Beyond Satisfaction: Learning Associations Between Content, Reviews, and Well-Being
Recommendation algorithms shape what people consume online, yet they are primarily optimized for star ratings, likes, and sentiment, signals of user satisfaction. Whether these signals capture what is good for users, beyond momentary preference, remains largely unexamined. Well-being offers an alternative target grounded in psychology that extends beyond satisfaction. We operationalize it as immediate and enduring expressions of the five PERMA facets in reviews. Empirically, immediate positive emotion and engagement cluster as more hedonic (pleasure-oriented), while relationships, meaning, accomplishment, and enduring positive emotion and engagement behave as more eudaimonic (meaning- and growth-oriented). Analyzing 13.7 million Goodreads reviews and the content of 16 thousand books, we study how PERMA expressions differ from satisfaction targets and which book content features are associated with them. First, satisfaction signals (star ratings and review sentiment) correlate moderately with immediate hedonic well-being (r = .43-.50) but less with its enduring expressions (r <= .28), and have almost no association with eudaimonic facets (r <= .09). Second, book content features are more strongly associated with PERMA facets (r up to .46) than with satisfaction signals (|r| <= .15). Content associated with immediate hedonic well-being differs from that associated with enduring and eudaimonic well-being: third-person pronouns and past-focused language relate positively to immediate engagement (d=1.14 and 1.05) but negatively to every enduring facet (d down to -1.58 and -1.32), while prosocial and achievement-oriented language relate to relationships and accomplishment (d=.45 and .96). These findings suggest that satisfaction signals capture only part of the well-being users express, and that well-being-informed targets could lead platforms to recommend substantially different content.