Exploring acquaintances of social network site users for effective social event recommendations

https://doi.org/10.1016/j.ipl.2015.11.013Get rights and content

Highlights

  • •

    An effective social event recommendation method is developed.

  • •

    The content of the attended events are useful for social event recommendations.

  • •

    The experiments reveal interesting behavior patterns of social network site users.

Abstract

In this paper, we propose a social event recommendation method that exploits a user's social interaction relations and collaborative friendships to recommend events of interest. A challenge of the social event recommendation is that social events, such as online seminars and meet-ups, are one-and-only items. They are only valid for a short period of time and the ratings made by participants are not available until the events are over. Hence, recommending useful events to users is challenging. Instead of using ratings, we analyzed the behavior patterns of social network users to measure their social and collaborative friendships. The friendships were aggregated to identify the acquaintances of a target user and social events relevant to the preferences of the acquaintances and the user were recommended.

Introduction

Social network sites (SNSs) such as Facebook are popular Internet services that enable people to define their social networks and share information. Many SNSs now provide social event functions that enable users to create or attend a social event. Basically, a social event groups SNS users together to achieve a specific goal at a certain time. Examples of social events are online seminars and meet-ups. Social event functions invariably receive a great deal of attention because they enrich people's social relationships, which explains why more than 16 million social events are created on Facebook each month.1 While these functions enhance SNSs significantly, users often have difficulty finding social events of interest. This is because many social events created on SNSs keep popping up and most of these simply fade away quickly from users' attention. It is therefore essential for SNSs to provide users with appropriate social event recommendations.

Recommendation systems constitute a practical research topic in the field of e-commerce. Methods like collaborative filtering [20] and content-based filtering [16] have been developed to recommend items of interest to users. Specifically, collaborative filtering utilizes the item ratings made by users to compute the similarity between items (or users). This similarity is then used to predict a relevance score of an item for a target user, with items of high relevance scores being similar to the user's preferences and therefore being recommended. Content-based filtering, on the other hand, does not use the item ratings, but rather determines the similarity based on item descriptions. Items whose descriptions are similar to those rated or purchased by a user are considered relevant and are thus recommended to the user. Both methods have been used successfully by several e-commerce sites like Amazon. The social event recommendation problem is significantly different from the traditional recommendation problem for movies or products [12]. This is because social events are ephemeral and one-and-only items [6] that are only valid for a short time (i.e., several hours or few days). And since recommending a finished social event is meaningless, it is difficult for collaborative filtering methods to collect enough event attendance records and ratings to make effective recommendations. Further, content-based methods tend to recommend social events that are similar to those a user has attended, rather than new events that may be of interest to the user. Recommending useful social events for users is thus challenging [4], [6], [12].

In this paper, we propose a novel social event recommendation method. We analyze users' social interactions, event attendance records, and the content of the attended events to measure their social interaction relations and collaborative friendships. The friendships are aggregated to identify a target user's acquaintances. A social event that attracts many of the user's acquaintances is recommended if its topic is relevant to the user's preferences. To evaluate the proposed social event recommendation method, we collected 21,101 event attendance records from two real world social network sites, and examined the effect of the above friendship aggregation. The experiment results reveal interesting behavior patterns of SNS users. Moreover, the proposed method outperforms a number of well-known recommendation methods in terms of the coverage rate.

Section snippets

Related work

Two widely used social event recommendation approaches are content-based recommendation and collaborative filtering. Basically, collaborative filtering methods examine events' attendance records or ratings to recommend events useful to users. For instance, Klamma et al. [11] developed an academic event recommendation system which examines researchers' event participation networks to measure the similarity between their research interests. Then, researchers with similar interests are considered

The proposed social event recommendation method

We first define the symbols used in our social event recommendation method. Let U={u1,u2,…,uN} be a set of users in a social network site, and let E={e1,e2,…,eM} be all past social events. En denotes all the events that user un has ever attended such that En⊆E. Our objective is to recommend an upcoming social event ef that a target user is going to attend. The proposed social event recommendation method is comprised of two key components, acquaintance identification and recommendation generation

Dataset and performance metrics

To the best of our knowledge, there are no official evaluation datasets for the social event recommendation task because the research subject is relatively new. We therefore compiled our own datasets for the performance evaluations (see Table 1). We collected social events from Meetup2 and Facebook to evaluate our method's performance. Meetup, a well-known social network site that facilitates offline group meetings all over the world, provides a group function that helps

Concluding remarks

In this paper, we have developed a social event recommendation method that analyzes the behavior of SNS users to identify their social interaction relations and collaborative friendships. The friendships are then combined to identify a user's acquaintances so that upcoming events relevant to the preferences of the acquaintances and the user can be recommended. The results of experiments on two real-word datasets show that the proposed method is able to identify representative acquaintances of

Acknowledgements

This research was supported in part by MOST 103-2221-E-002-106-MY2 from the Ministry of Science and Technology, Republic of China.

References (21)

  • C.C. Chen et al.

    An effective recommendation method for cold start new users using trust and distrust networks

    Inf. Sci.

    (2013)
  • D. Bollen et al.

    Understanding choice overload in recommender systems

  • L. Candillier et al.

    Designing specific weighted similarity measures to improve collaborative filtering systems

  • C. Cornelis et al.

    A fuzzy relational approach to event recommendation

  • E.M. Daly et al.

    Effective event discovery: using location and social information for scoping event recommendations

  • S. Dooms et al.

    A user-centric evaluation of recommender algorithms for an event recommendation system

  • J.L. Herlocker et al.

    Evaluating collaborative filtering recommender systems

    ACM Trans. Inf. Syst.

    (2004)
  • T. Hofmann

    Probabilistic latent semantic indexing

  • M. Kayaalp et al.

    A collaborative and content based event recommendation system integrated with data collection scrapers and services at a social networking site

  • H. Khrouf et al.

    Hybrid event recommendation using linked data and user diversity

There are more references available in the full text version of this article.

Cited by (0)

View full text
Copyright © 2015 Elsevier B.V. All rights reserved.