## Name
PICO extractor 

## Application domain
Natural Language Processing, Semantic Web 

## Keywords 
Biomedicine, Scientific Hypotheses, PICO, Information Extraction, Frame Extractor, Knowledge Graphs, RDF 

## Description 
If PICO graphs are to be seen as a representation of a narrative underlying a text, the automatic extraction of PICO graphs from texts can be seen as the extraction of narratives from them. One idea is to do this by starting from frames extracted from the text (using Brussel's frame extraction method, T1.4). We discussed the initial design of a method for updating and using a personal dynamic memory to build up and interpret the information contained in the frames. Concretely, the entities and relations between the entities that are part of the extracted semantic frames are incrementally added to and extended in the personal dynamic memory. At the same time this information is matched to perform the resolution of references to these entities. Based on the gold standard PICO annotations, it can then be learned how (a subset of) the relations contained in the PDM can be mapped to PICO relations based on the entities they contain. 

Katrien and Paul presented an initial idea of how a personal dynamic memory could be build up based on the combination of text analysis and information contained in knowledge graphs (see the Matilda microproject from the Venice meeting). The personal dynamic memory would contain here the referents of the entities evoked in a story, as well as the relations between them. The knowledge graphs are used as background information and can be used to resolve co-references or add additional relations that are not mentioned in the text to the personal dynamic memory. The representations captured in the PDM can then be used for a variety of tasks, including answering questions about the story, summarising the story, providing temporal, causal, etc. perspectives on the story, or explaining based on which prior information specific reasoning steps have been taken.  

In fact, the conceptual foundations and technical architecture that are required for the referent-based extraction of narratives and the automatic creation, extension and annotation of PICO graphs is to a large extent the same. We concluded that the PICO graphs are good to use in a first phase as the domain is more restricted and that gold annotations are abundantly available. In a second phase, we could work towards a new benchmark on the extraction of narratives from stories (T3.1 + T3.4).

This repository contains: 

/resources:
- example systematic reviews from the biomedical domain
- the PICO ontology (semantic script) 
- the annotations of the systematic reviews with the PICO ontology (small narrative graph, in accordance with the semantic script)

Online resources: 

[The PICO ontology documentation](https://linkeddata.cochrane.org/pico-ontology#d4e27)
[The Cochrane Linked Data vocabulary](https://data.cochrane.org/concepts/)


## Ontologies 
[ontology/pico-ontology.ttl](https://data.cochrane.org/pico-ontology.ttl)

## Programming languages
Common Lisp  

## License
Apache License 2.0 

## Contact 
Lise Stork, l.stork@uva.nl

## Contributors
Katrien Beuls, Paul van Eecke, Lise Stork, Annette ten Teije

## Owner 
muhai-project

## Owner type
Organization  

## Repository status 
Concept  

## Acknowledgment
This work was funded by the European MUHAI project (Horizon 2020 research and innovation program) under grant agreement
number 951846. 

---