bannersHon
  EN | FR |
img
About HON
img PATIENT / INDIVIDUAL img MEDICAL PROFESSIONAL img WEB PUBLISHER img
img
 
img
   

img
Health On the Net celebrates its 20th anniversary
img
img
img
img
About the
organization
 Mission & users 
 Who we are 
 Accredited by HAS 
 NGO status 
 Awards & Recognition 
 HON History  
img
Research &
Development
 In progress 
 Completed 
    European
       WRAPIN
       Active Health
       PIPS
    International
       Google Co-op
       Joslin Dubai
       iphone HealthPedia
    National
       MARVIN
       Hunt Services
    Internal
       HONcode
       HONselect
       Orphanet.ch
       Quality Detector
    Project for
       developing countries
       RUIG
       RAFT
 Publications 
img
International
collaboration
 French-speaking Africa 
 English-speaking Africa 
 Spanish-speaking
   countries
 
 Arabic-speaking countries 
img
 Media Corner 
 Collaborations 
 Helping HON and Funding 
 Jobs offers  
 Ethical policy 
 Contacts 
 

Donate: You can support our work by clicking on the button below:


Quality Detector
img



In the health information field, quality appears to be a high priority matter. The increasing volume of online health information covers a large specter of health related subjects, but most of the time the quality of health web content is uneven. There is no identification of doubtful sources. Existent methods consist of either, criteria evaluation or quality detection related to one specific theme. To guarantee the quality of all online health information, itís important to be able to evaluate not only the production process of a website, but also the content of any web page related to any health subject. In this study we attempt to classify health web pages and determine whether they have a high quality level or not. We use a learning machine to detect low and high quality web pages.


Here you will find the results obteained with the system. We have chosen the SVM (Support Vector Machine) classifier. Indeed, its provides better results than the NaiveBayes one. Then, a cross-validation has been realized. The average of our results is represented in the table below:

maR

maP

miR

miP

Erreur

0.979

0.985

0.982

0.982

0.017

 

Legend
maR / miR : macro Recall / micro Recall
maP / miP : macro Precision / micro Precision

We obtained a macro recall of 0.979 and a micro recall of 0.982 . This means that the system is able to return more than 97% of the reliable documents for all categories and over 98% of relevant documents mixed all together. For accuracy (precision), we obtained a value greater than 0.98 (micro and macro values). This means that over 98% of the documents returned by the system are properly classified . It is important to note that only 1.7% of documents are bad categorized by the system. To conclude, the system provides good results. This suggests its integration into a user interface for use in production.

 


Home img About us img MediaCorner img HON newsletter img Site map img Ethical policies img Contact