Performance of a Recommender System in Stimulating Viral Growth and Social Tie Formation in a Web-Based Health Intervention
|
If you are the presenter of this abstract (or if you cite this abstract in a talk or on a poster), please show the QR code in your slide or poster (QR code contains this URL). |
Abstract
Background: Daily Challenge is a publicly available web-based health intervention with an integrated social network derived from Facebook. Social ties within the system provide social support and influence, driving engagement and adherence. We aimed to promote user-generated invitations and ultimately, the formation of friendships within the program. Upon enrollment in the Daily Challenge the system dynamically creates a representation of their local friendship network pulled directly from Facebook’s data stores. We hypothesized that social network analysis could be used to enhance the process of tie formation by pre-selecting and suggesting Facebook friends to become contacts within the Daily Challenge.
Objective: We sought to compare a network based recommendation algorithm with a random algorithm on their ability to generate successful invites.
Methods: We randomized 6,020 Daily Challenge members over a 33-day period to be presented with either a random selection of suggested Facebook friends or a algorithm selected set. We analyzed past invites and determined the social network characteristics of successful and unsuccessful invites. This data was used to create recommendation algorithm designed to optimize the probability of generating a successful invite. For each of a Daily Challenge user’s Facebook friends, the algorithm evaluates four factors that influenced probability of invitation success: 1) social network position (betweeness); 2) social network density (local clustering coefficient); 3) number of invitations previously received; 4) whether s/he shares last name with the user. The algorithm computes a score and ranks the Facebook friends by probability of accepting an invite. The highest-ranked friends, i.e., the friends who are more likely to be sent and to accept an invite, are recommended to the user for invite.
Results: Our recommendation algorithm performed 237% better than the random algorithm in generating successful invites, although the difference was only marginally significant (3.77 vs. 1.59 conversions per thousand recommendations; p = 0.067). Daily Challenge users agreed to invite 74% more friends recommended by our algorithm than by a random algorithm (45.5 vs. 26.2 invites per thousand recommendations; p < 0.05). Finally, invitees accepted more invites initiated by our algorithm than by the random algorithm (8.3% vs. 6.1% p > 0.05) but did not meet criteria for statistical significance.
Conclusions: A social network analysis based algorithm generated more invites than a random selection algorithm, but the increased number of invitations did translate into a statistically significant increase in new connections. Implementation and study considerations will be discussed.
Objective: We sought to compare a network based recommendation algorithm with a random algorithm on their ability to generate successful invites.
Methods: We randomized 6,020 Daily Challenge members over a 33-day period to be presented with either a random selection of suggested Facebook friends or a algorithm selected set. We analyzed past invites and determined the social network characteristics of successful and unsuccessful invites. This data was used to create recommendation algorithm designed to optimize the probability of generating a successful invite. For each of a Daily Challenge user’s Facebook friends, the algorithm evaluates four factors that influenced probability of invitation success: 1) social network position (betweeness); 2) social network density (local clustering coefficient); 3) number of invitations previously received; 4) whether s/he shares last name with the user. The algorithm computes a score and ranks the Facebook friends by probability of accepting an invite. The highest-ranked friends, i.e., the friends who are more likely to be sent and to accept an invite, are recommended to the user for invite.
Results: Our recommendation algorithm performed 237% better than the random algorithm in generating successful invites, although the difference was only marginally significant (3.77 vs. 1.59 conversions per thousand recommendations; p = 0.067). Daily Challenge users agreed to invite 74% more friends recommended by our algorithm than by a random algorithm (45.5 vs. 26.2 invites per thousand recommendations; p < 0.05). Finally, invitees accepted more invites initiated by our algorithm than by the random algorithm (8.3% vs. 6.1% p > 0.05) but did not meet criteria for statistical significance.
Conclusions: A social network analysis based algorithm generated more invites than a random selection algorithm, but the increased number of invitations did translate into a statistically significant increase in new connections. Implementation and study considerations will be discussed.
Medicine 2.0® is happy to support and promote other conferences and workshops in this area. Contact us to produce, disseminate and promote your conference or workshop under this label and in this event series. In addition, we are always looking for hosts of future World Congresses. Medicine 2.0® is a registered trademark of JMIR Publications Inc., the leading academic ehealth publisher.
This work is licensed under a Creative Commons Attribution 3.0 License.