ISEP - DM – Engenharia de Inteligência Artificial
Permanent URI for this collection
Browse
Browsing ISEP - DM – Engenharia de Inteligência Artificial by Author "Bastos, Caroline"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- Development of a hybrid recommendation system focused on serendipity for e-commercePublication . Bastos, Caroline; Martins, António Constantino LopesIn the dynamic world of e-commerce, Recommender Systems have become crucial in guiding user choices and improving shopping experiences. Recommender Systems enhances ecommerce profits and increases user satisfaction. It has also been proved the positive influence in customer satisfaction of aspects like diversity, novelty and serendipity in recommendation. Despite it, most recommender systems rely mostly on accuracy metrics. To address this issue, this study proposes a novel approach to enhance serendipitous recommendations by leveraging machine learning and natural language processing techniques applied to item descriptions. Four different embedding models were employed to generate item embeddings from textual item descriptions. An initial analysis compared calculated serendipity, derived from the cosine similarity of item embeddings, with ground-truth serendipity. Followed by the development of a serendipity classification model. Three different recommendation models were constructed: the baseline XGBoost model, which functions as a content-based recommender system; the XGBoost-Seren model, which incorporates ground-truth serendipity data to prioritize serendipitous items; and the XGBoost-Seren + Classifier model, which replaces the ground-truth serendipity data with the predictions from the serendipity classifier. Each model was compared against SASRec, a state-of-the-art sequential recommender system, to evaluate both accuracy and serendipity metrics. The results demonstrate that the XGBoost-Seren + Classifier model outperforms SASRec in both accuracy and serendipity metrics. The study confirms that incorporating predicted serendipity into recommendation algorithms enhances the ability to suggest unexpected items without compromising accuracy. The findings indicate that using text-based embeddings not only enriches the recommendation process but also improves the overall user experience by integrating serendipity more effectively.