AceParse: A Comprehensive Dataset with Diverse Structured Texts for Academic Literature Parsing

Abstract: With the development of data-centric AI, the focus has shifted from model-driven approaches to improving data quality. Academic literature, as one of the crucial types, is predominantly stored in PDF formats and needs to be parsed into texts before further processing. However, parsing diverse structured texts in academic literature remains challenging due to the lack of datasets that cover various text structures. In this paper, we introduce AceParse, the first comprehensive dataset designed to support the parsing of a wide range of structured texts, including formulas, tables, lists, algorithms, and sentences with embedded mathematical expressions. Based on AceParse, we fine-tuned a multimodal model, named AceParser, which accurately parses various structured texts within academic literature. This model outperforms the previous state-of-the-art by 4.1% in terms of F1 score and by 5% in Jaccard Similarity, demonstrating the potential of multimodal models in academic literature parsing. Our dataset is available at https://github.com/JHW5981/AceParse.

• The paper introduces AceParse, the first dataset capturing diverse structured academic texts including formulas, tables, and algorithms.
• The paper presents AceParser, a fine-tuned multimodal model that improves parsing performance with a 4.1% F1 score and a 5% increase in Jaccard similarity.
• The paper emphasizes a data-centric AI approach, enhancing reliable extraction and analysis for academic research and literature processing.

"AceParse: A Comprehensive Dataset with Diverse Structured Texts for Academic Literature Parsing" presents a significant advancement in the field of data-centric AI, particularly focusing on the nuances of parsing structured texts from academic literature. The shift towards enhancing data quality over merely focusing on model-driven approaches is highlighted as a contemporary trend in AI development.

Dataset Overview

AceParse is introduced as the first of its kind, providing a comprehensive dataset tailored to address the complexities involved in parsing diverse structured texts. This dataset encompasses a wide array of text structures commonly encountered in academic documents, including:

• Mathematical formulas
• Tables
• Lists
• Algorithms
• Sentences embedding mathematical expressions

Model Development and Performance

In conjunction with the dataset, the paper introduces AceParser, a fine-tuned multimodal model designed to tackle the challenge of parsing structured academic texts. The model leverages the diverse instances provided by AceParse to improve its parsing accuracy.

Key Findings

The results of utilizing AceParser show a marked improvement in parsing performance:

• F1 Score: AceParser outperforms the previous state-of-the-art models by 4.1%.
• Jaccard Similarity: The model shows a 5% improvement compared to existing methods.

These improvements underscore the potential efficacy of multimodal models when trained on sufficiently diverse and high-quality datasets like AceParse.

Importance and Implications

The introduction of AceParse addresses a critical gap in the field of academic literature parsing. By providing a dataset that encompasses a variety of structured text forms, it enhances the ability to parse documents accurately, facilitating more reliable downstream processing and analysis. The implications are significant for research communities relying on the parsing of academic texts for tasks such as information extraction, literature review automation, and the development of knowledge graphs.

Availability

The AceParse dataset is made publicly available, which encourages further research and development in this area. Researchers and developers can access it via the provided GitHub repository (https://github.com/JHW5981/AceParse), fostering an open-source approach to tackling the challenges of academic literature parsing.

In summary, "AceParse: A Comprehensive Dataset with Diverse Structured Texts for Academic Literature Parsing" represents a notable contribution to the field, both in terms of its dataset and the accompanying AceParser model. The improvements in F1 score and Jaccard Similarity metrics highlight the robustness of their multimodal approach, paving the way for more sophisticated and accurate parsing techniques in the field of academic literature.