Tracking influence and opinions in social media


Monday, November 13, 2006, 12:00pm - Monday, November 13, 2006, 2:00am

325b ITE

blog, influence, opinion extraction, social media

Recently, social media such as forums, wikis and blogs, in particular, are playing a notable role in influencing the buying patterns of consumers. Often a person looks for opinions, user experiences and reviews on such sources before purchasing a product. Detecting influential nodes, opinion leaders and understanding their role in how people perceive and adopt a product or service provides a powerful tool for marketing, advertising and business intelligence. This requires new algorithms that build on social network analysis, community detection and opinion extraction.

We propose to study and characterize influence on the Blogosphere by combining many contributing factors, including topic, social structure, opinions, biases and time. Studies on influence in social networks and collaboration graphs have considered a static view of the network and are based purely on link analysis. However, influence on the Web is often a function of topic. We propose the notion of `topical influence' and extend existing techniques to make them topic sensitive. An important component in understanding influence is to detect sentiment and opinions. Changes in opinions, aggregated over many users, can be a predictor for an interesting trend in a community. We describe BlogVox, a testbed blog analytics system that we developed for TREC opinion retrieval task. This system finds opinionated blog posts about a topic. We propose to extend this system to detect bias and to aggregate opinions across communities. Finally, we propose to model influence as a temporal phenomenon. The Blogosphere, being a buzzy and dynamic environment, has new topics emerging constantly and blogs rising and falling in popularity. Tracking these changes over time allows us to find blogs that are influential versus something that is just briefly popular.

We will develop, implement and experimentally evaluate such a model to demonstrate its improved accuracy over models based on any one of these factors alone.

OWL Tweet

UMBC ebiquity