ECIR 2020 – The role of knowledge in supporting IR applications

There was one very clear takeaway from ECIR 2020, a theme that to me kept occurring throughout many of the presentations at ECIR. This was the role of explicit knowledge in support applications, in particular the role of ontologies. There are many search technologies that can benefit from ontologies e.g. entity linking and recognition where searches can be broadened out or narrowed down to the user needs for better search performance. However it is specific domains where the benefit can be mostly usefully accrued.

The first of these is health and related life science fields. The short paper section has a theme on medical IR with 8 papers in it, whilst the CLEF evaluation session had a number of papers on forum for investigating issues in health and related fields for a variety of different media. For example ImageCLEF specifies a task on extracting conditions from images e.g. finding tumours in X-Ray images. Having an ontology and being able to link to such conditions could clearly help a clinician identify particular health problems. Semantic indexing with the help of ontologies can help summarisation, question and answering as well as IR. Having a chemical ontology can help entity recognition from chemistry articles and identify specific issues such as type and characteristic of a chemical compound as well as events associated with such a compound. Clinical reports may have discrepancies and errors in them – ontologies could be used to identify and classify these discrepancies and rank them according for a clinician to act on i.e. by the level of impact or risk to the patient.

Medical Short paper track playlist

CLEF session recording

The Legal domain is also one in which ontologies could be used to support search applications. Pauline Chavallard of Doctrine gave a very interesting talk in the Industry day track on search in French law. She gave an overview of the scheme to build a knowledge graph to enhance document understanding and to help provide clarity to user query intent. The knowledge graph is in fact a ontology, with relations establish between courts and decisions, decisions with topics, commentary and legislation, and decision with entities (such as plaintiffs, lawyers and judges). In this way documents can be enriched e.g. associating decision with legislation and commentaries with decisions, allowing a much broader view of the impact of legislation for the searcher. Pauline felt that more work needs to be carried out in this area, and is not as developed as for example ontologies in health (see above). There is a clear need for work in this area, in particular tacking account of different legal jurisdictions e.g. common law systems prevalent in the anglosphere verses civil law inherent in many mainland European countries.

The talk is available here.

Leave a comment

Your email address will not be published. Required fields are marked *