Large language models have enabled automated agents to tackle complex scientific research tasks, yet most existing approaches struggle to deliver scientifically rigorous and verifiable outputs. In this work, we propose a novel agent framework, TrustSciAgent, which introduces a unified evidence–reasoning–validation pipeline explicitly governed by newly formulated scientific research trustworthiness principles. TrustSciAgent structurally organizes the entire research process into pre-research, in-research, and post-research phases, ensuring that each stage strictly adheres to these principles. This design compels the agent to generate transparent, logically sound reasoning chains and deliver auditable scientific conclusions. Comprehensive experiments across four scientific domains and three representative language models demonstrate that TrustSciAgent consistently improves both the structural completeness and the correctness of reasoning outputs, outperforming standard LLM-based agents. Our results provide strong evidence that embedding domain-agnostic trustworthiness principles into the agent workflow is critical for enabling credible, generalizable, and verifiable automated scientific research.

Named Entity Recognition (NER) is a key task to support biomedical research. In Biomedical Named Entity Recognition (BioNER), obtaining high-quality expert annotated data is laborious and expensive, leading to the development of automatic approaches such as distant supervision. However, manually and automatically generated data often suffer from the unlabeled entity problem, whereby many entity annotations are missing, degrading the performance of full annotation NER models. To conquer this issue, we undertake a systematic exploration of the efficacy of partial annotation learning methods for BioNER, which encompasses a comprehensive evaluation conducted across a spectrum of distinct simulated scenarios of missing entity annotations. Furthermore, we propose a TS-PubMedBERT-Partial-CRF partial annotation learning model. We standardize a compilation of 16 BioNER corpora, encompassing a range of five distinct entity types, to establish a gold standard. And we compare against the state-of-the-art partial annotation model EER-PubMedBERT, the widely acknowledged partial annotation model BiLSTM-Partial-CRF model, and the state-of-the-art full annotation learning BioNER model PubMedBERT tagger. Results show that partial annotation learning-based methods can effectively learn from biomedical corpora with missing entity annotations. Our proposed model outperforms alternatives and, specifically, the PubMedBERT tagger by 38% in F1-score under high missing entity rates. Moreover, the recall of entity mentions in our model demonstrates a competitive alignment with the upper threshold observed on the fully annotated dataset.

Scientific survey papers play a pivotal role in advancing knowledge and scientific progress by providing concise summaries and analyses of research trends and findings. To facilitate better knowledge organization and analysis, we have undertaken the challenge of defining the scientific entity recognition task for survey papers and carefully curated a dataset that closely emulates real-world scenarios. The scientific entity recognition task presents unique challenges, including multi-label, low-resource, and nested scenarios. To address these challenges, we propose a unified framework based on the machine reading comprehension (MRC) paradigm. This framework not only supports nested and multi-label settings but also enables the effective transfer of information from high-resource categories to low-resource ones, ensuring adaptability and robustness. To further enhance performance, we introduce the Knowledge-Driven Prototype Guidance (KDPG) module, seamlessly integrated into a two-phase learning strategy. The KDPG module leverages prior knowledge and acts as an initial prototype-based manifold constraint, effectively harnessing the power of few-shot learning capabilities. Through this integration, our approach complements the classification learning tasks for entity recognition, resulting in improved accuracy and efficiency. Our experimental results validate the effectiveness of the proposed KDPG-enhanced MRC framework, showcasing its leading performance on publicly available datasets and our collected scientific survey paper dataset.