Temporal Dynamics of Lexical and Semantic Novelty in Scientific Abstracts: Kazakhstanrelated Publications
6th IEEE International Conference on Smart Information Systems and Technologies, SIST 2026, Astana, Kazakhstan, 13 - 15 May 2026, (Full Text)
- Publication Type: Conference Paper / Full Text
- Doi Number: 10.1109/sist61674.2026.11596111
- City: Astana
- Country: Kazakhstan
- Keywords: science of science, scientific novelty, semantic embeddings, temporal analysis, text-based novelty
- Istanbul University Affiliated: Yes
Abstract
This study examines temporal trends in textbased novelty of scientific abstracts using a corpus of 10,000 English-language publications related to Kazakhstan from 2013 to 2025. Novelty is defined under a strict temporal protocol as one minus the maximum similarity between a target abstract and the set of previously published abstracts. Similarity is computed using two representations: (i) lexical TF-IDF vectors with cosine similarity and (ii) semantic sentence embeddings obtained with SentenceTransformer (all-mpnet-base-v2). In addition, principal component analysis (PCA) is applied to an interpretable feature set including lexical novelty, semantic novelty, maximum similarity to prior work, and abstract length, yielding an integrated indicator of template-like structure. The results reveal a statistically significant decline in both lexical and semantic novelty over time, with a substantially stronger decrease observed for the embedding-based metric. The PCA-based indicator exhibits the most stable temporal trend. These findings indicate increasing semantic consolidation in scientific writing and demonstrate the utility of embeddingbased and integrated metrics for scalable and reproducible novelty assessment.