DEVELOPMENT OF DEEP LEARNING MODEL BASED ON TRANSFORMER ARCHITECTURE (VASWANI ET AL.) FOR CREATING UNIQUE VECTOR REPRESENTATIONS AND CLUSTERING OF PANEL DATA

The purpose of the research in this article is to develop an adaptive methodology for generating semantic vector models that ensure the integration of multidimensional attributes of objects in conditions of temporal and structural heterogeneity of data. The present research is a development of a deep learning model based on the Transformer architecture to create vector representations and cluster panel data simultaneously. Often, one object on a temporary school is characterized by an impressive amount of descriptive statistical data, so it is quite natural that it becomes necessary to present information about the object in the form of a numerical vector. According to the results of this article, the obtained vector representations (embeddings) correspond to a set of related statistical data, which gives researchers the opportunity to study the general structure of the objects themselves using vector analysis instead of studying the "proxy" objects using the obtained statistical information (description of the time stamp: one object, many vectors – panel data).