[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14]

Journal of Information Science and Engineering, Vol. 37 No. 4, pp. 935-958

Mining Influential Who-to-Post and When-to-Post Curators on Social Networks

Department of Statistics
Institute of Data Science
National Cheng Kung University
Tainan, 701 Taiwan
E-mail: cchsia@gs.ncku.edu.tw; chengte@mail.ncku.edu.tw

Curators on the social networking sites become prominent and indispensable nowadays. Gradually, they come to be the voice in the business's online marketing field. The problem that how to find the most future-influential curators and plan the best posting time for them, notwithstanding, has been hidden and under-explored as yet. In this study, we initiate to analyze this problem with those two primary concerns from four distinct dimensions. To find the most future-influential curators, we consider this problem from the following two dimensions, Future Influence Ranking Prediction and Future Influential Leader Prediction. To plan the best posting time for the curators, similarly, we consider this part with two dimensions below, Accumulated Influence Post-time Scheduling and Limited Influence Post-time Scheduling. We aim at predicting the future influential curator with a series of basic and advanced self-defined features. Based on network embedding, we add learned features to capture the connection between users. To deal with the problem, we implement Learning-to-Rank algorithm and two newly devised ones, self-training algorithm and mutual-training algorithm, which are served to become the solution to the imbalanced data. With the experiments on large-scale Facebook data, we find the proposed methods significantly outperform the conventional prediction settings. The F1 score in predicting the most future influential curators can be up to 0.875. Also, in the part of planning the best posting time, the result shows in comparison with the overall performance of curators, the limited influence of the curators in our planned time can be boosted up to three times.

Keywords: social network analysis, feature engineering, influence prediction, node embedding, when-to-post scheduling

  Retrieve PDF document (JISE_202104_12.pdf)