step, R contains less than N video IDs, N − |R| video IDs from the head of the list L are added to R; these IDs correspond to the top N −|R| noncached videos that are directly related to video v (lines [10][11][12][13][14][15]. Remark: the operations in lines 4-9 and 10-15 could be merged in an implementation to slightly reduce complexity.…”

“…Despite the assumptions made in the model (e.g., independence between p i and the set of recommended videos), it can be generalizable to any service (e.g., to short-video services having considerably different user demand patterns than YouTube [13], [39]) given general user demand statistics, i.e., q C and p i . Caching optimization under CABaRet.…”

“…The joint network and recommendations paradigm has been recently introduced, in the context of soft cache hits [4], [6] or network-friendly recommendations [5], [9], aiming to jointly design the content caching policy and the recommendation policy in order to achieve higher cache hit rates. The promising gains in the caching efficiency (which comes "for free" from a technology point of view, e.g., without extra investment in equipment or new communication technologies) demonstrated by these early works, motivated more work on the topic [7], [8], [10]- [13], [16]- [22], [44], [46]. The majority of related works considers cache-aware recommendations in mobile edge caching [4]- [13] to improve the cache hit ratio by optimizing the recommendation and/or caching policies.…”

“…A recently proposed solution for increasing the efficiency of MEC is jointly considering caching and recommending content [4]- [13]. Recommendation Systems (RS) are integrated in many popular services (e.g., YouTube, Netflix) and significantly affect the user demand [14], [15].…”

“…Previous works generalize the above example [4]- [13], [16]- [22], by considering more general recommendation techniques and parameters (e.g., number and order of recommendations [19]), more general content delivery schemes (e.g., multiple caches [6], [20], broadcasting [16], [17], [22]), and more general user demand models (e.g., acceptance of recommendations [18], sequential requests [7]).…”