| Nome do aluno | Joel Machado Pires |
|---|---|
| Título do trabalho | Integrating Confidence into Embedding-Based Models for Learn-to-Rank in Recommender Systems |
| Resumo do trabalho | Recommendation systems (RecSys) enhance information retrieval efficiency across different domains by delivering personalized content that aligns with user preferences. These systems address user-item relationships through methods like matrix factorization and graph attention networks (GAT). Despite advancements in accuracy, existing approaches often narrowly focus on predictive performance while neglecting the broader utility of confidence estimation. This estimation is crucial for quantifying the certainty behind recommendations, particularly in fostering trust when a balance between risk and reward is needed. RecSys can mitigate uncertainties stemming from data noise and model limitations by utilizing confidence. Existing approaches to confidence integration face critical limitations. Parametric methods risk non-convergence, inflexible uncertainty design, and often degrade accuracy. Non-parametric techniques, such as neural network-based probabilistic calibration, remain confined to classification tasks, failing to address regression scenarios, including rating prediction and listwise learn-to-rank. Many confidence models operate independently of core recommendation processes, limiting their adaptability and calibration impact. Notably, the literature neglects the integration of confidence estimation into GAT-based models. Additionally, the literature lacks experimental evaluation of different distribution-based methods. Therefore, this study proposes an experimental evaluation of previous distribution-based methods and explores the integration of suitable confidence in GAT-based models. For instance, we evaluate four prior solutions in terms of rating prediction accuracy, ranking accuracy, and correlation with confidence and error. These solutions and our proposal are evaluated in public datasets from varying contexts and characteristics. Results reveal that distribution-based confidence integration often harms models’ accuracy and leaves room for improvement regarding the correlation between the confidence and error. Although these findings also hold for our method, it still achieves superior performance compared to all prior solutions and shows promising results in terms of negative confidence–error correlation. Furthermore, as a second part of this study, we propose and evaluate the integration of confidence into embedding models for learn-to-rank methods. This proposal and its baselines are also evaluated across various public datasets, using different ranking metrics, and the correlation with confidence and error. The results reveal that the proposed method consistently demonstrates competitive rank performances and even outperforms the baselines in some datasets. Therefore, the proposed method for learn-to-rank does not significantly harm the model’s performance. Additionally, we observed a cubic polynomial relationship between confidence and error. |
| Orientador | Frederico Araújo Durão |
| Membro Titular Externo (com afiliação) | André Levi Zanon (Insight Research Ireland Centre for Data Analytics) |
| Link para o curriculum lattes |
http://lattes.cnpq.br/ |
| Membro Titular Interno ou Titular Externo 2 (com afiliação) | Yuri de Almeida Malheiros Barbosa (UFPB) |
| Link para o curriculum lattes |
http://lattes.cnpq.br/ |
| Membro Suplente Externo (com afiliação) | Windson Viana de Carvalho (UFC) |
| Link para o curriculum lattes |
http://lattes.cnpq.br/ |
| Membro Suplente Interno ou Suplente Externo 2 (com afiliação) | Danilo Barbosa Coimbra (UFBA) |
| Link para o curriculum lattes |
http://lattes.cnpq.br/ |
| Data da defesa | 20 Jan, 2026 |
| Horário da defesa | 8:30 AM |
| Quais os principais impactos deste trabalho (social, tecnológico, científico, ambiental)? | O trabalho avança o estado da arte ao avaliar sistematicamente métodos de estimativa de confiança baseados em distribuição e propor sua integração em modelos de recomendação, incluindo abordagens baseadas em GAT, até então pouco exploradas. Os resultados demonstram que é possível incorporar confiança sem comprometer de forma relevante a precisão das previsões e do ranqueamento, oferecendo também evidências de correlação negativa entre confiança e erro, o que reforça o valor interpretativo da estimativa de incerteza. Além disso, o estudo amplia o uso de técnicas de confiança para cenários de regressão e aprendizado para ranqueamento, superando limitações frequentes na literatura. A descoberta de uma relação polinomial cúbica entre confiança e erro traz ainda um novo insight teórico, contribuindo para sistemas de recomendação mais seguros, informativos e sensíveis ao risco. |
Data da Defesa:
20/01/2026 - 08:30
Tipo de Defesa:
Defesa de Mestrado