Xiang Zhai

Zhaoxuan Tan, Xiang Zhai, Yan Zhu et al.

Personalized systems rely on user representations to connect behavioral history with downstream recommendation applications. Existing methods typically employ either supervised latent user embeddings, which are effective for retrieval but difficult to interpret, or textual user profiles, which are interpretable but challenging to optimize for downstream utility due to lack of direct supervision. To bridge this gap, we present BLUE, a reinforcement learning framework that unifies these two forms of user representation by aligning language-based user profiles with embedding-based recommendation objectives. Given a user interaction history, BLUE leverages a profiler Large Language Model (LLM) to generate textual profiles, while an embedding model provides reward signals. This encourages the resulting textual representations to move closer to positive items and farther from negative ones in the embedding space. We further introduce a text-space supervision signal based on next-item prediction, ensuring the learned profiles remain both semantically meaningful and highly effective for downstream retrieval. Experiments on Amazon Reviews 2023 and Google Local Reviews in zero-shot sequential recommendation settings demonstrate that BLUE consistently outperforms strong baselines under both frozen and trainable embedding conditions. Notably, BLUE achieves clear gains in cross-domain transfer, highlighting the strong generalization ability of the learned user profiles. Furthermore, these generated profiles provide superior personalized context for question answering compared to raw user histories or alternative profile optimization methods. Overall, these results show that BLUE provides an effective way to unify interpretable textual profiling with discriminative latent embeddings for personalization.

Yaoyiran Li, Xiang Zhai, Moustafa Alzantot et al. · cambridge

Traditional recommender systems such as matrix factorization methods have primarily focused on learning a shared dense embedding space to represent both items and user preferences. Subsequently, sequence models such as RNN, GRUs, and, recently, Transformers have emerged and excelled in the task of sequential recommendation. This task requires understanding the sequential structure present in users' historical interactions to predict the next item they may like. Building upon the success of Large Language Models (LLMs) in a variety of tasks, researchers have recently explored using LLMs that are pretrained on vast corpora of text for sequential recommendation. To use LLMs for sequential recommendation, both the history of user interactions and the model's prediction of the next item are expressed in text form. We propose CALRec, a two-stage LLM finetuning framework that finetunes a pretrained LLM in a two-tower fashion using a mixture of two contrastive losses and a language modeling loss: the LLM is first finetuned on a data mixture from multiple domains followed by another round of target domain finetuning. Our model significantly outperforms many state-of-the-art baselines (+37% in Recall@1 and +24% in NDCG@10) and our systematic ablation studies reveal that (i) both stages of finetuning are crucial, and, when combined, we achieve improved performance, and (ii) contrastive alignment is effective among the target domains explored in our experiments.

Rohan Thavarajah, Xiang Zhai, Zheren Ma et al.

Is a deep learning model capable of understanding systems governed by certain first principle laws by only observing the system's output? Can deep learning learn the underlying physics and honor the physics when making predictions? The answers are both positive. In an effort to simulate two-dimensional subsurface fluid dynamics in porous media, we found that an accurate deep-learning-based proxy model can be taught efficiently by a computationally expensive finite-volume-based simulator. We pose the problem as an image-to-image regression, running the simulator with different input parameters to furnish a synthetic training dataset upon which we fit the deep learning models. Since the data is spatiotemporal, we compare the performance of two alternative treatments of time; a convolutional LSTM versus an autoencoder network that treats time as a direct input. Adversarial methods are adopted to address the sharp spatial gradient in the fluid dynamic problems. Compared to traditional simulation, the proposed deep learning approach enables much faster forward computation, which allows us to explore more scenarios with a much larger parameter space given the same time. It is shown that the improved forward computation efficiency is particularly valuable in solving inversion problems, where the physics model has unknown parameters to be determined by history matching. By computing the pixel-level attention of the trained model, we quantify the sensitivity of the deep learning model to key physical parameters and hence demonstrate that the inversion problems can be solved with great acceleration. We assess the efficacy of the machine learning surrogate in terms of its training speed and accuracy. The network can be trained within minutes using limited training data and achieve accuracy that scales desirably with the amount of training data supplied.