2015 IEEE / WIC / ACM International Conference on Web Intelligence and Intelligent Agent Technology (WI-IAT) (2015)
Dec. 6, 2015 to Dec. 9, 2015
Twitter has been massively adopted by millions users and has become an essential tool for the information dissemination and to collect timely information on news and events. Due to the large volume of tweets, users need proper tools such as recommendation systems to quickly identify the most valuable contents among the information stream. However, traditional recommendation systems that analyze data and periodically update models cannot follow the highly dynamic nature of events and topics on Twitter. This limitation appeals for real-time recommendation systems which pose great challenges. State-of-the art recommendation systems which can be performed in real-time rely on centralized servers or cluster of nodes to support the cost of the recommendation operations, and possibly raise scalability issues when the amount of data increases. In this work, we explore another direction by leveraging the browser of users to compute real-time personalized recommendation at the edge of the network on user machines. In this demonstration, we present MYSTREAM, a mobile application to follow events from Twitter. MYSTREAM leverages the power of Twitter to have fresh and real time information related to an event or a topic identified through its hashtag. To help users to discover relevant contents through the associated stream of tweets, MYSTREAM implements a modular recommendation engine. To capture the highly dynamic nature of exchanges on Twitter, MYSTREAM uses stream processing algorithms to identify the relevant contents in real-time. In addition, to provide a user-centric service while tackling the scalability issue inherent to recommendation systems, MYSTREAM performs the computations dedicated to recommendations in the browser of users while the server forwards the stream of tweets related to the desired event to the users' machines. On the client side, MYSTREAM relies on web technologies to perform local computation totally transparently from users (i.e. before to present the data on the interface). The recommendation engine of MYSTREAM is highly modular. Each module provides a specific filtering scheme that processes the stream of tweets to highlight a specific content, from the computation of the most popular tweets or pictures to more complex personalized recommendation operations which highlight interesting tweets over time. Moreover, users assemble these modules to build a personalized dashboard to follow events. Users can also create their own recommendation modules to capture specific contents from the data stream. Finally, users can promote contents to create a personal journal on the event. In this demonstration, we present the capabilities of MYSTREAM to effectively help users to follow events from Twitter. This work conveys the feasibility of a real-time recommendation system exploiting the browser of each user to perform the personalized recommendation process. Real traces from the Twitter activity of the 2014 New York City Marathon will be replayed to allow attendees to follow this event through MYSTREAM. We show that MYSTREAM quickly identifies valuable contents in real-time from the data stream. Moreover, from a system perspective, we show that MYSTREAM remains lightweight on the clients' smartphone.
Twitter, Real-time systems, Browsers, Servers, Engines, Tagging, Yttrium
A. Boutet, F. Laforest, S. Frenot and D. Reimert, "MYSTREAM: An in Browser Personalization Service to Follow Events from Twitter," 2015 IEEE / WIC / ACM International Conference on Web Intelligence and Intelligent Agent Technology (WI-IAT), Singapore, Singapore, 2015, pp. 247-248.