Start Date
16-8-2018 12:00 AM
Description
Recently, live streaming, as a primary social media trend, has become more and more commonplace in businesses’ social media marketing. For live streaming platforms, how to keep viewers stay is always the central issue. To better predict viewer’s behaviors, we study a watching duration prediction problem in this paper. Different from prior literature, we take interactions between viewers and anchors into consideration and develop an integrated modeling framework for the prediction task. The proposed interaction-aware model combines probabilistic matrix factorization model with deep learning model to extract useful features from interactions for collaborative prediction. To the best of our knowledge, our study is the first one to analyze interactive behaviors for video watching prediction. Comprehensive experiments have been conducted on a real-world dataset to evaluate our predictive model. Experimental results demonstrate that our new model significantly outperforms baselines in watching duration prediction.
Recommended Citation
Chen, Jiawei; Liu, Hongyan; He, Jun; and Han, Sanpu, "Interaction-Aware Watching Duration Prediction on Live Streaming Platforms" (2018). AMCIS 2018 Proceedings. 10.
https://aisel.aisnet.org/amcis2018/DataScience/Presentations/10
Interaction-Aware Watching Duration Prediction on Live Streaming Platforms
Recently, live streaming, as a primary social media trend, has become more and more commonplace in businesses’ social media marketing. For live streaming platforms, how to keep viewers stay is always the central issue. To better predict viewer’s behaviors, we study a watching duration prediction problem in this paper. Different from prior literature, we take interactions between viewers and anchors into consideration and develop an integrated modeling framework for the prediction task. The proposed interaction-aware model combines probabilistic matrix factorization model with deep learning model to extract useful features from interactions for collaborative prediction. To the best of our knowledge, our study is the first one to analyze interactive behaviors for video watching prediction. Comprehensive experiments have been conducted on a real-world dataset to evaluate our predictive model. Experimental results demonstrate that our new model significantly outperforms baselines in watching duration prediction.