Lutz Bornmann

Hongkan Chen, Lutz Bornmann, Yi Bu

The temporal dimension of citation accumulation poses fundamental challenges for quantitative research evaluations, particularly in assessing disruptive and consolidating research through the disruption index (D). While prior studies emphasize minimum citation windows (mostly 3-5 years) for reliable citation impact measurements, the time-sensitive nature of D - which quantifies a paper' s capacity to eclipse prior knowledge - remains underexplored. This study addresses two critical gaps: (1) determining the temporal thresholds required for publications to meet citation/reference prerequisites, and (2) identifying "optimal" citation windows that balance early predictability and longitudinal validity. By analyzing millions of publications across four fields with varying citation dynamics, we employ some metrics to track D stabilization patterns. Key findings reveal that a 10-year window achieves >80% agreement with final D classifications, while shorter windows (3 years) exhibit instability. Publications with >=30 references stabilize 1-3 years faster, and extreme cases (top/bottom 5% D values) become identifiable within 5 years - enabling early detection of 60-80% of highly disruptive and consolidating works. The findings offer significant implications for scholarly evaluation and science policy, emphasizing the need for careful consideration of citation window length in research assessment (based on D).

Mengjia Wu, Yi Zhang, Robin Haunschild et al.

Assessing the quality of scientific research is essential for scholarly communication, yet widely used approaches face limitations in scalability, subjectivity, and time delay. Recent advances in large language models (LLMs) offer new opportunities for automated research evaluation based on textual content. This study examines whether LLMs can support post-publication peer review tasks by benchmarking their outputs against expert judgments and citation-based indicators. Two evaluation tasks are constructed using articles from the H1 Connect platform: identifying high-quality articles and performing finer-grained evaluation including article rating, merit classification, and expert style commenting. Multiple model families, including BERT models, general-purpose LLMs, and reasoning oriented LLMs, are evaluated under multiple learning strategies. Results show that LLMs perform well in coarse grained evaluation tasks, achieving accuracy above 0.8 in identifying highly recommended articles. However, performance decreases substantially in fine-grained rating tasks. Few-shot prompting improves performance over zero-shot settings, while supervised fine-tuning produces the strongest and most balanced results. Retrieval augmented prompting improves classification accuracy in some cases but does not consistently strengthen alignment with citation indicators. The overall correlations between model outputs and citation indicators remain positive but moderate.

Christian Leibel, Lutz Bornmann

Although bibliometrics has become an essential tool in the evaluation of research performance, bibliometric analyses are sensitive to a range of methodological choices. Subtle choices in data selection, indicator construction, and modeling decisions can substantially alter results. Ensuring robustness (meaning that findings hold up under different reasonable scenarios) is therefore critical for credible research and research evaluation. To address this issue, this study introduces multiverse analysis to bibliometrics. Multiverse analysis is a statistical tool that enables analysts to transparently discuss modeling assumptions and thoroughly assess model robustness. Whereas standard robustness checks usually cover only a small subset of all plausible models, multiverse analysis includes all plausible models. The benefits of multiverse analysis are illustrated by assessing the robustness of the findings reported by Wu et al. (2019), who observed that small teams tend to produce more disruptive research than large teams. While we found robust evidence of a negative effect of team size on disruption scores, the effect size depends substantially on the model specification. Our findings underscore the importance of assessing the multiverse robustness of bibliometric results to clarify their practical implications.

Christoph Schlager, Lutz Bornmann, Gerald Schweiger

Multiple organisational affiliations are an increasingly common feature of research systems, yet their implications for organisational performance had received limited systematic attention. We developed a scalable, network-based analytical framework that represents simultaneous researcher affiliations as relational links between organisations and applied it to bibliometric data from Austria. Using harmonised publication and affiliation metadata, we constructed two complementary co-affiliation networks: a complete network capturing all simultaneous affiliations and a temporally filtered network retaining only organisational pairs that recurred over time. Network regression analyses showed that geographical proximity remained an important determinant of co-affiliation formation, with spatial distance consistently reducing shared appointments. Clear sectoral differences emerged beyond geography. Universities formed a dense and persistent core of co-affiliations, whereas ties involving medical institutions, government, non-profit and private-sector organisations were often short-lived and attenuated under temporal filtering. Among crosssector links, co-affiliations between universities and research institutes were notably resilient, indicating a more structurally embedded form of organisational integration. We assessed the effect of concurrent affiliations on organisational citation impact across organisational types using field- and year-normalised indicators. Research institutes and universities consistently exhibited higher citation impact than organisations from other sectors, and persistent co-affiliations were associated with greater and more stable scientific visibility.

Robin Haunschild, Lutz Bornmann, Devendra Potnis et al.

One way to assess a certain aspect of the value of scientific research is to measure the attention it receives on social media. While previous research has mostly focused on the "number of mentions" of scientific research on social media, the current study applies "topic networks" to measure public attention to scientific research on Twitter. Topic networks are the networks of co-occurring author keywords in scholarly publications and networks of co-occurring hashtags in the tweets mentioning those scholarly publications. This study investigates which topics in opioid scholarly publications have received public attention on Twitter. Additionally, it investigates whether the topic networks generated from the publications tweeted by all accounts (bot and non-bot accounts) differ from those generated by non-bot accounts. Our analysis is based on a set of opioid scholarly publications from 2011 to 2019 and the tweets associated with them. We use co-occurrence network analysis to generate topic networks. Results indicated that Twitter users have mostly used generic terms to discuss opioid publications, such as "Opioid," "Pain," "Addiction," "Treatment," "Analgesics," "Abuse," "Overdose," and "Disorders." Results confirm that topic networks provide a legitimate method to visualize public discussions of health-related scholarly publications and how Twitter users discuss health-related scientific research differently from the scientific community. There was a substantial overlap between the topic networks based on the tweets by all accounts and non-bot accounts. This result indicates that it might not be necessary to exclude bot accounts for generating topic networks as they have a negligible impact on the results.

Sehrish Iqbal, Saeed-Ul Hassan, Naif Radi Aljohani et al.

Citation analysis is one of the most frequently used methods in research evaluation. We are seeing significant growth in citation analysis through bibliometric metadata, primarily due to the availability of citation databases such as the Web of Science, Scopus, Google Scholar, Microsoft Academic, and Dimensions. Due to better access to full-text publication corpora in recent years, information scientists have gone far beyond traditional bibliometrics by tapping into advancements in full-text data processing techniques to measure the impact of scientific publications in contextual terms. This has led to technical developments in citation context and content analysis, citation classifications, citation sentiment analysis, citation summarisation, and citation-based recommendation. This article aims to narratively review the studies on these developments. Its primary focus is on publications that have used natural language processing and machine learning techniques to analyse citations.

Robin Haunschild, Sven E. Hug, Martin P. Brändle et al.

In the context of a comprehensive Microsoft Academic (MA) study, we explored in an initial step the quality of linked references data in MA in comparison with Web of Science (WoS). Linked references are the backbone of bibliometrics, because they are the basis of the times cited information in citation indexes. We found that the concordance of linked references between MA and WoS ranges from weak to non-existent for the full sample (publications of the University of Zurich with less than 50 linked references in MA). An analysis with a sample restricted to less than 50 linked references in WoS showed a strong agreement between linked references in MA and WoS.