Xiaobin Wang

Tianyu Yu, Chengyue Jiang, Chao Lou et al.

Large language models (LLMs) have shown impressive ability for open-domain NLP tasks. However, LLMs are sometimes too footloose for natural language understanding (NLU) tasks which always have restricted output and input format. Their performances on NLU tasks are highly related to prompts or demonstrations and are shown to be poor at performing several representative NLU tasks, such as event extraction and entity typing. To this end, we present SeqGPT, a bilingual (i.e., English and Chinese) open-source autoregressive model specially enhanced for open-domain natural language understanding. We express all NLU tasks with two atomic tasks, which define fixed instructions to restrict the input and output format but still ``open'' for arbitrarily varied label sets. The model is first instruction-tuned with extremely fine-grained labeled data synthesized by ChatGPT and then further fine-tuned by 233 different atomic tasks from 152 datasets across various domains. The experimental results show that SeqGPT has decent classification and extraction ability, and is capable of performing language understanding tasks on unseen domains. We also conduct empirical studies on the scaling of data and model size as well as on the transfer across tasks. Our model is accessible at https://github.com/Alibaba-NLP/SeqGPT.

Qi Zhang, Zhaopeng Feng, Xiaonan Shi et al.

Agent skills, which consist of reusable strategies that guide agent reasoning and action, have shown strong potential for improving model capability at inference time. However, current skill construction methods treat the problem as one-shot extraction, overlooking a fundamental tension: a skill tailored to the specific task fails to transfer, while the abstracted skill often provides insufficient guidance. We attribute this fragility to the absence of explicit mechanisms for skill specification and generalization. To address this gap, we introduce SkillComposer, a framework that decomposes skill construction into three learnable operations: create, improve, and merge. Trained via systematic rejection sampling recipe, SkillComposer enables language models to self-evolve skills at inference time and supports three deployment modes: offline for building generalized libraries, online for task-specific refinement, and hybrid for combining both. Comprehensive experiments on $τ^2$-Bench, LiveCodeBench v6, and AppWorld show that SkillComposer consistently outperforms baselines. Our SkillComposer-4B improves a 27B executor by up to +4.5 on agent tasks and +3.4 on code tasks, while generalizing across domains and task types unseen during training. Analysis reveals that merge and improve address orthogonal quality dimensions and that skill composition is a transferable meta-ability, providing a practical recipe for skill-augmented inference.

Yangning Li, Shirong Ma, Xiaobin Wang et al.

Recently, instruction-following Large Language Models (LLMs) , represented by ChatGPT, have exhibited exceptional performance in general Natural Language Processing (NLP) tasks. However, the unique characteristics of E-commerce data pose significant challenges to general LLMs. An LLM tailored specifically for E-commerce scenarios, possessing robust cross-dataset/task generalization capabilities, is a pressing necessity. To solve this issue, in this work, we proposed the first e-commerce instruction dataset EcomInstruct, with a total of 2.5 million instruction data. EcomInstruct scales up the data size and task diversity by constructing atomic tasks with E-commerce basic data types, such as product information, user reviews. Atomic tasks are defined as intermediate tasks implicitly involved in solving a final task, which we also call Chain-of-Task tasks. We developed EcomGPT with different parameter scales by training the backbone model BLOOMZ with the EcomInstruct. Benefiting from the fundamental semantic understanding capabilities acquired from the Chain-of-Task tasks, EcomGPT exhibits excellent zero-shot generalization capabilities. Extensive experiments and human evaluations demonstrate that EcomGPT outperforms ChatGPT in term of cross-dataset/task generalization on E-commerce tasks.

Chengyue Jiang, Wenyang Hui, Yong Jiang et al.

Ultra-fine entity typing (UFET) predicts extremely free-formed types (e.g., president, politician) of a given entity mention (e.g., Joe Biden) in context. State-of-the-art (SOTA) methods use the cross-encoder (CE) based architecture. CE concatenates the mention (and its context) with each type and feeds the pairs into a pretrained language model (PLM) to score their relevance. It brings deeper interaction between mention and types to reach better performance but has to perform N (type set size) forward passes to infer types of a single mention. CE is therefore very slow in inference when the type set is large (e.g., N = 10k for UFET). To this end, we propose to perform entity typing in a recall-expand-filter manner. The recall and expand stages prune the large type set and generate K (K is typically less than 256) most relevant type candidates for each mention. At the filter stage, we use a novel model called MCCE to concurrently encode and score these K candidates in only one forward pass to obtain the final type prediction. We investigate different variants of MCCE and extensive experiments show that MCCE under our paradigm reaches SOTA performance on ultra-fine entity typing and is thousands of times faster than the cross-encoder. We also found MCCE is very effective in fine-grained (130 types) and coarse-grained (9 types) entity typing. Our code is available at \url{https://github.com/modelscope/AdaSeq/tree/master/examples/MCCE}.

