{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,24]],"date-time":"2026-06-24T06:51:07Z","timestamp":1782283867706,"version":"3.54.5"},"reference-count":24,"publisher":"Springer Science and Business Media LLC","issue":"5","license":[{"start":{"date-parts":[[2026,4,22]],"date-time":"2026-04-22T00:00:00Z","timestamp":1776816000000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"},{"start":{"date-parts":[[2026,4,22]],"date-time":"2026-04-22T00:00:00Z","timestamp":1776816000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"}],"funder":[{"DOI":"10.13039\/501100004115","name":"Gottfried Wilhelm Leibniz Universit\u00e4t Hannover","doi-asserted-by":"crossref","id":[{"id":"10.13039\/501100004115","id-type":"DOI","asserted-by":"crossref"}]}],"content-domain":{"domain":["link.springer.com"],"crossmark-restriction":false},"short-container-title":["Mach Learn"],"published-print":{"date-parts":[[2026,5]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>\n                    Generating context-aware, high-quality, and innovative scientific ideas remains a central challenge in AI-supported research. We introduce\n                    <jats:sc>SCI-IDEA<\/jats:sc>\n                    , a two-stage framework combining large language models (LLMs) with a specialised\n                    <jats:italic>Aha-Moment Detection<\/jats:italic>\n                    module for iterative idea refinement. The first stage extracts structured facets: objectives, methodology, evaluation, and future work from previous publications, compressing each paper to\n                    <jats:inline-formula>\n                      <jats:tex-math>$${\\sim }$$<\/jats:tex-math>\n                    <\/jats:inline-formula>\n                    200 tokens to enable scalable processing of researcher profiles that would otherwise exceed frontier LLM context windows. The second stage integrates these facets with a token-level embedding approach to identify novelty and context gaps, indicating candidate ideas as\n                    <jats:italic>Aha moments<\/jats:italic>\n                    when they surpass predefined thresholds for both novelty and surprise. We evaluate\n                    <jats:sc>SCI-IDEA<\/jats:sc>\n                    across 100 computer-science researcher profiles, 4 LLMs (GPT-4o, GPT\u22124.5, DeepSeek-32B, DeepSeek-70B), three embedding strategies, and 5 prompting configurations via a hybrid protocol of automated LLM-as-judge scoring with GPT\u22124.1 and a human evaluation with 15 PhD-level domain experts. These two evaluation signals diverge substantially: LLM-based scores exceed expert ratings by 3\u20134 points on a 10-point scale, with near-zero inter-rater correlation (\n                    <jats:inline-formula>\n                      <jats:tex-math>$$r = 0.02$$<\/jats:tex-math>\n                    <\/jats:inline-formula>\n                    \u20130.17,\n                    <jats:inline-formula>\n                      <jats:tex-math>$$p &gt; 0.05$$<\/jats:tex-math>\n                    <\/jats:inline-formula>\n                    ), indicating that automated scores reflect relative architectural comparisons rather than human-verified measures of absolute idea quality. Within the LLM-evaluated setting,\n                    <jats:sc>SCI-IDEA<\/jats:sc>\n                    with token-level embeddings achieves a mean quality score of 6.97, a modest but constant improvement over the LLM-only baseline\u00a0(\n                    <jats:inline-formula>\n                      <jats:tex-math>$$\\Delta = +0.20$$<\/jats:tex-math>\n                    <\/jats:inline-formula>\n                    points;\n                    <jats:inline-formula>\n                      <jats:tex-math>$$p &lt; 0.05$$<\/jats:tex-math>\n                    <\/jats:inline-formula>\n                    ) and driven primarily by improved feasibility