Facet-Aware Multi-Head Mixture-of-Experts Model with Text-Enhanced Pre-training for Sequential Recommendation

TL;DR

FAME addresses the limitation of monolithic item embeddings in sequential recommendation by introducing facet-aware multi-head predictions and a Mixture-of-Experts self-attention layer, enabling each head to specialize in a distinct item facet. A gating mechanism then dynamically fuses head-specific recommendations, while an MoE router disentangles user preferences within each facet. To overcome cold-start and semantic interpretability issues, the authors extend the framework with FAME+: a text-enhanced pre-training that uses a frozen language model to obtain facet-disentangled embeddings via alternating supervised contrastive learning and stratified PxK sampling. Empirical results across four public datasets show that FAME+ consistently outperforms strong baselines, with pronounced gains in sparse data settings (e.g., ML-20m and Sports), validating the utility of explicit facet modeling and semantic initialization for sequential recommendation.

Abstract

Sequential recommendation (SR) systems excel at capturing users' dynamic preferences by leveraging their interaction histories. Most existing SR systems assign a single embedding vector to each item to represent its features, adopting various models to combine these embeddings into a sequence representation that captures user intent. However, we argue that this representation alone is insufficient to capture an item's multi-faceted nature (e.g., movie genres, starring actors). Furthermore, users often exhibit complex and varied preferences within these facets (e.g., liking both action and musical films within the genre facet), which are challenging to fully represent with static identifiers. To address these issues, we propose a novel architecture titled Facet-Aware Multi-Head Mixture-of-Experts Model for Sequential Recommendation (FAME). We leverage sub-embeddings from each head in the final multi-head attention layer to predict the next item separately, effectively capturing distinct item facets. A gating mechanism then integrates these predictions by dynamically determining their importance. Additionally, we introduce a Mixture-of-Experts (MoE) network within each attention head to disentangle varied user preferences within each facet, utilizing a learnable router network to aggregate expert outputs based on context. Complementing this architecture, we design a Text-Enhanced Facet-Aware Pre-training module to overcome the limitations of randomly initialized embeddings. By utilizing a pre-trained text encoder and employing an alternating supervised contrastive learning objective, we explicitly disentangle facet-specific features from textual metadata (e.g., descriptions) before sequential training begins. This ensures that the item embeddings are semantically robust and aligned with the downstream multi-facet framework.

Figures (9)

• Figure 1: A motivating example illustrating multi-faceted user intent.
• Figure 2: Overview of the proposed model: (a) illustrates the original Transformer block, while (b) depicts the architecture of our proposed FAME model. For simplicity, the LayerNorm and Dropout operations following the FFN (FFN') are omitted from the Figure
• Figure 3: MoE Self-Attention Network: Integrated Item Representation Calculation. This diagram visualizes the computational process for determining the integrated item representation of the final item ($\mathbf{f}_{t}^{(h)}$) within a specific head ($h$) of our proposed model.
• Figure 4: An example on attention scores distribution and recommendation results among different experts on genre-focused head
• Figure 5: Overview of the Text-Enhanced Facet-Aware Pre-training Framework.