Leveraging Implicit Knowledge in Recommender Systems for Maternal Health

Previous research has indicated that pregnant women acquire information more effectively through watching videos rather than consulting with experts. Therefore, understanding what they search for on video content platforms has become progressively crucial in domains like pregnancy in maternal health. Learning their viewing habits can help providers and public health officials target messaging to ensure proper and fair use of this resource, while mitigating the impact of any harmful content. Although researchers have carried out several studies to answer this question, none so far has utilized data mining, specifically recommender systems themselves, which as we show in this paper can be a tremendous source of information. This paper presents a novel approach to learn the viewership habits of mothers-to-be by leveraging implicit knowledge embedded in recommendations with human annotators. We illustrate the value of this approach by presenting examples in pregnancy and maternal health and showing the value of the constructed search map based on key metrics and comparisons to alternative approaches. Interestingly while a lot of research in data mining and recommender systems focuses on algorithms *for* recommendations, this paper demonstrates that there may be great value in learning *from* recommendations to inform issues such as health and policy.