scores\u00a0(\n                    <jats:inline-formula>\n                      <jats:tex-math>$$+0.21$$<\/jats:tex-math>\n                    <\/jats:inline-formula>\n                    points,\n                    <jats:inline-formula>\n                      <jats:tex-math>$$p &lt; 0.001$$<\/jats:tex-math>\n                    <\/jats:inline-formula>\n                    ). Expert evaluators independently confirmed this trend while assigning lower absolute scores. These findings suggest an architectural advantage of facet-based context modelling and token-level novelty detection, though their practical importance deserves validation through longitudinal studies with larger domain experts. This work is validated in the computer science domain, but its application to other fields requires further investigation.\n                  <\/jats:p>","DOI":"10.1007\/s10994-026-07036-8","type":"journal-article","created":{"date-parts":[[2026,4,22]],"date-time":"2026-04-22T04:32:48Z","timestamp":1776832368000},"update-policy":"https:\/\/doi.org\/10.1007\/springer_crossmark_policy","source":"Crossref","is-referenced-by-count":2,"title":["SCI-IDEA: Context-Aware Scientific Ideation Using Token and Sentence Embeddings"],"prefix":"10.1007","volume":"115","author":[{"given":"Farhana","family":"Keya","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Gollam","family":"Rabby","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"S\u00f6ren","family":"Auer","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Sahar","family":"Vahdati","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Prasenjit","family":"Mitra","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Mohamad Yaser","family":"Jaradeh","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"297","published-online":{"date-parts":[[2026,4,22]]},"reference":[{"key":"7036_CR1","doi-asserted-by":"publisher","unstructured":"Baek, J., Jauhar, S.K., Cucerzan, S., & Hwang, S.J. (2025). Researchagent: Iterative research idea generation over scientific literature with large language models, 6709\u20136738 https:\/\/doi.org\/10.18653\/V1\/2025.NAACL-LONG.342","DOI":"10.18653\/V1\/2025.NAACL-LONG.342"},{"key":"7036_CR2","doi-asserted-by":"publisher","unstructured":"Baulin, V., Cook, A., Friedman, D., Lumiruusu, J., Pashea, A., Rahman, S., & Waldeck, B. (2025). The discovery engine: A framework for ai-driven synthesis and navigation of scientific knowledge landscapes. CoRR https:\/\/doi.org\/10.48550\/ARXIV.2505.17500. arXiv: abs\/2505.17500","DOI":"10.48550\/ARXIV.2505.17500"},{"key":"7036_CR3","unstructured":"Booker, L. D., Bontis, N., Burstein, F., Linger, H., & Serenko, A. (2013). Understanding the practical relevance of academic research in knowledge management: A lay epistemic theory approach."},{"key":"7036_CR4","unstructured":"Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., Agarwal, S., Herbert-Voss, A., Krueger, G., Henighan, T., Child, R., Ramesh, A., Ziegler, D. M., Wu, J., Winter, C., ... Amodei, D. (2020). Language models are few-shot learners. CoRR arXiv:2005.14165"},{"key":"7036_CR5","doi-asserted-by":"publisher","unstructured":"Cataneo, F. B. (2026). Resisting correction: How RLHF makes language models ignore external safety signals in natural conversation. CoRR https:\/\/doi.org\/10.48550\/ARXIV.2601.08842. 10.48550\/ARXIV.2601.08842","DOI":"10.48550\/ARXIV.2601.08842"},{"key":"7036_CR6","doi-asserted-by":"publisher","unstructured":"DeepSeek-AI. (2024). Deepseek-v3 technical report. CoRR arXiv:2412.19437https:\/\/doi.org\/10.48550\/ARXIV.2412.19437","DOI":"10.48550\/ARXIV.2412.19437"},{"key":"7036_CR7","doi-asserted-by":"publisher","unstructured":"DeepSeek-AI. (2025). Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. CoRR arXiv:2501.12948https:\/\/doi.org\/10.48550\/ARXIV.2501.12948","DOI":"10.48550\/ARXIV.2501.12948"},{"key":"7036_CR8","doi-asserted-by":"publisher","unstructured":"Gottweis, J., Weng, W., Daryin, A. N., Tu, T., Palepu, A., Sirkovic, P., Myaskovsky, A., Weissenberger, F., Rong, K., Tanno, R., Saab, K., Popovici, D., Blum, J., Zhang, F., Chou, K., Hassidim, A., Gokturk, B., Vahdat, A., Kohli, P., ... Natarajan, V. (2025). Towards an AI co-scientist. CoRR https:\/\/doi.org\/10.48550\/ARXIV.2502.18864. arXiv:2502.18864","DOI":"10.48550\/ARXIV.2502.18864"},{"key":"7036_CR9","unstructured":"Gu, X., & Krenn, M. (2024). Interesting scientific idea generation using knowledge graphs and llms: Evaluations with 100 research group leaders. arXiv preprint arXiv:2405.17044"},{"key":"7036_CR10","doi-asserted-by":"publisher","unstructured":"Guo, S., Shariatmadari, A. H., Xiong, G., Huang, A., Kim, M., Williams, C. M., Bekiranov, S., & Zhang, A. (2025). Ideabench: Benchmarking large language models for research idea generation, 5888\u20135899 https:\/\/doi.org\/10.1145\/3711896.3737419","DOI":"10.1145\/3711896.3737419"},{"key":"7036_CR11","doi-asserted-by":"crossref","unstructured":"Hu, X., Fu, H., Wang, J., Wang, Y., Li, Z., Xu, R., Lu, Y., Jin, Y., Pan, L., & Lan, Z. (2025). NOVA: an iterative planning framework for enhancing scientific innovation with large language models. In: Che, W., Nabende, J., Shutova, E., Pilehvar, M.T. (eds.) Findings of the association for computational linguistics, ACL 2025, Vienna, Austria, July 27 - August 1, 2025. Findings of ACL, vol. ACL 2025, pp. 21330\u201321359. Association for Computational Linguistics, Stroudsburg, PA. https:\/\/aclanthology.org\/2025.findings-acl.1099\/","DOI":"10.18653\/v1\/2025.findings-acl.1099"},{"key":"7036_CR12","doi-asserted-by":"publisher","unstructured":"Li, R., Jing, L., Han, C., Zhou, J., & Du, X. (2024). Learning to generate research idea with dynamic control. CoRR arXiv:2412.14626https:\/\/doi.org\/10.48550\/ARXIV.2412.14626","DOI":"10.48550\/ARXIV.2412.14626"},{"key":"7036_CR13","doi-asserted-by":"publisher","unstructured":"Lin, E., Peng, Z., & Fang, Y. (2024). Evaluating and enhancing large language models for novelty assessment in scholarly publications. CoRR arXiv:2409.16605https:\/\/doi.org\/10.48550\/ARXIV.2409.16605","DOI":"10.48550\/ARXIV.2409.16605"},{"key":"7036_CR14","doi-asserted-by":"publisher","unstructured":"Lu, C., Lu, C., Lange, R.T., Foerster, J.N., Clune, J., & Ha, D. (2024). The AI scientist: Towards fully automated open-ended scientific discovery. CoRR arXiv:2408.06292https:\/\/doi.org\/10.48550\/ARXIV.2408.06292","DOI":"10.48550\/ARXIV.2408.06292"},{"key":"7036_CR15","doi-asserted-by":"publisher","unstructured":"Michinov, N., & Jeanson, S. (2023). Creativity in scientific research: Multidisciplinarity fosters depth of ideas among scientists in electronic \u201cbrainwriting\u201d groups. Hum. Factors,65(7), 1542\u20131553. https:\/\/doi.org\/10.1177\/00187208211048301","DOI":"10.1177\/00187208211048301"},{"key":"7036_CR16","doi-asserted-by":"publisher","unstructured":"OpenAI. (2023). GPT-4 technical report. CoRR arXiv:2303.08774https:\/\/doi.org\/10.48550\/ARXIV.2303.08774","DOI":"10.48550\/ARXIV.2303.08774"},{"key":"7036_CR17","unstructured":"OpenAI. (2025). GPT-4.5 System Card. Accessed: 01-02-2025. https:\/\/cdn.openai.com\/gpt-4-5-system-card-2272025.pdf"},{"issue":"2","key":"7036_CR18","doi-asserted-by":"publisher","first-page":"947","DOI":"10.1007\/S00146-022-01462-7","volume":"38","author":"G Papagni","year":"2023","unstructured":"Papagni, G., Pagter, J., Zafari, S., Filzmoser, M., & K\u00f6szegi, S. T. (2023). Artificial agents\u2019 explainability to support trust: considerations on timing and context. AI Soc., 38(2), 947\u2013960. https:\/\/doi.org\/10.1007\/S00146-022-01462-7","journal-title":"AI Soc."