A Brief Account on the Recent Advances in the Use of Quantum Mechanics for Information Retrieval

Abstract

The key challenge of Information Retrieval (IR) is relevance, that is, the property of a document that makes the document relevant to the end user’s information need in a given context. Relevance is a key challenge because it is an unknown property – were every document at every context marked with a relevance assessment prior to any information need or were it indexed so that a set of terms is assigned if and only if the document is relevant, retrieval would be perfect. Unfortunately, this is not the case. Although an IR system is in principle able to optimally rank documents, retrieval effectiveness depends on how the document collection has been represented, e.g. which terms are associated to which documents.

Quantum Theory (QT) can be used in IR because its mathematics describes the properties of documents and queries which would outperform the current retrieval models if these properties could be observed. Using QT in a different domain than Physics is not obvious, thus making the use of QT in IR is useful and in fact necessary.

To this end, we describe how to use QT in, and we explain how QT can improve the current situation of IR. In particular, we suggest how measurement uncertainty can be leveraged in IR for designing novel retrieval functions constrasting current retrieval functions which are based on the uncertainty generated at indexing time. Moreover, we describe IR in terms of signal detection. By viewing retrieval as signal detection, it is possible to cast some results of quantum signal detection to IR and define the procedures for designing an indexing algorithms which better represents documents than traditional algorithms. Lastly, we describe some experimental results that suggest the potential of QT in IR.