Towards Neuro Relevance Feedback
Lead Research Organisation:
University of Strathclyde
Department Name: Computer and Information Sciences
Abstract
Summary: Relevance is a fundamental concept in information retrieval (IR), and a vast body of research exists that attempts to understand this concept so as to operationalise it for IR systems. Over the past 40 years, the orientation of research into the concept of relevance has developed from considering only the relevance of documents to a query to understanding and modelling a more user-oriented concept of relevance.
IR systems use an operationalised version of relevance. This concentrates on the retrieval algorithms that process information objects and match them with users' queries, attempting to maximise retrieval of relevant information objects (system side of relevance). Users of such systems then browse through retrieved results to find what they consider relevant (user side of relevance) depending on their context, cognition and affect. The system and user sides of relevance are complementary, and to improve the performance of IR systems, the user and system sides should work together.
Background: IR systems employ feedback techniques to integrate the user and system sides of relevance. An example of such an approach is the relevance feedback technique where feedback is gathered through explicit, implicit, or affective feedback. Despite the robustness of explicit feedback in improving retrieval effectiveness, it is not always applicable or reliable due to the cognitive burden that it places on users.
To overcome this cognitive burden, implicit feedback is proposed where relevance is inferred from the interactional data indirectly and unobtrusively. However, if we can identify which brain regions are activated during the explicit relevance judgement, and how these activations changes for positive (relevant) and negative (non-relevant) feedback, we can use these finding as a direct way of measuring relevance, and hence can use as a novel and effective feedback technique, "Neuro Relevance Feedback", without intruding the user cognitive process. This PhD project strives to investigate this challenging task.
Aims: This project will develop the underlying and core technology of "Neuro Relevance Feedback" (NRF), aiming to (i) infer relevance judgement phenomena happening in the brain while users are interacting with the retrieved results from a search engine and based on that (ii) personalise relevant information, as well as recommend new ones.
To achieve these aims, NRF will monitor changes in the brain and based on that, accurately infer and predict the relevance judgment in an information retrieval and seeking process. The predicted relevance judgment can be then incorporated by the search engine as a valuable source of information to improve the relevance of their retrieved results.
Objectives: This project has three main objectives:
(i) to detect relevance judgment from brain signals in real time;
(ii) to predict relevance judgment from brain signals in real time; and
(iii) to develop effective Neuro Relevance Feedback technique for an information retrieval and seeking process.
IR systems use an operationalised version of relevance. This concentrates on the retrieval algorithms that process information objects and match them with users' queries, attempting to maximise retrieval of relevant information objects (system side of relevance). Users of such systems then browse through retrieved results to find what they consider relevant (user side of relevance) depending on their context, cognition and affect. The system and user sides of relevance are complementary, and to improve the performance of IR systems, the user and system sides should work together.
Background: IR systems employ feedback techniques to integrate the user and system sides of relevance. An example of such an approach is the relevance feedback technique where feedback is gathered through explicit, implicit, or affective feedback. Despite the robustness of explicit feedback in improving retrieval effectiveness, it is not always applicable or reliable due to the cognitive burden that it places on users.
To overcome this cognitive burden, implicit feedback is proposed where relevance is inferred from the interactional data indirectly and unobtrusively. However, if we can identify which brain regions are activated during the explicit relevance judgement, and how these activations changes for positive (relevant) and negative (non-relevant) feedback, we can use these finding as a direct way of measuring relevance, and hence can use as a novel and effective feedback technique, "Neuro Relevance Feedback", without intruding the user cognitive process. This PhD project strives to investigate this challenging task.
Aims: This project will develop the underlying and core technology of "Neuro Relevance Feedback" (NRF), aiming to (i) infer relevance judgement phenomena happening in the brain while users are interacting with the retrieved results from a search engine and based on that (ii) personalise relevant information, as well as recommend new ones.
To achieve these aims, NRF will monitor changes in the brain and based on that, accurately infer and predict the relevance judgment in an information retrieval and seeking process. The predicted relevance judgment can be then incorporated by the search engine as a valuable source of information to improve the relevance of their retrieved results.
Objectives: This project has three main objectives:
(i) to detect relevance judgment from brain signals in real time;
(ii) to predict relevance judgment from brain signals in real time; and
(iii) to develop effective Neuro Relevance Feedback technique for an information retrieval and seeking process.
