QD-Rec: Open-Ended Evolution of Recommendations via Quality-Diversity
Takashi Ishida, Taro Makino, Issa Sugiura
Sakana AI
Published October 16, 2026 · on the feed October 20, 2026 · ▲ 6 · arXiv 2610.23470
The abstract
Open-ended search combined with foundation models has produced advances such as faster algorithms, new results on open mathematical problems, and self-improving agents. We extend this paradigm to recommendation for decision-making, focusing on decisions where the set of possible choices is fixed but the reasons for choosing them are not. In high-stakes decisions, e.g., which company to acquire or which customer to target, comparing diverse, well-reasoned options can be more useful than a single ranked list. We formalize open-ended recommendation as building a diverse, high-quality set of recommendations, each an item together with a hypothesis, a natural-language reason why the item fits the needs of a decision maker. We propose QD-Rec, a quality--diversity method that selects items with a contextual bandit and generates hypotheses with a foundation model from earlier ones kept in an archive as stepping stones. A new recommendation enters the archive if it is novel or better than its nearest neighbor. Across M&A, customer targeting, and paper and book recommendation, QD-Rec gives the highest mean quality and builds the largest archive in most settings.
The Brief
2:04A short podcast summary of the paper. Written and voiced by AI.
Transcript
QD-Rec. Open-Ended Evolution of Recommendations via Quality-Diversity. By Takashi Ishida and two others, from Sakana AI. Published on the sixteenth of October. Its code repository is worked out: about sixty source files, with a readme and a licence.
Open-ended search combined with foundation models has produced faster algorithms, new results on open mathematical problems, and self-improving agents. The authors carry that paradigm over to recommendation for decision-making. Their focus is decisions where the set of possible choices is fixed but the reasons for choosing them are not. For high-stakes decisions, such as which company to acquire or which customer to target, comparing diverse and well-reasoned options can be more useful than a single ranked list.
They formalize open-ended recommendation as building a set of recommendations that is both diverse and high quality. Each recommendation is a pair: an item, and a hypothesis, meaning a natural-language reason why that item fits the needs of the decision maker. The reason is part of the recommendation, not a caption attached afterwards.
QD-Rec is a quality-diversity method. It selects items with a contextual bandit. It generates hypotheses with a foundation model, working from earlier hypotheses kept in an archive, which serve as stepping stones. The archive's admission rule is short: a new recommendation enters if it is novel, or if it is better than its nearest neighbour already in the archive. Over time the archive fills with recommendations that are either new or an improvement on the closest one held.
Evaluation covers mergers and acquisitions, customer targeting, and paper and book recommendation. Across these, QD-Rec gives the highest mean quality, and builds the largest archive, in most settings.
The reframing carries the paper. A ranked list answers which item is best. QD-Rec answers a different question: what are the good reasons one might have for a choice, and which items do they point to. Making the reason part of what is recommended is what lets quality-diversity search, already productive on algorithms, mathematics and agents, apply to decisions whose menu of options is fixed in advance.
More with an account
Listen to every day’s papers
Daily playlists, deep dives, quests for the papers, and quizzes on their key terms.
Join free