Xiang Wei, Xingyu Cui, Ning Cheng et al.

Zero-shot information extraction (IE) aims to build IE systems from the unannotated text. It is challenging due to involving little human intervention. Challenging but worthwhile, zero-shot IE reduces the time and effort that data labeling takes. Recent efforts on large language models (LLMs, e.g., GPT-3, ChatGPT) show promising performance on zero-shot settings, thus inspiring us to explore prompt-based methods. In this work, we ask whether strong IE models can be constructed by directly prompting LLMs. Specifically, we transform the zero-shot IE task into a multi-turn question-answering problem with a two-stage framework (ChatIE). With the power of ChatGPT, we extensively evaluate our framework on three IE tasks: entity-relation triple extract, named entity recognition, and event extraction. Empirical results on six datasets across two languages show that ChatIE achieves impressive performance and even surpasses some full-shot models on several datasets (e.g., NYT11-HRL). We believe that our work could shed light on building IE models with limited resources.

Ning Ding, Yulin Chen, Ganqu Cui et al. · tsinghua

Metric-based meta-learning is one of the de facto standards in few-shot learning. It composes of representation learning and metrics calculation designs. Previous works construct class representations in different ways, varying from mean output embedding to covariance and distributions. However, using embeddings in space lacks expressivity and cannot capture class information robustly, while statistical complex modeling poses difficulty to metric designs. In this work, we use tensor fields (``areas'') to model classes from the geometrical perspective for few-shot learning. We present a simple and effective method, dubbed hypersphere prototypes (HyperProto), where class information is represented by hyperspheres with dynamic sizes with two sets of learnable parameters: the hypersphere's center and the radius. Extending from points to areas, hyperspheres are much more expressive than embeddings. Moreover, it is more convenient to perform metric-based classification with hypersphere prototypes than statistical modeling, as we only need to calculate the distance from a data point to the surface of the hypersphere. Following this idea, we also develop two variants of prototypes under other measurements. Extensive experiments and analysis on few-shot learning tasks across NLP and CV and comparison with 20+ competitive baselines demonstrate the effectiveness of our approach.

Xinyu Wang, Yongliang Shen, Jiong Cai et al.

The MultiCoNER shared task aims at detecting semantically ambiguous and complex named entities in short and low-context settings for multiple languages. The lack of contexts makes the recognition of ambiguous named entities challenging. To alleviate this issue, our team DAMO-NLP proposes a knowledge-based system, where we build a multilingual knowledge base based on Wikipedia to provide related context information to the named entity recognition (NER) model. Given an input sentence, our system effectively retrieves related contexts from the knowledge base. The original input sentences are then augmented with such context information, allowing significantly better contextualized token representations to be captured. Our system wins 10 out of 13 tracks in the MultiCoNER shared task.

Yongliang Shen, Xiaobin Wang, Zeqi Tan et al.

Named entity recognition (NER) is a fundamental task in natural language processing. Recent works treat named entity recognition as a reading comprehension task, constructing type-specific queries manually to extract entities. This paradigm suffers from three issues. First, type-specific queries can only extract one type of entities per inference, which is inefficient. Second, the extraction for different types of entities is isolated, ignoring the dependencies between them. Third, query construction relies on external knowledge and is difficult to apply to realistic scenarios with hundreds of entity types. To deal with them, we propose Parallel Instance Query Network (PIQN), which sets up global and learnable instance queries to extract entities from a sentence in a parallel manner. Each instance query predicts one entity, and by feeding all instance queries simultaneously, we can query all entities in parallel. Instead of being constructed from external knowledge, instance queries can learn their different query semantics during training. For training the model, we treat label assignment as a one-to-many Linear Assignment Problem (LAP) and dynamically assign gold entities to instance queries with minimal assignment cost. Experiments on both nested and flat NER datasets demonstrate that our proposed method outperforms previous state-of-the-art models.

Xin Zhang, Guangwei Xu, Yueheng Sun et al.

Recent works of opinion expression identification (OEI) rely heavily on the quality and scale of the manually-constructed training corpus, which could be extremely difficult to satisfy. Crowdsourcing is one practical solution for this problem, aiming to create a large-scale but quality-unguaranteed corpus. In this work, we investigate Chinese OEI with extremely-noisy crowdsourcing annotations, constructing a dataset at a very low cost. Following zhang et al. (2021), we train the annotator-adapter model by regarding all annotations as gold-standard in terms of crowd annotators, and test the model by using a synthetic expert, which is a mixture of all annotators. As this annotator-mixture for testing is never modeled explicitly in the training phase, we propose to generate synthetic training samples by a pertinent mixup strategy to make the training and testing highly consistent. The simulation experiments on our constructed dataset show that crowdsourcing is highly promising for OEI, and our proposed annotator-mixup can further enhance the crowdsourcing modeling.

