PROJECT TITLE :

On Summarization and Timeline Generation for Evolutionary Tweet Streams

ABSTRACT:

Tweets and other short-text messages are being created and disseminated at an unprecedented rate. While tweets are instructive in their raw form, they may also be overwhelming. Plowing through millions of tweets, which contain tremendous amounts of noise and duplication, is a pain for both end-users and data analysts. We offer Sumblr, a new continuous summarization framework, to solve the problem in this research. Sumblr is designed to deal with dynamic, rapidly arriving, and large-scale tweet streams, as opposed to typical Document Summarization algorithms that focus on static and small-scale data sets. There are three important components to our proposed framework. To begin, we present an online tweet stream Clustering algorithm for clustering tweets and storing distilled statistics in a tweet cluster vector data structure (TCV). Second, we create a TCV-Rank summarization technique that can be used to generate online summaries and historical summaries for any time period. Third, we develop a subject evolution identification method that uses summary-based/volume-based variations to automatically generate timelines from tweet streams. Our studies on large-scale real tweets show that our system is efficient and effective.