Програмне забезпечення для генерації текстів українською мовою

The work is devoted to the usage of large language models in combination with morphological and thematic analysis for the generation of texts in the Ukrainian language. As a result of the research, a scalable software architecture was built, consisting of generation modules (LLM), morphological analysis (LanguageTool), thematic analysis (LDA) and process control (Apache Airflow). A specialized dataset of bachelor's and master's theses was prepared, which includes structured text sections, topics and keywords. Its pre-processing was performed - text cleaning, lemmatization, filtering of parts of speech, and construction of N-grams.nbsp;A test set was created for fine-tuning the GPT-3.5-turbo and LLaMA-3-8B models. A study of the generation efficiency for both models was conducted, as well as testing the quality of error correction using a modified LanguageTool. According to the testing results, the GPT model showed a higher quality level of text generation in the Ukrainian language, and the new type of LanguageTool rules increased the quality of grammatical checking, especially for complex syntactic constructions. Ref. 19, fig. 5, tab. 3