Conference Proceeding Article
Diffusion is an important dynamics that helps spreading information within an online social network. While there are already numerous models for single item diffusion, few have studied diffusion of multiple items, especially when items can interact with one another due to their inter-similarity. Moreover, the well-known homophily effect is rarely considered explicitly in the existing diffusion models. This work therefore fills this gap by proposing a novel model called Topic level Interaction Homophily Aware Diffusion (TIHAD) to include both latent factor level interaction among items and homophily factor in diffusion. The model determines item interaction based on latent factors and edge strengths based on homophily factor in the computation of social influence. An algorithm for training TIHAD model is also proposed. Our experiments on synthetic and real datasets show that: (a) homophily increases diffusion significantly, and (b) item interaction at topic level boosts diffusion among similar items. A case study on hashtag diffusion in Twitter also shows that TIHAD outperforms the baseline model in the hashtag adoption prediction task.
Computer Sciences | Databases and Information Systems
Data Management and Analytics
Machine Learning and Knowledge Discovery in Databases: European Conference, ECML PKDD 2015, Porto, Portugal, September 7-11, 2015, Proceedings, Part II
City or Country
LUU, Duc Minh and Ee-peng LIM.
Latent Factors Meet Homophily in Diffusion Modelling. (2015). Machine Learning and Knowledge Discovery in Databases: European Conference, ECML PKDD 2015, Porto, Portugal, September 7-11, 2015, Proceedings, Part II. 9285, 701-718. Research Collection School Of Information Systems.
Available at: http://ink.library.smu.edu.sg/sis_research/3108