Authors:
B. S. A. S. Rajita
;
Yaganti Vikas
;
Pritish Prashant Moharir
and
Subhrakanta Panda
Affiliation:
CSIS, BITS-Pilani Hyderabad Campus, Hyderabad, India
Keyword(s):
Social Networks, Influence Score, Community, Derived Feature, Event Prediction, ML Models.
Abstract:
In real-life social networks (SN), dynamic community evolution changes the structure of that network. Hence, a comprehensive framework is imperative for predicting community evolution, which this research refers to as an ’event’. This research studies how the influence of peer nodes in a social network often triggers community evolution. Therefore, this paper proposes calculating the communities’ new derived feature called Influence Score (IS) , to predict their events. Thus, it is imperative to compute the communities’ influence score (as a derived feature) and study its suitability for accurately predicting events using Machine Learning (ML) models. The experimental results show that derived features together with community features are more effective in predicting community events. The implementation and significance of the presented approach on the dataset show that IS, as an added feature, improved the accuracy of the ML models by approximately 6.6%. Additionally, it considerabl
y improved other parameters, including F-measure, recall, and precision. This paper also presents a comparative analysis with other derived features. It shows an improvement in the accuracy by approximately 1.5% and 0.8%. The results also indicate that the IS score improved the accuracy of the logistic regression by 2.53% compared to an existing similar approach. Thus, this paper infers that IS as a derived feature is considerably effective in improving the accuracy of ML models in predicting events in SN communities.
(More)