Twitter is Faster

Zhengyu Deng, Ming Yan, Jitao Sang, Changsheng Xu
2015 ACM Transactions on Multimedia Computing, Communications, and Applications (TOMCCAP)  
Traditional personalized video recommendation methods focus on utilizing user profile or user history behaviors to model user interests, which follows a static strategy and fails to capture the swift shift of the short-term interests of users. According to our cross-platform data analysis, the information emergence and propagation is faster in social textual stream-based platforms than that in multimedia sharing platforms at micro user level. Inspired by this, we propose a dynamic user modeling
more » ... strategy to tackle personalized video recommendation issue in the multimedia sharing platform YouTube, by transferring knowledge from the social textual stream-based platform Twitter. In particular, the cross-platform video recommendation strategy is divided into two steps: (1) Real-time hot topic detection: the hot topics that users are currently following are extracted from users' tweets, which are utilized to obtain the related videos in YouTube. (2) Time-aware video recommendation: for the target user in YouTube, the obtained videos are ranked by considering user profile in YouTube, time factor and quality factor to generate the final recommendation list. In this way, the short-term (hot topics) and long-term (user profile) interests of users are jointly considered. Carefully designed experiments have demonstrated the advantages of the proposed method. tremendous data become extremely difficult. The personalized services, including personalized search, subscription, recommendation, etc., stand out for solution and play a vital role in tackling the issue of information overload [Sang and Xu 2012] [Gao et al. 2013 ]. Most of the traditional personalized video recommendation methods are devoted to static user modeling, which utilizes the user profile or the history behaviors to understand the long-term interest of the user. However, user interest often distributes dynamically, which differs from time to time. Especially, when surrounded by the tremendous fresh messages every day, user's short-term interest may change continuously with the current hot events 3 . For example, the fact that a user has read the news about "US presidential election" may lead to the consequent action of searching for videos on "US presidential election debate" to gain further details about this event. In this case, the user may not really have great interest in politics from the long-term perspective, but his/her short-term interest is largely influenced by the popularly acquired information around him/her. Therefore, the personalized video recommendation strategies which cannot capture the swift drift of the user interest will fail to push the timely videos to desired users. Existing personalized video recommendation work referring to the short-term interests of users mainly focuses on the single multimedia sharing platform itself. The limitations of the short-term interest extraction in single multimedia sharing platform are as follows. 1) Firstly, the users' behaviors in single platform are often limited, resulting in that it is difficult to exactly capture the swift drift of user interest. 2) Besides, recommendation based on the users' short-term interests inferred from their current behaviors in the same platform may make the recommendation always lags behind users' actual behaviors, leading to duplicated recommendation. 3) Moreover, information emergence and propagation in multimedia sharing platforms is slower than that in social textual stream platforms and users' short-term interests extracted from multimedia sharing platforms are less time-aware. Therefore, we propose to enrich users' short-term interests by introducing social textual stream platforms and we are dedicated to investigating whether a user has consistent behaviors between different platforms and their temporal relations. Notably, social textual stream-based platforms (such as Twitter, Weibo) are widely regarded as a source of realtime breaking news, and information emergence and propagation is faster in these platforms than multimedia sharing platforms (such as YouTube, Flicker). It was reported that the news about "Virginia earthquake" appeared in Twitter almost at the same time when the earthquake happened, and it propagated throughout America in the following five minutes even faster than the earthquake waves 4 . Existing work analyzed the temporal patterns of user behaviors between different platforms on global level. In this paper, we are interested in investigating whether there is a consistent conclusion on a micro user level, e.g., for a specific user, is there any activity pattern that he/she has come across a piece of news on Twitter before they search the related videos on YouTube? To answer this question and explore the temporal characteristics of different platforms on user level, we further investigate into the temporal patterns across different platforms based on each single user and find that information emergence and propagation is also faster in social textual stream-based platforms than multimedia sharing platforms on user level. From the perspective of information inquiry, it is highly possible that users have come across a piece of news in Twitter before they search the related videos in YouTube. In other words, if we know which topic a user is following currently in social textual stream platforms, we can recommend the relevant videos to him/her on YouTube to help get deeper insight into this topic. Enlightened by this, we designed a personalized time-aware video recommendation solution for the multimedia sharing platforms by exploiting users' activities in social textual stream platforms to capture users' short-term interests. In this paper, we address the time-aware personalized video recommendation issue by cross-platform collaboration from Twitter to YouTube: we use YouTube as the video sharing platform to perform the recommendation task, and Twitter as the social textual stream platform to extract the real-time hot topics users followed. We first conduct a crossplatform data analysis to examine the evolution of topics between Twitter and YouTube and conclude that information propagation in Twitter is faster than that in YouTube on both global level and user level. Based on this observation, we 3 Hot event is defined as a subject discussed and shared frequently in many documents and platforms. Examples are like "Olympic opening ceremony 2012", "US election day 2012", "Super bowl game", etc. In this paper, hot event is equivalent to hot topic. 4
doi:10.1145/2637285 fatcat:3qcvb2vcxjebvetmjx3d6kaj54