The eXtreme Multi-label Classification (XMC) aims at accurately assigning large-scale labels to instances, and is challenging for learning, managing, and predicting over the large-scale and rapidly growing set of labels. Traditional XMC methods, like one-vs-all and tree-based methods struggle with the growing set of labels due to their static label assumptions, and embedding-based methods struggle with the complex mapping relationships due to their late-interaction paradigm. In this paper, we propose a large language model (LLM) powered agent framework for extreme multi-label classification – XMC-Agent, which can effectively learn, manage and predict the extremely large and dynamically increasing set of labels. Specifically, XMC-Agent models the extreme multi-label classification task as a dynamic navigation problem, employing a scalable hierarchical label index to effectively manage the unified label space. Additionally, we propose two algorithms to enhance the dynamic navigation capabilities of XMC-Agent: a self-construction algorithm for building the scalable hierarchical index, and an iterative feedback learning algorithm for adjusting the agent to specific tasks. Experiments show that XMC-Agentachieves the state-of-the-art performance on three standard datasets.

Under the umbrella of anonymous social networks, many women have suffered from abuse, discrimination, and other sexist expressions online. However, exsiting methods based on keyword filtering and matching performed poorly on online sexism detection, which lacked the capability to identify implicit stereotypes and discrimination. Therefore, this paper proposes a System of Ensembling Fine-tuning Models (SEFM) at SemEval-2023 Task 10: Explainable Detection of Online Sexism. We firstly use four task-adaptive pre-trained language models to flag all texts. Secondly, we alleviate the data imbalance from two perspectives: over-sampling the labelled data and adjusting the loss function. Thirdly, we add indicators and feedback modules to enhance the overall performance. Our system attained macro F1 scores of 0.8538, 0.6619, and 0.4641 for Subtask A, B, and C, respectively. Our system exhibited strong performance across multiple tasks, with particularly noteworthy performance in Subtask B. Comparison experiments and ablation studies demonstrate the effectiveness of our system.