chatpaper

ChatPaper is an open-source paper processing tool designed for researchers. Its core goal is to accelerate literature reading and writing workflows using large language models. The project was developed by a PhD student in reinforcement learning at USTC, and the README is primarily in Chinese, with an English version also available. Main capabilities: - Paper summarization: Automatically searches and downloads the latest papers on arXiv based on keywords, then calls the ChatGPT API to summarize papers into a fixed structure, including research background, existing approaches and their problems, the proposed method and steps, and experimental tasks and results, enabling quick initial screening. - Local PDF processing: Directly specify a local PDF file or folder path for batch summarization. - Full-text translation: Provides a local PDF full-text translation solution that relies on scipdf_parser and the grobid service. It requires a Java environment on Ubuntu or macOS before running. - Companion tools: The README lists related projects such as ChatReviewer (paper strengths/weaknesses analysis and improvement suggestions), ChatImprovement (polishing and translation), ChatResponse (review responses), and ChatGenTitle (title generation from abstracts), as well as community extensions like one-click literature review and ChatPaper2Xmind. - Google Scholar organization: Includes google_scholar_spider.py, which can search by keyword, sort by citations, and export to CSV. Running modes: - Script mode: Fill in the OpenAI key in apikey.ini, install the dependencies in requirements.txt, then run chat_paper.py or chat_arxiv.py. Supports parameters such as query, filter_keys, max_results, sort, save_image, and file_format. - Flask service: Run app.py to provide a web interface on local port 5000, wrapping multiple script interfaces. Results are saved in directories such as export, pdf_files, and response_file. - Docker: Provides docker-compose.yaml. Replace the OpenAI key inside before starting. The README notes that this path may have issues. - HuggingFace Space: Provides online deployment instructions, but the README notes that this feature is currently disabled. The README also warns that search terms cannot recognize hyphens, review-type articles are not supported, a network proxy is required, and emphasizes academic ethics and proper usage. All project features are free and the code is open source.