NLP_Gutenberg_Book_Classification

Problem Overview

In this task, we are using semantically-close books from the Gutenberg project and we aim to classify text segments to its corresponding book name.

Libraries and Dependencies

The versions are the defaults set by colaB

• Python
• NLTK
• SKlearn
• Eli5
• Lime
• Matplotlib
• Jupyter/spider/colaB

Output Example:

An example for the output of eli5 for the top 10 words of 5 books:

Steps:

1. We start by 5 books that are semantically close to each other.
2. Extract 200 samples from each book, each sample comprises 100 words.
3. Data preprocessing is performed on these segments:

• Tokenization
• Punctuation and stop words Removal
• Lowercasing
• Lemmatization

4. Feature Engineering on the clean data from (3):

• Bag of Words
• TF-IDF

5. Splitting the data into train/test splits (80/20) and 10 fold cross validation.
6. Modelling for TF-IDF features:

• Decision Tree
• KNN
• SVM
• Logistic Regression

7. Evaluation: Accuracy, Bias-Variance tradeoff
8. Error Analysis for Misclassified segments:

• eli5
• Lime

9. Insights, Analysis then modify some hyperparameters (e.g. number of words per segment) and retrain.