Topic aware fair representation for finding influential members in social network

Influence maximization (IM) [1] aims to select a few key individuals to maximize the spread of influence.For instance, in Meetup, a popular Event-Based Social Network (EBSN), users transition from online interactions to real-world events.An individual who influences others to attend sports-related events may not be effective for business-related events, highlighting the importance of considering event topics when selecting influencers.Thus, when recommending influencers, it's crucial to also recommend event topics or badges where these influencers will be most effective.However, this approach has limitations.For example, if the recommended badges are Women Professional, Business, the percentage of the influenced population who align with 'Women Professional' may be much lower than the overall proportion of Meetup members.This underrepresentation could lead to lower event attendance.Therefore, it's essential not only to recommend the top influencers and badges, but also to ensure fairness in the influenced population with respect to these badges, as they are not fixed like traditional sensitive attributes.In this paper we recommend Fair2vec, which computes the badge wise embedding of influencers and followers and thus subsequently recommending (i) top-k influential members , and (ii) top-r influence badges , such that becomes most effective for topics in , to maximize the total influenced population and also ensuring that (iii) the influenced population are fair with respect to the recommended badges.Fair2vec mainly consists of 3 parts.We evaluate the performance of Fair2vec on two real-world datasets and demonstrate that Fair2vec outperforms the baselines in terms of the volume of the influenced population and fairness.