Generate and Instantiate What You Prefer: Text-Guided Diffusion for Sequential Recommendation

TL;DR

Experimental results on four datasets demonstrate that iDreamRec not only outperforms existing diffusion-based generative recommenders but also facilitates the incorporation of intention instructions for more precise and effective recommendation generation.

Abstract

Recent advancements in generative recommendation systems, particularly in the realm of sequential recommendation tasks, have shown promise in enhancing generalization to new items. Among these approaches, diffusion-based generative recommendation has emerged as an effective tool, leveraging its ability to capture data distributions and generate high-quality samples. Despite effectiveness, two primary challenges have been identified: 1) the lack of consistent modeling of data distribution for oracle items; and 2) the difficulty in scaling to more informative control signals beyond historical interactions. These issues stem from the uninformative nature of ID embeddings, which necessitate random initialization and limit the incorporation of additional control signals. To address these limitations, we propose iDreamRec to involve more concrete prior knowledge to establish item embeddings, particularly through detailed item text descriptions and advanced Text Embedding Models (TEM). More importantly, by converting item descriptions into embeddings aligned with TEM, we enable the integration of intention instructions as control signals to guide the generation of oracle items. Experimental results on four datasets demonstrate that iDreamRec not only outperforms existing diffusion-based generative recommenders but also facilitates the incorporation of intention instructions for more precise and effective recommendation generation.

Figures (6)

• Figure 1: The t-SNE visualization of item embeddings under different guidance signals. The color bar scale represents the relative density. When the oracle items generated based on the user's interaction history do not align with the user's interests, introducing user intention control signals can improve the recommendation effectiveness.
• Figure 2: Overall framework of iDreamRec. Firstly, iDreamRec employs Text Embedding Models (TEM) to encode the item description acquired from ChatGPT into unified text embeddings, with variance-preserving transformation to make text embeddings more suitable for diffusion models. Moreover, the generation of iDreamRec can be guided by intention instruction embeddings translated from intention instructions by TEM, allowing to involve more informative control signals. For detailed algorithms, please refer to Appendix \ref{['alg']}.
• Figure 3: Two cases of intention guidance. $k$-intention indicates that the intention strength $\rho$ in Equation \ref{['mixG']} is set to $k$.
• Figure 4: The left figure shows the performance of iDreamRec under different datasets with varying types of item embeddings: 1) iDreamRec: text embeddings are derived as the method. 2) iDreamRec w/o GPT: text embeddings are merely derived from the name of items. 3) iDreamRec w/o TEM: iDreamRec with pretrained ID embeddings from a SASRec. 4) iDreamRec w/o PT: iDreamRec with learnable ID embeddings. The right figure is the results of intention guidance on Goodreads.
• Figure 5: Analyzing the performance (NDCG@10) of different recommenders with text embeddings. Avg. refers to the case where we compute the cosine similarity between the target item embeddings and the mean of the text embeddings in the interaction history, as indicated by logic.