Analyzing stakeholder review of an industrial lubricant refinery using a combination of topic modeling and keyword co-occurrence analysis
Pages 109-140
https://doi.org/10.48313/jqem.2026.557985.1582
Zahra Bayat, Bakhtiar Ostadi, Ehsan Nikbakhsh
Abstract Purpose: Analyzing customer feedback is crucial for developing effective market strategies. Given the research gap in analyzing stakeholder reviews for Persian data, this study aims to present a combined approach for analyzing stakeholder opinions of a lubricant refinery using topic modeling and keyword co-occurrence analysis. Methodology: This research utilizes the Latent Dirichlet Allocation (LDA) algorithm and keyword co-occurrence network to analyze the opinions of three stakeholder groups (sales agents, company website users, and Digikala buyers). The main steps include Persian text preprocessing, LDA implementation with specified parameters (optimal number of topics based on c_v coherence score, passes=40, iterations=200, alpha='symmetric'), co-occurrence matrix construction, and key phrase extraction from high-frequency edges using a scoring method. Findings: The highest prevalence rate of extracted phrases from sales agents' comments (31.01%) was related to the weakness of chelik (steel drum) container packaging. Among website users' comments, the ambiguity about the container material of motor oil No. 2 had the highest prevalence rate of 36.67%. Digikala users' comments indicated satisfaction with appropriate price and quality of products. Based on the findings, recommendations such as packaging improvement (product development strategy) and creating an online recommender system (market penetration strategy) are proposed. Originality/Value: The proposed approach provides a practical method for systematic analysis of organizational stakeholder opinions and increases the accuracy of key phrase extraction by combining two text analysis methods.