Repository files navigation

• Our goal is to determine what class(es) the customer feedback sentences should be annotated with five-plus-one-classes categorization (comment, request, bug, complaint, meaningless and undetermined) as in four languages i.e. English, French, Japanese and Spanish.
• This is one of the shared tasks ofIJCNLP - 2017. For more details about the task, please visithere.

If you are using this code for any sort of research, please cite our paper

• IITP at IJCNLP-2017 Task 4: Auto Analysis of Customer Feedback using CNN and GRU Network.

• Command :python3 train.py training.tsv parameters.json
• A directory will be created during training, and the best model will be saved in this directory.

• Provide the model directory (created when runningtrain.py) and test data topredict.py
• Command :python3 predict.py trained_model_1505467324/ test.tsv

• Command :python3 train.py training.tsv training_config.json
• A directory will be created during training, and the best model will be saved in this directory.

• Provide the model directory (created when runningtrain.py) and test data topredict.py
• Command :python3 predict.py trained_results_1505468375/ test.tsv

• For any queries, please drop me an email atpabitra.lenka18@gmail.com.
• Please refer to the publication for detailed results and model performances.

• I would like to thankJie Zhang andDenny Britz for sharing their code.
• We have used their code and modified according to our need by incorporating pre-trainedWord2Vec embedding.
• Deepak Gupta has also contributed to this code repository.