Guilherme Holdack - quartz-itn.eu fileDesign, evaluation and application of an interactive quantum theoretical polyrepresentativeframework (ESR-4) Guilherme Holdack University of Bedfordshire,

Design, evaluation and application of an interactive quantum theoretical polyrepresentative framework

(ESR-4)Guilherme Holdack

University of Bedfordshire, UKMTRM, Padova, 07/12/2018

H2020-MSCA-ITN Grant Agreement N. 721321

My Background

MTRM, Padova, 07/12/2018 H2020-MSCA-ITN Grant Agreement N. 721321 1

Training Experience

MTRM, Padova, 07/12/2018 H2020-MSCA-ITN Grant Agreement N. 721321 2

QUARTZ Project• QUARTZ Winter School and Workshop 1, 07/02/2018 –14/02/2018, Padua, Italy.• QUARTZ Summer School and Workshop 2, 20/06/2018 – 22/06/2018, Luton, UK• QUARTZ Autumn School and Workshop 3, 24/09/2018 – 26/09/2018, Cottbus, Germany

University of Bedfordshire• AI in the wild – Erik Graf (Cortical.io), University of Bedfordshire – July 2018.• The theory of open quantum systems in cognitive psychology by Irina Basieva, University

of Bedfordshire – June 2018.• Passed first PhD Pre-examination. Interview with two local examiners (University), and

submission of thesis report.

Conferences• ICTIR, 2018 - Tianjin, China.

Research Goals (1/2)

MTRM, Padova, 07/12/2018 H2020-MSCA-ITN Grant Agreement N. 721321 3

Main Goal: Design, evaluation and application of an interactive and

polyrepresentative IR framework, i.e. a framework capable of combining

different sources of evidences, on different representations, from documents.

The session-based interaction should change IR System state to provide

personalized and relevant documents on the user perspective.

Research Goals (2/2)

MTRM, Padova, 07/12/2018 H2020-MSCA-ITN Grant Agreement N. 721321 4

Research Goals (2/2)

MTRM, Padova, 07/12/2018 H2020-MSCA-ITN Grant Agreement N. 721321 5

1) Polyrepresentation: identification of different sources of evidencesGathering contextual and cognitive aspects from textual input, from different actors, like:a) Making use of techniques to identify terms spaces, like word2vec.b) Mapping the term/vector spaces into quantum inspired representations,

reproducing the work of [3] (publication from the same network) to detect dependencies between contexts and cognitive aspects.

c) Apply same methodology of b) to possibly detect interference in betweendifferent representations.

2) User interaction and change of system state:a) deployment of a prototype capable of reflecting

feedbacks, providing content to users on ‘infinitescroll’, populating the feed with new contentbased on session interaction.

Research Methodology (1/2)

MTRM, Padova, 07/12/2018 H2020-MSCA-ITN Grant Agreement N. 721321 6

Research Methodology (2/2)

MTRM, Padova, 07/12/2018 H2020-MSCA-ITN Grant Agreement N. 721321 7

b) divide the work in 2.a into two stages; first to create a dataset of interactions and

relevance of diferent representations, on user perspective. Second to evaluate the model

served as an outcome of the former. (distributed log analysis and experiments)

3) evaluation against traditional IR baseline:a) collectively conduct experiment with two different groups, having

polyrepresentative+interactive feature activated against classical models in IR community.

Conduct all evaluations against proper evaluation framework; such as TREC

Achieved Results (1/2)

MTRM, Padova, 07/12/2018 H2020-MSCA-ITN Grant Agreement N. 721321 8

A. Reproduction of Complex Word Embeddings [3] to understand interference between terms in a complex term space.• An outcome was a paper to optimize the generation of such

structures (LWDA 2018).

B. Currently ongoing development of a prototype capable of addressing the design of the polyrepresentative / interactive framework.• The pilot should be deployed by February/2019 on a distributed

manner (web search + native app for Android).

Achieved Results (2/2)

MTRM, Padova, 07/12/2018 H2020-MSCA-ITN Grant Agreement N. 721321 9

Future Work

MTRM, Padova, 07/12/2018 H2020-MSCA-ITN Grant Agreement N. 721321 10

Short term• Pilot ‘distributed lab tests’ to gain insights on users and their correlations

with different representations;• Analyze the provided data to propose a polyrepresentative model;

Long term• A quantum-inspired polyrepresentative and interactive Information

Retrieval framework (application of short term goals + evaluation of the proposed model).

Project Deliverables

[1] Jaiswal, A. K., Holdack, G., Frommholz, I., & Liu, H. (2018). Quantum-like Generalization of Complex Word Embedding: a lightweight approach for textual classification. In Proceedings of the Conference "Lernen, Wissen, Daten, Analysen" (LWDA). Mannheim, Germany: CEUR-WS.

[2] Holdack, G. (2018). On polyrepresentation, user interaction mapping and dynamic relevance. In Proceedings FDIA 2018.

References:

[3] Li, Q., Uprety, S., Wang, B., & Song, D. (2018). Quantum-inspired Complex Word Embedding. In ACL Third Wordshop on Representation Learning for NLP.

MTRM, Padova, 07/12/2018 H2020-MSCA-ITN Grant Agreement N. 721321 11