Named Entity Recognition
DOCA - Database of Variables for Content Analysis · 08.08.2026
DOI: 10.34778/qat0wh79 ↗Kurz gefasst
Fachartikel in der Zeitschrift DOCA - Database of Variables for Content Analysis, online erschienen am 08.08.2026. Verfasst von Jonas Baur und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Named Entity Recognition (NER) is the task of identifying and categorizing references to entities — such as persons, organizations, places, or events — within text (Jehangir et al., 2023). State-of-the-art NER models are usually transformer-based (Balluff et al., 2024), though rule-based, unsupervised, and more recently Large Language Model-based approaches are also in use. In the social sciences, NER is commonly applied as a pre-processing step, for instance to study the salience or sentiment of actors in media coverage. Field of Application/Theoretical Foundation Named Entity Recognition is the “task of recognizing and categorizing named entities that are presented in a text document” (Jehangir et al., 2023, p. 1), where anything denotable by a proper name can count as a named entity (Jurafsky & Martin, 2026, p. 387). In the social sciences, named entities are often actors broadly defined: individual persons as well as collective actors such as political parties or social movements (Schwarz, 2025), but also events (e.g., “Swiss national election 2023”), legal constructs (e.g., “EU AI Act”), or infrastructure projects (e.g., “Nord Stream”) (Balluff et al., 2024). NER is generally regarded as a pre-processing technique (Munnangi, 2024) useful for many social science applications. Common applications in communication research include studying the salience of actors in media coverage (Ort et al., 2023; Wozniak et al., 2026; Zai et al., 2025) or actor visibility on social media (Kruikemeier et al., 2018). To identify entities, there exists a plethora of NER approaches that can be categorized in many ways. Seow et al. (2025) provide the most recent and most comprehensive review. We use a broad categorization for the purpose of this entry. First, rule-based NER approaches use a carefully designed collection of rules to identify entities by relying on tailored dictionaries or sets of task-specific heuristics (Seow et al., 2025). This approach works well for narrow, well-defined tasks, such as identifying specific politicians in text (e.g., Gattermann, 2018). However, rule-based approaches don’t generalize well across languages or domains or even research questions (Jehangir et al., 2023). Second, unsupervised approaches rely on techniques that identify entities without labeled data (Jehangir et al., 2023). Computer science research has developed such approaches (e.g., Iovine et al., 2022; Zhang & Elhadad, 2013), whereby the model learns to identify entities from the data alone. One common application is the detection of drug names in biomedical sciences (Zhang & Elhadad, 2013), yet such approaches are rarely used in the social sciences. Unsupervised approaches are versatile and don’t require labeled data. However, their biggest difficulty is that their application typically lacks a ground truth (“gold standard”), which makes validation difficult (Seow et al., 2025). Third, supervised approaches rely on labeled data to classify entities. Typical applications use feature extraction methods such as part-of-speech tagging and subsequently learn these patterns using machine learning algorithms (Munnangi, 2024). A variety of algorithms can be used, ranging from more classical models like Hidden Markov models (Zhou & Su, 2002) to deep learning models (Jehangir et al., 2023). More recently, transfer learning approaches have gained traction. Balluff et al. (2024) find that these transformer-based approaches perform best on social science applications. Models trained in a supervised manner are sometimes applied in different contexts without additional labeled data. A common ready-to-use pipeline is the spaCy text processing toolkit for Python or R (Honnibal et al., 2020). Fourth, computer science literature has developed NER approaches that use a Large Language Model (LLM) with zero-shot or few-shot learning, where the LLM is prompted to identify and categorize entities without examples (zero-shot) or with a small number of examples (few-shot) (e.g., Wang et al., 2025; Zhu & Liu, 2025). However, existing tools still face several shortcomings: most are optimized for English text and specific domains, and they generally cannot disambiguate entities that share the same surface form (e.g., two different people both named "John Smith") without an additional entity-linking step (Schwarz, 2025). Of note, the literature on natural language processing treats this last step as a different but related problem called Named Entity Disambiguation (Al-Moslmi et al., 2020). References / Combination with other Methods Because NER is a pre-processing technique (Munnangi, 2024), it rarely stands alone. Often, the extracted entities are subsequently used in combination with other methods. The output of NER can be