Titre : Eye-Tracking Assisted Search Engine

Sujet proposé dans : M1 Informatique, TER --- M1 Mosig, TER --- Magistere, M1

Responsable(s) :

Mots-clés : Information retrieval, Eye-tracking
Durée du projet : TER
Nombre maximal d'étudiants : 1
Places disponibles : 1
Interrogation effectuée le : 24 novembre 2020, ŕ 12 heures 11


Description

 

Eye-tracking is a technique for detecting where a user gaze at, and it makes it possible to determine what his/her interests are (https://en.wikipedia.org/wiki/Eye_tracking).

 

The scientific objective of the project are to study the impact of the use of eye-tracking on the performance of a Web search engine. Our hypothesis is that "knowing the information read by the user on the results page returned by a search engine could effectively improve the relevance of the results provided it".

 

After the query of a user, classic Web search engines usually return a list of results composed of extracts (called snippets) of supposedly relevant documents. Usually the user then chooses the document that seems most interesting by clicking on the corresponding snippet, and then consults the document. If the document is not relevant, the user can return to the results page, and choose another document, etc.

 

The eye-tracking can study how the user reads each snippet of the results page, and determine if some words are more particularly watched by the user. It is thus possible to extend the initial query of the user with these words, to send a new more precise query, and thus to automatically improve the search results without explicit intervention by the user.

 

A mockup has already been realized and gave interesting results [Sungeelee et al. 2020]. We would like to improve this mockup by using a new and more powerful eye-tracking device (Tobii 4C) and to connect it to more powerful search engines (Google, Bing, Qwant, Terrier V5). This is the goal of the project.

 

Experiments with users could be organized to validate the interest (for the information search on the web) of the proof of concept realized.

 

References: