eCommerce and AI

 Some good examples of typical uses in the space.

How eCommerce leaders use AI to woo their customers with Personalization

Posted by Kateryna Reshetilo   in DSC

Once upon a time, customers simply walked into stores and met a friendly shopkeeper who helped them out. No one hears you walking the halls of those buildings... not anymore. The world shifted to online shopping.    Unprecedented growth occurred in the eCommerce sector amid the COVID-19 crisis. And Amazon has spoiled us to the point that we expect extreme personalization. If we don’t get it, we just go elsewhere.  

If you don't have a personalization strategy, you are losing 75% of shoppers who leave your website because they get frustrated by irrelevant offers. Website personalization helps to make the purchasing process faster, get better deals, and curb information overload. Before we dive in, let’s make sure we’re all on the same page.

What is personalization?

Personalization is the process of creating customized, pleasing experiences for visitors. Every customer sees a slightly different version of your website which is defined by the particular audience segment the shopper falls into. These segments are based on:

        Past purchase history and average order value 

        Demographics (age, ethnicity, location, income level)

        Psychographics (interests, habits, emotions, attitudes)

        Visitor type (new visitors or returning visitors)

        Traffic source (referral sites, social media, direct or from paid link ads)

        Site engagement (browsing time or number of pages viewed)

        Current cart profile  .... 

When a store suggests products that supplement the items in your cart, that’s personalization. It allows you to create unique shopping experiences tailored to each particular user. 

7 brilliant eCommerce personalization examples to boost sales

Ever hopped on a website and thought: “That’s exactly what I was looking for!” There’s a good chance that the retailer knew who you were, why you were visiting, and what you were seeking. Let’s take a look at eCommerce personalization done well. .... '

Product Recommendation using a Knowledge Graph

 An example of an interaction between a knowledge graph and a specific task. Have used Neo4j Graphs, so this was instructive. A number of reco examples are shown at the link, beyond this intro.

Amazon Product Recommendation using Neo4j Knowledge Graph

Leveraging knowledge graph capabilities in building a product recommendation system

By Shyam Pratap Singh  in   TowardsdataScience

As we see all around, our choices are driven by recommendations. Our decision about online shopping or going out for dinner or looking for college or watching movies on Netflix, we often need recommendations from our close circles and, based on that, we make the best decisions possible.

Building a product recommendation system is not something new, but leveraging a knowledge graph to achieve this is exciting and amazing. We are going to use Neo4j Graph Database to analyze 100,000 Amazon product records.

This is the satellite view of our galaxy, just kidding :P. Our Amazon product dataset in the graph

So, before talking about graphs, let’s understand a bit about a recommendation system.

Recommendation System:

A recommender system seeks to predict the “rating” or “preference” a user would give to an item. It helps the user to make the best choices.

Recommender systems are used in a variety of areas, with commonly recognized examples taking the form of playlist generators for video and music services, product recommenders for online stores, or content recommenders for social media platforms and open web content recommenders. These systems can operate using a single input, like music, or multiple inputs within and across platforms like news, books, and search queries.

Recommender systems are a useful alternative to search algorithms since they help users discover items they might not have found otherwise..... '

On Recommendation Systems

 Anther well considered piece, on recommendation systems, link to original post includes more useful references.  Note the inclusion of elements of 'trust'.  Intro below. 

The Increasing Influence of Recommendation Systems in Our Everyday Lives by Irving Wladlensky-Berger in his blog ... 

A few years ago, I attended a seminar by University of Toronto professor Avi Goldfarb on the economic value of AI. Goldfarb explained that the best way to assess the impact of a new radical technology is to look at how the technology reduces the cost of a widely used function. Computers, for example, are powerful calculators whose cost of arithmetic and other digital operations have dramatically decreased over the past several decades. As a result, we’ve learned to define all kinds of tasks in terms of digital operations, e.g., financial transactions, inventory management, word processing, photography. Similarly, the Internet and World Wide Web have drastically reduced the cost of communications and of access to all kinds of information, - including numbers, text, pictures, music and videos.

