Recommendation Systems & Text Analytics in Excel

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.

• Read Bramble Books' ratings matrix like a recommender does: user and book averages, how empty the matrix is, and how many books two readers have in common.
• Measure how similar every Bramble Books reader is to Ana with cosine similarity and find her closest "taste neighbour".
• Predict how Ana would rate two books she has not read, from the ratings of similar readers, first as a weighted average, then mean-centred.
• Build the "customers who bought Salt & Ember also bought …" box from purchase data, and see why cosine beats a raw co-purchase share.
• Check on eight held-out ratings whether collaborative filtering really beats the simplest possible recommender: the book's average rating.
• Turn six Bramble Books reviews into numbers: how often each word appears, how long each review is, and in how many reviews each word shows up.
• Weight the words of five reviews with TF-IDF and find the term that characterises each review best, not just the most frequent one.
• Score eight Bramble Books reviews with a small positive/negative word list, label them with IFS and check the labels against a human.