AutoRelatedWork

• A GPT-based Tool for Automatic Related Work Generation

Introduction

• Feel difficult to write the related work? This project is help you automatically generate related work / literature review.
• Compared to traditional LLMs (ChatGPT), ours can generate text with reference and avoid fabrication.
• Currently, the source related work all origin from Arxiv.
• Feel free to modify prompt and improve it!

Quick Start

1. Put your OPENAI_API_KEY in the file OPENAI_API_KEY.txt.
2. Configure the environment pip install requirements.txt.
3. Run and generate by python run.py!
4. Get output: in output.txt.

Example

Keyword: large language models

Output file:

Text:

In the field of language modeling and processing, several significant works have been conducted. Quesada et al. proposed Fence, an efficient bottom-up parsing algorithm with lexical and syntactic ambiguity support that enables the use of model-based language specification in practice \cite{luis2011fence}. This work has applications in the implementation of language processors, the design of domain-specific languages, model-driven software development, data integration, text mining, natural language processing, and corpus-based induction of models.

Östling and Tiedemann proposed using continuous vector representations of language, showing that these can be learned efficiently with a character-based neural language model \cite{robert2016continuous}. They demonstrated that these representations can improve inference about language varieties not seen during training.

Asgari and Mofrad introduced a new measure of distance between languages based on word embedding, called word embedding language divergence (WELD) \cite{ehsaneddin2016comparing}. They performed language comparison for fifty natural languages and twelve genetic languages, showing that in many cases languages within the same family cluster together.

Cotterell et al. developed an evaluation framework for fair cross-linguistic comparison of language models, using translated text so that all models are asked to predict approximately the same information \cite{ryan2018are}. They conducted a study on 21 languages, demonstrating that in some languages, the textual expression of the information is harder to predict with both $n$-gram and LSTM language models.

Bibtex:

@misc{luis2011fence,
    title={Fence - An Efficient Parser with Ambiguity Support for Model-Driven\n  Language Specification}, 
    author={Luis Quesada, Fernando Berzal, Francisco J. Cortijo}, 
    year={2011},
    eprint={1107.4687v2},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

@misc{robert2016continuous,
    title={Continuous multilinguality with language vectors}, 
    author={Robert Östling, Jörg Tiedemann}, 
    year={2016},
    eprint={1612.07486v2},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

@misc{ehsaneddin2016comparing,
    title={Comparing Fifty Natural Languages and Twelve Genetic Languages Using\n  Word Embedding Language Divergence (WELD) as a Quantitative Measure of\n  Language Distance}, 
    author={Ehsaneddin Asgari, Mohammad R. K. Mofrad}, 
    year={2016},
    eprint={1604.08561v1},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

@misc{ryan2018are,
    title={Are All Languages Equally Hard to Language-Model?}, 
    author={Ryan Cotterell, Sabrina J. Mielke, Jason Eisner, Brian Roark}, 
    year={2018},
    eprint={1806.03743v2},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

Acknowledgements

• Based on Arxiv API and Arxiv API Example