Nanping Luo, Xiaobin Wang, Shenghong Gu et al.

We study the surface composition of asteroids with visible and/or infrared spectroscopy. For example, asteroid taxonomy is based on the spectral features or multiple color indices in visible and near-infrared wavelengths. The composition of asteroids gives key information to understand their origin and evolution. However, we lack compositional information for faint asteroids due to limits of ground-based observational instruments. In the near future, the Chinese Space Survey telescope (CSST) will provide multiple colors and spectroscopic data for asteroids of apparent magnitude brighter than 25 mag and 23 mag, respectively. For the aim of analysis of the CSST spectroscopic data, we applied an algorithm using artificial neural networks (ANNs) to establish a preliminary classification model for asteroid taxonomy according to the design of the survey module of CSST. Using the SMASS II spectra and the Bus-Binzel taxonomy system, our ANN classification tool composed of 5 individual ANNs is constructed, and the accuracy of this classification system is higher than 92 %. As the first application of our ANN tool, 64 spectra of 42 asteroids obtained in 2006 and 2007 by us with the 2.16-m telescope in the Xinglong station (Observatory Code 327) of National Astronomical Observatory of China are analyzed. The predicted labels of these spectra using our ANN tool are found to be reasonable when compared to their known taxonomic labels. Considering the accuracy and stability, our ANN tool can be applied to analyse the CSST asteroid spectra in the future.

Xin Zhang, Yong Jiang, Xiaobin Wang et al.

Successful Machine Learning based Named Entity Recognition models could fail on texts from some special domains, for instance, Chinese addresses and e-commerce titles, where requires adequate background knowledge. Such texts are also difficult for human annotators. In fact, we can obtain some potentially helpful information from correlated texts, which have some common entities, to help the text understanding. Then, one can easily reason out the correct answer by referencing correlated samples. In this paper, we suggest enhancing NER models with correlated samples. We draw correlated samples by the sparse BM25 retriever from large-scale in-domain unlabeled data. To explicitly simulate the human reasoning process, we perform a training-free entity type calibrating by majority voting. To capture correlation features in the training stage, we suggest to model correlated samples by the transformer-based multi-instance cross-encoder. Empirical results on datasets of the above two domains show the efficacy of our methods.

Songbo Hu, Yinhong Liu, Ej Zhou et al.

Creating spoken dialogue datasets is methodologically challenging, and these challenges are amplified when the goal is to build multilingual, multi-parallel datasets at scale. This work introduces HEALTHDIAL, a large-scale, multilingual, and multi-parallel dataset for developing and evaluating retrieval-augmented generation (RAG)-based spoken dialogue systems. The dataset comprises 6,000 information-seeking dialogues (1,500 per language) grounded in trusted content from the World Health Organization (WHO) and 163 hours of user speech recorded from native speakers of diverse dialects across four official WHO languages: Arabic, Chinese, English, and Spanish. Each speaker is annotated with demographic (e.g., gender, age) and sociolinguistic (e.g., primary language, region of origin) variables. We report benchmark results across key dialogue tasks, which reveal consistent performance disparities across languages, even among high-resource ones. To support future research, we release the dataset, a prototype system, and a toolkit for data collection and system evaluation.

Peng Wang, Xiaobin Wang, Chao Lou et al.

In-context learning (ICL) is a few-shot learning paradigm that involves learning mappings through input-output pairs and appropriately applying them to new instances. Despite the remarkable ICL capabilities demonstrated by Large Language Models (LLMs), existing works are highly dependent on large-scale labeled support sets, not always feasible in practical scenarios. To refine this approach, we focus primarily on an innovative selective annotation mechanism, which precedes the standard demonstration retrieval. We introduce the Language Model-based Determinant Point Process (LM-DPP) that simultaneously considers the uncertainty and diversity of unlabeled instances for optimal selection. Consequently, this yields a subset for annotation that strikes a trade-off between the two factors. We apply LM-DPP to various language models, including GPT-J, LlaMA, and GPT-3. Experimental results on 9 NLU and 2 Generation datasets demonstrate that LM-DPP can effectively select canonical examples. Further analysis reveals that LLMs benefit most significantly from subsets that are both low uncertainty and high diversity.

Shen Huang, Yuchen Zhai, Xinwei Long et al.

