Research I: AI4IA

The huge volume of information available to us through open sources is an incredibly rich and useful resource, but making sense of it all, knowing what to focus on and linking concepts together across it all is overwhelming. Sifting through it and synthesising it to extract relevant information and data points - or entities - is humanly impossible, but with the help of the right computing power and algorithms, the impossible becomes possible. Not to replace us, but to augment us. This is the topic of this Artificial Intelligence for Intelligence Augmentation session. Join us in the first of two applied research-based sessions as we explore how machine learning can be used to augment public health intelligence and make the otherwise impossible possible.

What's missing in geographic parsing?

Nigel Collier, Lecturer in Computational Linguistics, University of Cambridge, UK

This talk will report on research with Milan Gritta and Taher Pilehvar that outlines current capabilities and challenges in Natural Language Processing for automated place name understanding ('geo-parsing'). The ability to geo-locate events in textual reports represents a valuable source of information in many real-world applications such as emergency responses. However, geo-parsing is still widely regarded as a challenge because of domain language diversity, place name ambiguity, metonymic language and limited leveraging of context. Geo-parsing task, its challenges, evaluation metrics and a comparison of current methods will briefly be introduced.

Developing information extraction algorithms for open-source data from event-based surveillance systems

Victoria Ng, Senior Scientific Evaluator, Public Health Agency Canada

The Internet-based Surveillance Informing Global Health Threats (InSIGHT) project led by PHAC will integrate advanced analytics into event-based surveillance systems. To achieve this, the extraction of key information from open-source data is necessary to convert unstructured data into structured data for epidemiological risk modelling and assessment. Using Zika virus as a pilot candidate disease, we are developing algorithms to extract information about infection events, including the time of events and the number of infected individuals associated with each event. This presentation will focus on the current developments of these algorithms and an early evaluation of the algorithms against human performance.

From Publications to Knowledge Graphs

Panos Constantopoulos, Professor, Department of Informatics, Athens University of Economics and Business

This talk will address the potential for knowledge access and integration offered by ontology-driven semantic graph indexing. More effective knowledge access becomes possible when data- and process- oriented approaches are combined, especially in view of the increasing ability to use automatic knowledge extraction and indexing techniques. In particular, it will review the Scholarly Ontology, specifically designed to represent different aspects of scientific work processes, and how this can be used to drive the extraction from publications with natural language processing methods and subsequent linking of information about 'who' does 'what', 'why' and 'how'. Tests to date yield promising results.