Skip to content

A short intent classification project with CLINC 150 dataset using Machine Learning

Notifications You must be signed in to change notification settings

shriadke/Intent_Classification_using_CLINC150_Dataset

Repository files navigation

Intent_Classification_using_CLINC150_Dataset

This project shows various machine learning methods that can be used for the classification of intents. Please make sure that libraries listed in requirements are installed within the working environment.

The data can be downloaded from CLINC150 that contains 10 real-world problem domains having 150 distinct intent classes(with 15 classes in each domain), each with 100 train, 20 validation, 30 test samples. This project uses data_full.json as input.

For this implementation, I have selected 2 distinct intents from each of the 10 domains resulting in 20 distinct intent-classes (The selection is explained in notebooks).

Notebook Intent_classification demonstrates an implementation of 4 basic ML classifiers along with the training and evaluation of the data.

Notebook Intent_classification_with_BiLSTM shows the use of a simple Bidirectional LSTM classifier along with the training and evaluation of the data. This model can be further tuned and performance can be significantly improved. The sample model is uploaded along with this repo.

Overall, The class “IntentClassifier” provided in IntentClassifier.py can be initialized with the json file in given CLINC150 format. It takes several optional arguments such as “classes to consider”, “classifier models”, “train to be performed?”, “models to be evaluated?”. Each of which can be set at the time of instantiation.

To test the class run IntentClassifier.py as:

python3 IntentClassifier.py

About

A short intent classification project with CLINC 150 dataset using Machine Learning

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published