Influencer Ranking Framework Using TH-DCNN for influence maximization

Abstract

As the influencer gains more significance in social media marketing, companies raise their budgets for influencer campaigns. With business increasing day by day, finding efficient influencers is becoming the most prominent factor for success, but choosing the right influencer from these social media users is quite a challenge. This manuscript proposes a novel method to rank influencers by their effectiveness based on their posting behavior and social relations over time. Initially, the data from Twitter is collected from the Indian politics tweets and reactions dataset. This raw data undergoes preprocessing using various techniques including, tokenization, stemming, lemmatization, stop word removal, and data normalization using the Min-Max normalization approach to ensure the data is relevant and suitable format for analysis. Next, construct a heterogeneous network to represent the complex interactions between entities like users, tweets, hashtags, and mentions. Then Tree Hierarchical Deep Convolutional Neural Network (TH-DCNN) is applied to these networks to derive information representation for each influencer at each period. Finally, a Cosine similarity (CS) is used to learn from the network and predict the influencer rankings. The performance metrics such as accuracy, f1-score, mean average precision (MAP), Normalized Discounted Cumulative Gain (NDCG), Receiver Operating characteristic (ROC), Mean Reciprocal Rank (MRR), and Hit Rate are analyzed in experimental evaluations. The proposed method improved the accuracy compared with existing techniques.