},{"key":"7036_CR19","doi-asserted-by":"publisher","unstructured":"Radensky, M., Shahid, S., Fok, R., Siangliulue, P., Hope, T., & Weld, D. S. (2024). Scideator: Human-llm scientific idea generation grounded in research-paper facet recombination. CoRR arXiv:2409.14634https:\/\/doi.org\/10.48550\/ARXIV.2409.14634","DOI":"10.48550\/ARXIV.2409.14634"},{"key":"7036_CR20","doi-asserted-by":"publisher","unstructured":"Shahid, S., Radensky, M., Fok, R., Siangliulue, P., Weld, D. S., & Hope, T. (2025). Literature-grounded novelty assessment of scientific ideas. CoRR arXiv:2506.22026https:\/\/doi.org\/10.48550\/ARXIV.2506.22026","DOI":"10.48550\/ARXIV.2506.22026"},{"issue":"5","key":"7036_CR21","doi-asserted-by":"publisher","first-page":"711","DOI":"10.1177\/1745691619850561","volume":"14","author":"EL Uhlmann","year":"2019","unstructured":"Uhlmann, E. L., Ebersole, C. R., Chartier, C. R., Errington, T. M., Kidwell, M. C., Lai, C. K., McCarthy, R. J., Riegelman, A., Silberzahn, R., & Nosek, B. A. (2019). Scientific utopia iii: Crowdsourcing science. Perspectives on Psychological Science, 14(5), 711\u2013733.","journal-title":"Perspectives on Psychological Science"},{"key":"7036_CR22","doi-asserted-by":"publisher","unstructured":"Wu, Z., Weber, T., & M\u00fcller, F. (2025). One does not simply meme alone: Evaluating co-creativity between llms and humans in the generation of humor, 1082\u20131092 https:\/\/doi.org\/10.1145\/3708359.3712094","DOI":"10.1145\/3708359.3712094"},{"key":"7036_CR23","doi-asserted-by":"publisher","unstructured":"Yu, Y., & Romero, D. M. (2024). Does the use of unusual combinations of datasets contribute to greater scientific impact? CoRR arXiv:2402.05024https:\/\/doi.org\/10.48550\/ARXIV.2402.05024","DOI":"10.48550\/ARXIV.2402.05024"},{"key":"7036_CR24","doi-asserted-by":"publisher","unstructured":"Zemke, H., Stahmann, P., & Janiesch, C. (2025). Leveraging generative artificial intelligence for design thinking in creative processes: A literature review. In: Bui, T.X. (ed.) 58th Hawaii International Conference on System Sciences, HICSS 2025, Hilton Waikoloa Village, Hawaii, USA, January 7-10, 2025, pp. 1\u201310. ScholarSpace, Honolulu, HI, USA. https:\/\/doi.org\/10.24251\/HICSS.2025.018","DOI":"10.24251\/HICSS.2025.018"}],"container-title":["Machine Learning"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/link.springer.com\/content\/pdf\/10.1007\/s10994-026-07036-8.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/link.springer.com\/article\/10.1007\/s10994-026-07036-8","content-type":"text\/html","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/link.springer.com\/content\/pdf\/10.1007\/s10994-026-07036-8.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,6,24]],"date-time":"2026-06-24T06:35:19Z","timestamp":1782282919000},"score":1,"resource":{"primary":{"URL":"https:\/\/link.springer.com\/10.1007\/s10994-026-07036-8"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,4,22]]},"references-count":24,"journal-issue":{"issue":"5","published-print":{"date-parts":[[2026,5]]}},"alternative-id":["7036"],"URL":"https:\/\/doi.org\/10.1007\/s10994-026-07036-8","relation":{},"ISSN":["0885-6125","1573-0565"],"issn-type":[{"value":"0885-6125","type":"print"},{"value":"1573-0565","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,4,22]]},"assertion":[{"value":"31 October 2025","order":1,"name":"received","label":"Received","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"21 March 2026","order":2,"name":"revised","label":"Revised","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"23 March 2026","order":3,"name":"accepted","label":"Accepted","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"22 April 2026","order":4,"name":"first_online","label":"First Online","group":{"name":"ArticleHistory","label":"Article History"}},{"order":1,"name":"Ethics","group":{"name":"EthicsHeading","label":"Declarations"}},{"value":"The authors declare no conflict of interest.","order":2,"name":"Ethics","group":{"name":"EthicsHeading","label":"Conflict of interest"}}],"article-number":"103"}}