Published on: 2026-06-23
Source: Novosibirsk State University –
An important disclaimer is at the bottom of this article.
An intelligent system for automated collection, analysis, and updating of data on investment projects in the Russian Federation from open internet sources with subsequent integration of results into the geoinformation system database of the Institute of Economics and Industrial Production Organization of the SB RAS (IEOPP SB RAS) was developed by a graduate of the Faculty of Information Technologies of Novosibirsk State University, engineer-technologist of the Center for Financial Technologies.Polina Braerunder the scientific supervision of a Candidate of Economic Sciences, Associate ProfessorAndrey Kostin. Her project is based on the use of several neural network models depending on the task set by the user.
The system developed by a young researcher analyzes open internet sources, not only providing basic information about the investment project but also auditing up-to-date data on its potential, determining the current stage of the project, and issuing the current investment value in it. There is no analog of this intellectual system at the moment, despite the active application and development of neural networks, as well as the growing need for their use when working with investment projects.This development, in addition to scientific research in the field of economics, may be in demand by government management bodies, investment agencies, and analytical departments for monitoring investment activity and updating the databases of investment projects of the Russian Federation.
—Currently, information about investment projects is posted in a variety of open sources. These include official government websites, regional investment portals, news publications, corporate resources, as well as specialized analytical platforms. The data characterizing the projects is often fragmentary and heterogeneous, presented in different formats, and may not be updated for quite a long time or updates may not be made at all.Descriptions of projects, information about stages of implementation and its temporal parameters, project participants, geographic location, volume and sources of financing, and a number of other relevant characteristics are quite often contained in unstructured text. All this significantly complicates the formation of a single and up-to-date database on investment projects, as well as reduces the efficiency of analytical research. At the same time, traditional geoinformation systems are mostly oriented towards the use of structured data., — toldPolina Braer. Â
Investment project monitoring is carried out manually, which requires significant time expenditure, complicates scaling, and reduces the operational speed of information updates.
An expert necessarily needs to find the text in an open internet source, analyze the reliability of the site where it is posted, guided by a list of reliable sources, compare the obtained information with that posted on other reliable sites. This is monotonous and routine work that takes a lot of time. The human factor is important here — when overtired, a person can make mistakes, miss important facts, or skip the site containing the target information.
Ready-made databases of regional investment projects are provided by specialized aggregators on a subscription basis. However, the data obtained in this way is often insufficient because they contain brief information about the project name and its location, industry description, and implementation dates. Moreover, the service or portal that provides this data may update it quite rarely. Frequently updated services usually provide information at a fairly high cost. As it turned out during this study, regional portals, as a rule, do not update information in a timely manner.
Therefore, there is a need to develop an intellectual system capable of automatically searching for and collecting data on investment projects from open sources, analyzing unstructured textual materials, determining the relevant characteristics of projects, forming meaningful descriptions, and ensuring the update of the database used for further economic and spatial analysis.
—The scientific novelty of our research lies in the development of an intelligent analysis system for investment projects, incorporating automated data collection from open web sources, processing of unstructured texts, neural network analysis of publication content, and the formation of an updated database suitable for subsequent geoinformational analysis.Unlike traditional approaches based on manual monitoring or exclusively using pre-structured data, the proposed approach ensures the transformation of scattered text messages into a unified analytical representation of the investment project,— explained Polina Brauer.
When creating the intelligent system, methods of natural language processing, machine learning and deep learning, transfer learning, and web parsing were used, and the instrumental base consisted of Python, HuggingFace Transformers, Scikit-learn, BeautifulSoup, and Selenium. An intelligent system was developed, including four specialized analytical modules: a project implementation stage classifier based on XLM-RoBERTa; an investment prospectivity assessment model based on logistic regression with TF-IDF; a text description generator based on ruT5; as well as a hybrid module for extracting the investment volume.Thus, the developers applied an integrated approach that combines automated data collection from open sources with neural network processing and the formation of an updated analytical database for geoinformation analysis.
—In the code we implemented, a list of certain websites-sources is protected, which are considered reliable and most prioritized. If necessary, the developer can make changes to it – add some sources, exclude others, change the priority, or add keywords. At the same time, the system’s main functionality is not violated. The system generates a detailed description of a particular investment project based on the texts of articles it has collected from reliable internet sources.This is exhaustive information and the user no longer needs to manually search for the missing information on other sites. In one study, it was calculated that the work which would take a person a full work week, the system performs in 3-4 hours. For completing this task, an intelligent model T5 was used,— explained Polina Braer.
The T5 model was also used for another task — extracting the value of investments in the project at the given moment. It gathers information on the mention of prices or values stated in the article, published in open internet sources, and analyzes which of these values is precisely the investment one, based on context, since the article may mention other prices as well.
The experimental task was to determine the indicator of the project’s investment attractiveness.
—To determine the investment attractiveness of the project, we used the TF-IDF model in combination with logistic regression. This is a combination of a neural network and statistical data processing. The model identifies key words in the text that characterize the project. For example, “successfully completed,” “built,” “made,” “executed,” or “finished.” By analogy, it also selects negative key words.When performing this task, the neural network plays a more technical role — it converts these words into vectors, and as a result, the word becomes a certain vector that is processed by logistic regression, and based on the results of this processing, the investment project falls into one of three groups according to the degree of prospectiveness: high, medium, or low.It should be specified that investment attractiveness is an experimental indicator, since the understanding of investment attractiveness is subjective and the real classification may differ among various specialists — for one project it may be attractive, and for another not.— said Polina Brauer.
The intelligent system is also capable of providing information about the current stage of projects. And there are nine: from initiation to commissioning or freezing/cancellation. The system also relies on data from open internet sources. It produces results by analyzing texts, using keywords, the list of which is expanded because the article does not always describe an obvious stage of the project, and additional keywords are needed to define it. It is also possible to use several sources.For this case, the system provides a protective mechanism that prevents mentioning a stage lower than the preceding one. For example, the “initiation” stage cannot follow the “construction” stage because, in reality, this is impossible.
—In modern conditions of digitalization of the economy, the role of information-analytical systems that provide collection, systematization, processing, and interpretation of data, essential for making well-founded managerial decisions, is increasing. The theoretical significance of our research lies in the development of approaches to the integration of artificial intelligence methods, natural language processing, and geoinformation systems to tackle the analysis of territorially distributed economic objects.The results we have obtained expand the understanding of the possibilities of applying large language and neural network models for extracting information from unstructured sources and including it in the scope of geospatial data processing. The intellectual system developed by us can be used for monitoring investment activity, supporting project portfolios, updating databases of investment projects, and for further spatial analysis., — summarized Polina Braer.
Please note; this information is raw content received directly from the source. It represents an accurate report of what the source claims and does not necessarily reflect the position of MIL-OSI or its clients.