After this lesson, you will be able to: Read research papers efficiently and build a literature review that situates your idea and finds the right baselines.
Research starts with reading. This lesson teaches the three-pass reading method, how to build a focused literature review, and how reading reveals the baselines and gap your paper must address.
Do not read papers front to back. Pass 1 (5 minutes): title, abstract, figures, conclusion, decide if it is relevant. Pass 2 (~30 minutes): read for the main claim, the method at a high level, and the experimental setup, ignore proofs/details. Pass 3 (hours, only for papers central to your work): reconstruct the method well enough that you could reimplement it, and note every assumption. You will three-pass dozens of papers, deep-read only a handful. Keep notes in a reference manager (Zotero) with your own one-line summary of each paper's contribution.
Start from a few seminal papers in your area, then expand via 'cited by' (Semantic Scholar, Google Scholar) and the related-work sections of recent papers. The ACL Anthology is searchable and complete for NLP. Organize papers by the question they answer, not chronologically. Your goal is to be able to say: here is what has been tried, here is what each approach got, and here is the specific gap my work addresses. That paragraph becomes your paper's related-work section and justifies your contribution.
Deep-reading every paper (you will read three and burn out). Citing papers you only read the abstract of as if you understood them. Missing a closely-related recent paper, then having a reviewer point out you reinvented it. Building a chronological 'X did this, then Y did that' related-work section instead of one organized around the research question. Choosing weak baselines because they are easy to beat. Not using a reference manager, then losing track of what you read.
Pick the best answer.
Sign in and purchase access to unlock this lesson.