Intent Classification & Slot Tagging

Task Desciription

Intent Classification

• Input: Text
• Output: Intent

Example 1

• Input: i dont like my current insurance plan and and want a new one.
• Output: insurance_change

Example 2

• Input: when will my american express credit card expire
• Output: expiration_date

Example 3

• Input: how would i get to city hall via bus
• Output: directions

Slot Tagging

• Input: Text
• Output: Tags of each word

Example 1

• Input: a table today for myself and 3 others
• Output: O O B-date O B-people I-people I-people O

Example 2

• Input: my three children and i are in the party
• Output: B-people I-people I-people I-people I-people O O O O

Data format

Intent Classification

• id: Unique id
• text: Input sentence
• intent: A string that denotes the intent of the input sentence

Example

{
  "text": "i need you to book me a flight from ft lauderdale to houston on southwest",
  "intent": "book_flight",
  "id": "train-0"
}

Slot Tagging

• id: Unique id
• text: A list of input tokens preprocessed from the input sentence
• tags: A list of strings, each denotes the tag of its corresponding token in the input sentence

Example

{
  "tokens": [
    "i",
    "have",
    "three",
    "people",
    "for",
    "august",
    "seventh"
  ],
  "tags": [
    "O",
    "O",
    "B-people",
    "I-people",
    "O",
    "B-date",
    "O"
  ],
  "id": "train-0"
}

Instructions

Step 1. Download Model

bash download.sh

Step 2. Testing

For testing intent classifiation model:

python test_intent.py

The predictions are saved in pred.intent.csv.

For testing slot tagging model:

python test_slot.py

The predictions are saved in pred.slot.csv.

Training

For training intent classification model:

python train_intent.py

For training slot tagging model:

python train_slot.py