Large Language Models (LLMs) have exhibited significant promise in recommender systems by empowering user profiles with their extensive world knowledge and superior reasoning capabilities. However, LLMs face challenges like unstable instruction compliance, modality gaps, and high inference latency, leading to textual noise and limiting their effectiveness in recommender systems. To address these challenges, we propose UserIP-Tuning, which uses prompt-tuning to infer user profiles. It integrates the causal relationship between user profiles and behavior sequences into LLMs' prompts. It employs Expectation Maximization (EM) to infer the embedded latent profile, minimizing textual noise by fixing the prompt template. Furthermore, a profile quantization codebook bridges the modality gap by categorizing profile embeddings into collaborative IDs pre-stored for online deployment. This improves time efficiency and reduces memory usage. Experiments show that UserIP-Tuning outperforms state-of-the-art recommendation algorithms. An industry application confirms its effectiveness, robustness, and transferability. The presented solution has been deployed in Huawei AppGallery's Explore page since May 2025, serving 2 million daily active users, delivering significant improvements in real-world recommendation scenarios. The code is publicly available for replication at https://github.com/Applied-Machine-Learning-Lab/UserIP-Tuning.

Abstract Social recommender systems (SRS) have garnered adequate attention due to the supplementary information provided by social network, which aids in making recommendations. However, social network information contains noise, which can be detrimental to recommendation performance. Current social recommendation models are deficient in feature validation and extraction of social data. To fill that gap, we propose a novel model called Social View Explorer Collaborative Filtering (SVE-CF) which aims to extract significant consistent signals from the noisy social network. First, SVE-CF correlates users’ social and interaction behaviors, creating follow, joint, and interaction views to represent all interaction patterns. Second, it samples unlabeled examples from users to assess consistency across the three views, assigning pseudo-labels as evidence of social homophily. Third, it selects top-k pseudo-labels to amplify significant consistent signals and minimize noise through tri-view joint learning. Extensive experiments are conducted to demonstrate the effectiveness of the proposed model over the commonly used state-of-the-art (SOTA) methods.

Inferential modeling has been of significance for modern manufacturing in estimating the quality-related process variables. As an effective inferential model, probabilistic slow feature analysis (PSFA) has gained attention in regression tasks to interpret dynamic properties with a slowness preference. However, PSFA is often challenged by the nonlinear sequential data due to its linear state-space structure. In this article, a new nonlinear extension of PSFA is proposed under the deep learning framework to enhance the dynamic feature extraction with limited labels, incorporating variational inference and Monte Carlo inference to derive the objective function. The proposed model considers the relevance of inputs with outputs as the input weights to upgrade prediction performance. The proposed model is verified through an industrial hydrocracking process to predict diesel yield with missing labels ranged from 0% to 50%, and the root mean squared error is reduced by at least 8.78% compared to PSFA.

In soft sensing, relationship variation of process variables and quality indicators may cause the model trained from the training datasets unsuitable for the prediction on the testing datasets. As the model-agnostic meta-learning can utilize the supporting datasets to strengthen the prediction performance of the query samples, it can maintain reliable prediction performance in relationship variation. However, the traditional model-agnostic meta-learning contains inconsistencies between the parameters evaluated in the training stage and those adapted in the predicting stage. The phenomenon is inferred as the dilemma of getting valuable evaluated parameters related to the initial parameters and accurate parameters representing the parameters adapted in the predicting stage. In this article, we propose the stage-related adaption block to use the model-agnostic meta-learning modularly. Finally, the model-agnostic meta-learning method based on the optimal alternative scaling value is proposed and verified in a numerical example and an industrial application.