The project involves classifying user questions into one of ten predefined categories using machine learning, with a focus on Natural Language Generation (NLG). You are working with a dataset of 2169 labeled questions, evenly distributed among 10 classes (labels 0–9).
- Format: CSV with two columns:
text(question),label(category). - Size: 2169 questions.
- Goal: Predict the correct label for a new question based on its text.
You explored various ML classification models and chose Logistic Regression with TF-IDF vectorization due to:
- Dataset size (small)
- Balanced classes
- Simplicity and interpretability
- Good performance for multi-class text classification
- Preprocessing: Handled missing values and vectorized text using TF-IDF with
max_features=5000. - Model: Trained a Logistic Regression model (
max_iter=1000). - Evaluation: Achieved 91% accuracy on the test set.
- Additional insights: Used a confusion matrix and classification report for evaluation.
This model could be used in real-world applications like:
- Chatbots: Automatically classifying incoming user queries into relevant categories.
- Customer support: Routing queries to the right department.
- Helpdesk systems: Automating question triage.
- Collect more and better-balanced training data.
- Experiment with advanced NLP techniques: n-grams, lemmatization, word embeddings.
- Try deep learning models like BERT for richer semantic understanding.
- Use ensemble models to reduce misclassification.
- Apply data augmentation to boost underrepresented labels.