used to measure similarity between texts through shared named entities, for example to study news influence (Welbers et al., 2018) or news story chains (Gedikli et al., 2021). In a similar vein, co-occurring entities within documents can be used to construct networks of actors (Traag et al., 2015). With such an approach networks of political elites can be constructed from news coverage (e.g., Bro, 2025). Other applications use NER to analyze targeted sentiment towards actors (e.g., Fogel-Dror et al., 2019). Example Studies To illustrate how the rule-based, unsupervised, supervised, and LLM-based approaches discussed above translate into applied social science research, the table below compares four studies that differ in scope, entity taxonomy, and language, but together span this continuum (Licht & Sczepanski, 2025; Nenno, 2025; Ort et al., 2023; Zaratiana et al., 2023). Three of the four studies develop and validate a custom NER approach; the fourth illustrates the common case in which NER is used as an off-the-shelf pre-processing step rather than as the object of methodological evaluation. Table 1: Comparison of Named Entity Recognition approaches. Article Approach Entity types/domain Language(s) Validation/performance Nenno (2025) Domain-specific supervised NER via continuous fine-tuning (CFT): a transformer (XLM-R) first fine-tuned on weakly labeled data (dictionary matching against a database of public speakers), then further fine-tuned on a small manually annotated set 5 custom categories specific to public discourse: politicians, parties, authorities, media, journalists ("Public Entity Recognition") German (training); adapted to English and Spanish Outperforms the rule-based dictionary baseline (F1 .37–.61) by a wide margin, reaching F1 ≈ .8 on in-domain, out-of-domain, and Twitter data; with only ~100–150 additional labeled sentences, performance on English/Spanish improves from ≈ .6 (zero-shot) to ≈ .8 Licht & Sczepanski (2025) Supervised token classification (BIO-tagging) by fine-tuning transformer encoders (DeBERTa/RoBERTa) on human-annotated word spans Broadly defined "social group mentions" in political rhetoric (e.g., workers, refugees, "ordinary working people") English (UK party manifestos, parliamentary speech); cross-lingual transfer tested on German Mention-level F1 = .87 (cross-validated); document-level counts correlate strongly with an independent manual content analysis (Thau, 2019; r = .94); reliable few-shot transfer to new parties, domains, and languages with a few hundred additional labeled sentences Zaratiana et al. (2023) Generalist zero-/few-shot NER: a small bidirectional transformer matches entity-type embeddings to text-span embeddings, avoiding costly autoregressive LLM prompting Open/arbitrary entity types, specified in natural language at inference time rather than fixed in advance Trained on English data; generalizes zero-shot to several other languages Outperforms ChatGPT and several fine-tuned LLMs (e.g., UniNER) on zero-shot out-of-domain NER benchmarks, despite being orders of magnitude smaller (a 50M-parameter version still beats multi-billion-parameter models) Ort et al. (2023) Off-the-shelf NER (spaCy) applied without custom training, followed by manual classification of the most frequent entities into broader actor categories General-purpose entities (persons, organizations, places), grouped post hoc into four societal actor types (politics, government, economy, science) German and French (Swiss news coverage) No dedicated NER validation reported; relies on spaCy's established benchmark performance plus a manual face-validity check of the 150 most frequent entities; NER serves purely as a pre-processing step feeding into a downstream statistical analysis References Al-Moslmi, T., Gallofré Ocaña, M., Opdahl, A. L., & Veres, C. (2020). Named Entity Extraction for Knowledge Graphs: A Literature Overview. IEEE Access, 8, 32862–32881. https://doi.org/10.1109/ACCESS.2020.2973928 Balluff, P., Boomgaarden, H. G., & Waldherr, A. (2024). Automatically Finding Actors in Texts: A Performance Review of Multilingual Named Entity Recognition Tools. Communication Methods and Measures, 18(4), 371–389. https://doi.org/10.1080/19312458.2024.2324789 Bro, N. (2025). A frustratingly easy way of extracting political networks from text. PLOS ONE, 20(1), e0313149. https://doi.org/10.1371/journal.pone.0313149 Fogel-Dror, Y., Shenhav, S. R., Sheafer, T., & Van Atteveldt, W. (2019). Role-based Association of Verbs, Actions, and Sentiments with Entities in Political Discourse. Communication Methods and Measures, 13(2), 69–82. https://doi.org/10.1080/19312458.2018.1536973 Gattermann, K. (2018). Mediated Personalization of Executive European Union Politics: Examining Patterns in the Broadsheet Coverage of the European Commission, 1992–2016. The International Journal of Press/Politics, 23(3), 345–366. https://doi.org/10.1177/1940161218779231 Gedikli, F., Stockem Novo, A., & Jannach, D. (2021). Semi-automated identification of news story chains: A new dataset and entity-based labeling