CiteAssist: A System for Automated Preprint Citation and BibTeX Generation

• Citation information is often inaccessible alongside preprints, and papers can receive many downloads but few citations.The passage describes this as a discrepancy between reading and citing.Source: §1 · p. 1
• The growing volume of preprints makes efficient retrieval challenging, while citation accessibility is identified as one reason reading and citing can be imbalanced.Source: §1 · p. 1
• Researchers lack a simple public system that places citation details within the PDF and a common way to recommend literature beyond the related-work section.Potential recommendations include related ideas from other fields, books, or talks.Source: §1 · p. 2
• CiteAssist automatically generates citation annotations from metadata and appends them to preprints as a new PDF or a LATEX file for integration into the source.The annotation can include author names, publication details, abstracts, and custom metadata such as keywords.Source: §1 · p. 2
• CiteAssist also suggests additional literature on potentially overlooked works, and researchers can manually add resources for readers.Source: §1 · p. 2

2.1 Preprint Platforms

• Researchers publish preprints across repositories, while some prefer personal websites or do not know LATEX; CiteAssist offers a simplified interface to enter citation details into papers.Source: §2.1 · p. 2

2.2 Annotation Generation Tools

• CiteAssist directly attaches BibTeX annotations and suggests related papers, unlike CERMINE, which primarily extracts metadata for structured XML output.CiteAssist also provides a more direct annotation workflow within the paper.Source: §2.2 · p. 2
• Neural ParsCit extracts citation metadata but does not capture publication date or journal, whereas CiteAssist combines user input and keyword searches for BibTeX and related-paper integration.Source: §2.2 · p. 3
• Grobid extracts titles, authors, and abstracts but not journal information, and lacks CiteAssist’s direct BibTeX annotation and related-paper suggestions (Lopez, 2008–2024).Source: §2.2 · p. 3

• The frontend accepts PDF uploads, while the backend parses them with Grobid and presents merged results for user correction.Source: §3 · p. 3
• Users can add related papers supplied through keyword matching or identified by DOI or arXiv ID.Source: §3 · p. 3
• After user approval, CiteAssist generates a PDF or LATEX file with BibTeX annotation and related papers, stores the PDF locally, and forwards metadata to the database.Source: §3 · p. 3

• Users upload a preprint, then review or edit extracted information and add metadata such as keywords.Keywords can be selected from suggestions or added by the user.Source: §4 · p. 3
• The system lemmatizes refined keywords to search for related preprints, which users review and select before annotation generation.Source: §4 · p. 3
• CiteAssist attaches the compiled BibTeX annotation at the document’s end and adds a title-page button linking to it; users can download PDF or LATEX output.With user consent, annotated PDFs are stored for later viewing, sharing, or downloading.Source: §4 · p. 3

5 BibTeX Annotation Generation

• PDF-LIB accesses saved document metadata, while a custom function extracts text from the first page and the rest of the document.The first-page text supports title and author extraction; the remaining text is used for keyword extraction.Source: §5 · p. 3
• Extracted metadata is merged into a formatted BibTeX annotation, with editable fallback values when neither source provides a field.The annotation can include author, title, journal, volume, pages, year, and DOI.Source: §5 · p. 4
• Grobid is sent the PDF in parallel to extract the title, author, publication date, and keywords for the annotation process.Article type defaults to “article”; date defaults to the current date when unavailable, and title can fall back to the PDF title or filename.Source: §5 · pp. 3–4

6 Integration of Related Papers

• The system extracts five frequent keywords from the full text after stopword removal and lemmatization; this extraction currently supports English texts.Grobid can also extract author-provided keywords, and users can enter additional keywords or tags.Source: §6 · p. 4
• Lemmatized keyword queries return five suggested papers ranked by the number of matching keywords.The suggestions are intended to provide complementary resources beyond works already cited as related work.Source: §6 · p. 4
• Users can select suggested preprints or add papers by DOI or arXiv ID for integration into the annotation.The authors say the added resources broaden existing related work and can save researchers time and effort.Source: §6 · p. 4

7 Final Considerations

• CiteAssist generates BibTeX annotations in a new PDF or directly in a LATEX source document, with a title-page button linking to the annotation.The authors say this keeps citation information accessible regardless of the publication repository.Source: §7 · p. 4
• The system also suggests related papers, expanding the resources available to researchers around a publication.Source: §7 · p. 4
• The authors describe the system as helping authors receive proper credit and be cited correctly with the publication venue rather than the preprint server.Source: §7 · p. 4
• Future enhancements could recommend papers by semantic similarity and represent a paper’s content more comprehensively.Source: §7 · p. 4

Limitations

• Keyword extraction may become more relevant with multilingual stopword removal or direct keyword extraction using language models.Source: Limitations · p. 5
• Large language models may improve metadata accuracy and coverage across document layouts and formats.The authors also identify overall efficiency as a potential area for improvement.Source: Limitations · p. 5
• The authors plan to improve the interface and usability through user feedback sessions and iterative refinement.Source: Limitations · p. 5

A Detailed System Architecture

• The system is organized into a frontend, backend, and database, with each component supporting semiautomatic annotation generation.Its interface is described as designed for a responsive, user-friendly experience.Source: A Detailed System Architecture · p. 5

A.1 Frontend and UI/UX Design

• CiteAssist combines frontend and backend metadata extraction with Grobid, merges the results, and lets users correct them before submission.Users can also paste an existing BibTeX entry for metadata extraction and provide keywords to the backend.Source: §A.1 · pp. 5–6
• The frontend integrates BibTeX annotations and related papers into a new PDF or creates a LATEX file for importing into a source document.This supports source-code-based repositories such as arXiv.Source: §A.1 · p. 6

A.2 Backend

• The backend stores preprint metadata and searches previously uploaded preprints for related works using user-provided keywords, without processing preprint content directly.It also stores PDF files and serves them to the frontend when requested.Source: §A.2 · pp. 6–7

A.3 Database

• The database stores each preprint's title, author, DOI, publication URL, publication year, and associated keywords.It serves as a central repository for preprint metadata and supports storage and retrieval.Source: §A.3 · p. 7
• CiteAssist uses PostgreSQL for data integrity, indexing, and transaction support in managing preprint information.Source: §A.3 · p. 7

B CiteAssist Screenshots

• The screenshots cover the homepage, metadata editing, advanced metadata editing, related papers, BibTeX parsing, generation, and a preprint webview.Source: Figure 2 · p. 8, Figure 3 · p. 8, Figure 4 · p. 9

C CiteAssist Enhanced Document

• The enhanced-document illustrations show a generated annotation button on the first page and a citation box on the last page.Source: Figure 10 · p. 12, Figure 11 · p. 13

AI Usage Card for Enhanced Preprint Generator

• The AI Usage Card lists ChatGPT for literature review and finding literature, and reports rewriting the authors' own text to improve writing style.It states that corresponding authors verify and agree with modifications or generated content.Source: AI Usage Card for Enhanced Preprint… · pp. 14–15

Related Papers

• The Related Papers section lists Behera, Jain, and Kumar's 2023 work on visual exploration of literature using Connected Papers.Source: Related Papers · p. 16