Emotion and sentiment enriched decision transformer for personalized recommendations

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Recommender Systems

Background:

• Traditional recommendation systems struggle to incorporate user emotions and sentiment from reviews.
• Enhanced user engagement and satisfaction depend on understanding nuanced emotional content.

Purpose of the Study:

• To propose a novel framework, the Emotion and Sentiment-Enriched Decision Transformer (ESE-DT), for sequential recommendation.
• To integrate sentiment and emotion analysis into the Decision Transformer (DT) framework for improved contextual understanding.

Main Methods:

• Reimagining recommendation tasks as trajectory modeling problems using the Decision Transformer (DT) framework.
• Leveraging advanced Natural Language Processing (NLP) for sentiment and emotion extraction to enrich user and item embeddings.
• Experimentally validating ESE-DT on the Yelp and Google Local Reviews datasets.

Main Results:

• ESE-DT demonstrated superior performance compared to state-of-the-art baselines on both datasets.
• Achieved significant improvements in key metrics: up to 11.76% in nDCG@10 and 11.58% in HR@10 on Yelp.
• Showcased substantial reductions in RMSE by up to 42.11% on Yelp and 17.43% on Google Local Reviews.

Conclusions:

• The proposed ESE-DT framework effectively integrates emotion and sentiment analysis into sequential recommendation.
• ESE-DT enhances personalized recommendations by capturing temporal dynamics and contextual user understanding.
• Findings underscore the potential of ESE-DT in advancing personalized services in smart cities.