Exploiting Live Feedback for Tweet Real-time Push Notifications

Abstract

Twitter has been developed as an immense information creation and sharing network through which users post diverse information. Although a user would regularly check her Twitter timeline to stay up-to-date on her topics of interest, it is impossible to survive with manual topic tracking techniques while tackling the challenges that emerge from the Twitter timeline nature. Among these challenges are the big volume of posted tweets, noise (e.g., spam), redundant information (e.g., retweets), and the rapid development of topics over time. This necessitates the development of real-time summarization (RTS) systems that automatically track predefined topics of interest and summarize the stream while considering the relevance, novelty, and freshness of the selected tweets. We tackle this problem as part of Qatar University's participation in TREC-2017 Real-Time Summarization (RTS) track. Our RTS system adopts a light-weight and conservative filtering strategy that monitors the continuous stream of tweets over a pipeline of multiple phases including pre-qualification, preprocessing, indexing, relevance filtering, novelty filtering, and tweets nomination. The system also exploits life (explicit) feedback to update profiles and pushing criteria (e.g., relevance threshold). The experimental results show that the runs that exploit the live explicit feedback exhibited a better performance in comparison to the baseline run that has been the best (among our runs) for the last two years. Additionally, all submitted runs have scored above the median provided by the track organizers in the batch evaluation.