Speech Entity Linking aims to recognize and disambiguate named entities in spoken languages. Conventional methods suffer gravely from the unfettered speech styles and the noisy transcripts generated by ASR systems. In this paper, we propose a novel approach called Knowledge Enhanced Named Entity Recognition (KENER), which focuses on improving robustness through painlessly incorporating proper knowledge in the entity recognition stage and thus improving the overall performance of entity linking. KENER first retrieves candidate entities for a sentence without mentions, and then utilizes the entity descriptions as extra information to help recognize mentions. The candidate entities retrieved by a dense retrieval module are especially useful when the input is short or noisy. Moreover, we investigate various data sampling strategies and design effective loss functions, in order to improve the quality of retrieved entities in both recognition and disambiguation stages. Lastly, a linking with filtering module is applied as the final safeguard, making it possible to filter out wrongly-recognized mentions. Our system achieves 1st place in Track 1 and 2nd place in Track 2 of NLPCC-2022 Shared Task 2.

Jingheng Ye, Yong Jiang, Xiaobin Wang et al.

This paper introduces the task of product demand clarification within an e-commercial scenario, where the user commences the conversation with ambiguous queries and the task-oriented agent is designed to achieve more accurate and tailored product searching by asking clarification questions. To address this task, we propose ProductAgent, a conversational information seeking agent equipped with abilities of strategic clarification question generation and dynamic product retrieval. Specifically, we develop the agent with strategies for product feature summarization, query generation, and product retrieval. Furthermore, we propose the benchmark called PROCLARE to evaluate the agent's performance both automatically and qualitatively with the aid of a LLM-driven user simulator. Experiments show that ProductAgent interacts positively with the user and enhances retrieval performance with increasing dialogue turns, where user demands become gradually more explicit and detailed. All the source codes will be released after the review anonymity period.

Shihao Cai, Runnan Fang, Jialong Wu et al.

Conducting reinforcement learning (RL) in simulated environments offers a cost-effective and highly scalable way to enhance language-based agents. However, previous work has been limited to semi-automated environment synthesis or tasks lacking sufficient difficulty, offering little breadth or depth. In addition, the instability of simulated users integrated into these environments, along with the heterogeneity across simulated environments, poses further challenges for agentic RL. In this work, we propose: (1) a unified pipeline for automated and scalable synthesis of simulated environments associated with high-difficulty but easily verifiable tasks; and (2) an environment level RL algorithm that not only effectively mitigates user instability but also performs advantage estimation at the environment level, thereby improving training efficiency and stability. Comprehensive evaluations on agentic benchmarks, including tau-bench, tau2-Bench, and VitaBench, validate the effectiveness of our proposed method. Further in-depth analyses underscore its out-of-domain generalization.

Zhaopeng Feng, Liangcai Su, Zhen Zhang et al.

As large language models (LLMs) evolve into autonomous agents for long-horizon information-seeking, managing finite context capacity has become a critical bottleneck. Existing context management methods typically commit to a single fixed strategy throughout the entire trajectory. Such static designs may work well in some states, but they cannot adapt as the usefulness and reliability of the accumulated context evolve during long-horizon search. To formalize this challenge, we introduce a probabilistic framework that characterizes long-horizon success through two complementary dimensions: search efficiency and terminal precision. Building on this perspective, we propose AgentSwing, a state-aware adaptive parallel context management routing framework. At each trigger point, AgentSwing expands multiple context-managed branches in parallel and uses lookahead routing to select the most promising continuation. Experiments across diverse benchmarks and agent backbones show that AgentSwing consistently outperforms strong static context management methods, often matching or exceeding their performance with up to $3\times$ fewer interaction turns while also improving the ultimate performance ceiling of long-horizon web agents. Beyond the empirical gains, the proposed probabilistic framework provides a principled lens for analyzing and designing future context management strategies for long-horizon agents.

Jin Su, Runnan Fang, Yeqiu Li et al.

Large language model (LLM)-based agents have been successfully deployed in many tool-augmented settings, but their scalability is fundamentally constrained by context length. Existing context-folding methods mitigate this issue by summarizing past interactions, yet they are typically designed for single-query or single-intent scenarios. In more realistic user-centric dialogues, we identify two major failure modes: (i) they irreversibly discard fine-grained constraints and intermediate facts that are crucial for later decisions, and (ii) their summaries fail to track evolving user intent, leading to omissions and erroneous actions. To address these limitations, we propose U-Fold, a dynamic context-folding framework tailored to user-centric tasks. U-Fold retains the full user--agent dialogue and tool-call history but, at each turn, uses two core components to produce an intent-aware, evolving dialogue summary and a compact, task-relevant tool log. Extensive experiments on $τ$-bench, $τ^2$-bench, VitaBench, and harder context-inflated settings show that U-Fold consistently outperforms ReAct (achieving a 71.4% win rate in long-context settings) and prior folding baselines (with improvements of up to 27.0%), particularly on long, noisy, multi-turn tasks. Our study demonstrates that U-Fold is a promising step toward transferring context-management techniques from single-query benchmarks to realistic user-centric applications.

