Domain Classification for Haptik Dataset

• The dataset contains two separate files for training and testing/evaluation of data.
• Each human query/message is categorized into single/multiple domains by human annotators.
  • T: message belongs to the corresponding class
  • F: message does not belong to the corresponding class (column heading represents the class name)

Haptik Domain Classification

• Task: train a multi-label classifier to classify "message" into one or more classes/domains
• Suggested evaluation metrics
  • Precision
  • Recall
  • Accuracy per class
  • Overall Accuracy

NOTE:

• Please note that the messages are in raw format, it also contains system specific content which may not be important for identifying domain.
• It is important to preprocess messages and remove such information for better training.

The Setting

Here is a nice function that will fetch the dataset and format it in the correct way for further processing:

import pandas as pd
import numpy as np

def fetch_grid():
    train = pd.read_csv('./data/domain_classification/train_data.csv')
    test = pd.read_csv('./data/domain_classification/test_data.csv')
    return train, test

train_haptik, test_haptik = fetch_grid()
train_haptik.head()
test_haptik.head()

The column "message" contains the text to be analysed and the following columns contain the information about which category the text falls in.

Task 1: Constructing the Datasets

Write a fucntion feature_target_separator() with following parameter:

• df: non-separated dataset

The function splits the dataset into the features and target dataframes

Task 2: Reversing One-Hot

It can be seen that the target variables are One-hot-encoded.

Write a function reverse_OHE() with follwing parameters

• df: target dataframe

And returns labeled pandas Series

Task 3: Word Cloud

Word cloud is a technique that helps us visualise raw data

Write a function word_cloud() with follwing parameters

• X: pandas series containing text data to be plotted
• max_font_size: maximum size of the font desired in the word cloud

And returns the word cloud.

The function should perform following tasks:

• tokenization
• removal of stop words
• Stemming

Task 4: Classification

Write a function classification() with follwing parameters

• X_train: 1 column pandas dataframe with text
• X_test: 1 column pandas dataframe with text
• y_train: 1 column pandas dataframe with labels
• y_test: 1 column pandas dataframe with labels
• tokenizer: string
• stop_words: string
• ngram_range: tuple with (min_val, max_val) format
• max_df: ignore terms that appear in more than given fraction of the documents[0,1]
• min_df: only keep terms that appear in at least n documents (int)

The function uses MultinomialNB classifier and should return accuracy and confusion matrics