In recent years social and news media have increasingly been used to explain patterns in disease activity and progression. Social media data, principally from the Twitter network, has been shown to correlate well with official disease case counts. This fact has been exploited to provide advance warning of outbreak detection, forecasting of disease levels and the ability to predict the likelihood of individuals developing symptoms. In this paper we introduce DEFENDER, a software system that integrates data from social and news media and incorporates algorithms for outbreak detection, situational awareness and forecasting. As part of this system we have developed a technique for creating a location network for any country or region based purely on Twitter data. We also present a disease nowcasting (forecasting the current but still unknown level) approach which leverages counts from multiple symptoms, which was found to improve the nowcasting accuracy by 37 percent over a model that used only previous case data. Finally we attempt to forecast future levels of symptom activity based on observed user movement on Twitter, finding a moderate gain of 5 percent over a time series forecasting model.

Influential agents in networks play a pivotal role in information diffusion. Influence may rise or fall quickly over time and thus capturing this evolution of influence is of benefit to a varied number of application domains such as: digital marketing, counter-terrorism or policing. In this paper we investigate the influence of users in programming communities on Twitter.We propose a new model for capturing both time-invariant influence and also temporal influence. The unified model is a combination of network topological methods and observation of influence-relevant events in the network. We provide an application of Hidden Markov Models (HMM) for capturing this effect on the network. There are many possible combinations of influence factors, hence we required a ground-truth for model configuration. We performed a primary survey of our population users to elicit their views on influential users. The survey allowed us to validate the results of our classifier. We introduce a novel reward-based transformation to the Viterbi path of the observed sequences which provides an overall ranking for users. Our results show an improvement in ranking accuracy over using solely topology-based methods for the particular area of interest we sampled. Utilising the evolutionary aspect of the HMM we attempt to predict future states using current evidence. Our prediction algorithm significantly outperforms a collection of naive models, especially in the short term (1-3 weeks).

No abstract

BackgroundTwitter updates now represent an enormous stream of information originating from a wide variety of formal and informal sources, much of which is relevant to real-world events. They can therefore be highly useful for event detection and situational awareness applications.ResultsIn this paper we apply customised filtering techniques to existing bio-surveillance algorithms to detect localised spikes in Twitter activity, showing that these correspond to real events with a high level of confidence. We then develop a methodology to automatically summarise these events, both by providing the tweets which best describe the event and by linking to highly relevant news articles. This news linkage is accomplished by identifying terms occurring more frequently in the event tweets than in a baseline of activity for the area concerned, and using these to search for news. We apply our methods to outbreaks of illness and events strongly affecting sentiment and are able to detect events verifiable by third party sources and produce high quality summaries.ConclusionsThis study demonstrates linking event detection from Twitter with relevant online news to provide situational awareness. This builds on the existing studies that focus on Twitter alone, showing that integrating information from multiple online sources can produce useful analysis.

The reconstruction of analysts' reasoning processes (reasoning provenance) during complex sensemaking tasks can support reflection and decision making. One potential approach to such reconstruction is to automatically infer reasoning from low-level user interaction logs. We explore a novel method for doing this using machine learning. Two user studies were conducted in which participants performed similar intelligence analysis tasks. In one study, participants used a standard web browser and word processor; in the other, they used a system called INVISQUE (Interactive Visual Search and Query Environment). Interaction logs were manually coded for cognitive actions based on captured think-aloud protocol and posttask interviews based on Klein, Phillips, Rall, and Pelusos's data/frame model of sensemaking as a conceptual framework. This analysis was then used to train an interaction frame mapper, which employed multiple machine learning models to learn relationships between the interaction logs and the codings. Our results show that, for one study at least, classification accuracy was significantly better than chance and compared reasonably to a reported manual provenance reconstruction method. We discuss our results in terms of variations in feature sets from the two studies and what this means for the development of the method for provenance capture and the evaluation of sensemaking systems.