Modeling Marketing Strategies for Small and Medium-Sized Enterprises in the Food Industry Using Reinforcement Learning and Natural Language Processing

Abstract

Artificial intelligence (AI) has become a strategic enabler for redesigning marketing practices, particularly for small and medium-sized enterprises (SMEs) in the food industry, where firms face intense competition, resource constraints, heterogeneous customer preferences, and infrastructural limitations. This study developed a context-specific model for improving marketing strategies in food-sector SMEs by integrating Deep Q-Network (DQN) and DistilBERT algorithms. A mixed-methods design was used. In the qualitative phase, semi-structured interviews were conducted with 12 experts and analyzed through thematic network analysis. In the quantitative phase, a questionnaire developed from 25 organizing themes was administered to 384 managers and specialists. Reliability was acceptable to excellent (Cronbach's alpha = 0.78-0.88), and normality was confirmed using the Kolmogorov-Smirnov test (Sig. = 0.08-0.19). The highest mean scores were found for senior management technology adoption (4.25) and marketing-strategy personalization capability (4.25). DQN achieved accuracy of 0.94, MSE of 0.15, F1-score of 0.92, and mean cumulative reward of 98.5. DistilBERT achieved accuracy of 0.91, cross-entropy loss of 0.12, precision of 0.89, and recall of 0.90. The findings suggest that DQN is better suited to dynamic marketing optimization, whereas DistilBERT is more appropriate for text-based customer analytics. The proposed framework provides a practical AI-driven model tailored to Iranian food-industry SMEs. The algorithmic analyses were based on structured questionnaire-derived features and coded textual materials from expert and managerial narratives; therefore, the reported performance should be interpreted as internal model validation rather than evidence from deployed real-time marketing campaigns.