Search and Discovery

AcaTrove's search goes beyond simple keyword matching. The platform uses semantic search powered by vector embeddings to find papers and documents by meaning, not just by the exact words they contain. This means searching for "machine learning applications in genomics" returns relevant results even if the documents use terms like "deep learning," "neural networks," or "computational biology."

How Search Works

AcaTrove combines two search approaches:

Semantic search -- Your query is converted into a vector embedding (using Sentence-Transformers) and compared against the embeddings of all indexed documents. Results are ranked by semantic similarity, so conceptually related content surfaces even without exact keyword overlap.

Full-text search -- Traditional keyword search that matches exact terms in titles, abstracts, full text, and metadata. This is useful when you need to find a specific paper by title or locate documents containing a particular technical term.

By default, AcaTrove blends both methods for the best results. You can switch to exclusively semantic or full-text mode using the search mode toggle.

Performing a Search

1. Navigate to /search or use the global search bar available in the top navigation on every page.
2. Type your query. It can be a natural language question, a set of keywords, or a paper title.
3. Press Enter or click the search icon.
4. Results appear ranked by relevance, showing the title, authors, a snippet of matching text, and a relevance score.

Search results page with relevance scores and snippets

Filtering Results

Narrow your results using the filter panel:

• Document type -- Papers, proposals, reports, presentations, or all types.
• Date range -- Limit results to a specific time period.
• Project/Lab -- Search within a specific project or lab scope.
• Authors -- Filter by specific author names.
• Tags -- Filter by tags you or your team have applied to documents.

Filters are applied in real time, and the result count updates instantly.

Search filter panel with document type and date range options

Understanding Relevance Scores

Each result includes a relevance score from 0 to 1. Scores above 0.8 indicate a strong semantic match. Scores between 0.5 and 0.8 suggest moderate relevance. Results below 0.5 may be tangentially related. The blended mode boosts results that score well in both semantic and keyword matching.

Saving and Reusing Searches

Click Save Search on any results page to save your query and filters for quick reuse. Saved searches appear in the search sidebar under My Searches. You can also set up search alerts to be notified when new documents matching your query are uploaded.

Searching Within a Project

From any project page, click the search icon in the project header to search only within that project's documents. This uses the same semantic and full-text search but scopes results to the project's library.

Tips

• Phrase your query as a natural language question for best semantic results (e.g., "What methods are used to detect protein-protein interactions?" rather than "protein protein interaction methods").
• Use quotes around exact phrases in full-text mode to find precise matches.
• If results are too broad, add filters rather than making the query longer.