PPLRE RR Algorithm - Cooccurrence

From GM-RKB
Jump to: navigation, search

This page describes the configuration of the Coocurrence Relation Recognition Algorithm for its application to the PPLRE Evaluation task. (For information on this algorithm's Performance please see PPLRE Evaluation - Cooccurrence).


Overview

To apply PPLRE Naive Algorithm to the task the following approach was taken:

These configuration decisions are implemented via the configuration file.


v2.3

  • Release Date: 070318

v2.1

  • Release Date: 060511
  • 57 of them.
  • Passages involve two sentences.
  • The pattern is: any two consecutive sentences; with one Organism; one or two Localizations; one, two or three proteins; one or two Spatial terms; and one or two Laboratory_Procedure terms.
  • None of the documents tested overlap with any of the ones in v1.0 of the database.
  • LocalizationID is started at 151. (See also DocumentReference attribute).
  • ProteinID starts at 10000. (See also PredictedProteinName attribute)


  • For Future
    • In the next version we should remove PMID and PMCID from DocumentReference and have them instead looked up in the concordance table
      • I had to match my documents to the latest version of the concordance (which I have also attached)
    • One of the challenges with populating the database (the way that we are now) is that LocalizationID is AutoNumbered. :-/
    • We should increase the corpora otherwise we will run out of predictions.
    • How about proposing that the next version be release in two weeks in order to give us time to take the time to automate the process. There is a lot of time spent on manual work.