method. Proceedings of the 9th International Workshop on News Recommendation and Analytics (INRA 2021), 29–42. https://ceur-ws.org/Vol-3143/paper3.pdf Iovine, A., Fang, A., Fetahu, B., Rokhlenko, O., & Malmasi, S. (2022). CycleNER: An Unsupervised Training Approach for Named Entity Recognition. Proceedings of the ACM Web Conference 2022, WWW ’22, 2916–2924. https://doi.org/10.1145/3485447.3512012 Jehangir, B., Radhakrishnan, S., & Agarwal, R. (2023). A survey on Named Entity Recognition—Datasets, tools, and methodologies. Natural Language Processing Journal, 3, 100017. https://doi.org/10.1016/j.nlp.2023.100017 Jurafsky, D., & Martin, J. H. (2026). Speech and Language Processing: An Introduction to Natural Language Processing, Computational Linguistics, and Speech Recognition with Language Models (3rd ed.). https://web.stanford.edu/~jurafsky/slp3/ Kruikemeier, S., Gattermann, K., & Vliegenthart, R. (2018). Understanding the dynamics of politicians’ visibility in traditional and social media. The Information Society, 34(4), 215–228. https://doi.org/10.1080/01972243.2018.1463334 Licht, H., & Sczepanski, R. (2025). Detecting Group Mentions in Political Rhetoric A Supervised Learning Approach. British Journal of Political Science, 55, e119. https://doi.org/10.1017/S0007123424000954 Munnangi, M. (2024). A Brief History of Named Entity Recognition (arXiv:2411.05057). arXiv. https://doi.org/10.48550/arXiv.2411.05057 Nenno, S. (2025). Bootstrapping public entities. Domain-specific NER for public speakers. Communication Methods and Measures, 19(2), 131–156. https://doi.org/10.1080/19312458.2024.2388098 Ort, A., Rohrbach, T., Diviani, N., & Rubinelli, S. (2023). Covering the Crisis: Evolution of Key Topics and Actors in COVID-19 News Coverage in Switzerland. International Journal of Public Health, 67, 1605240. https://doi.org/10.3389/ijph.2022.1605240 Schwarz, P. (2025). Named entity recognition and entity linking. In P. Bański, U. Heid, & L. Herzberg (Eds.), Harmonizing language data: Standards for linguistic resources (pp. 89–114). De Gruyter. https://doi.org/doi:10.1515/9783112208212-005 Seow, W. L., Chaturvedi, I., Hogarth, A., Mao, R., & Cambria, E. (2025). A review of named entity recognition: From learning methods to modelling paradigms and tasks. Artificial Intelligence Review, 58(10), 315. https://doi.org/10.1007/s10462-025-11321-8 Traag, V. A., Reinanda, R., & Klinken, G. van. (2015). Elite Co-Occurrence in the Media. Asian Journal of Social Science, 43(5), 588–612. https://doi.org/10.1163/15685314-04305005 Wang, Z., Chen, H., Xu, G., & Ren, M. (2025). A novel large-language-model-driven framework for named entity recognition. Information Processing & Management, 62(3), 104054. https://doi.org/10.1016/j.ipm.2024.104054 Welbers, K., Van Atteveldt, W., Kleinnijenhuis, J., & Ruigrok, N. (2018). A Gatekeeper among Gatekeepers: News agency influence in print and online newspapers in the Netherlands. Journalism Studies, 19(3), 315–333. https://doi.org/10.1080/1461670X.2016.1190663 Wozniak, A., Liu, Z., & Lind, F. (2026). Media coverage of the Russo–Ukraine war beyond the West: Geopolitics and mainstream news in Brazil, India and South Africa. Media, War & Conflict, 19(1), 226–250. https://doi.org/10.1177/17506352241302244 Zai, F., Rohrbach, T., & Hänggli Fricker, R. (2025). Voices and media frames in the public debate on artificial intelligence: Comparing results from manual and automated content analysis. Frontiers in Communication, 10, 1599854. https://doi.org/10.3389/fcomm.2025.1599854 Zaratiana, U., Tomeh, N., Holat, P., & Charnois, T. (2023). GLiNER: Generalist Model for Named Entity Recognition using Bidirectional Transformer (arXiv:2311.08526). arXiv. https://doi.org/10.48550/arXiv.2311.08526 Zhang, S., & Elhadad, N. (2013). Unsupervised biomedical named entity recognition: Experiments with clinical and biological texts. Journal of Biomedical Informatics, Special Section: Social Media Environments, 46(6), 1088–1098. https://doi.org/10.1016/j.jbi.2013.08.004 Zhou, G., & Su, J. (2002). Named Entity Recognition using an HMM-based Chunk Tagger. In P. Isabelle, E. Charniak, & D. Lin (Eds.), Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics (pp. 473–480). Association for Computational Linguistics. https://doi.org/10.3115/1073083.1073163 Zhu, Y., & Liu, Y. (2025). LLM-NER: Advancing Named Entity Recognition with LoRA+ Fine-Tuned Large Language Models. 2025 11th International Conference on Computing and Artificial Intelligence (ICCAI), 364–368. https://doi.org/10.1109/ICCAI66501.2025.00063
Zitieren
APA
Baur, J., & Rohrbach, T. (2026). Named Entity Recognition. DOCA - Database of Variables for Content Analysis. https://doi.org/10.34778/qat0wh79
Vancouver
Baur J, Rohrbach T. Named Entity Recognition. DOCA - Database of Variables for Content Analysis. 2026. doi:10.34778/qat0wh79
BibTeX
@article{Baur2026,
author = {Baur, Jonas and Rohrbach, Tobias},
title = {Named Entity Recognition},
journal = {DOCA - Database of Variables for Content Analysis},
year = {2026},
doi = {10.34778/qat0wh79}
}
EUVIMED