Text Classification by Hand: Naive Bayes Spam Filter & Sentiment

What you'll build

What you will be able to do

Each lab ends with a concrete result that you build yourself and the checker confirms, cell by cell.

• Count the spam and ham messages in a labelled training set, turn the counts into prior probabilities and their natural logs, and read the prior log-odds.
• Count how often each of the 8 vocabulary words appears in spam and in ham with one SUMIFS formula, total the words per class and spot the zero-count problem.
• Turn word counts into smoothed likelihoods P(word | class) with add-one (Laplace) smoothing, so no word ever gets a probability of zero.
• Convert likelihoods to logs, score a new message for both classes with one SUMPRODUCT each, and turn the log-odds back into a spam probability.
• Score 12 unseen test messages for both classes, predict Spam or Ham, and count true/false positives and negatives into a confusion matrix.
• Classify 16 messages at two log-odds thresholds, count TP, FP and FN for each, and compare precision, recall and F1 to choose a safer spam threshold.
• Score 12 product reviews with a hand-made sentiment lexicon and with Naive Bayes log-ratios, and compare the accuracy of both against the true labels.
• Compute each word's log-likelihood ratio, rank the vocabulary, pick the strongest spam and ham signals with INDEX/MATCH and chart the ranking.