Practical Recommender Systems

This book is a real gem. I learn so much from it very quickly. It’s even better that an online courses on the same topic. Highly recommended!

This books give you a complete sense of how design a recommender system. It breaks down the steps and digs into details supported with figures and code! Great job Kim Falk!

I like this book because of the practicality. I bought this book because it teaches you the whole flow of gathering data to all the way to recommending items to users. I wanted a book that paints this picture and it does it for me in a very simple and approachable way.

This book is outstanding, it provides enough theory mixed with real-world examples and code to learn what is necessary to build and curate recommender systems. If you are a beginner, the book will help you learn. If you are an old dog, you can learn new tricks.

Disclaimer: I am just over half way through this book, and I expect to update this review when I have finished reading.This book is divided into two parts. The first part, slightly less than half the book, introduces the topic, describes data collection, data monitoring, personalized vs. non-personalized recommendations, etc. This portion of the book wasn't very helpful for me personally, as I was already familiar with most of this material. However, if you're brand new to the topic, the author did a really fantastic job diving deep into what "kinds" of signals might be collected to build a recommender system. There are plenty of examples describing Amazon, Netflix, and more. The end result is that it should help the reader build a strong, intuitive sense of what kind of data must be collected in order to build a recommender, in addition to learning about instrumentation (collecting metrics) and why it's crucial.Part 2 of the book goes into specific algorithms. There's some mathematical notation involved, but nothing too bad. I've only finished up to chapter 7 (similarities) and 8 (collaborative filtering) so far, though I found the information generally meaningful and easy to digest.At times, the author does a great disservice. If you don't want to cover matrices in detail, that's fine. But to say on page 185: "Matrix is a fancy word for a table with numbers ..." Just.. wow. I understand this isn't the most mathematically rigorous book, but with these statements, the author is egregiously over simplifying important concepts and doing a great disservice to the reader.Later on the same page when briefly mentioning pre-calculating item-item similarities, he states, that its "important when you’re talking about a catalog the size of Amazon (you can think about both the size of the Amazon River or Amazon the internet store!)." That last statement in the parenthesis – the author has a writing style where he keeps padding the length of this book with superfluous statements. When the author finally does get to a formula, he spends little time on the math or providing a detailed technical explanation. The details are hand waved. IMO, without these details, it's unlikely a serious practitioner can make use of the information in this book in a production use case.The early chapters in this book ask the reader to pull down some Python Django application that works around the Movie Geeks dataset. I didn't do any of this work – it's really besides the point.For the author of this book, my question is – WHO is your audience? If it's developers, are you trying to teach them django? Unless the reader is extremely familiar with the framework, it's just more cognitive load for the reader to figure out this tooling that you've unnecessarily imposed. It doesn't make it any easier to learn recommender systems. Any why a full web app? If the expected audience is a developer, you can explain the key concepts and leave all the django and web application portions out. Otherwise, you're just adding noise.If I was the author, I would have provided crisp instructions on setting up a Jupyter notebook, installing Python 3 and any dependencies, and have the reader write all the code in Jupyter. And, use this as an opportunity to teach software developers about data science notebooks, hammer through the idea that building recommenders (or any kind of ML) is an iterative process. Despite how the code is laid out in Chapter 7, realistically, no one is ever going to build a Web front-end that also contains logic for calculating Pearson scores or measures the cosine distance. This logic would most likely be executed in a separate micro service or possibly in some offline/batch job.

It is an excellent book to get started on recommendation systems. A book I wish I had read before building my first recommendation system.Ideas on association rules, weights for implicit ratings and a summary of landscape on recommender systems by big companies was particularly useful.

I can see that it tries its best to avoid any mathematics. But in the same time, the author fails to give intuitive figure on the ideas such as manifold learning (P 286). Instead, the author puts in whole bunch of code which makes the book more like an doc rather than a serious book published. All in all, I can see that the author targets the undergraduate reader and yet fails to use the figure to explain the idea behind machine learning model. My suggestion is that: if you have a graduate level of math and statistics, then avoid this book. If you are an undergraduate who is about join a recsys team as an pure SDE in the company, then you may buy and take a few glimpse.

This books is predicated on using and standing up a web application... For a beginner, the GitHub documentation is bad. I give it 1 out of 5 stars simply because completing Chapter 1 is non-trivial.