Russian Railways creates a neural network that will restore order in disparate regulatory documentation

Russian Railways decided to create a system for the development, intelligent search and meaningful analysis of regulatory documents using neural networks. On August 31, 2021, the company announced a tender for the development of such a system with a maximum contract price of 174.4 million rubles. Open^[1]

The prerequisites for the project were the identified problems. The tender documentation published by Russian Railways says that the company has more than 25 thousand regulatory documents managed by internal information systems. However, their functionality in modern realities does not allow effective management of regulatory documents in Russian Railways, since these systems provide only part of its life cycle: coordination of draft documents, their approval, archiving and storage.

The problem of effective management of regulatory documents in the company is also associated with the "independence" of the systems used and the lack of communication between them. Therefore, when developing documents, employees search for the necessary information in each of the sources individually. At the same time, the regulatory documentation management systems used in Russian Railways do not have modern tools for searching and analyzing text information (natural language processing, natural language processing).

There is no unified system that would allow managing the life cycle of regulatory documentation in Russian Railways "(photo - Izvestia/Alexander Kazakov)"

Another important problem of effective work with regulatory documentation in Russian Railways is the lack of automation in identifying duplication and contradictions in existing and developed documents, which leads to significant labor costs for company employees when manually searching for all documents, on a specific issue, and comparing disagreements between the documents under consideration, the procurement documentation says. - This approach has a high risk of incomplete coverage of the analyzed documentation and taking into account all parameters in it, which may in the future lead to inconsistencies in the activities of structural divisions of the Russian Railways holding.

The "bottlenecks" in the process of developing Russian Railways regulatory documents are also the use of a classic approach to the development of documents, including local creation of a document, loading into the electronic document management system, collecting comments, repeating iterations of coordination, and the absence of a "single window" for searching and studying regulatory documents for employees.

A unified system that would allow managing the life cycle of regulatory documents is not yet available in Russian Railways. Therefore, the company wants to create a system that provides a single information space for managing regulatory documents, contains an intelligent search tool for a single information space of regulatory documents. It must also manage the full lifecycle of regulatory documents.

Neural networks in the system should be used for pre-processing of text, semantic search of data in documents, semantic search of created vector spaces, a requirement is indicated in the specification.

The system should be implemented using software from the register of domestic software or freely distributed software. At the same time, the system will not be designed to work with documents containing information constituting trade secrets, as well as having the heading "For official use" or constituting state secrets.

The total number of users of the system will be up to 15 thousand employees.

Russian Railways plans to summarize the selection of a supplier on October 7, 2021, and the deadlines for the implementation of the work are indicated on October 31, 2022.

Notes