Lab 7: CNN Text Classification

DATA 441/641 · Week 7 · PDF

A small 1-D CNN, then parallel kernels, on SMS spam.

Work in a Jupyter notebook. Number the exercises. Do notes 7.1 through 7.3 first. You will need pandas, numpy, and tensorflow (Keras). A laptop CPU is enough if you keep the network small.

Download spam.csv into the same folder as the notebook. Open it with encoding="latin1". v2 is the message; v1 is ham / spam.

When you are done: File → Download as → HTML, then upload the HTML on Canvas. Save the notebook before you export.

import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report
from tensorflow.keras.preprocessing.text import Tokenizer
from tensorflow.keras.preprocessing.sequence import pad_sequences
from tensorflow.keras import Sequential
from tensorflow.keras.layers import Embedding, Conv1D, GlobalMaxPooling1D, Dense, Dropout

If tensorflow is not installed: pip install tensorflow.

1. Load and encode (1 point)

1. Read spam.csv with encoding="latin1". Keep v1 and v2. Drop empty messages.

2. Map ham → 0 and spam → 1. Print the class counts. Spam is the minority class.

3. Split 70% train / 30% test, stratified on the label.

2. Tokens to a matrix (1 point)

1. Fit a Keras Tokenizer on the training messages only (num_words around 3000–5000 is enough).

2. Convert train and test text to integer sequences. Pad / truncate to a fixed maxlen (40–80 is plenty for SMS).

Print vocab size, maxlen, and the shape of the two padded matrices. A row should look like a list of integers, not a TF–IDF vector.

3. A small 1-D CNN (1 point)

Build a Keras model with this spine (you may change the widths):

1. Embedding (50–100 dimensions is enough; you may learn it from scratch).
2. At least one Conv1D (for example 64 filters, kernel size 3).
3. GlobalMaxPooling1D.
4. A Dense hidden layer with dropout if you want it.
5. A single sigmoid unit for binary spam/ham.

Print model.summary(). In a comment: what does the kernel of size 3 see, and what does global max keep?

4. Train and score (1 point)

Train for a few epochs (batch_size 32 or 64). Use validation data or a validation split. Predict on the test set (threshold 0.5 unless you choose another and say why).

Print accuracy, precision, recall, and F1 for the spam class, plus a classification report.

The class split is uneven. A model that always says ham will look accurate and still be useless.

5. Parallel kernels (1 point)

Add a second model that follows note 7.3: two or three Conv1D branches with different kernel sizes (for example 2, 3, and 5), concatenate the pooled vectors, then classify.

You can use the Keras functional API for the branches. Train it the same way. Print the test F1 next to the single-kernel model.

In two or three sentences: did the extra kernel sizes help, and would you keep them on this SMS set?