With the rapid evolution of Large Language Models (LLMs), generative recommendation is gradually reshaping the paradigm of recommender systems. However, most existing methods remain confined to the interaction-driven next-item prediction paradigm, struggling to keep pace with the latest evolving trends or address the diverse recommendation tasks along with business-specific requirements in real-world scenarios. To this end, we present SIGMA, a Semantic-Grounded Instruction-Driven Generative Multi-Task Recommender deployed at AliExpress. Specifically, we first ground item entities in a unified latent space capturing both general semantics and collaborative signals. Building upon this, we introduce a hybrid item tokenization method for both precise modeling and efficient generation. Moreover, we construct a large-scale multi-task supervised fine-tuning dataset empowering SIGMA to fulfill various recommendation demands via instruction-following. Finally, we design a three-step item generation procedure integrated with an adaptive probabilistic fusion mechanism to calibrate the output distributions based on task-specific requirements for recommendation accuracy and diversity. Extensive offline experiments and online A/B tests demonstrate the effectiveness of SIGMA across various real-world recommendation tasks.

Synthetic minority oversampling technique (SMOTE) has been widely used in dealing with the imbalance classification problem in the machine learning field. However, classical SMOTE implements the oversampling by linear interpolation between adjacent minority class samples, which may fail to consider the uneven distribution of the samples. This article proposes a minority clustering SMOTE (MC-SMOTE) method that involves the clustering of minority class samples to improve the imbalance classification performance. First, samples from the minority class are clustered into several clusters. Second, oversampling is performed by linear interpolation between adjacent clusters to create new samples from different clusters that contain additional information of the entire minority class. Then classical classification techniques can be employed to achieve efficient classification. The superiority of the MC-SMOTE is first verified by experiments on some benchmark datasets from various application domains. The proposed method is then applied to the real industrial SCADA data of wind turbine blade icing. Classification results indicate that the MC-SMOTE exhibits a better performance than that of the classical SMOTE.