Deep Learning-based Short Video Recommendation and Prefetching for - PowerPoint PPT Presentation

口令： RAPID708 Deep Learning-based Short Video Recommendation and Prefetching for Mobile Commuting Users Qian Li 1 , Yuan Zhang 1 , Hong Huang 2 , Jinyao Yan 1 1 Communication University of China 2 Huazhong University of Science and Technology 2 0 1 9 . 8 1 BBS1113

Short Video Applications  What is short video application?  User-generated short video clips  Usually <1min  Vlog, advertising, self-media, etc.  Can only be watched online  Rapid growth of short video applications 1  Number of short video users is 648 million in China by the end of 2018  Number of daily active users is growing at a speed of > 800% each year since 2016 1 The 43rd statistical report on the development of Internet in China.2018 2 BBS1113

Commuting Scenario Watching short videos using public transportation (bus, subway, etc.) during way to work In a survey 1 with 190 respondents, over 46% reported watching short videos on their daily commuting time by public transport service High moving speed, unstable network connection 2 Over 71% out of the 46% respondents reported having disconnections when watching short videos on public transportation Regular, predictable 1 https://www.wjx.cn/m/37307561.aspx 2 H. Deng, et. al., Mobility Support in Cellular Networks: A Measurement Study on Its Configurations and Implications. IMC '18 3 BBS1113

Problem and Objective User QoE Watching the preferred video Watching it in time Accurately recommend the Accurately prefetch the recommended interested video clips to the user video clips to the BS where the user would connect to 4 BBS1113

Overall system architecture 5 BBS1113

Short video recommendation Google Inception: YouTube-8M: A Large-Scale Video Classification Benchmark. CVPR’16. 6 BBS1113

Mobility prediction LSTM model Mobility prediction structure 7 BBS1113

Evaluation: Dataset and Preprocessing User mobility trace: 5,000 users event ‐ driven trace from one of the largest ISPs in China for one week Contains start time, user ID, downlink speed, cell ID etc. Short video dataset: Crawled from Douyin of 78170 records from 233 users Includes user ID, avatar, nickname, location, constellation; video ID, release time, associate user ID, # of likes\comments\forwardings\shares and video file 70% for training, the rest 30% for testing Sample balancing: Classify short videos using YOLO3 with COCO coefficients Obtain user preference of types according to her mark Randomly extract unmarked video from the least liked types as negative examples BBS1113 8

Results Performance of short video recommendation using different Recommendation results using different neural networks layers loss function and number of nodes Layers Acc F1 Acc F1 0.69 3 0.817 0.664 Mean absolute error 0.798 0.698 4 Loss function 0.822 0.667 Cross entropy 0.808 (nodes=50) 0.704 5 0.826 0.688 Mean square error 0.815 0.641 6 0.77 0.67 10 0.808 0.61 7 0.75 0.69 30 0.817 0.67 8 0.8 Nodes 0.664 50 0.798 (Loss function 0.68 128 0.814 = MAE) 0.65 160 0.791 0.66 256 0.796 9 BBS1113

Results Comparison to original Inception structure w.r.t. computation time Comparison to original Inception structure w.r.t. accuracy PCAdrop PCA Inception PCAdrop PCA Inception accuracy for one short video Average recommendation Average computation time for 0.73 100 0.71 80 one short video (ms) 0.69 60 0.67 40 0.65 0.63 20 0.61 0 0.59 3 4 5 6 7 8 3 4 5 6 7 8 Number of layers Number of layers 10 BBS1113

Results User mobility prediction using different sequences Mobility prediction using different layers 0.8 1500 0.8 1500 0.6 Accuracy 0.6 1000 Time (ms) 1000 Time (ms) Accuracy 0.4 0.4 accuracy 500 accuracy 500 0.2 0.2 Time Time 0 0 0 0 5 7 9 11 13 15 17 19 21 23 25 27 29 5 7 9 11 13 15 17 19 21 23 25 27 29 # of layers Sequences 11 BBS1113

Conclusion Conclusion  A two-stage recommendation and prefetching scheme for short video application in mobile commuting scenario  Trace-driven analysis to evaluate the accuracy and efficiency of the proposed approach 12 BBS1113

Thank you for your time! Q&A 13 BBS1113

Related work Deep Neural Networks for YouTube Recommendations ， RecSys’ 16 14 BBS1113

Results User mobility prediction using different sequences 0.8 1500 0.6 1000 Time (ms) Accuracy Comparison of user mobility prediction algorithms 0.4 accuracy 500 0.2 Time 1.1 0 0 1 GRU 5 7 9 11 13 15 17 19 21 23 25 27 29 0.9 Accuracy LSTM 0.8 Sequences RNN 0.7 0.6 Mobility prediction using different layers 0.5 0.4 0.8 1500 0 255 510 765 1020 1275 1530 1785 2040 2295 2550 2805 3060 0.6 Accuracy 1000 Time (ms) 0.4 Users accuracy 500 0.2 Time 0 0 5 7 9 11 13 15 17 19 21 23 25 27 29 Training times 15 BBS1113

Feasibility Assessment We test the network condition in commuting case for both subway and bus. In our experiment, the network condition is very unstable. the highest download rate can reach 7.5Mbps the lowest is only 4.8Mbps, and the average download rate is 5.15 Mbps The maximum size of the crawled short video files is 5.2MB, and is on average 2.5MB While the crawled video length is between 7.2s to 25.3s therefore, in the worst case, the overall recommendation and transmission time of the short videos is 12ms+5.2*8/4.8s=8.7s It means that for the worst case, the user would wait for 8.7s-7.2s=1.5s to get her preferred short video clips using our prefetching scheme, which is quite tolerant compared to the 8.7s waiting time without prefetching. 16 BBS1113