Extracting Interaction-Aware Monosemantic Concepts in Recommender Systems
This provides a practical tool for interpretable and controllable personalization in recommender systems, though it is incremental as it adapts existing methods to a new domain.
The authors tackled the problem of extracting interpretable concepts from user and item embeddings in recommender systems, resulting in neurons that capture properties like genre and popularity, enabling control operations such as filtering without modifying the base model.
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.