Runnan Fang, Xiaobin Wang, Yuan Liang et al.

In the interaction between agents and their environments, agents expand their capabilities by planning and executing actions. However, LLM-based agents face substantial challenges when deployed in novel environments or required to navigate unconventional action spaces. To empower agents to autonomously explore environments, optimize workflows, and enhance their understanding of actions, we propose SynWorld, a framework that allows agents to synthesize possible scenarios with multi-step action invocation within the action space and perform Monte Carlo Tree Search (MCTS) exploration to effectively refine their action knowledge in the current environment. Our experiments demonstrate that SynWorld is an effective and general approach to learning action knowledge in new environments. Code is available at https://github.com/zjunlp/SynWorld.

Shuofei Qiao, Zhisong Qiu, Baochang Ren et al.

Large Language Models (LLMs) have achieved considerable performance across various agentic planning tasks. However, traditional agent planning approaches adopt a "flood irrigation" methodology that indiscriminately injects gold trajectories, external feedback, and domain knowledge into agent models. This practice overlooks the fundamental human cognitive principle of situational self-awareness during decision-making-the ability to dynamically assess situational demands and strategically employ resources during decision-making. We propose agentic knowledgeable self-awareness to address this gap, a novel paradigm enabling LLM-based agents to autonomously regulate knowledge utilization. Specifically, we propose KnowSelf, a data-centric approach that applies agents with knowledgeable self-awareness like humans. Concretely, we devise a heuristic situation judgement criterion to mark special tokens on the agent's self-explored trajectories for collecting training data. Through a two-stage training process, the agent model can switch between different situations by generating specific special tokens, achieving optimal planning effects with minimal costs. Our experiments demonstrate that KnowSelf can outperform various strong baselines on different tasks and models with minimal use of external knowledge. Code is available at https://github.com/zjunlp/KnowSelf.

Songbo Hu, Xiaobin Wang, Zhangdie Yuan et al. · cambridge

We present DIALIGHT, a toolkit for developing and evaluating multilingual Task-Oriented Dialogue (ToD) systems which facilitates systematic evaluations and comparisons between ToD systems using fine-tuning of Pretrained Language Models (PLMs) and those utilising the zero-shot and in-context learning capabilities of Large Language Models (LLMs). In addition to automatic evaluation, this toolkit features (i) a secure, user-friendly web interface for fine-grained human evaluation at both local utterance level and global dialogue level, and (ii) a microservice-based backend, improving efficiency and scalability. Our evaluations reveal that while PLM fine-tuning leads to higher accuracy and coherence, LLM-based systems excel in producing diverse and likeable responses. However, we also identify significant challenges of LLMs in adherence to task-specific instructions and generating outputs in multiple languages, highlighting areas for future research. We hope this open-sourced toolkit will serve as a valuable resource for researchers aiming to develop and properly evaluate multilingual ToD systems and will lower, currently still high, entry barriers in the field.

Ziyang Chen, Xiaobin Wang, Yong Jiang et al.

Question Answering (QA) systems face challenges in handling complex questions that require multi-domain knowledge synthesis. The naive RAG models, although effective in information retrieval, struggle with complex questions that require comprehensive and in-depth answers. The pioneering task is defined as explanatory answer generation, which entails handling identified challenges such as the requirement for comprehensive information and logical coherence within the generated context. To address these issues, we refer to systematic thinking theory and propose SynthRAG, an innovative framework designed to enhance QA performance. SynthRAG improves on conventional models by employing adaptive outlines for dynamic content structuring, generating systematic information to ensure detailed coverage, and producing customized answers tailored to specific user inquiries. This structured approach guarantees logical coherence and thorough integration of information, yielding responses that are both insightful and methodically organized. Empirical evaluations underscore SynthRAG's effectiveness, demonstrating its superiority in handling complex questions, overcoming the limitations of naive RAG models, and significantly improving answer quality and depth. Furthermore, an online deployment on the Zhihu platform revealed that SynthRAG's answers achieved notable user engagement, with each response averaging 5.73 upvotes and surpassing the performance of 79.8% of human contributors, highlighting the practical relevance and impact of the proposed framework. Our code is available at https://github.com/czy1999/SynthRAG .

Tongyi DeepResearch Team, Baixuan Li, Bo Zhang et al.

