Knowledge graphs for question answering over structured and unstructured data

Join us and get a fully-funded PhD scholarship in this EPSRC ICASE collaboration between King’s College London and BT on multimodal knowledge graphs!

Knowledge graphs (KGs) are on the rise to connect structured and unstructured data in organisations, answer questions, and discover insights. Many enterprise information services (search engines, chatbots, recommender systems) use KGs to extract, link, and contextualise answers. Through the reuse of public resources and semantic technologies, KGs can manage and integrate disparate organisational data sources, foster application interoperability, and meaningfully represent structured enterprise knowledge (e.g. spreadsheets, databases, etc.).

However, structured data is rarely the only source of key organisation knowledge: unstructured text (e.g. memoranda, reports, communications, etc.) is becoming increasingly critical in knowledge markets and decision making. The use, extraction, and combination of knowledge from structured and unstructured sources for question answering for organisational decision-making is often not understood and typically unautomated.

This project proposes an end-to-end system for enterprise question answering over structured and unstructured data, combining ETL, information extraction, multimodal querying, and KG embeddings. Of particular interest will be linking Internet of Things sensor data to other structured and unstructured data sources.

This is a collaborative doctoral training project between King’s College London (KCL) and British Telecom (BT), in which the following research questions will be addressed: (a) How can current knowledge graph querying paradigms be extended to include multimodality, and enable the simultaneous querying of structured and unstructured sources? (b) What are the requirements for extending current ETL and information extraction workflows with explainability models? What is an adequate evaluation framework for assessing the quality of these ETL and extraction explanations? What is the role of knowledge engineers in such an evaluation? (c) How effective are existing provenance models and provenance generation systems in documenting the knowledge creation and curation processes in organisations? How useful are those provenance traces for accountability purposes? (d) How can various data sources and scenarios from BT projects be integrated into a multimodal question answering dataset? How can such a dataset be used for maximising data retrieval from structured and unstructured sources simultaneously? (e) What metrics and benchmarks are adequate to evaluate such a multimodal question answering system using offline and online (user-based) evaluation techniques?

How to apply

Candidates must apply via King’s Apply online application system. Details are available at How to apply - King's College London (kcl.ac.uk).

Please indicate Professor Elena Simperl and Dr. Albert Meroño-Peñuela as the supervisors and quote the project title “Knowledge graphs for question answering over structured and unstructured data” within your application and in all correspondence.

The selection process will involve a pre-selection on documents and, if selected, will be followed by an invitation to an interview. If successful at the interview, an offer will be provided in due time.

Queries

Please direct all queries regarding this project to Dr. Albert Meroño-Peñuela, albert.merono@kcl.ac.uk.

(Again for applications - please read the 'How to Apply' and submit via King's Apply)