ECO

This repository releases the necessary code and the evaluation data of ECO method proposed at: “Focused Query Expansion with Entity Cores for Patient-Centric Health Search” research paper. 

ECO is a novel method based on query expansion that generates and executes patient-centric focused queries. It is designed to overcome the topical drift present in this kind of queries by computing entity cores that identify the most relevant and coherent terms for focused expansion.  
 

ECO aims to answer patient-centric information needs consisting of

1. An informational search query: a short phrase or few keywords, e.g., mentioning a disease or side effect

2. A patient-specific case description: free text about the user's individual health situation and anamnesis.
 

An example of such query coming from forum.alzheimers.org.uk  is:

Subforum: Alzheimer

Post Title: Aricept has made my mum worse 

Post: My mum was diagnosed with early stages Alzheimers at the end of June but she has been suffering loss of memory for probably 3 years but as usual we put it down to 'old age'. My dad is 85 and in wonderful health, mum is 76. She had her first appointment with the local memory clinic and they prescribed 5mg Aricept. Day 1 she was fine, no side effects; day 2 she started feeling sick and then was very ill for the best part of 24 hours. She also hallucinated did not recognise her home, my dad, saw my gran (deceased) etc. We stopped the medication immediately, called the memory clinic (who didn't call back) so we have been trying to cope, for the first time, with my mum in a horrible state of confusion and anger, where she thinks we are poisoning her, trying to steal her money, not let her go home... the paranoia is endless. Can anyone please tell me how this has happened?
 

The pipeline to generate and execute such a query using ECO approach would be:

  1. Perform Named Entity Recognition and Disambiguation (NERD) to extract all medical entities present the patient query. To perform NERD we have used the method proposed in http://resources.mpi-inf.mpg.de/d5/bebe/ repository. 
  2. After extracting all medical entities, construct the Query-Graph (QG) which a directed and labelled graph with medical entities found in the patient query as nodes and all relational statements that exist between these entities in the Knowlife Knowledge Graph (http://knowlife.mpi-inf.mpg.de/). 
  3. Extract the Entity-Core (EC) out of the QG. 
  4. Expand the initial query expressed in the Post Title with all the entities that are part of the EC. 
  5. Execute the expanded query over the full textual corpus. 

Detailed information on how to run the ECO framework can be found here.

All the evaluation data gathered from the crowdsourcing experiment presented in the paper can be found here.