AI-Powered Upload

When you upload a publication file to AcaTrove, the AI-powered upload system goes beyond basic metadata extraction. It analyzes the document's content, identifies key findings, suggests categories and tags, and provides recommendations for organizing the work within your library.

How AI-Powered Upload Works

The AI analysis runs automatically when you upload a PDF, DOCX, or text file through the Publications upload interface. The process involves several steps:

1. Text extraction -- AcaTrove extracts the full text from the uploaded document using document parsing tools optimized for academic papers.
2. Metadata identification -- The AI identifies structured metadata including title, authors, abstract, journal, DOI, publication date, and references.
3. Content analysis -- The AI reads the full text to identify the research topic, methodology, key findings, and contribution to the field.
4. Recommendations -- Based on the analysis, AcaTrove suggests how to organize and use the publication within your workflow.

What the AI Extracts

The AI-powered upload extracts and populates the following fields:

• Title and authors -- Identified from the document header and author list.
• Abstract -- Extracted from the abstract section of the paper.
• DOI and references -- DOIs found in the document are used to cross-reference external databases for additional metadata.
• Keywords -- Subject keywords extracted from the text or identified through topic analysis.
• Publication type -- The AI classifies the document as a journal article, conference paper, review, preprint, thesis, or other type based on its structure and content.

AI Recommendations

After analysis, AcaTrove presents recommendations on the upload review screen:

Categorization suggestions -- The AI suggests which projects, labs, or collections the publication might belong to, based on topical similarity with your existing library.

Related publications -- If your library contains publications on similar topics, the AI identifies them so you can explore connections or organize them together.

Tag suggestions -- The AI proposes tags based on the paper's subject matter, methodology, and research domain. You can accept, modify, or dismiss each suggested tag.

Quality indicators -- For publications from recognized journals, the AI may note the journal's impact factor or the paper's publication status (peer-reviewed, preprint, etc.).

Reviewing AI Results

The AI analysis results appear on a review screen before the publication is saved. You have full control over what gets stored:

• Each extracted metadata field is editable. If the AI misidentified the title or an author name, correct it before saving.
• Recommendations are optional. Accept the ones that are useful and dismiss the rest.
• The AI's content summary can be saved as a note attached to the publication for future reference.

When AI Extraction Needs Help

The AI performs best on well-structured academic documents. It may produce incomplete results for:

• Scanned documents without OCR text.
• Heavily formatted or multi-column layouts where text extraction order is ambiguous.
• Non-standard document types such as slide decks or posters.

In these cases, supplement the AI extraction with manual edits on the review screen.

Tips

• Upload the publisher's PDF version when available. Publisher PDFs have more consistent formatting than author manuscripts, which improves extraction accuracy.
• Review the AI's author list carefully. Author name parsing can be tricky with non-Western name formats or unusual formatting conventions.
• Accept tag suggestions liberally. Tags make your publications easier to find later through search and filtering, and excess tags can always be removed.
• The AI analysis is stored with the publication record. You can revisit it at any time by opening the publication and clicking AI Analysis.