Trait Analysis to Facilitate Children’s Books Recommender Systems

Abstract

Recommender systems are a useful tool for match- ing readers with books. However, the lack of user data from children, both due to privacy concerns as well as a low incentive to leave reviews, results in existing systems proving inadequate at recom- mending to the youth. It has been shown that chil- dren have a different relation regarding emotion at different ages. We also theorize that the perspec- tive that a book is told from varies depending on target audience age. In this paper we examine traits that could be used for recommendation instead of review data. By examining both the description of books as well as the fulltext we obtain a set of traits relating to perspective and sentiment. Com- paring these traits among books written for differ- ent age groups we observe a subset of these traits that show potential significance in age categoriza- tion. By performing a combination of empirical examination and significance testing we find that both book descriptions and book fulltext contain perspective and sentiment traits that show signifi- cance when comparing books written for different ages. Because traits obtained from book fulltext present a higher quantity of significant instances when compared those obtained from descriptions, we conclude that analysing fulltext for traits shows promise when considering a recommender system that aims to use book content.