README.md

Improving Multi-Label Emotion Classification (MLEC) via Sentiment Classication (SC)

Author

Jianfei YU

[email protected]

Dec 26, 2018

Data and Code for:

Improving Multi-Label Emotion Classification via Sentiment Classification with Dual Attention Transfer Network EMNLP 2018 http://aclweb.org/anthology/D18-1137

(Here the labels for MLEC contain 11 emotions, which are respectively anger, anticipation, disgust, fear, joy, love, optimism, pessimism, sadness, surprise, trust; and the labels for SC are respectively neutral, positive and negative)

This repository contains the following components:

• [Data Preprocessor for changing the data format of SC to be consistent with MLEC] at ec_preprocess_4_model.py
• [Data Preprocessor for creating the pickle files for both MLEC and SC] at ec_data_build_kl_stdtok_py3.py
• [Base Model without SC] at ec_main.py (https://github.com/jefferyYu/emnlp18_codes/tree/master/ec_lstm_kl_att_topic_other/)
• [Proposed Dual Attention Model with SC] at ec_main.py (https://github.com/jefferyYu/emnlp18_codes/tree/master/ec_lstm_kl_att_topic_other+sentiment_dasts2/)

Steps to run the codes:

• python ec_preprocess_4_model.py (https://github.com/jefferyYu/emnlp18_codes/tree/master/ec_lstm_kl_att_topic_other+sentiment_dasts2/)
• python ec_data_build_kl_stdtok_py3.py (https://github.com/jefferyYu/emnlp18_codes/tree/master/ec_lstm_kl_att_topic_other+sentiment_dasts2/)
• python ec_main.py (https://github.com/jefferyYu/emnlp18_codes/tree/master/ec_lstm_kl_att_topic_other/)
• python ec_main.py (https://github.com/jefferyYu/emnlp18_codes/tree/master/ec_lstm_kl_att_topic_other+sentiment_dasts2/)

Data for MLEC and SC

• [Data for MLEC] (https://github.com/jefferyYu/emnlp18_codes/tree/master/data/) train, dev and test files from SemEval 2018 Task 1C.
• [Data for SC] (https://github.com/jefferyYu/emnlp18_codes/tree/master/data_twitter/) merge the train, dev and test files from SemEval 2016 Task 4A together to have the 'twitter-2016_all.txt' file.

Note that since the format of [Data for SC] is different from [Data for MLEC], we need to run this as shown above - [Data Preprocessor for changing the data format of SC to be consistent with MLEC].

Requirements

• Python 3.x
• TensorFlow
• Scikit-Learn
• Numpy
• ekphrasis
• ftfy
• ujson
• termcolor

Running Examples and Results

• [Log files] (https://github.com/jefferyYu/emnlp18_codes/tree/master/log_files/) To show the runing procedure and results, I also attach the log files of running our preprocessing, base model and proposed model codes under this folder.