Event Recommendations in Social Networks


Student thesis: Doctoral Thesis

View graph of relations

Related Research Unit(s)


Awarding Institution
Award date6 Sep 2018


Technological advancement and increasing use of smartphones paved the way for event-based social networks (EBSNs) (e.g., Meetup.com) while the application of recommender systems leads to its major breakthrough because they help users to discover events that align with their preferences from a pool of events available. In an EBSN, users can join various groups and any member of a group can create physical events (e.g., party, boat trip, and hiking) and invites other members of the group. However, an invitation to members to participate in an event does not necessarily guarantee participation in the event except it serves their interests. Users preferences of events in an EBSN are reflected in their historical records of participated events, which are available in the community-contributed data. Some of the information revealed in the community-contributed data include various groups that users belong to, events' information such as geographical locations, category and time of events and users' response to invitations (i.e., RSVPs information). In this thesis, user preferences such as geographical, categorical, social, and temporal preferences are exploited for event personalized recommendations.

First, real-world examples are used to show that a user's behavior is unique through data analysis and this thesis illustrates how the geographical, categorical, and temporal influences can significantly affect users' preferences in EBSNs. Specifically, some users are selected and behavioral analyses are conducted on them. Secondly, this thesis proposes EventRec and SoCaST which investigate users' historical records of events' participation and model their preferences for personalized event recommendations. To this end, four major influences are modeled based on users' preferences that are implied in the community-contributed data, they include, the geographical influence modeled by utilizing an adaptive Kernel Density Estimation (KDE) on personalized two-dimensional geographical location, categorical influence modeled based on the relevance of an event's category to a user and its popularity, social influence modeled as the relevance of a group to a user and her friends, and the temporal influence which is modeled through the KDE method to generate event recommendations.

Finally, this thesis investigates the problem of finding personalized weights for users' preferences which can be used for improving personalized recommendations and proposes a personalized event recommendation framework called SoCaST* that employs multi-criteria decision making (MCDM) approach to evaluate and rank events. In SoCaST*, a user's preference models are built to compute geographical, categorical, social and temporal influences and a personalized weight is estimated for each criterion (i.e., each influence). By utilizing the personalized criterion's weight, dominance intensity measures (i.e., dominating and dominated measures) are computed for alternatives of each criterion and the set of alternatives are ranked based on the estimated dominance intensity measures to recommend top-k events.

The experimental results on real-world data sets crawled from Meetup.com data sets show that the proposed event recommendation techniques achieve significantly superior recommendation performance compared to other existing event recommendation methods.