You can organize papers by algorithm, dataset, and performance metric by creating a structured tagging system in your reference manager and building custom filters or views based on these three dimensions.
Set Up a Multi-Dimensional Tagging System
Start by defining consistent tags for each category. For algorithms, use standardized names (e.g., "BERT," "ResNet-50," "Random Forest"). For datasets, include both the dataset name and domain (e.g., "ImageNet-vision," "GLUE-NLP"). For metrics, use abbreviations that match what papers report (e.g., "accuracy," "F1," "BLEU," "mAP").
When you add a paper to your library, apply all relevant tags from these three categories. A single paper might get tagged with multiple algorithms if it's a comparison study, or multiple metrics if the authors report several benchmarks.
Create Filtered Views and Collections
Most reference managers let you filter by multiple tags simultaneously. Create saved searches or smart folders that combine your criteria — for example, "all papers using Transformer AND trained on WMT14 AND reporting BLEU scores." This turns your library into a queryable database where you can instantly surface papers that match specific experimental setups.
WisPaper's My Library feature supports this workflow with automatic metadata extraction and AI-powered organization, letting you chat with your entire collection to find papers matching complex criteria like "show me all CNN architectures tested on medical imaging datasets with AUC above 0.90."
Build Comparison Tables
For active research projects, maintain a spreadsheet or markdown table that extracts key results from each paper. List papers in rows and create columns for algorithm variant, dataset details, metric values, and any notable experimental conditions (training time, hardware, hyperparameters). This makes it easy to spot performance trends, identify gaps in the literature, and choose baselines for your own experiments.
The key is consistency — decide on your taxonomy early and apply it uniformly. Over time, this structured approach transforms a pile of PDFs into a searchable knowledge base that directly supports literature reviews, experiment design, and writing your related work sections.

