The Anti-Recommendation Framework: A Person-Led Method for Widening Taste Across Domains

Recommender systems are built to predict what a person will like, and in doing so they tend to return more of what the person has already chosen. Evidence on how much they narrow what people consume is mixed: a randomised field experiment found a clear drop in individual diversity, while studies of news and political content find that people's own choices account for much of the narrowness they see. This paper proposes the Anti-Recommendation Framework, a method people can run for themselves, with or without software, to step deliberately outside their own patterns in any domain, from music and books to food, places and ideas. The framework maps a person's habits along eight dimensions in a Box Card, sets the size of each step on a felt-distance dial that requires a familiar handhold, generates candidates with nine operators, screens them with six filters including a check against simple inversion of taste, and runs a try, repeat and reflect loop whose only measure of success is range gained. Each component is motivated by research on recommender feedback loops, curiosity, repeated exposure and variety seeking, including evidence that complicates it. A minimal measurement kit is proposed from established measures of diversity, novelty and interest development. A review of research prototypes, commercial features and one relevant patent finds that person-set direction and distance already exist within single catalogues, but found no work that combines a per-dimension personal map, a handhold rule, an inversion check and range as the outcome. The paper closes with an agenda of three follow-up studies. This is a preprint and has not been peer reviewed.