Viewed through this lens, the data and AI revolution can be viewed as reducing the cost of predictions. Predictions mean anticipating what is likely to happen in the future. Over the past decade, increasingly powerful computers, advanced machine learning algorithms, and the explosive growth of big data have enabled us to extract insights from the data and turn them into valuable predictions. As was previously the case with digital operations, communications and access to information, - we’re now able to reframe all kinds of applications as prediction problems. A major such family of applications are recommendation engines or recommender systems, which Wikipedia defines as “a subclass of information filtering system that seeks to predict the ‘rating’ or ‘preference’ a user would give to an item.”  ... '

ACM Talk on Recommender Systems

Of interest, the topic is broadly addressed:

Register now for the next free ACM Webinar, "Recommender Systems: Beyond Machine Learning" presented on Tuesday, October 8, at 4pm ET by Joseph A. Konstan, Distinguished McKnight University Professor and Distinguished University Teaching Professor at the University of Minnesota. Bart Knijenburg, Assistant Professor at the Clemson University School of Computing, will moderate the questions and answers session. Continue the discussion on ACM's Discourse Page. You can view our entire archive of past ACM TechTalks on demand at https://learning.acm.org/techtalks-archive.

Technology of Choice and Defining Selections

Though I have never worked in the apparel space.  I have worked in spaces where consumers make many choices in context, and companies aim to insert new choices to maximize engagement, while strengthening demand by marketing influence.

I happened on this article in Stitch Fix,  an online personal styling service in the United States, talking about their use of technology.

Some fascinating things here. both decision oriented and mathematically defined.  Algorithms of choice.  Useful beyond the realm of apparel selection?  I think so:

In their Multithreaded Blog:

WELCOME TO Stitch Fix  (If you read the whole thing at the link below the math is covered)

We are reinventing the retail industry through innovative technology.

Simulacra and Selection: Styling at Stitch Fix ... 

Modern retailers aid and influence customer decisions, using techniques like recommender systems and market basket analysis to deliver personalized and contextual item suggestions. While such methods typically just augment a traditional browsing experience, Stitch Fix goes a step further by exclusively delivering curated selections of items, via algorithmically-assisted stylist recommendations1.

For most of the history of Stitch Fix, stylists have worked with a styling platform that functions in a fairly straightforward way. For simplicity and concreteness, imagine an e-commerce platform with various filters, wherein stylists are able to browse for clothing items and add them to a cart. The role of our styling algorithm in this system is to rank the items based on information the client provides us, and the role of the stylist is to select an assortment of items that a client will love.

I’m going to gloss over all the details of the styling algorithm except for one important and nuanced point: it is trained to estimate the probability that a particular client will like a given item if a stylist decides to send it. This is a natural and useful framing: we observe what happens to items that stylists select, but not the counterfactual for items they don’t select. However, this selection bias comes with some occasionally perplexing caveats. 

One observation of long-standing Stitch Fix lore is the “shorts in winter” problem: the styling algorithm tends to assign a high score to shorts during the dead of winter. Do we somehow think that everyone wants to stock up on shorts despite the chilly weather? Of course not: this only means that shorts are likely to be successful if a stylist chooses to send them, which they won’t do without a very good reason—e.g., a client requesting them for a tropical cruise. This is an amusing example but the problem is broader: stylists need to spend a decent fraction of their time browsing through items that, for one reason or another, are clearly a poor match. .... " 

McDonald's Acquires Dynamic Yield for Decision Logic

Companies leveraging advanced analytic methods for their data by acquisition.  Have recently stood in front of McDonald's store kiosk and after looking for the coffee I wanted, wondered  what recommendation it should suggest, and what data and context it had at hand to make that recommendation.

McDonald's Bites on BigData with $300 Million Acquisition By Brian Barrett in Wired  

Mention McDonald’s to someone today, and they're more likely to think about Big Mac than Big Data. But that could soon change: The fast-food giant has embraced machine learning, in a fittingly super-sized way.

