Finding topical experts in Twitter via query-dependent personalized PageRank

Abstract

Finding topical experts on micro-blogging sites, such as Twitter, is an essential information-seeking task. In this paper, we introduce an expert-finding algorithm for Twitter, which can be generalized to find topical experts in any social network with endorsement features. Our approach combines traditional link analysis with text mining. It relies on crowd-sourced data from Twitter lists to build a labeled directed graph called the endorsement graph, which captures topical expertise as perceived by users. Given a text query, our algorithm uses a dynamic topic-sensitive weighting scheme, which sets the weights on the edges of the graph. Then, it uses an improved version of query-dependent PageRank to find important nodes in the graph, which correspond to topical experts. In addition, we address the scalability and performance issues posed by large social networks by pruning the input graph via a focused-crawling algorithm. Extensive evaluation on a number of different topics demonstrates that the proposed approach significantly improves on query-dependent PageRank, outperforms the current publicly-known state-of-the-art methods, and is competitive with Twitter's own search system, while using less than 0.05% of all Twitter accounts.