We present Tongyi DeepResearch, an agentic large language model, which is specifically designed for long-horizon, deep information-seeking research tasks. To incentivize autonomous deep research agency, Tongyi DeepResearch is developed through an end-to-end training framework that combines agentic mid-training and agentic post-training, enabling scalable reasoning and information seeking across complex tasks. We design a highly scalable data synthesis pipeline that is fully automatic, without relying on costly human annotation, and empowers all training stages. By constructing customized environments for each stage, our system enables stable and consistent interactions throughout. Tongyi DeepResearch, featuring 30.5 billion total parameters, with only 3.3 billion activated per token, achieves state-of-the-art performance across a range of agentic deep research benchmarks, including Humanity's Last Exam, BrowseComp, BrowseComp-ZH, WebWalkerQA, xbench-DeepSearch, FRAMES and xbench-DeepSearch-2510. We open-source the model, framework, and complete solutions to empower the community.

Shuofei Qiao, Yanqiu Zhao, Zhisong Qiu et al.

Data-analytic agents are emerging as a key catalyst for automated scientific discovery and for the vision of Innovating AI. Current approaches, however, rely heavily on prompt engineering over proprietary models, while open-source models struggle to face diverse-format, large-scale data files and long-horizon, multi-step reasoning that real-world analytics demands. This paper introduces DataMind, a scalable data synthesis and agent training recipe designed to build generalist data-analytic agents. DataMind tackles three key challenges in building open-source data-analytic agents, including insufficient data resources, improper training strategy, and unstable code-based multi-turn rollout. Concretely, DataMind applies 1) a fine-grained task taxonomy and a recursive easy-to-hard task composition mechanism to increase the diversity and difficulty of synthesized queries; 2) a knowledge-augmented trajectory sampling strategy followed by model-based and rule-based filtering; 3) a dynamically adjustable training objective combining both SFT and RL losses; 4) a memory-frugal and stable code-based multi-turn rollout framework. Built on DataMind, we curate DataMind-12K, a high-quality trajectory set spanning diverse domains, task categories, and data file formats for data-analytic tasks. Trained on DataMind-12K, our DataMind-14B achieves state-of-the-art with an average score of 71.16% on multiple data analysis benchmarks, outperforming the strongest proprietary baselines DeepSeek-V3.1 and GPT-5. Our DataMind-7B also performs best among all open-source models with a score of 68.10%. We also incorporate some empirical insights gained from our exploratory trials into the analysis experiments, aiming to provide actionable insights about agentic training for the community. We will release DataMind-12K and DataMind-7B,14B for the community's future research.

Boli Chen, Guangwei Xu, Xiaobin Wang et al.

Named Entity Recognition (NER) from speech is among Spoken Language Understanding (SLU) tasks, aiming to extract semantic information from the speech signal. NER from speech is usually made through a two-step pipeline that consists of (1) processing the audio using an Automatic Speech Recognition (ASR) system and (2) applying an NER tagger to the ASR outputs. Recent works have shown the capability of the End-to-End (E2E) approach for NER from English and French speech, which is essentially entity-aware ASR. However, due to the many homophones and polyphones that exist in Chinese, NER from Chinese speech is effectively a more challenging task. In this paper, we introduce a new dataset AISEHLL-NER for NER from Chinese speech. Extensive experiments are conducted to explore the performance of several state-of-the-art methods. The results demonstrate that the performance could be improved by combining entity-aware ASR and pretrained NER tagger, which can be easily applied to the modern SLU pipeline. The dataset is publicly available at github.com/Alibaba-NLP/AISHELL-NER.

Runnan Fang, Shihao Cai, Baixuan Li et al.

Advanced agentic intelligence is a prerequisite for deploying Large Language Models in practical, real-world applications. Diverse real-world APIs demand precise, robust function-calling intelligence, which needs agents to develop these capabilities through interaction in varied environments. The breadth of function-calling competence is closely tied to the diversity of environments in which agents are trained. In this work, we scale up environments as a step towards advancing general agentic intelligence. This gives rise to two central challenges: (i) how to scale environments in a principled manner, and (ii) how to effectively train agentic capabilities from experiences derived through interactions with these environments. To address these, we design a scalable framework that automatically constructs heterogeneous environments that are fully simulated, systematically broadening the space of function-calling scenarios. We further adapt a two-phase agent fine-tuning strategy: first endowing agents with fundamental agentic capabilities, then specializing them for domain-specific contexts. Extensive experiments on agentic benchmarks, tau-bench, tau2-Bench, and ACEBench, demonstrate that our trained model, AgentScaler, significantly enhances the function-calling capability of models.

