Artificial Intelligence

Using ChatGPT to Generate & Translate Search Strings: An Exploration

At this point (February 2024), most academic librarians are aware that ChatGPT will create citations to books and articles that don’t exist to satisfy a query. However, does that mean that ChatGPT is of no use in the literature searching process? 

When creating and conducting a comprehensive search strategy, there are two major, time-consuming tasks that ChatGPT may be able to help with. The post will detail my exploration into ChatGPT’s utility for:

  1. Keyword/Term Harvesting: The process of developing a list of keywords that correspond to the concept(s) you are researching. These may be synonyms, terms that fall under the same umbrella, and/or various spellings of the same term (British vs. American English, plural vs. singular, etc.). 
  2. Boolean Structuring: The process of using Boolean Operators (AND, OR, NOT) as well as correctly placed parentheses to structure the search logically. 

The Exploration

I am going to use ChatGPT to complete the above tasks for the following question: What is the relationship between socioeconomic status and traumatic brain injury?

Keyword/Term Harvesting

Typically, Keyword/Term Harvesting is an iterative process, where the searcher uses the terms that they know to begin their search, and then gathers more terms by reviewing these results. The search will likely grow as the searcher builds their search, but some terms may be tested out and not included. 

When asking ChatGPT for suggestions for synonyms of “Socioeconomic Status”, it provided some useful terms, including “Income level” and “Financial status.” However, the terms were very repetitive, most of them including an identical term to many of the others (for example, most contained “status” or “level”). See screenshot below. 

Figure 1: Example of prompt in ChatGPT requesting synonyms for a search.

I next asked ChatGPT to provide a list of terms that fall under the umbrella of “Socioeconomic Status.” It provided terms such as Occupation, Social Capital, and Health Status. While many of these would not work as search terms themselves, this question did a much better job at opening up the term “Socioeconomic Status” and digging into subtopics that we may want to search. 

Figure 2: Example of prompt in ChatGPT requesting specific terms in search topic.

ChatGPT worked in a similar way when asked the same two questions for Traumatic Brain Injury. That said, the umbrella terms question led to many more specific medical terms that may be useful in a search. 

Figure 3: Example of prompt in ChatGPT requesting specific terms for research topic.

Would I Recommend? I would only recommend using ChatGPT for Keyword/Term Harvesting to a searcher who is already familiar with the subject area. This research could use ChatGPT’s suggestions as a starting point, and then use their own knowledge and resources to fill in any gaps. I worry that an unfamiliar searcher would stop here and unknowingly limit their search based on ChatGPT’s suggestions. 

Boolean Structuring

Within the same “conversation”, I took a chance and asked ChatGPT to combine all of the keywords that it came up with for each concept into a Boolean search. I was excited that ChatGPT knew exactly how to structure the search in terms of parentheses, AND, and OR. 

Figure 4: Example of prompt in ChatGPT asking for Boolean search strings.

However, it did not include the umbrella terms for Socioeconomic Status. I asked for a revision and it complied, including the terms. An important thing to note is that ChatGPT puts all terms in “quotes”, even if they are just one word. As the first database I plan to search with this strategy is PubMed, I want to remove the quotation marks to ensure that Automatic Term Mapping is enabled. 

Would I Recommend? I would recommend this workflow to searchers, however as with almost all automated processes, I would recommend that they check ChatGPT’s work before running the search. In my case it inexplicably left off terms that I wanted to include, and had asked for! 

Conclusions

ChatGPT is a useful tool for combining search terms with Boolean operators, however should be used sparingly and only in conjunction with established methods for keyword/term harvesting. 

Samantha Walsh, Research Services Librarian, CUNY Hunter College

We welcome your comments and suggestions. If you have a resource that you would like to see highlighted please leave us a comment.

References

OpenAI. (2023). ChatGPT (Mar 14 version) [Large language model]. https://chat.openai.com/chat

Leave a comment