Retrieval-augmented generation (RAG) depends critically on the quality and granularity of retrieved evidence. Large retrieval units preserve context but often introduce irrelevant content, which can dilute answer-bearing evidence and worsen long context utilization. Fine-grained units are more compact, but they may be difficult to retrieve reliably because short chunks can lack semantic, lexical, or bridging cues needed to match the query. We propose **Uncertainty-aware Multi-Granularity RAG (UMG-RAG)**, a training-free hybrid retrieval framework that adapts retrieval across chunk granularities based on query-specific reliability estimates. Instead of training a new retriever or modifying the generator, UMG-RAG uses existing dense and sparse retrievers as complementary experts across multiple chunk granularities. For each query, it converts each expert-granularity score list into an evidence distribution, estimates reliability from distribution entropy, and fuses candidates according to query-specific semantic, lexical, and granularity confidence. We further introduce **UMGP-RAG**, a parent promotion variant that uses fine-grained hits to locate relevant evidence while returning broader non-redundant parent chunks for local coherence. Experiments on question answering benchmarks show that uncertainty-aware fusion and parent promotion improve generation quality in a training-free, plug-and-play retrieval pipeline.

Effectively representing long documents is a persistent challenge in natural language processing, as foundational encoders are constrained by limited context windows. Prevailing methods like truncation and independent chunk aggregation can discard content or lose cross-chunk context, hindering downstream performance. To address this problem, we introduce **Spectral Attention Token Pooling (SATPool)**, a novel, encoder-agnostic module that generates a single, fixed-size vector from variable-length sequences of locally encoded tokens. SATPool operates in two stages: it first uses efficient linear attention to model global token interactions after chunk encoding, then employs **Spectral Token Compression (STC)** to compress these globally-aware token representations into a compact, context-aware vector. Across long-document classification, Retrieval-Augmented Generation (RAG), and multimodal RAG, SATPool achieves strong overall improvements and robust top-tier performance across datasets and encoder backbones. The pretrained encoder remains frozen, while the lightweight SATPool module and task-specific heads are trained for each downstream task.

Machine learning applications for text classification are increasingly used in domains such as toxicity and misinformation detection in online settings. However, obtaining precisely labeled data for training remains challenging, particularly because not all problematic instances are reported. Positive-Unlabeled (PU) learning, which uses only labeled positive and unlabeled samples, offers a solution for these scenarios. A significant concern in PU learning, especially in online settings, is fairness: specific groups may be disproportionately classified as problematic. Despite its importance, this issue has not been explicitly addressed in research. This paper aims to bridge this gap by investigating the fairness of PU learning in both offline and online settings. We propose a novel approach to achieve more equitable results by extending PU learning methods to online learning for both linear and non-linear classifiers and analyzing the impact of the online setting on fairness. Our approach incorporates a convex fairness constraint during training, applicable to both offline and online PU learning. Our solution is theoretically robust, and experimental results demonstrate its efficacy in improving fairness in PU learning in text classification.