The rapid expansion of scientific literature makes it increasingly difficult to acquire new knowledge, particularly in specialized domains where reasoning is complex, full-text access is restricted, and target references are sparse among a large set of candidates. We present a Deep Reinforcement Learning framework for sparse reference selection that emulates human knowledge construction, prioritizing which papers to read under limited time and cost. Evaluated on drug--gene relation discovery with access restricted to titles and abstracts, our approach demonstrates that both humans and machines can construct knowledge effectively from partial information.

Full-text error correction with Large Language Models (LLMs) for Automatic Speech Recognition (ASR) is attracting increased attention for its ability to address a wide range of error types, such as punctuation restoration and inverse text normalization, across long context. However, challenges remain regarding stability, controllability, completeness, and fluency. To mitigate these issues, this paper proposes the Chain of Correction (CoC), which uses a multi-turn chat format to correct errors segment by segment, guided by pre-recognized text and full-text context for better semantic understanding. Utilizing the open-sourced ChFT dataset, we fine-tune a pre-trained LLM to evaluate CoC's performance. Experiments show that CoC significantly outperforms baseline and benchmark systems in correcting full-text ASR outputs. We also analyze correction thresholds to balance under-correction and over-rephrasing, extrapolate CoC on extra-long ASR outputs, and explore using other types of information to guide error correction.

Academic misconduct detection in biomedical research remains challenging due to algorithmic narrowness in existing methods and fragmented analytical pipelines. We present BMDetect, a multimodal deep learning framework that integrates journal metadata (SJR, institutional data), semantic embeddings (PubMedBERT), and GPT-4o-mined textual attributes (methodological statistics, data anomalies) for holistic manuscript evaluation. Key innovations include: (1) multimodal fusion of domain-specific features to reduce detection bias; (2) quantitative evaluation of feature importance, identifying journal authority metrics (e.g., SJR-index) and textual anomalies (e.g., statistical outliers) as dominant predictors; and (3) the BioMCD dataset, a large-scale benchmark with 13,160 retracted articles and 53,411 controls. BMDetect achieves 74.33% AUC, outperforming single-modality baselines by 8.6%, and demonstrates transferability across biomedical subfields. This work advances scalable, interpretable tools for safeguarding research integrity.

The volume and diversity of digital information have led to a growing reliance on Machine Learning techniques, such as Natural Language Processing, for interpreting and accessing appropriate data. While vector and graph embeddings represent data for similarity tasks, current state-of-the-art pipelines lack guaranteed explainability, failing to determine similarity for given full texts accurately. These considerations can also be applied to classifiers exploiting generative language models with logical prompts, which fail to correctly distinguish between logical implication, indifference, and inconsistency, despite being explicitly trained to recognise the first two classes. We present a novel pipeline designed for hybrid explainability to address this. Our methodology combines graphs and logic to produce First-Order Logic representations, creating machine- and human-readable representations through Montague Grammar. Preliminary results indicate the effectiveness of this approach in accurately capturing full text similarity. To the best of our knowledge, this is the first approach to differentiate between implication, inconsistency, and indifference for text classification tasks. To address the limitations of existing approaches, we use three self-contained datasets annotated for the former classification task to determine the suitability of these approaches in capturing sentence structure equivalence, logical connectives, and spatiotemporal reasoning. We also use these data to compare the proposed method with language models pre-trained for detecting sentence entailment. The results show that the proposed method outperforms state-of-the-art models, indicating that natural language understanding cannot be easily generalised by training over extensive document corpora. This work offers a step toward more transparent and reliable Information Retrieval from extensive textual data.

Many peer-review venues are using algorithms to assign submissions to reviewers. The crux of such automated approaches is the notion of the "similarity score" -- a numerical estimate of the expertise of a reviewer in reviewing a paper -- and many algorithms have been proposed to compute these scores. However, these algorithms have not been subjected to a principled comparison, making it difficult for stakeholders to choose the algorithm in an evidence-based manner. The key challenge in comparing existing algorithms and developing better algorithms is the lack of publicly available gold-standard data. We address this challenge by collecting a novel dataset of similarity scores that we release to the research community. Our dataset consists of 477 self-reported expertise scores provided by 58 researchers who evaluated their expertise in reviewing papers they have read previously. Using our dataset, we compare several widely used similarity algorithms and offer key insights. First, all algorithms exhibit significant error, with misranking rates between 12%-30% in easier cases and 36%-43% in harder ones. Second, most specialized algorithms are designed to work with titles and abstracts of papers, and in this regime the SPECTER2 algorithm performs best. Interestingly, classical TF-IDF matches SPECTER2 in accuracy when given access to full submission texts. In contrast, off-the-shelf LLMs lag behind specialized approaches.

