Score: 1

Extracting Interaction-Aware Monosemantic Concepts in Recommender Systems

Published: November 22, 2025 | arXiv ID: 2511.18024v1

By: Dor Arviv , Yehonatan Elisha , Oren Barkan and more

Potential Business Impact:

Makes movie suggestions better by understanding what you like.

Business Areas:
Semantic Search Internet Services

We present a method for extracting \emph{monosemantic} neurons, defined as latent dimensions that align with coherent and interpretable concepts, from user and item embeddings in recommender systems. Our approach employs a Sparse Autoencoder (SAE) to reveal semantic structure within pretrained representations. In contrast to work on language models, monosemanticity in recommendation must preserve the interactions between separate user and item embeddings. To achieve this, we introduce a \emph{prediction aware} training objective that backpropagates through a frozen recommender and aligns the learned latent structure with the model's user-item affinity predictions. The resulting neurons capture properties such as genre, popularity, and temporal trends, and support post hoc control operations including targeted filtering and content promotion without modifying the base model. Our method generalizes across different recommendation models and datasets, providing a practical tool for interpretable and controllable personalization. Code and evaluation resources are available at https://github.com/DeltaLabTLV/Monosemanticity4Rec.

Repos / Data Links

Page Count
9 pages

Category
Computer Science:
Information Retrieval