A carregar...
Publicação
Recommender Systems Reimagined: A Systematic Review of Quality, Trade-offs, and Evaluation Frameworks
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 936.23 KB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Recommender Systems have become essential mechanisms in our digital era, and while
traditional evaluation emphasizes accuracy, there is an emerging demand for comprehensive
system performance assessment. This study aims to systematically identify frameworks and
techniques that evaluate and balance multidimensional quality in Recommender Systems.
Specifically, it addresses three core questions: What quality evaluation measures exist in
Recommender Systems? Which methods effectively convey and balance multidimensional
quality? What novel techniques address current evaluation challenges across various
domains? The literature review was conducted using Scopus as the primary source. The
inclusion criteria required peer-reviewed journal articles published in English between 2020
and 2024. Studies were excluded if they lacked focus on quality assessment, domain
application, or trade-off balancing. Thirty-one studies were identified through a multi-stage
process that included automated text mining, NLP-based abstract screening, and full-text
analysis guided by the PRISMA 2020 framework. Risk of bias was mitigated by combining
automated semantic filtering with manual review, ensuring a high-precision selection process.
Results were synthesized through thematic coding and descriptive mapping. The selected
studies encompassed a broad range of quality dimensions, in addition to accuracy, with hybrid
and deep learning-based models prevailing, although with limited practical validation. Several
integrative frameworks have been identified for balancing conflicting metrics. However, their
empirical deployment remains scarce. Limitations of the evidence include an overreliance on
offline evaluations, a lack of standardized metrics across studies, and limited guidance for realworld implementation. Despite advancements, real-time testing and stakeholder-driven
metric weighting remain underexplored. This literature review introduces a layered visual
framework to support systematic, stakeholder-aligned evaluation of Recommender Systems.
By mapping contextual objectives to measurable outcomes and explicitly balancing trade-offs,
it provides a practical guide for both researchers and practitioners seeking to design
Recommender Systems that are effective, fair, and transparent across domains.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Business Analytics
Palavras-chave
Recommender Systems Artificial Intelligence Natural Language Processing Systematic Literature Review SDG 9 - Industry, innovation and infrastructure