TY - JOUR

T1 - Learning behavior rate models on social network data

AU - Toropova, Aleksandra V.

AU - Tulupyeva, Tatiana V.

N1 - Funding Information: The research was carried out in the framework of the project on state assignment SPIIRAS No. 0073-2019-0003, with financial support from the Russian Foundation for Basic Research, projects No. 19-37-90120, No. 18-01-00626 and No. 20-07-00839. Publisher Copyright: © 2020 Copyright for this paper by its authors. Copyright: Copyright 2020 Elsevier B.V., All rights reserved.

PY - 2020

Y1 - 2020

N2 - Intensity is one of the main characteristics of human behavior, using data about behavior intensity we can make high enough quality predictions about future human behavior. But it is often impossible to get a direct behavior rate, because of high cost, time consumption or restrictions for monitoring private lives, so we need tools to estimate it indirectly. We offer two models for behavior rate evaluation with expert-defined and learned structures. These models are Bayesian belief networks. They include information about the intervals in days between the last three behavior episodes of the study period, the minimum and maximum intervals between episodes, and the interval between the last episode of the study period and the next episode, respectively, after the end of the study period. As we need for the models approbation an example of behavior allowing us to get direct behavior rate, we take users' posting behavior in social network. For learning parameters and structure one of the models, testing models, data from the social network Vkontakte for December 2019 was collected. This data includes an information about posting on own users' "walls" for this month, i.e. posts quantity, time of last three posts, maximum and minimum time interval between posts for December 2019, and time of the first post starting from January 2020.

AB - Intensity is one of the main characteristics of human behavior, using data about behavior intensity we can make high enough quality predictions about future human behavior. But it is often impossible to get a direct behavior rate, because of high cost, time consumption or restrictions for monitoring private lives, so we need tools to estimate it indirectly. We offer two models for behavior rate evaluation with expert-defined and learned structures. These models are Bayesian belief networks. They include information about the intervals in days between the last three behavior episodes of the study period, the minimum and maximum intervals between episodes, and the interval between the last episode of the study period and the next episode, respectively, after the end of the study period. As we need for the models approbation an example of behavior allowing us to get direct behavior rate, we take users' posting behavior in social network. For learning parameters and structure one of the models, testing models, data from the social network Vkontakte for December 2019 was collected. This data includes an information about posting on own users' "walls" for this month, i.e. posts quantity, time of last three posts, maximum and minimum time interval between posts for December 2019, and time of the first post starting from January 2020.

UR - http://www.scopus.com/inward/record.url?scp=85092323049&partnerID=8YFLogxK

M3 - Conference article

AN - SCOPUS:85092323049

VL - 2648

SP - 200

EP - 209

JO - CEUR Workshop Proceedings

JF - CEUR Workshop Proceedings

SN - 1613-0073

T2 - 2020 "Russian Advances in Artificial Intelligence", RAAI 2020

Y2 - 10 October 2020 through 16 October 2020

ER -