Deriving Interpersonal Role Identities from Social Network Interactions Md. Saddam Hossain Mukta Bangladesh University of Engineering & Technology (BUET) and United International University (UIU) saddam@cse.uiu.ac.bd Mohammed Eunus Ali Bangladesh University of Engineering & Technology (BUET) eunus@cse.buet.ac.bd Jalal Mahmud IBM Research-Almaden jumahmud@us.ibm.com ABSTRACT In recent years, social networking sites (SNS) have been trans- formed into virtual societies where users express their feelings, share opinions, and socialize with friends, families, and co-workers. Users in these sites are still connected with each other with so called friends/followers relationships which are not representative of the real life role identities. However, a single person can play multiple roles in a society. For example, a person has a family mem- ber role identity with his wife, professional member role identity with his colleagues and academic member role identity with his class fellows. Thus, the same person can interact diferently with diferent people based on her role identity with her counterpart. In this paper, we have predicted interpersonal role identities (e.g., family members, academic members, professional members, friends, and acquaintances) of a user with other connected members in an egocentric network (e.g., Facebook) from their word use patterns during interactions. We have proposed a weighted hybrid machine learning based model to predict the role identities from users’ word usage patterns. We have also validated our experiment results by using the datasets of both Facebook and Twitter. CCS CONCEPTS · Human-centered computing → Social network analysis; So- cial recommendation; · Computing methodologies → Classif- cation. KEYWORDS Social Networking Sites, Social Roles, Interactions, LIWC, Empath ACM Reference Format: Md. Saddam Hossain Mukta, Mohammed Eunus Ali, and Jalal Mahmud. 2019. Deriving Interpersonal Role Identities from Social Network Interactions. In 6th International Conference on Networking, Systems and Security (6th NSysS 2019), December 17ś19, 2019, Dhaka, Bangladesh. ACM, New York, NY, USA, 9 pages. https://doi.org/10.1145/3362966.3362974 Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for proft or commercial advantage and that copies bear this notice and the full citation on the frst page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specifc permission and/or a fee. Request permissions from permissions@acm.org. 6th NSysS 2019, December 17ś19, 2019, Dhaka, Bangladesh © 2019 Association for Computing Machinery. ACM ISBN 978-1-4503-7699-0/19/12. . . $15.00 https://doi.org/10.1145/3362966.3362974 1 INTRODUCTION A person plays multiple roles in a society, e.g., a person has a family member role with his wife, professional member role with his colleagues and academic member role with his class fellows. These roles are important features that characterize the behaving patterns among the members in a society. Depending on social identities and situations, the roles guide us to behave diferently with diferent people. For example, we respect teachers, care family members, share our feelings with friends and build positive relationship with colleagues [5]. In this paper, we are the frst to develop techniques to automatically derive a user’s role identities with other connected users from their psycholinguistic attributes obtained from their social media interactions. Social networking sites (SNS) have been transformed into virtual societies where users express their feelings, share opinions, and socialize with friends, families, and co-workers. Though SNS have put their eforts to make their sites as close as possible to the real life interactions, users in these SNS are still connected with each other with so called friends/followers relationships which are not representatives of the real life role identities (e.g., father and son are friends to each other in Facebook). However, we observe that interactions among members inside SNS vary depending on their role identities in real life. For example, we may post, comment, and tag a class fellow instantaneously with a sarcastic meme, whereas we are likely to give advice, share afection and talk about health concerns with a family member in Facebook [7]. Thus, it might be possible to identify individual role identities among the SNS members from their word usage patterns. A number of studies have been conducted to identify role identi- ties among the members in SNS. Authors in [29, 30, 32, 35] largely identify users’ role identities by analyzing features such as inter- action frequency and time, text, etc. None of these works identify the role identities between ego and alters by analyzing the patterns of word use of Facebook comments. Note that Facebook is an ego- centric social network where a person (an ego) is connected with other friends (the alters) [14]. In this paper, we primarily connect our work with the notion of sociology-diferent strokes from diferent folks [33]. Wellman and Wortley describe in their paper that people receive diferent social supports from their friends and relatives. They state that parent and adult children exchange fnancial and emotional aid; friends, neigh- bors, and sibling provide majority of the social supports, i.e., small and large services, companionship, and emotional aid. However, the authors also describe that not all community ties, i.e., relationships, provide similar kinds of support. The relationship and the kinds