The tendency of generative artificial intelligence (AI) sys tems to "hallucinate" false information is well-known; AI-generated cit ations to nonexistent sources have made their way into the reference list s of peer-reviewed publications. Here, I propose a solution to this pr oblem, taking inspiration from the T ransparency and Openness Promotion ( TOP) data sharing guidelines, the clash of generative AI with the Amer ican judiciary, and the precedent set by submissions of prior art to the Unite d States Patent and T rademark Office. Journals should require authors to sub mit the full text of each cited source along with their manuscripts, ther eby preventing authors from citing any material whose full text they cannot produce. This solution requires limited additional work on the part of aut hors or editors while effectively immunizing journals against hallucinat ed references. Within the same month, commenters on Pub-Peer raised concerns regarding the article's reference list.

This study explores strategies for efficiently classifying scientific full texts using both small, BERT-based models and local large language models like Llama-3.1 8B. We focus on developing methods for selecting subsets of input sentences to reduce input size while simultaneously enhancing classification performance. To this end, we compile a novel dataset consisting of full-text scientific papers from the field of invasion biology, specifically addressing the impacts of invasive species. These papers are aligned with publicly available impact assessments created by researchers for the International Union for Conservation of Nature (IUCN). Through extensive experimentation, we demonstrate that various sources like human evidence annotations, LLM-generated annotations or explainability scores can be used to train sentence selection models that improve the performance of both encoder- and decoder-based language models while optimizing efficiency through the reduction in input length, leading to improved results even if compared to models like ModernBERT that are able to handle the complete text as input. Additionally, we find that repeated sampling of shorter inputs proves to be a very effective strategy that, at a slightly increased cost, can further improve classification performance.

While previous studies have demonstrated that Large Language Models (LLMs) can predict peer review outcomes to some extent, this paper builds on that by introducing two new contexts and employing a more robust method - averaging multiple ChatGPT scores. The findings that averaging 30 ChatGPT predictions, based on reviewer guidelines and using only the submitted titles and abstracts, failed to predict peer review outcomes for F1000Research (Spearman's rho=0.00). However, it produced mostly weak positive correlations with the quality dimensions of SciPost Physics (rho=0.25 for validity, rho=0.25 for originality, rho=0.20 for significance, and rho = 0.08 for clarity) and a moderate positive correlation for papers from the International Conference on Learning Representations (ICLR) (rho=0.38). Including the full text of articles significantly increased the correlation for ICLR (rho=0.46) and slightly improved it for F1000Research (rho=0.09), while it had variable effects on the four quality dimension correlations for SciPost LaTeX files. The use of chain-of-thought system prompts slightly increased the correlation for F1000Research (rho=0.10), marginally reduced it for ICLR (rho=0.37), and further decreased it for SciPost Physics (rho=0.16 for validity, rho=0.18 for originality, rho=0.18 for significance, and rho=0.05 for clarity). Overall, the results suggest that in some contexts, ChatGPT can produce weak pre-publication quality assessments. However, the effectiveness of these assessments and the optimal strategies for employing them vary considerably across different platforms, journals, and conferences. Additionally, the most suitable inputs for ChatGPT appear to differ depending on the platform.

Topic models are used to identify and group similar themes in a set of documents. Recent advancements in deep learning based neural topic models has received significant research interest. In this paper, an approach is proposed that further enhances topic modeling performance by utilizing a pre-trained large language model (LLM) to generate summaries of documents before inputting them into the topic model. Few shot prompting is used to generate summaries of different lengths to compare their impact on topic modeling. This approach is particularly effective for larger documents because it helps capture the most essential information while reducing noise and irrelevant details that could obscure the overall theme. Additionally, it is observed that datasets exhibit an optimal summary length that leads to improved topic modeling performance. The proposed method yields better topic diversity and comparable coherence values compared to previous models.

Evaluating the quality of academic journal articles is a time consuming but critical task for national research evaluation exercises, appointments and promotion. It is therefore important to investigate whether Large Language Models (LLMs) can play a role in this process. This article assesses which ChatGPT inputs (full text without tables, figures and references; title and abstract; title only) produce better quality score estimates, and the extent to which scores are affected by ChatGPT models and system prompts. The results show that the optimal input is the article title and abstract, with average ChatGPT scores based on these (30 iterations on a dataset of 51 papers) correlating at 0.67 with human scores, the highest ever reported. ChatGPT 4o is slightly better than 3.5-turbo (0.66), and 4o-mini (0.66). The results suggest that article full texts might confuse LLM research quality evaluations, even though complex system instructions for the task are more effective than simple ones. Thus, whilst abstracts contain insufficient information for a thorough assessment of rigour, they may contain strong pointers about originality and significance. Finally, linear regression can be used to convert the model scores into the human scale scores, which is 31% more accurate than guessing.