Targeting Online Communities to Maximise Information Diffusion

Belák, V., Lam, S. & Hayes, C., 2012. Targeting online communities to maximise information diffusion. In Proceedings of the 21st international conference companion on World Wide Web. WWW ™12 Companion. New York, NY, USA: ACM, pp. 1153—1160.

Introduction

The authors investigate how to maximize information flows across communities and introduce a novel measure for cross-community influence.

1. users are continuously flooded with news, and may , therefore, become increasingly inert to further stimuli
2. there are specific hight influence communities that engage a substantial part of the system. The paper shows how to identify these communities.
3. impact defines how likely a user will respond to a certain stimuli. These responses are measured in terms of replies.

Related Work

1. spread of diseases in a population
2. spread of influence across a social network
3. cross-posts in USENET
4. Wu et al. use a reply-to network as a proxy for information flow. Such reply-to relations have similar properties than friendship relations within social networks.

The Independent Cascade Model (ICM) for Diffusion

1. introduced by Kempe et al. (2003)
2. similar to spreading activation:
• Graph: G=(V,E)
• Vertices (V) refer to actors and have a certain (probabilistic) activation function
• Edges (E) refer to the probability of an information flow between an actor i and an actor j.
• the diffusion process starts with seed nodes that try to activate their neighbors at every iteration

Method

Experimental setup

1. the method has been applied to the boards.ie forum based on 51 temporal snapshots that have been taken within a year
2. an experiment has let to a wait time $$\tau$$ = 1 week between original post and reply (84% of all responding users responded to a post within one week)
3. the edge's weights reflect the likelihood of a flow between user i and j (reply from j to i divided by the total number of replies from j).
4. all measures have been computed based on sets of sample users per community

Influence Measures

1. impact measure considers
• a user's degree of membership, and
• its centrality within a community
2. the authors introduce
• impact (centrality * membership)
• normalized impact that normalizes this measure with the community size
• total impact (subtracts a communities self-impact values)
• entropy (how many distinct communities are influenced by the seed community)