STM

Overview

Implementation of the model described in the ACL'19 short paper:

• Interpretive Spatio-Temporal Features for Multi-TurnResponses Selection by Junyu Lu, Chenbin Zhang, Zeying Xie, Guang Ling, Chao Zhou, Zenglin Xu

Files in this repository:

• dataset folder: the dataset.
• tmp folder: path to save log and models.
• cache folder: path to save pre-processed data while running at the first time.
• model.py folder: code of STM.
• main.py: code entrance.
• processor.py: code for processing the Ubunutu Corpus and the Advising Dataset that release by DSTC7 Track1.
• STMDataLoader.py: dataloader.

Dataset

dataset/ubuntu_train_subtask_1.json, dataset/ubuntu_dev_subtask_1.json and dataset/ubuntu_test_subtask_1.json are the training, development and test sets, respectively. The format of them is as follows:

[
  {
        "data-split": "dev",
        "example-id": 1100001,
        "messages-so-far": [
            {
                "speaker": "participant_1",
                "utterance": "hey guys, does your livecd have chroot installed? and bash?"
            },
            {
                "speaker": "participant_2",
                "utterance": "sure"
            },
            {
                "speaker": "participant_1",
                "utterance": "does it have everything I need to format a partition ext2?. and ext3?"
            },
            {
                "speaker": "participant_2",
                "utterance": "yep"
            },
            {
                "speaker": "participant_1",
                "utterance": "yay I can use it to install gentoo. !"
            },
            {
                "speaker": "participant_2",
                "utterance": "lol. LOL"
            },
            {
                "speaker": "participant_1",
                "utterance": "=-). brb rebooting into ubuntu"
            },
            {
                "speaker": "participant_2",
                "utterance": "form last week:. 04:21:47] <findme> this is a big crowd here. [04:21:53] <findme> have all gentoo users moved here ?"
            },
            {
                "speaker": "participant_1",
                "utterance": "to bad its still using apt I would switch in a heart beat if it had its own package manager"
            }
        ],
       "options-for-correct-answers": [
            {
                "candidate-id": "TLSHF16Y4J4L",
                "utterance": "what are you missing in apt ?"
            }
        ],
        "options-for-next": [
            {
                "candidate-id": "YWOA49156J9P",
                "utterance": "issues with msn?. I'm experiencing them on windows atm, current msn version"
            },
            {
                "candidate-id": "RYBI7QRD9QZN",
                "utterance": "<> AmaroqWolf: alias='sudo admincommand'.  <AmaroqWolf>  aw, can't make myself type sudo? I like it better that way."
            },
            {
                "candidate-id": "TLSHF16Y4J4L",
                "utterance": "what are you missing in apt ?"
            },
            ...
       ],
       ...
]

Usage

1. Download the subtask1 dataset from the website and put it into dataset/
2. Train a model using default hyperparameters (Excuting at the first time may need a little time to prepare the data and save in cache/).

python main.py --save_path 'tmp/' --encoder_type 'GRU' --pretrain_embedding glove.42B.300d.txt

3. Evaluated results are saved to tmp/eval_result.txt and tmp/test_result.txt respectively.

Note: You can change hyperparameters in main.py

Citation

@inproceedings{lu-etal-2019-constructing,
    title = "Constructing Interpretive Spatio-Temporal Features for Multi-Turn Responses Selection",
    author = "Lu, Junyu  and
      Zhang, Chenbin  and
      Xie, Zeying  and
      Ling, Guang  and
      Zhou, Tom Chao  and
      Xu, Zenglin",
    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/P19-1006",
    pages = "44--50",
    abstract = "Response selection plays an important role in fully automated dialogue systems. Given the dialogue context, the goal of response selection is to identify the best-matched next utterance (i.e., response) from multiple candidates. Despite the efforts of many previous useful models, this task remains challenging due to the huge semantic gap and also the large size of candidate set. To address these issues, we propose a Spatio-Temporal Matching network (STM) for response selection. In detail, soft alignment is first used to obtain the local relevance between the context and the response. And then, we construct spatio-temporal features by aggregating attention images in time dimension and make use of 3D convolution and pooling operations to extract matching information. Evaluation on two large-scale multi-turn response selection tasks has demonstrated that our proposed model significantly outperforms the state-of-the-art model. Particularly, visualization analysis shows that the spatio-temporal features enables matching information in segment pairs and time sequences, and have good interpretability for multi-turn text matching.",
}