Content-based recommendations

We first start with content-based recommendations, as they are the most similar to what we previously discussed in this book. The term content refers to the usage of only an item's or user's content information in the shape of a (numeric) feature vector. The way to arrive at a feature vector from an item (an article in a web shop) or a user (a browser session in a web service) is through data mining, data pre-processing and feature engineering—skills you learned in Chapter 4, ETL, data preparation, and feature extraction, and Chapter 6, Advanced feature extraction with NLP.

Using users' and items' feature vectors, we can divide content-based recommendations into roughly two approaches:

• Item-item similarity
• User-user similarity

Hence, recommendations are based on the similarity of items or the similarity of users. Both approaches work great in cases where little to no interaction data between user and items...