Enhancing Local Live Tweet Stream to Detect News

Twitter captures invaluable information about real-world news, spanning a wide scale from large national/international stories like a presidential election to small local stories such as a local farmers market. Detecting and extracting small news for a local place is a challenging problem and the focus of this work. The main challenge lies in identifying these small stories that correspond to a local area of interest, which are typically harder to detect compared to national stories in the sense that there may be just a handful of tweets about a local story. A system, called Firefly, is proposed that overcomes the data sparsity and captures thousands of local stories per day from a metropolitan area (e.g., Boston). The key idea lies in combining the enhancement of a local live tweet stream in Twitter, the identification of "locality-aware" keywords, and using these keywords to cluster tweets. Experiments show that the proposed system has a significantly higher recall over a set of representative local news agencies, and at the same time, outperforms the baseline approach TwitterStand. More importantly, the results also demonstrate that our system, by utilizing the enhanced local live tweet stream, discovers much more local news than the methods working only on geotagged tweets, i.e., those with embedded GPS coordinate values.