EMNLP 2024 Conference Paper Summarizer

This project automates the crawling and abstract summarization of papers presented at the EMNLP 2024 conference. The goal is to create a two sentence summary of all accepted papers to help researchers quickly identify papers of interest. The final PDF can be found at data/emnlp_2024_summary_demo.pdf.

List of Papers

The list of papers is presented here: https://2024.emnlp.org/program/accepted_main_conference/

Paper Matching

We crawl this page and search for the matching papers via the arXiv API and via Google search (using the Serper API).

Abstract Summarization and Category Assignment

If we find a matching paper by string-matching the title, we extract the abstract and categorize the paper into one of the categories defined by the EMNLP conference: https://2024.emnlp.org/calls/main_conference_papers/ We then generate a two-sentence summary of the paper and assign a score to the paper based on how general or domain-specific the paper is.

PDF Report Generation

Finally, we generate a PDF report that lists all papers, their summaries, categories, and scores. The PDF also includes direct links to the full papers on arXiv.

Stats

We were able to find and summarize 974 papers out of 1269 papers listed on the EMNLP 2024 website.

Found: 71 (arXiv), 903 (Serper), Not Found: 295: 100%|██████████| 1269/1269 [1:24:16<00:00,  3.98s/it]

Overview

The repository fetches paper titles from the EMNLP 2024 conference website, searches for their abstracts on arXiv and Serper (Google), and categorizes them into relevant NLP categories. Summaries and scores are generated to highlight general vs. domain-specific contributions, which are then compiled into a PDF for easy review.

Key Features

• Automated Paper Title Extraction: Extracts paper titles from the EMNLP 2024 accepted papers page.
• Multi-Source Abstract Retrieval: Searches for abstracts on arXiv and Google (via Serper API).
• Paper Categorization: Categorizes papers into predefined EMNLP categories.
• PDF Report Generation: Creates a PDF of categorized papers, including summaries and links to full texts.

Setup

Prerequisites

• Python 3.x
• Required libraries: beautifulsoup4, pandas, tqdm, requests, fpdf, pydantic, openai, arxiv, rapidfuzz

Install the dependencies:

pip install requirements.txt

Environment Variables Add your Serper API key as an environment variable:

export SERPER_API_KEY='Your_Serper_API_key_here'
export OPENAI_API_KEY='Your_OpenAI_API_key_here'

Directory Structure The repository is structured as follows:

graphql

├── main.py                # Main script to extract, search, categorize, and summarize papers
├── utils/
│   ├── crawler.py         # Contains search functions for arXiv and Serper API
│   ├── pdf.py             # Generates a PDF report from summarized data
│   └── summarizer.py      # Categorizes and summarizes papers based on abstracts
├── data/                  # Directory for storing intermediate and final output files
└── README.md              # This readme file

Usage Run the Main Script:

python main.py

This script fetches all paper titles, searches for abstracts, categorizes and summarizes the content, and saves the output to a PDF.

Script Workflow:

Step 1: Extracts paper titles from the EMNLP 2024 website.

Step 2: Searches for abstracts on arXiv and Google (using the Serper API).

Step 3: Categorizes papers into predefined categories based on the abstract.

Step 4: Summarizes the content, scores generality, and generates a categorized PDF report.

Checkpoints: Intermediate files are saved as Parquet files in the data/ directory every 50 papers to handle large datasets.

Logging and Debugging

To view detailed logs, uncomment and set the logging level to DEBUG in main.py. Logs provide insights into missing papers, API search results, and errors.

Output

The final categorized summary is saved as a PDF at ./data/emnlp_2024_summary.pdf. This PDF includes:

Paper titles Summaries (if available) Categories and scores Direct links to full texts (if available)