METHOD OF SEMANTIC DATA ANALYSIS FOR DETERMINING MARKER WORDS IN THE PROCESSING OF VISITORS' EVALUATION RESULTS IN INTERACTIVE ART

Abstract

The subject of the study is in-depth semantic data analysis based on the integration of the methodologies of latent Dirichlet distribution (LDA) and bidirectional encoding representation from transformers (BERT). This research focuses on processing textual data, in particular, visitors' evaluations of interactive art, to identify marker words that highlight key emotional and thematic elements. The goal is to deepen the understanding of visitors' experiences and perceptions of interactive art installations by identifying significant marker words using a combined LDA and BERT approach. This combination aims to capture both general thematic content and the nuanced context of feedback. Objectives: collection and preprocessing of textual data - visitor ratings, consisting of tokenization, normalization and lemmatization steps with the implementation of LDA to extract common themes from the collected data, providing insights into the main themes present in visitor feedback; integration of BERT to analyze contextual nuances and extract deeper meanings from individual words in the feedback; combining the results of LDA and BERT to create a comprehensive understanding of the textual data, focusing on identifying the most significant marker words. The following results were achieved successful extraction of key themes from visitors' ratings using LDA, which allowed us to identify broad thematic categories present in the reviews; a deep learning approach BERT was proposed, which provided nuanced contextual embeddings, emphasizing specific emotions and sentiments expressed by visitors; the results of LDA and BERT were integrated, which provided a rich set of marker words that effectively reflect the essence of the experience and perception of visitors to interactive art; the accuracy and depth of analysis in identifying key emotional and thematic elements was improved, as evidenced by the consistency and relevance of marker words in relation to visitors' ratings. Conclusions: The integration of LDA and BERT for semantic data analysis in interactive art contexts demonstrates a powerful approach for understanding complex visitor feedback. This method provides a two-level analysis, where LDA offers insights into general themes and BERT contributes to detailed contextual understanding. The study successfully identifies specific marker words that effectively capture the essence of visitors' impressions and ratings. This methodology can be useful for artists, curators, and researchers in measuring public reception and improving interactive art experiences. The adaptability of the methodology creates real prospects for its application in other areas that require a detailed semantic analysis of textual feedback.