Report

Teaser, summary, work performed and final results

Periodic Reporting for period 1 - QUARTZ (Quantum Information Access and Retrieval Theory)

Teaser

All of us are users of Information Access and Retrieval (IAR) systems such as Web search engines. The peculiar difficulty of IAR is the fact that relevance cannot be precisely and exhaustively described by the data itself. Quantum Mechanics (QM) shares with IAR the intrinsic...

Summary

All of us are users of Information Access and Retrieval (IAR) systems such as Web search engines. The peculiar difficulty of IAR is the fact that relevance cannot be precisely and exhaustively described by the data itself. Quantum Mechanics (QM) shares with IAR the intrinsic and unavoidable uncertainty of measurement. As a quantum mechanical system such as particle cannot be fully known through measurement because of incompatibility between observables, the user’s relevance of an information object to his/her information need cannot be fully known through keywords, clicks, or views because of incompatibility between context and data. The quantum mechanical framework aims, at a more foundational level, to study corpuses of documents in order to identify their underlying quantum structures. This will not only enable to better understand optimal mathematical structures to model semantic spaces, thus consistently and efficiently capturing the more abstract level of meaning conveyed by the documents and texts of human language, but also to use quantum structures to enhance the effectiveness of the different models in IAR applications. QUARTZ started from the idea of transferring the scientific research results and expertise from senior researchers in the utilisation of the theory of QM in IAR to the junior researchers, thus stimulating the birth and growth of a networked European community of scholars with a larger, stronger, and deeper expertise in IAR. Thanks to previous research work started from van Rijsbergen’s seminal book and recently culminated with another book by the coordinator, QUARTZ is based on the main pillars of the quantum mechanical framework to build a new approach to IAR. The senior researchers involved in this project as organisation leaders or independent advisory committee members are leading researchers in multidisciplinary fields of QM and IAR.

Work performed

We investigated and evaluated the geometry of IAR at the level of high-dimensional word embedding vectors and at the level of high-level conceptual relevance dimensions.
* We investigated polyrepresentation, which consists of using the geometry of abstract vector spaces to embed a variety of contextual dimensions, information objects and information needs.
* We proposed a geometric model inspired by the quantum mechanical framework to capture the user\'s importance given to each dimension of relevance.
* We discussed our methodology to construct incompatible basis for documents from real world query log data, the experiments to test Bell inequalities on this dataset and possible reasons for the lack of violation.
* We put forward that quantum structures can be useful in the decision-making processes that follow from the retrieval of relevant information.
* We proposed a methodology for carrying out such an investigation in large scale and real world data using query logs of a web search engine, and device a test to detect the presence of an irrational user behavior in relevance judgment of documents.
* We further validated this behavior through a Quantum Cognitive explanation of the order and context effects.
* We aimed to understand the extent in which Information Scent, Information Patch and Information Diet, features of Information Foraging Theory, address the conceptual issues of information behaviour research by reviewing approaches to information interaction in the context of information seeking and retrieval.
* We proposed a plan for the application of IFT to information access and retrieval in the process of describing key IFT concepts in an information environment.

* We investigated a query expansion framework on the Quantum Language Modeling (QLM) basis.
* We proposed a binary classification model inspired by quantum signal detection theory in an effort to investigate how much benefit it brings as compared to classical models.
* We gave a few examples of irrational behavior and use a generalized probabilistic model inspired by the quantum mechanical frameworkto model and explain such behavior.
* We examined a multimodal fusion scenario that might be similar to that encountered in physics by firstly measuring two observables (i.e., text-based relevance and image-based relevance) of a multi-modal document without counting on an ensemble of multi-modal documents already labeled in terms of these two variables.
* We investigated the existence of non-classical correlations between pairs of multi-modal documents.
* We investigated the existence of this kind of non-classical correlation through the Bell inequality violation.
* We experimentally tested several novel association methods in a small-scale experiment.
* We presented a series of interesting discussions, which may provide theoretical and empirical insights and inspirations for future development of this direction.

We performed the following training activities:
• Winter school: http://www.quartz-itn.eu/training/winter-school
• Autumn school (added to the original plan to work around the issues of visa): http://www.quartz-itn.eu/training/autumn-school
• Workshop 1 (W1): http://www.quartz-itn.eu/training/winter-school/program
• Workshop 2 (W2), which was doubled to work around the issues of visa):
o http://www.quartz-itn.eu/training/workshop2-ts1-ts3
o http://www.quartz-itn.eu/training/autumn-school

We also provided the following complementary training:
• Transversal skills (Agile for knowledge workers: practical hints, Team building, Research in video: Main guidelines to communicate scientific results with video) provided at the Winter school.
• Transversal skills (Academic writing) provided at the Autumn school.

Final results

* To replace the conventional IAR paradigm based on keywords and the classical probabilistic and statistical models, by a geometry of IAR, in which non-classical methods are based on complex and heterogeneous feature vectors in abstract vector spaces.

* To extend the usual interaction models, logic and languages utilised by IAR systems to model and make non-distributive nor commutative operators for complex features and interaction which require a different logic from a classical set-based logic, in particular, represented by subspaces and relations thereof.

* To generalise the probability theory adopted by current IAR systems.

Website & more info

More info: https://www.quartz-itn.eu.