Project description

The Alan Turing Institute Studentship.

Systematic reviews provide a robust approach to the comprehensive identification of relevant sources of evidence used by Health Technology Assessment agencies such as the National Institute for Health and Care Excellence. AI methods, which automate the development of evidence-based systematic reviews have mostly focused on search and screening. The topic of this PhD is to conduct novel research on synthesis through automatic text summarization (ATS) methods, which will create a concise and coherent summary containing the salient points of a document, and/or document collection. The research will be applied to decision-analytic model based cost effectiveness analysis of stratified medicine. ATS provides a solution to information overload driven by massive amounts of textual data on the Web. Multi-document summarisation has attracted more attention given the need to filter out articles with identical or similar content and reduce redundancy. Depending on the output, ATS can be categorized into extractive and abstractive summarisation. Extractive summarisation produces a summary by choosing a subset of sentences related to the main idea of the input document. Abstractive summarisation, in contrast, generates summaries by modifying phrases and sentences from the input. Most ATS systems have focused on extractive methods. Depending on the context, text summarisation can be categorized into generic and query-based summarisation. Generic summarisation determines the importance of information only with respect to the content of the input alone. Query-focused summarisation outputs a short summary answering the query according to the data in the provided documents. To support the process of synthesis and the population of evidence tables for systematic reviews for economic evidence for stratified medicine, we will investigate new methods for query-focused abstractive multi-document summarization. Traditionally, human judges are needed to assess the quality of the summary. To avoid this time consuming and labour intensive evaluation, we will use automatic measuring metrics (such as ROUGE scores), which we will compare with gold standard summaries produced by humans. Creating such a summary is a subjective task because the content of the final summary depends to a great degree on the human annotator. This PhD will be validated with use cases from economic evidence for stratified medicine.


Hashimoto, K., Kontonatsios, G., Miwa, M., and Ananiadou, S. (2016) Topic Detection using Paragraph Vectors to Support Active Learning in Systematic Reviews, Journal of Biomedical Informatics 62, 59-65

Nenkova, A. and K. McKeown. Automatic Summarization. Foundations and Trends in Information Retrieval. Vol. 2, Nos. 2-3, 2011, 103-233

See, A., Peter J. Liu, and C. D. Manning. (2017) Get to the point: summarisation with pointer-generator networks. Proceedings 55th ACL, 1073-1083

Christopoulou, F., Miwa, M. and Ananiadou, S. (2018) A Walk-based model on Entity Graphs for Relation Extraction, Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, 81-88

For more information on the Alan Turing Studentship please visit the Turing website
https://www.turing.ac.uk/phd-at-turing
For information on how to apply and documentation needed please contact engineering-pgr-admissions@manchester.ac.uk
Deadline for application is 12th January 2019