Aaker Brand Personality Classification for Indonesian Brands using Fine-Tuned Indonesian BERT-Variants

Traditional methods for assessing Aaker’s brand personality dimensions (Sincerity, Excitement, Competence, Sophistication, Ruggedness) often rely on time-consuming surveys and manual analysis. This research investigates the feasibility of automating this classification for Indonesian brands by leveraging deep learning models on social media data. We collected text posts from the X and Instagram accounts of selected Indonesian brands representing the five dimensions. After preprocessing, we fine-tuned three Indonesian BERT-based transformer models (IndoBERT, IndoDistilBERT, IndoRoBERTa) and compared their performance against a baseline BiLSTM model. The results demonstrate that fine-tuned models, particularly IndoDistilBERT, significantly outperform the baseline, achieving high f1-scores (often exceeding 0.99) across the personality dimensions, even with imbalanced data. Model interpretation using SHAP further confirmed the models' ability to identify relevant textual cues. This study validates the use of fine-tuned BERT models as an effective and efficient approach for automatically classifying Aaker brand personality from social media text in the Indonesian context, offering a valuable tool for brand analysis.