1. [M2-7] Intention Classifier

2. Package summery

Intention Classifier is a module that analyzes the intention of the user’s utterance. This module modifies and combines “bi-RNN” and “Attention mechanism” to implement an Intention classification model.

• 2.1 Maintainer status: maintained
• 2.2 Maintainer: Wonhyuk Choi, [email protected]
• 2.3 Author: Wonhyuk Choi, [email protected]
• 2.4 License (optional):
• 2.5 Source git: https://github.com/DeepTaskHY/DM_Intent_2

3. Overview

To analyze the intention of the user's utterance, this module consists of two parts: 1)keyword extraction, 2)intention analysis. To extract the keywords of the user's utterance, we used Google Dialogflow. This module use KoBERT(https://github.com/SKTBrain/KoBERT.git) to implement an Intention classification model.

4. Hardware requirements

None

5. Quick start

5.1 Install dependency:

requirements

torch == 1.7.1
mxnet >= 1.4.0
gluonnlp >= 0.6.0
sentencepiece >= 0.1.6
onnxruntime >= 0.3.0
transformers < 4
scikit-learn >= 1.0
google-cloud-dialogflow==2.7.1
https://github.com/SKTBrain/KoBERT/archive/master.zip
https://github.com/DeepTaskHY/ROS/archive/main.zip

5.2 Start the module

• Go to Author's Google Drive and download the JSON file (authorization key), and put it in the authorization folder.

docker environment

$ git clone $rpository_rul
$ docker-compose build
$ docker-compose up dm

local environment(linux)

$ git clone $rpository_url
$ cd $repository
$ sh dm_intent/model_download.sh
$ pip install -r requirements.txt
$ roslaunch dm_intent dm_intent_launcher.launch

6. Input/Subscribed Topics

{  
   "header": {
        "id": 1,
        "source": "perception",
        "target": ["dialog_intent", "planning"],
        "timestamp": "1563980552.933543682",
        "content": ["human_speech"]
   },
   "human_speech":{ 
      "speech": "안녕하세요"
   }
}

○ header (header/recognitionResult): contain information about published time, publisher name, receiver name and content.

• timestamp: published time
• source: publish module name
• target: receive module name
• content: role of this ROS topic name

○ dialog_intent (dialog_intent/recognitionResult): contain human speech and user name.

• speech: human speech
• name: user name
• information: keyword to use for intention classification

7. Output/Published Topics

{
    "header": {
        "id": 1,
        "timestamp": "1563980561.940629720",
        "source": "dialog",
        "target": ["planning"], 
        "content": ["dialog_intent"],
    }, 
    "dialog_intent": {
        "speech": "좋아진 것 같아.", 
        "intent": "단순 정보 전달",
        "information": {}
    }
}

○ header (header/taskExecution): contain information about published time, publisher name, receiver name.

• timestamp: published time
• source: publish module name
• target: receive module name
• content: role of this ROS topic name

○ dialog_generation (dialog_generation/taskExecution): contain generated robot speech sentence, id, result.

• id: id
• dialog: generated robot speech
• result: task progress result

8. Parameters

There are one category of parameters that can be used to configure the module: deep learning model.

8.1 model parameters

• Editing

9. Related Applications (Optional)

• KoBERT. https://github.com/SKTBrain/KoBERT.git

10. Related Publications (Optional)

• Mikolov, T., Chen, K., Corrado, G., & Dean, J. Efficient estimation of word representations in vector space. arXiv:1301.3781. Retrieved from https://arxiv.org/abs/1301.3781 , 2013
• Jeongmin Yoon and Youngjoong Ko. Speech-Act Analysis System Based on Dialogue Level RNNCNN Effective on the Exposure Bias Problem. Journal of KIISE, 45, 9 (2018), 911-917, 2018