QH Libraries Search - Tips: Research Assistant (AI)

The generative AI tool Research Assistant allows you to perform a search using natural language. Instead of searching keywords on your topic, such as heart attack AND prevention AND diet you can ask a question, e.g. What diets assist in the prevention of heart attacks?

You can refine your search by resource type (books, journal articles and peer-reviewed) and date (last 5 years, last 10 years, or a custom date range).

The tool will provide you with 5 relevant sources and generate an overview of your question by analysing the description and abstracts of the sources.  In addition, it will provide suggested research questions to further explore your topic.

How does it work?

• The tool will turn your natural language question into a complex boolean query using a large language model (LLM).  
• The boolean query is sent to the Central Discovery Index (a massive index of articles and other documents) to perform the search. 
• The top results are returned (a maximum of 30), with the top 5 of these sent to the LLM to create the overview with inline references. 
• The overview response and the sources are displayed to you. 
• The LLM will only cite what is found in a search, so it does not make up non-existent citations as can occur with other LLMs such as ChatGPT. 
• The actions provided allow you to send a thumbs up or a thumbs down rating with feedback to the vendor, copy the information to your clipboard or try again. You may receive a slightly different answer due to the nature of LLMs. 
• The 5 sources are displayed with their abstracts and links to any full text availability. 
• If you click on View more results from your library search, this takes you to a QH Libraries Search page with more results. 
• You can also see the boolean query that was used for your search. 
• The Research Assistant will save your last 200 queries so that you can come back to them.  

What doesn’t it include?

The Research tool does not include QH Libraries collections – books and local collections (e.g. hospital history collections). Anything in the Library Catalogue search scopes are not included. 

Only sources that you get from an ‘Journal Articles’ search scope are included. Excluded are the following sources: 

• News content 
• Collections from Elsevier, APA, DataCite, Kogan Page, and Conde Nast 
• Databases that are not free to search without a library subscription e.g. Web of Science 

Keep in mind 

• As a summarisation tool, it is useful for quick searches but should not be used for comprehensive research. 
• The overview is based on 5 sources it determines to be relevant rather than based on objective evaluation methods so it may miss sources that are important to your topic.  
• Each question stands by itself, so follow-up questions are not currently supported.
• The overviews may also miss content contained in the full text of articles as only the description and abstracts are used. 
• The LLM may misinterpret the information in the abstracts and descriptions sourced especially with dates, so it is important check the accuracy of the responses provided.  
• Consider that biases that are present in research literature can lead to a biased response.  
• Search results may include sources that aren't available in our library and may recommend resources with full-text restriction.

Risks and privacy 

• Your input is not stored by the LLM nor is it used to train the LLM. 
• The vendor’s strict data privacy policies are adhered to.  
• All analytics and user feedback data are aggregated and anonymised. 

For more information about the use of AI within Queensland Health, refer to the Use of generative AI in Queensland Government guideline.