What is it about?
When you type something into a search engine, have you ever wondered how the search engine knows in advance whether your search is likely to find good results or not? That's exactly what a task called Pre-Retrieval Query Performance Prediction (QPP) tries to solve by estimating how well a search will go before it even happens. This paper asks: can we improve those predictions by reading the brain activity of the person doing the searching? To explore this, we built a system that combines traditional text analysis of a search query with brainwave data (recorded using EEG, which measures electrical activity on the scalp) and, in some cases, eye-tracking data. We tested this across three different datasets, covering people who were either reading or listening to text, and trained machine learning models that learned to predict how well a given search query would perform. The results showed that, under the right conditions, adding brainwave data to the mix can genuinely improve these predictions, particularly for vague or fragmentary queries where text analysis alone struggles.
Featured Image
Photo by Sumaid pal Singh Bakshi on Unsplash
Why is it important?
Search engines today are powerful, but they are essentially flying blind when it comes to understanding how a user is cognitively engaging with what they're searching for. A search system that could recognise when a query is likely to return poor results and adapt accordingly would be far more helpful. It could, for example, prompt the user to refine their query, suggest alternative search strategies, or automatically adjust how it ranks results. This research is a step toward that vision. It shows that the brain's response to language contains real, measurable signals that can meaningfully improve prediction accuracy. As EEG technology becomes smaller, cheaper, and more wearable, the idea of neurophysiologically-informed search systems moves from science fiction toward something practically achievable. Beyond search engines, the broader implication is significant: human cognition contains information about information needs that no algorithm can fully infer from words alone, and tapping into it directly could reshape how we design intelligent systems.
Perspectives
At NeuraSearch, we have long believed that there is a direct and meaningful connection between what is happening inside a person's brain and the information they are looking for. When someone searches for something, that search is not just a string of words; it is the outward expression of an inner cognitive state: a question forming in the mind, a gap in knowledge, a moment of curiosity or confusion. EEG gives us a window into that inner state. We wanted to demonstrate that this neurophysiological signal, something as intimate as a person's brainwaves as they read or hear language, carries genuine, extractable information that is relevant to information retrieval. In this paper, we chose to focus on pre-retrieval query performance prediction as a concrete and well-defined task to test this idea: if the brain's response to language can tell us something about how well that query will perform in a search, then we built a case that neurophysiology and information seeking can be linked. This work is, in a sense, a small preliminary step towards a bigger vision in which future information retrieval systems could be built not just around what users type but around how they think.
Yashar Moshfeghi
University of Strathclyde
Read the Original
This page is a summary of: On The Use Of Electroencephalography In Query Performance Prediction, ACM Transactions on Information Systems, May 2026, ACM (Association for Computing Machinery),
DOI: 10.1145/3816244.
You can read the full text:
Resources
Contributors
The following have contributed to this page
