GA-LDA Approach for Topic Modeling in Turkish Accounting and Finance Articles: Performance Optimization in Text Classification

The volume of research in the social sciences is expanding rapidly, creating significant challenges in extracting meaningful insights from unstructured text, particularly from articles lacking a classification system. Analysing these high-volume texts offers numerous advantages, including the ability to automatically identify topic relevance and track thematic trends over time. Such insights are valuable for journal management and enable researchers to access detailed information about evolving areas of study. Latent Dirichlet Allocation (LDA) is a widely used method for topic modelling, effectively extracting topics from textual data. However, its performance can be further enhanced through optimization techniques such as Genetic Algorithms (GA). This study introduces an intelligent GA-LDA framework designed to optimize word subsets for LDA, thereby improving its predictive capabilities. The proposed system is applied to a dataset of 928 abstracts from a Turkish-language academic journal specializing in accounting and finance, covering publications from 2005 to 2020. The genetic algorithm selects optimal word subsets for LDA analysis, with perplexity scores serving as the fitness function to guide the optimization process. Experimental results demonstrate that the GA-enhanced LDA significantly improves classification accuracy and topic modelling performance. This study not only underscores the potential of GA-LDA in handling unstructured text but also provides a robust tool for advancing automated content analysis in Turkish academic literature.