What the brain tells us about search engines

Joemon Jose is studying how the brain reacts when people peruse relevant documents on the Internet. In association with a psychologist, he is using an MRI scanner which can provide  high-resolution anatomical images of brain structure, and lower resolution  images of functional brain activity, to understand the process.

In today’s computerised world, we create huge amounts of data that are distributed widely on a daily basis. Our data creation prowess is increasing steadily, with the amount of stored data doubling every year, and in 2011 a colossal 1.8 Zetabytes (1.821 bytes) of data was created and replicated.

To help people find relevant bits of information as and when they need it amongst this vast quantity of data increasingly sophisticated search engines have been developed and have become an indispensible part of life. Nowadays, searching the internet is a ubiquitous activity and is performed by people from all walks of life all over the planet every single minute of the day.

With so much data stored on servers around the world, how do we actually find the information we are looking for? During a search process a user issues a query and the search system ranks potentially relevant documents in order of relevance. 

The average query posed by a user usually contains less than three terms and the search engines sift through billions of documents to identify the relevant documents for the user. However, this seemingly simple process introduces a number of uncertainties. There is an agreement in general that the ideal retrieval systems should provide users what they need in any given situation irrespective of their specified query.

Search engines

At present, web search engines use a number of techniques for modeling this retrieval process, which mainly exploit the search logs. Understanding relevance and clearly defining it is a prelude to building effective search systems.

Uncertainties involved in an information retrieval process and the subsequent deficiencies in the effectiveness of the system are dealt with by using a technique called relevance feedback.

Relevance feedback techniques exist in many different forms like implicit feedback, explicit feedback, affective feedback, etc. The basic idea is that the system gets searchers’ feedback on the relevance of documents and these are used for subsequent retrieval. However, capturing this feedback unobtrusively and reliably is a challenge. For this purpose, it is important to detect user reactions to relevant documents, which then the system can exploit to finding more relevant documents.

In this context, we are studying the connection between relevance and brain activity. The objective is to understand how does relevance happen and how does the brain react when users see relevant documents?

Despite the fact we often have difficulty specifying what we want, it is relatively easy for us to identify relevant information when we see it. Therefore, if we know how the brain reacts when we encounter relevant bits of information then this will open ways of building systems that are much more effective.

Brain activity

Working with Professor Frank Pollick in the School of Psychology at the University of Glasgow, I am studying how the brain reacts when searchers peruse relevant documents. Using an MRI scanner we study how the brain reacts during the information retrieval process. The MRI scanner can provide high-resolution anatomical images of brain structure, and lower resolution images of functional brain activity. The high spatial resolution and whole brain coverage that characterizes MRI images allows spatial localization of brain activation during the relevance assessment process.

The initial findings identified three brain regions in the frontal, parietal, and temporal cortex where brain activity had differed when searchers pursued relevant and non-relevant documents. These results are the starting point of building much more effective search systems.

Given the proliferation of smart phones and the gadgets, there is an opportunity to fetch and retrieve documents for users as and when they need it. However, in order to build such systems it is important to know what documents users find relevant and how they react to it. Such reactions will then be exploited to find more relevant documents in real-time and unobtrusively.