McDonald’s is set to announce that it has reached an agreement to acquire Dynamic Yield, a startup based in Tel Aviv that provides retailers with algorithmically driven "decision logic" technology. When you add an item to an online shopping cart, it’s the tech that nudges you about what other customers bought as well. Dynamic Yield reportedly had been recently valued in the hundreds of millions of dollars; people familiar with the details of the McDonald’s offer put it at over $300 million. That would make it the company's largest purchase since it acquired Boston Market in 1999.  ... " 

and also:

McDonald’s to use A.I. to tempt you into extra purchases at the drive-thru  By Trevor Mogg in DigitalTrends   ...  "

Alternatives to Quantum Recommendation

In our early look at Quantum Computing we tried to make these comparisons, and and found little then.  Is this fair?   Combinatorially difficult.   Following.  I note that the title of the article claims 'Quantum-Inspiration' in the classical algorithim, forgotten by most all of the commenters, would like to follow that thread.   Could such inspiration take us further?

Major Quantum Computing Advance Made Obsolete by Teenager  In Quanta Mag

18-year-old Ewin Tang has proven that classical computers can solve the “recommendation problem” nearly as fast as quantum computers. The result eliminates one of the best examples of quantum speedup.

 A teenager from Texas has taken quantum computing down a notch. In a paper posted online earlier this month, 18-year-old Ewin Tang proved that ordinary computers can solve an important computing problem with performance potentially comparable to that of a quantum computer.

In its most practical form, the “recommendation problem” relates to how services like Amazon and Netflix determine which products you might like to try. Computer scientists had considered it to be one of the best examples of a problem that’s exponentially faster to solve on quantum computers — making it an important validation of the power of these futuristic machines. Now Tang has stripped that validation away.

“This was one of the most definitive examples of a quantum speedup, and it’s no longer there,” said Tang, who graduated from the University of Texas, Austin, in spring and will begin a Ph.D. at the University of Washington in the fall.   .... " 

Full paper: https://arxiv.org/abs/1807.04271     Considerable discussion.

Who Builds Big Recommendation Engines

Breadth and particular participants were surprising.   How these operate essentially acts as the 'algorithm' for suggestion based upon data.   How well they work, and how this is perceived by the customer, is a key aspect of their operation.

From Wharton Customer Analytics Institute on Linkedin

E-commerce and Retailer are Adopting Recommendation Engine: Amazon and IBM to Generate More Revenue

Download PDF Brochure@ https://tinyurl.com/y8bzx2pn

Recommendation engines use a variety of technologies and techniques that enable them to filter large amounts of data and provide a smaller, focused body of suggestions for the user. 

Recommendation engines are common among e-commerce, social media and content-based websites. Amazon was one of the first sites to use a recommendation system. When the company was essentially an online book store, it began using software to suggest books the user might be interested in, based on data gathered about their previous activity, as well as the activity of other users who made similar choices.

The major vendors in the global recommendation engine, are IBM (US), SAP (Germany), Salesforce (US), HPE (US), Oracle (US), Google (US), Microsoft (US), Intel (US), AWS (US), and Sentient Technologies (US). New product launches, and product enhancements and partnerships are the key growth strategies adopted by these market players to offer feature-rich products and services to their customers.. ... "

AI Based Recommendation Engines

Makes much sense to broaden the approach.  Link to PDF.

Recommendation Engine Based on AI: High Revenue Margin by 2022
 By Akshita Banker in Linkedin

 With the help of Recommendation engine more than 100 business logic and personalization configurations make best offers to their customers. Recommendation engine are common among e-commerce, social media and content-based websites. It Analyses customer behavior, order history, and like-minded shopper intent. This information enables e-retailers to offer the right products to the right customers and at the appropriate time. ... " 

(Updated) Graph Databases for Dynamic Recommender Systems and Outlier Detection

Interesting ideas  Slides and recording now at link below:

Talk by: Denis Vrdoljak & Gunnar Kleemann  at Cisco & UC Berkeley

“Graph Databases for Dynamic Recommender Systems and Outlier Detection”

Abstract: At Berkeley Data Science Group (BDSG), we set out to commercialize Data Science projects and ideas developed at UC Berkeley. From our first project (predicting biotech IPO’s) to our most recent work (AI assisted copywriting and NLP), we’ve developed several tools and techniques using Graph Databases to solve several of the challenges encountered. In this presentation, we’ll share some of the lessons we’ve learned and some of the techniques we’ve used.

