Using Artificial Intelligence to Maximize the Spread of Sexual Health Information in a Multimodal Communication Network of Young Black Women

Background: Young Black women (YBW) between the ages of 18 and 25 have the highest rates of HIV as compared with groups of women of other ages, and races or ethnicities in the United States. Within the field of health behavior change and social work, past HIV prevention research has demonstrated the benefit of peer leaders to effect behavior change among high-risk groups. Furthermore, the application of computational models that utilize advances in network analysis and simulation has proven to be cost-effective when there are limited resources for peer leader training. However, there has yet to be any research that uses computational modeling to find the optimal selection of peer leaders from multimodal communication networks of YBW. We present the results of a Los Angeles-based network of 73 YBW’s multimodal communication behaviors, with a particular focus on (a) how information about sexuality and sexual health was spread among this network, and via which communication modes; (b) how we leveraged this network data to simulate and model how to optimize information flow within different communication modalities in a hypothetical peer leader intervention program; and (c) how different modalities of peer leaders impact the spread of information, and which optimal individuals should be selected to maximize this spread.

Methods: Quantitative survey dataset examining sexual health communication habits in a social network of 73 YBW residing in Los Angeles County was used. Five modes of communication were considered in this study (i.e. in-person, text, phone calls, social media, and a combination of the 4 modes). To determine which method of communication is optimal for disseminating sexual health information through a network of YBW, the Greedy algorithm (a maximum influence approximation algorithm) was used.

Results: We found that a texting-based intervention would yield similar results to a multimodal intervention for our surveyed network of YBW in the case of limited training resources. Social media, in-person, and phone-based networks yielded decent results, indirectly influencing 33-39% of all connections. In the texting-based network, the average number of people indirectly influenced by peer-leaders was 41.55% (n=72), while in the all combination network 45.73% of the people were indirectly influenced (n=73). When examining only YBW who already discuss sexual health via texting with network-members (n=64), approximately 45% of YBW would be indirectly influenced.

Conclusion: Currently, many social network-based interventions are designed with limited consideration on the best mode of dissemination for effective and efficient diffusion of information and behavioral changes within targeted social networks. We propose that for future peer-led intervention studies, knowledge of the communication modalities and their prevalence within the target network is critical to maximize the diffusion of information. With knowledge of the network structure and the communication mode for each edge of the network, computer-based simulations of social network-based interventions can be completed at virtually no cost to assist in determining the best mode of dissemination.