It is a practical recommendation system (retrieval and ranking tasks) using TensorFlow Recommenders and Keras and deploy it using TensorFlow Serving.
A recommendation system is a system that gives a query (context) which is what we know about the liking list, and filters the corpus (full catalog of items) to a shortlist of candidates (items, documents). A query (context) can be a user id, the user's geographical location, or the user's history of previous purchases, and the resulting candidates can be some new items that we guess are interesting for the user.
In practice, dealing with a large corpus and filtering it to a shortlist is an intractable and inefficient task. So practical recommender systems have two (or three) filtering phase
1. Retrieval (Candidate Generation)
2. Ranking (Scoring)
3. Re-ranking or optimazation or ...
A query or a candidate has lots of different features. For example a query can be constructed by these features:
customer_id
customer_id_history
etc.
And a candidate can have features like:
item_description
item_title
item_price
posted_vote
etc.
These obviouse features can be numerical variables, categorical variables, bitmaps or raw texts. However, these low-level features are not enough and we should extract some more abstract latent features from these obvious features to represent the query or the candidate as a numerical high-dimensional vector - known as Embedding Vector.
To involve side-features as well as ids while learning latent features (embeddings), we can use deep neural network (DNN) architectures like softmax or two-tower neural models. YouTube two-tower neural model uses side-features to represent queries and candidates in an abstract high-dimentional embedding vector.
The book dataset is a benchmark dataset in the field of recommender system research containing a set of voting given to book by a set of users, collected from the amazon_us_review website.