scholaroctopus

ScholarOctopus is a search and exploration tool for academic papers in computer vision and machine learning. Instead of using a traditional keyword search, it organizes thousands of papers visually on a 2D map so you can see how research topics cluster together and relate to each other.

The tool has indexed about 7,000 papers from major conferences like CVPR, ICCV, NeurIPS, and ICML published between 2006 and 2014. It analyzes the language and terminology used in paper titles and abstracts, then uses a technique called t-SNE to arrange them spatially. Papers that use similar vocabulary and concepts end up near each other on the map, so you can browse a topic area and discover related work visually rather than scrolling through search results.

This kind of tool is useful for researchers trying to get oriented in a new field, students exploring what questions people are actually working on, or anyone trying to understand the landscape of a research area. Instead of guessing the right keywords to search for, you can click around a visual map and see what your neighbors are studying. The project also includes basic statistics about the papers, giving you a sense of how active different research areas have been over time.

The creator has kept the data-gathering scripts private for now, but welcomes contributions from others who want to build new visualizations or views of this paper collection. If someone wanted to create a different way of exploring the same papers—say, organized by author, institution, or filtered by year—they could contribute that as an additional visualization without needing access to the underlying scraping code.