← News

ResearchSources: arXiv · 16 Nov 2026

CPGRec+ adds LLM reasoning to balance accuracy and diversity in game recommendations

Researchers introduce CPGRec+, a video game recommender that pairs graph-based filtering with large language models to weigh player preferences and improve both accuracy and diversity.

In plain words

The authors of CPGRec+ built a system that suggests video games to players using graph neural networks, a technique that models relationships between players and games. Their two new modules use large language models to judge how strongly a player likes or dislikes a game and to write richer descriptions of both. In tests on two Steam datasets, the system reportedly matched or beat existing models on accuracy and diversity.

What was announced

The paper, an arXiv abstract, explains that earlier graph neural network recommenders, including the authors' own CPGRec, prioritized accuracy over diversity and overlooked the trade-off between the two. CPGRec+ addresses this with two modules. The Preference-informed Edge Reweighting module assigns signed edge weights that separate significant player interests from disinterests and measure preference strength, which the authors say mitigates over-smoothing in graph convolutions. The Preference-informed Representation Generation module uses large language models to produce contextualized descriptions of games and players by reasoning about personal preferences against global interests.

The authors report experiments on two Steam datasets showing superior accuracy and diversity over state-of-the-art models, and they point to a public code repository for the system. The claims come from the paper's abstract alone, so the experimental setup, baselines and dataset details are not visible in the supplied material.

What it does not show

  • Evidence is limited to the paper's abstract; experimental details, baselines and dataset characteristics are not available for verification.
  • The reported accuracy and diversity gains are the authors' own claims and have not been independently confirmed.

For security and IT teams

Access
The authors state the code is available in a public GitHub repository (HsipingLi/CPGRec-Plus).

Sources

Source-support review completed 17 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.

Coverage: arXiv: abstract.

Updates and corrections

No corrections so far.

First published 17 Nov 2026

New to the subject? AI from the Ground Up starts from the beginning.