By Kasper Løvschall, Aalborg University Library

With Primo Research Assistant, you do not have to think about keywords or building a search to find the information you need. You simply ask a subject or research question in plain language and the assistant turns it into a search in the library's Primo search system using a large language model (LLM).

And it does not stop here! The assistant searches and retrieves the top 30 search results and applies what is known in machine learning as an embedding model. Basically, embeddings are a way for the machine to "understand" the meaning and context of words in a sentence, much the way humans do when reading a text.

This insight is used to select five sources that are considered the best to provide an answer to the question at hand. The language model is then used again to answer the user's original question.

Unlike a traditional chatbot response, the assistant's result resembles an encyclopaedia or textbook article with a clear indication of which sources that support its statement. You can quickly validate the results and use them directly in your future work.

Can it be trusted?

It is well known that language models can fail by fabricating or hallucinating content, which is a challenge in academic work. Primo Research Assistant greatly reduces this risk, as the model must base its answer on abstracts or summaries from the five sources, rather than on its own knowledge from the model's training data.

The assistant also reserves the right not to answer if the sources do not meet the requirements or if there are too few to provide a proper answer.

Why not just use Copilot or ChatGPT?

This is of course a relevant question, and a chatbot will be happy to answer the same question. What makes Primo Research Assistant different is that it is based on material provided by the library - i.e. predominantly academic material that has been through a peer review or subject specialist selection. Furthermore, there are specific safeguards in the assistant to minimize the risk of fabricated answers. It is fully transparent about the sources used to answer the question.

When should you use the Primo Research Assistant?

The assistant is a brand-new tool in the information retrieval toolbox. It can be used to get started quickly, for example, to provide an initial answer to a problem formulation, to explain a concept, or to provide an introduction to a research or subject area.

At the same time, you get a search in Primo that is already quite specific and limited to the questions you ask. In this way, you are already well into the "funnel" compared to a search where you manually refine the keywords on an ongoing basis based on the results you get from the search terms.

Try it!

We encourage you to try the Primo Research Assistant next time you use Primo. Get your own experiences with the tool, and you will quickly discover how AI-assisted services can complement traditional information searching with some really stong results. If you have questions about the research assistant you are welcome to contact us.