Every Book Tells a Value

Abstract

Recommender Systems (RS) play a crucial role in shaping children's digital experience, by selecting which media they interact with. Through this selection process, RS not only influence immediate choices but may also contribute to the formation of preferences and worldviews. As is the case for all technologies, RS embody implicit and explicit design choices that reflect particular values. In this paper, we investigate the extent to which RS reflect children's values in the domain of books. Since books represent a popular medium of introducing moral concepts to children, we will focus on this domain, but our approach can be applied to other textual content as well. Given that children are in a formative developmental stage, their values are both distinct from those of adults and subject to change, making them especially susceptible to being influenced by RS. We first analyze children's book reviews to capture how values are expressed in practice. Based on these value expressions, we construct value profiles and obtain user segments through clustering, representing groups of children with similar value priorities and expressions. We use this in our analysis to gain insight into emerging patterns. Multiple recommendation strategies are employed to examine how well recommended items align with users' value profiles. By comparing the values reflected in recommended books with those expressed by users, we assess whether RS amplify, suppress, or remain neutral with respect to children's values. Our findings provide insight into how children manifest value preferences online and into the value alignment of RS for young users.