?
Automated Event Logs Generation Based on Unstructured Internet Sources for Process Analysis Tasks
This paper presents an approach to automated structuring event-related information extracted from unstructured textual Internet sources for process mining tasks. In many practical cases, information on events associated with various processes is not presented in the form of ready-made event logs, but is distributed among news publications, reports and other textual materials. This significantly complicates applying process analysis tools that use data presented in certain formats (for example, XES), since it requires searching for relevant texts, interpreting and transforming them into a structured view suitable for analysis with using existing Process Mining tools. The task is even more complicated if it is necessary to extract additional attributes associated with described events, which makes it possible to identify deeper regularities and patterns that characterize the processes under study, to use new methods for analyzing processes that take these characteristics into account. The proposed approach is based on combination of ontological representations of domain knowledge and information on data sources and large language models, generative artificial intelligence. The ontological layer is used to describe domains of the studied processes, open sources of information about them, user queries and results of processing the found information, while generative models provide searching for relevant information, extraction of data on events and their primary processing and structuring. To implement the approach, a general solution architecture has been proposed that allows automating the receipt of text data from various sources based on user requests generated using ontology, preprocessing the received data using generative models, structuring and normalizing the extracted event data, as well as preparing data for generating event logs in specifying formats. To illustrate possibilities of the proposed approach, a controlled experimental scenario is considered. Experiments have shown that the described software solution allows researchers to find relevant text materials, highlight events and their attributes, as well as form a structured view suitable for further logging events, transformation data into event log. At the same time, the results showed necessity additional normalization of data and expert interpretation of part of the extracted information.
Keywords: information retrieval; natural language processing; text mining; event information extraction; event logging; ontology-driven approach; AI-based approach; generative models; large language model; process mining.
For citation: Voronova K.D., Lyadova L.N. Automated Generation of Event Logs for Process Analysis Tasks
Based on Unstructured Internet Sources. Trudy ISP RAN/Proc. ISP RAS, vol. 38, issue 4, part 1, 2026, pp.153-170. DOI: 10.15514/ISPRAS-2026-38(4)-8.