FAQs about Collections Explorer

Collections Explorer currently includes over 6.5 million records from ArchiveSpace, Alma, LibraryCloud, and JSTOR Digital Stewardship.  Finding aids published in Harvard’s ArchivesSpace, including nearly 13,000 finding aids and 2.9 million child-level records. About 195,000 of these records include links to digitized materials. This metadata comes from ArchivesSpace and LibraryCloud. About 2.2 million bibliographic records for special collections materials at Harvard Library. This metadata comes from Alma. Around 1.4 million image records from JSTOR Digital Stewardship. We are continuing to add new data sources, and this description will be updated as additional materials become available.

Yes. You can search Collections Explorer using languages other than English, including non-Latin scripts. Results may include materials in any language represented in the collections, not just the language of your search.

Collections Explorer uses a hybrid search approach that combines keyword search and AI-based semantic search. This helps the system find both exact word matches and results that are similar in meaning, even when the wording does not exactly match your search. Each result is given a relevance score, and higher-scoring results appear closer to the top of the list.

Search results for the same search query may vary in Collections Explorer due to the probabilistic nature of semantic search. Variations in the search results should be small and typically only impacts the order of results.

No. Collections Explorer does not currently include personalization features, and users do not have accounts. Searches are not connected to information that identifies an individual user.  Please see Harvard Library’s policy on Privacy, Terms of Use & Copyright Information/ for more information.

Yes. In some cases, both records for the same collection may appear in search results. Records from Alma are labeled “Manuscripts/Archives.” Records from finding aids are labeled “Collection Guide” or “Part of Collection Guide.” This may change as we continue refining records and system functionality in Collections Explorer.

Collections Explorer uses AI in two ways: An embedding model supports semantic search and helps find materials related in meaning to your search terms. Large language models (LLMs) support some search features: Mistral generates the “About this Material” summaries. Claude generates the “You Might Also Try” suggestions and the keyword searches used for the “Try HOLLIS Catalog” link.

Collections Explorer uses a pre-trained AI model for search embeddings. The system does not “learn” from user queries.

The search results in Collections Explorer come from Harvard systems of record, such as ArchivesSpace and Alma. The AI-generated features, such as the search result summaries and the “You Might Also Try” suggestions may occasionally contain inaccurate or misleading information due to the nature of AI.

Collections Explorer uses a hybrid search approach that includes AI-based semantic search. This means the system may sometimes return results that are not closely related to your search term, particularly when there are few strong matches in the catalog. A large language model (LLM) generates the summary shown for each result. In some cases, the summary may note that a result does not appear to be relevant. The development team is exploring ways to better limit search results to display only those that meet a certain level of relevancy.

The following ArchivesSpace metadata fields are searchable in Collections Explorer: ead_id title creators agents dates notes (including bioghist, scopecontent, and abstract) lang_materials subjects repo_code