Multi-interest User Profiling in Short Text Microblogs.

Discourse on short text platforms like Twitter shapes the design of underlying knowledge-based recommendation engines. The resulting recommendations are powered by user connections as social network nodes as well as with shared interests. Twitter as a platform provides a complex mesh of users’ interest levels where some users tend to consume certain topical content to a lesser or greater extent. This consumption of content is usually considered a defining factor in curation of their online identity. Our aim in this paper is to quantify the multi-interests of users based on the tweets they disseminate. We do this by (i) representing all tweets as vectors for computations (ii) generating cluster centroids representative of the topics of interest. (iii) computing a responsibility matrix to depict their interest levels in the topics (iv) aggregating intra-user interest levels to define the user’s multi-topic affinities . We use a Twitter dataset geolocated to Kenya to validate users’ intra-topical interests. Our experimental results demonstrate the effectiveness of our approach in terms of capturing their multi-interests and in turn generate their multi-topic interest profiles.