Leveraging on a Recommender to Boost your e-Commerce

The Problem 

If you already own an e-commerce storefront like Amazon.com or Cold Stone Creamery, then you probably recognize the utmost importance of having a sound recommendation system built into your e-commerce site—so that when customers make purchases, they are provided with additional purchase suggestions or offers based on their past selections. A good recommender almost always help sales growth. One problem is that very often, e-commerce business owners do not have in-house data scientists or product developers. On the other hand, these entrepreneurs may not have a business large enough to justify outsourcing the development project for a recommendation system. On rare occasions, the owner himself might be a data scientist or a keen programmer, but that does not guarantee that he will possess the right insights to develop an effective recommender.

So how can this common and growing problem be solved?

Two educated minds collaborate to solve the problem

The e-book titled Practical Machine Learning Innovations in Recommendation published by O’Reilly upholds some tested programming practices that can kick start any average product-development team to build reasonably effective recommendation systems for e-commerce. Co-authored by Ted Dunning and Ellen Friedman, this book discusses extensive practical techniques for developing recommenders in the Apache Hadoop environment. 

Dunning, Chief Applications Architect at MapR Technologies, happens to be a PMC member of the Apache Mahout, Apache Zoo Keeper, and Apache Drill projects, and is also the mentor for Apache Storm. Ellen Friedman is a consultant and commentator with experience writing about big data. With Ted’s technical background and Ellen’s Bioscience research and publications background— they have jointly created a mini bible on building recommenders with Hadoop.

As you may have noticed from your own online shopping experiences that an appealing recommendation system not only helps you make buying choices, but they may actually enhance the seller’s business image in your eyes. In a digital age, a tool like a product or service recommender is an implicit marketing tool that does its job silently.

This book indicates that strangely enough, the mathematical algorithms form a small part of the total development effort. The main thrust behind building a good recommendation system comes from providing the right data to the recommendation engine!

How to build an effective recommender 

If you can afford it, an easier way to install a recommender is by outsourcing the project to a reputed machine-learning consultancy company. Many consultants can supply quick and effective simulations along with recommenders. The strategy they usually employ is

throwing a huge collection of algorithms at each problem, and—based on extensive experience in analyzing such situations—selecting the algorithm that gives the best outcome.

As the Hadoop technology platform continues to evolve, big data projects will gradually become more cost-friendly.

The theory of co-occurrence

Without going into unnecessary technical details, it is worth mentioning that central to the theme of creating an effective recommender model is the theory of co-occurrence. This theory relies on capturing substantial user histories to study purchase-behavior patterns. The book provides an in-depth discussion on how co-occurrence works with simulations.

Practical Machine Learning Innovations in Recommendation contains the following chapters:

Chapter 1: Practical Machine Learning

Chapter 2: Careful Simplification

Chapter 3: What I Do, Not What I Say

Chapter 4: Co-occurrence and Recommendation

Chapter 5: Deploy the Recommender

Chapter 6: Example: Music Recommender

Chapter 7: Making It Better

Chapter 8: Lessons Learned

Appendix A: Additional Resources

If you have decided to go in-house

You have a small, development team in place and are willing to try building a system in-house, provided you get reliable guidance—then this book is for you!  The book models a tutorial approach—taking readers through a step-by-step process from beginning to end of building a recommendation system. The early chapters introduce simple, understandable concepts, then slowly progress towards more difficult concepts of making a system work. Chapter 3 presents an excellent overview of how to select user input data for the predictive analysis part of the project. Some nifty methods for collecting user behavior data has been provided in this chapter. Chapters 7 and 8 are the troubleshooting chapters of the book —they skillfully navigate the readers into systematic, product improvement techniques. The book can serve as a constant reference or guide while constructing a developmental model. Chapter 6 showcases an actual working system, which can serve as an inspiration to hesitant innovators! The appendix provides some highly acclaimed publications that can certainly expand an inexperienced developer’s intellectual horizon.

In closing . . .

If you dearly wish your online business can greatly benefit from a recommender, and you do not have a budget to outsource the developmental effort, then you can think of putting your data experts to work with this book in hand. The simple and lucid writing style of the book can easily entice any reader, whether a developer or not, to at least skim through the pages till the end of the book.