The paper describes the TrendCarbon web service for the automated analysis and monitoring of scientific publications in the field of carbon research. It compares topic modeling methods (LDA, BERTopic) and evaluates the effectiveness of large language models (LLaMA 3, Mistral, YandexGPT-5 Lite, T-Lite) for the generation of interpretable topic labels. The service is implemented using a client–server architecture based on Elasticsearch, FastAPI, and React. Experimental evaluation on a corpus of Russian-language publications confirmed the effectiveness of TrendCarbon in identifying research trends and in supporting information analysis.
Kobylkina et al. (Wed,) studied this question.