Introduction:

This is the implementation in PyTorch for my ACL2019 paper:

"Joint Effects of Context and User History for Predicting Online Conversation Re-entries"

Requirement:

• Python: 2.7+

• Pytorch: 0.4+

• Sklearn: 0.20.0

Before running:

You need to download Glove pre-training embeddings from: https://nlp.stanford.edu/projects/glove/

"glove.twitter.27B.200d.txt" for twitter.data.

"glove.6B.200d.txt" for reddit.data.

Usage:

python train.py [filename] [modelname]

[filename]: "twitter.data" or "reddit.data".
[modelname]: "AVG", "CNN", "LSTM", "LSTMMerge", "LSTMAtt", "LSTMMEM", "LSTMBiA".

optional arguments:
  --cuda_dev          choose to use which GPU (default: "0")
  --max_word_num      max word number in vocabulary, -1 means all words appear in dataset (default: -1)
  --entrytime         given how many entries as context  (default: 1)
  --train_pc          percentage of conversations for training (default: 0.8)
  --embedding_dim     dimension for word embedding (default: 200)
  --hidden_dim        dimension for hidden states (default: 200)
  --kernal_num        number of kernals for CNN encoder (default: 50)
  --hop_num           number of hops for MEM network (default: 3)
  --batch_size        batch size during training (default: 8)
  --max_epoch         maximum iteration times (default: 200)
  --history_size      maximum number of messages for user history (default:50)
  --lr                learning rate during training (default: 0.001)
  --dropout           dropout rate (default: 0.2)
  --num_layer         number of layers for LSTM (default: 1)
  --model_num_layer   number of layers for LSTM in bi-attention mechanism (default: 2)
  --need_history      whether current model needs history (action="store_true")
  --bi_direction      whether change LSTM to BiLSTM (action="store_true")
  --runtime           record the current running time (default: 0)
  --train_weight      weights in loss function during training (default: 2)

Datasets:

format in each line:

[Conv ID] \t [Msg ID] \t [Parent ID] \t [Original sentence] \t [words after preprocessing] \t [User ID] \t [posting time]

(twitter dataset doesn't have time infos, but the conversations are ordered by posting time)

Citation:

@inproceedings{zeng-etal-2019-joint,
    title = "Joint Effects of Context and User History for Predicting Online Conversation Re-entries",
    author = "Zeng, Xingshan  and
      Li, Jing  and
      Wang, Lu  and
      Wong, Kam-Fai",
    booktitle = "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics",
    month = jul,
    year = "2019",
    address = "Florence, Italy",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/P19-1270",
    pages = "2809--2818",
}