Shirong Ma, Shen Huang, Shulin Huang et al.

Large Language Models (LLMs) pre-trained on massive corpora have exhibited remarkable performance on various NLP tasks. However, applying these models to specific domains still poses significant challenges, such as lack of domain knowledge, limited capacity to leverage domain knowledge and inadequate adaptation to domain-specific data formats. Considering the exorbitant cost of training LLMs from scratch and the scarcity of annotated data within particular domains, in this work, we focus on domain-specific continual pre-training of LLMs using E-commerce domain as an exemplar. Specifically, we explore the impact of continual pre-training on LLMs employing unlabeled general and E-commercial corpora. Furthermore, we design a mixing strategy among different data sources to better leverage E-commercial semi-structured data. We construct multiple tasks to assess LLMs' few-shot In-context Learning ability and their zero-shot performance after instruction tuning in E-commerce domain. Experimental results demonstrate the effectiveness of continual pre-training of E-commerce LLMs and the efficacy of our devised data mixing strategy.

Runnan Fang, Yuan Liang, Xiaobin Wang et al.

Large Language Models (LLMs) based agents excel at diverse tasks, yet they suffer from brittle procedural memory that is manually engineered or entangled in static parameters. In this work, we investigate strategies to endow agents with a learnable, updatable, and lifelong procedural memory. We propose Memp that distills past agent trajectories into both fine-grained, step-by-step instructions and higher-level, script-like abstractions, and explore the impact of different strategies for Build, Retrieval, and Update of procedural memory. Coupled with a dynamic regimen that continuously updates, corrects, and deprecates its contents, this repository evolves in lockstep with new experience. Empirical evaluation on TravelPlanner and ALFWorld shows that as the memory repository is refined, agents achieve steadily higher success rates and greater efficiency on analogous tasks. Moreover, procedural memory built from a stronger model retains its value: migrating the procedural memory to a weaker model yields substantial performance gains.

Haiquan Zhao, Chenhan Yuan, Fei Huang et al.

As large language models (LLMs) become more capable and widely used, ensuring the safety of their outputs is increasingly critical. Existing guardrail models, though useful in static evaluation settings, face two major limitations in real-world applications: (1) they typically output only binary "safe/unsafe" labels, which can be interpreted inconsistently across diverse safety policies, rendering them incapable of accommodating varying safety tolerances across domains; and (2) they require complete model outputs before performing safety checks, making them fundamentally incompatible with streaming LLM inference, thereby preventing timely intervention during generation and increasing exposure to harmful partial outputs. To address these challenges, we present Qwen3Guard, a series of multilingual safety guardrail models with two specialized variants: Generative Qwen3Guard, which casts safety classification as an instruction-following task to enable fine-grained tri-class judgments (safe, controversial, unsafe); and Stream Qwen3Guard, which introduces a token-level classification head for real-time safety monitoring during incremental text generation. Both variants are available in three sizes (0.6B, 4B, and 8B parameters) and support up to 119 languages and dialects, providing comprehensive, scalable, and low-latency safety moderation for global LLM deployments. Evaluated across English, Chinese, and multilingual benchmarks, Qwen3Guard achieves state-of-the-art performance in both prompt and response safety classification. All models are released under the Apache 2.0 license for public use.

Yulin Chen, Ning Ding, Xiaobin Wang et al.

Consistently scaling pre-trained language models (PLMs) imposes substantial burdens on model adaptation, necessitating more efficient alternatives to conventional fine-tuning. Given the advantage of prompting in the zero-shot setting and the observed performance fluctuation among different prompts, we explore the instance-level prompt and their generalizability. By searching through the prompt space, we first validate the assumption that for every instance, there is almost always a lottery prompt that induces the correct prediction from the PLM, and such prompt can be obtained at a low cost thanks to the inherent ability of PLMs. Meanwhile, we find that some strong lottery prompts have high performance over the whole training set, and they are equipped with distinguishable linguistic features. Lastly, we attempt to generalize the searched strong lottery prompts to unseen data with prompt ensembling method without any parameter tuning. Experiments are conducted on various types of NLP classification tasks and demonstrate that the proposed method can achieve comparable results with other gradient-free and optimization-free baselines.

Shuqing Shi, Xiaobin Wang, Zhiyou Yang et al.

