Segment-wise Users' Response Prediction based on Activity Traces in Online Social Networks

Users on the social network have both different levels of involvement and preferred topics. In this paper, we try to understand how the behavior of users in different activity segments varies and how predictable the behavior of segments is. To identify types of user behavior, we used a modified RFD approach (Recency, Frequency, Duration), which dynamically determines the number and activity of users in different segments in one community based on previous activity. To analyze the reactions to posts, we used topic modeling and post tonality analysis. We proposed a method for the response prediction to posts based on the prediction for separate activity segments and combining results. This method gives more accurate results than the general forecast for all subscribers in the community allows for each type of user to determine the most important characteristics of the post that affect the likelihood of a reaction. This allows you to determine the behavior of different segments of users depends on the activity.