Combining generative AI and SenticNet for aspect-based sentiment analysis of restaurant reviews on Google Maps

This paper proposes a pipeline for aspect-based sentiment analysis (ABSA) that combines Generative AI (Gemini API) and SenticNet to measure customer satisfaction from restaurant reviews on Google Maps. The pipeline operates in three stages: (1) automatic translation of Vietnamese reviews into English using Gemini 2.5 Flash; (2) extraction of aspect–sentiment pairs through few-shot prompting; and (3) polarity scoring using a SenticNet lexicon of 135 adjectives and 62 adverbs. Experiments on 688 reviews collected from 80 buffet and hot-pot restaurant branches across Vietnam yielded 1,432 sentiment words distributed over six aspects. Food received the most mentions (570 words) and the highest satisfaction score (3.86/5.0), whereas Service scored lowest (3.42/5.0). The overall positive-sentiment ratio was 73.0%. Pearson correlation between the SenticNet-derived scores and original Google Maps star ratings reached r = 0.676 (p < 0.001), confirming the validity of the approach.