We consider the trade-off problem between exploration and exploitation under finite discounted Markov Decision Process, where the state transition matrix of the underlying environment stays unknown. We propose a double Thompson sampling reinforcement learning algorithm(DTS) to solve this kind of problem. This algorithm achieves a total regret bound of $\tilde{\mathcal{O}}(D\sqrt{SAT})$in time horizon $T$ with $S$ states, $A$ actions and diameter $D$. DTS consists of two parts, the first part is the traditional part where we apply the posterior sampling method on transition matrix based on prior distribution. In the second part, we employ a count-based posterior update method to balance between the local optimal action and the long-term optimal action in order to find the global optimal game value. We established a regret bound of $\tilde{\mathcal{O}}(\sqrt{T}/S^{2})$. Which is by far the best regret bound for finite discounted Markov Decision Process to our knowledge. Numerical results proves the efficiency and superiority of our approach.

Ning Ding, Guangwei Xu, Yulin Chen et al.

Recently, considerable literature has grown up around the theme of few-shot named entity recognition (NER), but little published benchmark data specifically focused on the practical and challenging task. Current approaches collect existing supervised NER datasets and re-organize them to the few-shot setting for empirical study. These strategies conventionally aim to recognize coarse-grained entity types with few examples, while in practice, most unseen entity types are fine-grained. In this paper, we present Few-NERD, a large-scale human-annotated few-shot NER dataset with a hierarchy of 8 coarse-grained and 66 fine-grained entity types. Few-NERD consists of 188,238 sentences from Wikipedia, 4,601,160 words are included and each is annotated as context or a part of a two-level entity type. To the best of our knowledge, this is the first few-shot NER dataset and the largest human-crafted NER dataset. We construct benchmark tasks with different emphases to comprehensively assess the generalization capability of models. Extensive empirical results and analysis show that Few-NERD is challenging and the problem requires further research. We make Few-NERD public at https://ningding97.github.io/fewnerd/.

Ning Ding, Xiaobin Wang, Yao Fu et al.

Recognizing relations between entities is a pivotal task of relational learning. Learning relation representations from distantly-labeled datasets is difficult because of the abundant label noise and complicated expressions in human language. This paper aims to learn predictive, interpretable, and robust relation representations from distantly-labeled data that are effective in different settings, including supervised, distantly supervised, and few-shot learning. Instead of solely relying on the supervision from noisy labels, we propose to learn prototypes for each relation from contextual information to best explore the intrinsic semantics of relations. Prototypes are representations in the feature space abstracting the essential semantics of relations between entities in sentences. We learn prototypes based on objectives with clear geometric interpretation, where the prototypes are unit vectors uniformly dispersed in a unit ball, and statement embeddings are centered at the end of their corresponding prototype vectors on the surface of the ball. This approach allows us to learn meaningful, interpretable prototypes for the final classification. Results on several relation learning tasks show that our model significantly outperforms the previous state-of-the-art models. We further demonstrate the robustness of the encoder and the interpretability of prototypes with extensive experiments.

Ning Ding, Dingkun Long, Guangwei Xu et al.

Fully supervised neural approaches have achieved significant progress in the task of Chinese word segmentation (CWS). Nevertheless, the performance of supervised models tends to drop dramatically when they are applied to out-of-domain data. Performance degradation is caused by the distribution gap across domains and the out of vocabulary (OOV) problem. In order to simultaneously alleviate these two issues, this paper proposes to couple distant annotation and adversarial training for cross-domain CWS. For distant annotation, we rethink the essence of "Chinese words" and design an automatic distant annotation mechanism that does not need any supervision or pre-defined dictionaries from the target domain. The approach could effectively explore domain-specific words and distantly annotate the raw texts for the target domain. For adversarial training, we develop a sentence-level training procedure to perform noise reduction and maximum utilization of the source domain information. Experiments on multiple real-world datasets across various domains show the superiority and robustness of our model, significantly outperforming previous state-of-the-art cross-domain CWS methods.

Roy Adams, Yuelong Ji, Xiaobin Wang et al.

Measurement error in observational datasets can lead to systematic bias in inferences based on these datasets. As studies based on observational data are increasingly used to inform decisions with real-world impact, it is critical that we develop a robust set of techniques for analyzing and adjusting for these biases. In this paper we present a method for estimating the distribution of an outcome given a binary exposure that is subject to underreporting. Our method is based on a missing data view of the measurement error problem, where the true exposure is treated as a latent variable that is marginalized out of a joint model. We prove three different conditions under which the outcome distribution can still be identified from data containing only error-prone observations of the exposure. We demonstrate this method on synthetic data and analyze its sensitivity to near violations of the identifiability conditions. Finally, we use this method to estimate the effects of maternal smoking and opioid use during pregnancy on childhood obesity, two import problems from public health. Using the proposed method, we estimate these effects using only subject-reported drug use data and substantially refine the range of estimates generated by a sensitivity analysis-based approach. Further, the estimates produced by our method are consistent with existing literature on both the effects of maternal smoking and the rate at which subjects underreport smoking.