Code for the AAAI paper: Linguistic Properties Matter for Implicit Discourse Relation Recognition: Combining Semantic Interaction, Topic Continuity and Attribution.

Note that the code use python2.

sklearn
nltk

You may first need to configure the training data path: -train_path and testing data path: -test_path before you run this code. Here is an example: python autoNLP.py -evaluate_relation [the discourse relation you want to predict] -train_path [path/to/your/training/data] -test_path [path/to/your/testing/data] -log_path [path/to/save/logs] -confusionM_path [path/to/save/confusion/matrix]

For example, if you do fourway prediction, you will see a output like this:

First you will see the features used by this prediction. ['FourWayExpansion', 'FourWayContingency', 'FourWayComparison', 'FourWayTemporal', 'arg1NegPure+Arg1SubjRepeat_Arg2andAttr2Rela', .... ]

Next the output will show the most informative features. The most efficient features are displayed in reverse order. -----------------------Most informative Features--------------------- Most Informative Features none-arg2NegPure+Arg2PrediRepeat_Arg1andAttr1Rela = None Compar : Tempor = 16.6 : 1.0 arg2NegPure+Arg2PrediRepeat_Arg1andAttr1Rela = 1 Compar : Tempor = 16.6 : 1.0

Final you will see the prediction result: P: 0.548148, R: 0.542125, F: 0.545120 P, R, F represent precision, recall and F1 score respectively.

This project is licensed under the MIT License - see the LICENSE.md file for details