Paper Titler combines classic NLP with learned language patterns from arXiv to turn a description of your work into candidate titles.
Paste your abstract directly, or upload your manuscript in PDF, Word DOCX, TXT, or Markdown format. The system intelligently isolates the abstract and key methods automatically.
The input is cleaned, tokenized, and normalized — stopwords are removed, phrases are lemmatized, and domain-specific terms are detected using a scientific vocabulary built from arXiv metadata.
A frequency- and position-weighted extractor pulls out candidate method names, problem domains, and techniques — the building blocks every academic title is assembled from.
Extracted concepts are slotted into title patterns learned from tens of thousands of published papers — colon constructs, "A/An X Approach to Y," "Towards Z," and more — to produce natural, field-appropriate phrasing.
Each candidate is scored for semantic overlap with your original text and filtered for length, clarity, and novelty, so you only see options that are genuinely representative of your work.
The top 3–5 titles are returned with their keywords and scores. Copy any of them, save your favorites, or regenerate for a new batch.
The generation pipeline is model-agnostic: it ships with a fast built-in generator, and can be pointed at a large language model for even richer phrasing.
Title templates and phrasing patterns are mined from a sample corpus of published paper titles and abstracts.
Swap in an LLM API key to move from the built-in heuristic generator to full language-model generation, no code changes needed.
Relevance scores are computed from transparent keyword-overlap statistics, not a black box, so you always know why a title ranked where it did.