Effective influence estimation in twitter using temporal, profile, structural and interaction characteristics

Influence diffusion is extensively studied in social networks for product or service promotion and viral-marketing applications. This paper proposes two models for social influence estimation, namely Time Decay Features Cascade Model (TDF-C) and Time Decay Features Cascade Threshold Model (TDF-CT). These models overcome three main existing challenges - first, measure the strength of user's influence as an influencer; second, identify the set of users influenced by an influencer; third, estimate the time frame of the influence. TDF-C is an M-TAP based diffusion model, which learns influence probabilities between users using four types of features, namely temporal, interaction, structural, and profile features, and uses Independent Cascade (IC) model for influence estimation. TDF-CT is an extension of the TDF-C model, which uses temporal and interaction features to calculate the diffusion through the Progressive Feedback Estimation (PFE) model in place of IC model. PFE model is a fusion of two diffusion models, i.e., Linear Threshold (LT) and Independent Cascade. TDF-CT handles the limitations of the contemporary diffusion models, i.e., IC and LT. The efficacy of proposed models is evaluated with respect to existing models Independent Cascade (IC), Time Constant Cascade (TC-C), Time Decay Cascade (TD-C), and Time-Depth Decay Cascade (TDD-C). Experimental evaluation over two benchmark datasets namely Darwin and MelCup17 reveal that proposed models are able to make the predictions very close to the real-time in a given time frame. TDF-CT and TDF-C are most suitable for applications requiring high precision and high recall, respectively. Results of spread shape establish the efficacy of models to spread the influence with good coverage of the social network. Results are obtained with improved accuracy by up to 39%.