YOUR TEAM: You will work in the Department of Mathematics and Computer Science of the University of Cagliari
GOAL: The overall goal of this thesis is to identify and understand the personal health of the user from the emotions expressed along the text or said from speech, for example in applications in Mother&Child care and healthy-living and assisted care. We aim at: (i) using ontology of personal health terms which includes terms related to body organs, symptoms, treatment, medical professional designations; (ii) using lexical and semantic resources to identify terms that hold only health-related meaning and more ambiguous terms; (iii) leveraging NLP services defined within RO1 and semantics to come up with a system for emotion detection and associated REST APIs. For such a research objective, NLP techniques and tools will be analysed to detect those that can be applied to the Sentiment Analysis problem in personal health. Supervised and unsupervised approaches will be identified to have a clear overview of the current state of art. Then, lexical resources and frameworks such as (not limited to) FrameNet, SentiWordNet, WordNet, VerbNet, FrameBase, BabelNet, FRED, FrameSter will be taken into account with the purpose of defining a multi-disciplinary approach to sentiment analysis at the cross-roads between affective computing and common sense computing focused in personal health. The overall goal is to better recognize, interpret and process opinions and sentiment out of a text document or speech of a certain user within the personal health domain.
ESR activities and expected results
- To bridge the semantic gap between word-level natural language data and the concept-level opinions conveyed by these so as to provide a more efficient passage from unstructured textual information to structured machine-processable data. For such a purpose, model ontologies (especially the existing ones in personal health domain) will be defined and developed using best practices of Semantic Web.
- Identify the opinions (statement by somebody, called the holder), the opinion holders, and facts/topics (e.g. ailment, treatment, medications, and other topics within the personal health domain) where the opinions are expressed on by representing the semantics of a sentence and modelling the roles played by its elements with respect to a model of opinion sentences.
- Defining a new Semantic Role Labelling exploiting new lexical and semantic resources in personal health context.
- Build a new emotion detection system focused in personal health on top of SENTILO, a Sentiment Analysis system that we recently built, http://wit.istc.cnr.it/stlab-tools/sentilo/service, that performs sentence-based sentiment analysis and relies on FRED, a machine reader for the Semantic Web. FRED relies on Combinatory Categorical Grammar, Discourse Representation Theory, Frame Semantics (e.g. Semantic Role Labelling), and Ontology Design Patterns and performs Named Entity Resolution, Coreference Resolution, Word Sense Disambiguation.
- Creation of new lexical resources for Sentiment Analysis (e.g. annotated dataset, word embeddings within the health domain, etc.)
- Provision of REST services for the provided Emotion detection system.
- Degree in Computer Science, Information Engineering (or equivalent).
- A degree with distinction (cum laude) is an advantage;
- Prior knowledge in Machine Learning and Deep Learning is an advantage;
- Prior publications at international conferences or journals are desirable;
- Ability to program in Python is an advantage;
- Communication skill and team play are desirable.
Principal Investigator: Prof. Diego Reforgiato Recupero, (UNICA)
Academic PhD Supervisor: Diego Reforgiato Recupero (UNICA)
Academic PhD co-Supervisor: Daniele Riboni (UNICA)
Industrial PhD Supervisor: Rim Helaoui (PHILIPS)
Industrial PhD co-Supervisor: Aki Härmä (PHILIPS)
Main contact: Prof. Diego Reforgiato Recupero