Scientific support systems based on the constructing user profile
Year: 2025
Type: article
Abstract: This article provides an overview of modern scientific support systems (SSS), which increases the effectiveness of scientific activity of specialists due to timely receipt of relevant publications. Special attention is paid to the application of SSS in rapidly changing subject areas, in particular Computer Science. The main approaches to the creation of SSS are outlined, it is noted that the key direction of the development of SSS is the personalization of information search and analysis. One of the main ways to implement a personal approach is to build a user profile (PP). Various technologies are used for this purpose. This paper shows the advantages of constructing PP based on keywords (KW) — unigrams, bigrams, trigrams and n-grams. This allows to create a reliable «information portrait» of the text, to obtain a compressed semantic core for it. KW describe short documents most effectively and are well suited for analyzing bibliographic descriptions of scientific articles that include titles, annotations (and other supporting sections). The paper makes a formal statement of the task of extracting KW from scientific documents, systematizes methods for detecting KW, and justifies the choice of the neural network model KeyBERT for practical use in SSS. The high versatility of the KeyBERT and its effectiveness for detecting KW in short scientific texts are noted. An algorithm implementing KeyBERT is given (embedding vectors are based on the RoBERTa model). To demonstrate the capabilities of the developed SSS (in particular, the construction and updating), research was conducted on a sample of «UIT-2024», formed at the Department of Control and Intelligent Technologies of the Moscow Power Engineering Institute. It includes 10,000 documents on ten topics in the field of Computer Science. After additional learning, the KeyBERT algorithm extracts informative KW from this sample and automatically generates an expanded list of one-word and multi-word descriptors that allow the specialists to form and update their profiles, increasing the relevance of recommendations in the SSS. The possibilities of the proposed approach are considered in detail using the example of a hypothetical specialist in Computer Science.
Authors: N. A. Nazarov, V. O. Tolcheev
Institution: Moscow Power Engineering Institute
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4414519784