Daniel Carlos Guimarães Pedronette

Foram comparados três paradigmas de recomendação - LightGCN, KGAT e SASRec - sob a perspectiva de manifold learning, com o intuito de verificar como cada arquitetura molda a geometria dos embeddings aprendidos. Foram adotados dois protocolos de avaliação: previsão cronológica do próximo item e recomendação geral top-K, ambos com ranqueamento full-corpus. As métricas de Preservação de Vizinhança (NP@10) e Distorção de Distância Local (LDD@10) são definidas com o grafo de interação de itens (Jaccard) como espaço de referência. Em MovieLens-1M, Yelp2018 e Amazon-Book, o SASRec domina a previsão sequencial, enquanto o LightGCN iguala ou supera o KGAT em recomendação geral, contrariando a hipótese de que grafos de conhecimento aprimoram sistematicamente o desempenho. A análise geométrica esclarece esse resultado.

Content-based image retrieval (CBIR) has evolved significantly with the advent of deep learning models, yet effectively ranking similar images remains a challenging task, particularly in high-dimensional feature spaces where pairwise distance measures often fail to capture complex contextual relationships and the semantic gap between visual features and high-level concepts persists. In this scenario, manifold learning and rank-based refinement methods have emerged as complementary strategies, respectively improving feature representations and exploiting contextual information embedded in ranked lists, such as neighborhood relationships among images. However, combining projection-based and rank-based strategies to exploit their complementary properties remains a challenging research problem. To address this, this paper proposes a framework that combines neighbor embedding projections with rank-based manifold learning through rank aggregation. Specifically, Uniform Manifold Approximation and Projection (UMAP) is used to generate alternative low-dimensional feature representations, while ranked lists obtained from UMAP projections and rank-based re-ranking methods are combined using the Borda Count aggregation strategy. The experimental evaluation was conducted on several public datasets using deep learning features extracted from ResNet152, Swin Transformer, and DINOv2 models. The results show that the proposed approach can improve retrieval effectiveness in several scenarios, particularly when the baseline representation struggles to achieve high precision values. In addition, the aggregation strategy often improves the quality of the top-ranked positions, leading to competitive Mean Average Precision (MAP) and Precision values across different datasets and feature extractors. These findings suggest that combining projection-based and rank-based manifold learning strategies through rank aggregation can provide complementary contextual information for image retrieval tasks.

Feature extraction involves the identification and extraction of salient characteristics or patterns, including edges, textures, shapes, and color attributes. Contemporary feature extractors predominantly leverage deep learning architectures, such as Convolutional Neural Networks (CNNs) and Vision Transformers (VITs). The availability of diverse feature extractors in the literature provides a wide range of feature representations. Features extracted from an image depend on the specific application, the chosen extractor, and its configuration. Therefore, integrating complementary information by combining distinct extractors offers a promising way to enhance performance. Graph Neural Networks (GNNs), particularly Graph Convolutional Networks (GCNs), have emerged as powerful and widely adopted approaches for semi-supervised image classification, as they effectively leverage both labeled and unlabeled data while exploiting the underlying graph structures that capture relationships among samples. This study proposes a novel approach for GNNs in scenarios where labeled data is scarce, by integrating diverse sets of feature and graph representations derived from various extractors in classification scenarios. Experimental investigations were conducted, encompassing combinations of distinct feature and graph extractors, as well as rank aggregation strategies. The primary contributions of this work are underscored by the experimental findings, which demonstrate that the strategic combination of feature and graph representations, coupled with the application of manifold learning for graph processing, leads to significant improvements in classification accuracy across the majority of experimental conditions. Furthermore, the utilization of rank aggregation techniques to integrate features from different extractors was shown to enhance classification accuracy.

Information Retrieval (IR) systems are fundamental across various domains, from web search engines to enterprise data management, enabling efficient access to relevant information within large-scale digital collections. While research in IR has traditionally prioritized retrieval effectiveness, ensuring the efficiency and scalability of these systems is equally critical, particularly as data volumes and computational demands continue to rise. Over the past decades, substantial progress has been made in optimizing IR systems, with studies addressing indexing strategies, parallelization techniques, and resource-efficient retrieval. However, given the growing need for real-time and large-scale processing, further advancements in optimization techniques remain essential. This systematic review explores key methods that enhance IR performance through memory management, query execution improvements, and algorithmic efficiency. By synthesizing research from 2018 to 2024, we highlight approaches that significantly improve response time, resource utilization, and scalability in IR applications. Our goal is to provide researchers and practitioners with a comprehensive overview of optimization strategies in IR, emphasizing their critical role in modern retrieval systems. While optimization research may not always receive the same level of attention as other IR topics, it remains a foundational aspect of building high-performance, scalable search technologies. This review underscores the ongoing efforts in this area and identifies opportunities for further innovation.