Denis Vrdoljak: Denis is a Data Scientist/Marketing Analyst at Cisco and also a co-founder of Berkeley Data Science Group. His current work and research focus is in predicting edges in network graphs for recommender systems, and in using graph topologies to identify outliers for security.

Gunnar Kleemann, Ph.D.: Gunnar is a Data Science Instructor at UC Berkeley and co-founder of Berkeley Data Science Group. His current research focus and interest lies in analyzing collaboration networks of research scientists to identify patterns in community topologies and to identify outliers. He is currently analyzing geographic and domains.
--------------------

Slides and Recording are here: http://cognitive-science.info/community/weekly-update/

Join the CSIG LinkedIn Group to get reminders about talks and discuss them. Use twitter: #CSIGNews & #OpenTechAI

Replays before Dec 2015:  Dial 877.471.6587 or 402.970.2667 and enter the call’s Replay ID when prompted for a program ID number.   The Replay ID is listed in the Recording column of each date.

Better Recommendation Engines

Researchers Devise Better Recommendation Algorithm 
Improved recommendation algorithm should work especially well when ratings data are “sparse.”

MIT News  By Larry Hardesty

Researchers at the Massachusetts Institute of Technology (MIT) have developed a new recommendation algorithm based on a theoretical analytic framework using cosine similarity, which they say should work better than current algorithms. The researchers note the algorithm should be especially effective when ratings data is "sparse." Sparse data means there may be so little overlap between users' ratings that cosine similarity is rendered meaningless, making it necessary to aggregate the data of many users. MIT professor Devavrat Shah says the framework assumes the relative weight a user assigns to ratings remains the same, and each user's function is running on the same set of features. Shah notes this yields sufficient consistency to extrapolate statistical inferences about the probability that one user's ratings will predict another's. The team used the framework to demonstrate that, in instances of sparse data, their "neighbor's-neighbor" algorithm should return more accurate predictions than any known algorithm.... " 

AI Fashion Designer

Note the use of a classic recommendation approach.  Does this embody creativity as well?  Would like to see this compared direct to human design and measures like sales.

Amazon Has Developed an AI Fashion Designer 
in Technology Review  by  Will Knight 

Amazon researchers are working on machine-learning algorithms for analyzing, predicting, and leveraging trends in fashion design, with one Israel-based team having developed software that can infer how stylish a particular look is by analyzing only a few labels attached to images. Meanwhile, Amazon's Lab126 group has developed an algorithm that gleans a specific fashion style from example images so it can produce new items in similar styles from scratch, via a generative adversarial network. 

A recent Amazon-hosted workshop on machine learning and fashion highlighted research with broader applications to human behavior, including a project conducted by Cornell University professor Kavita Bala and colleagues. Bala says her team is using data collected from Instagram "to understand how people live their daily lives." In addition, a team from the University of Illinois in Urbana-Champaign developed an algorithm for identifying fashion-focused social-network accounts, and Myntra unveiled software for guessing a person's correct apparel size based on past purchases.  ... " 

Google and Walmart Team Up

New multi retailer giant teaming proposed.

Watch out Amazon: Google and Walmart form e-commerce alliance to battle online retail giant  By Nat Levy

" ... Starting in late September, customers will be able to make Walmart purchases through Google Express, an online marketplace where customers can purchase and set up delivery from stores like Target and Walgreens. Walmart items will also be available through the Google Assistant and Google Home smart speaker, which competes with Amazon’s Echo, the market leader for voice-controlled devices. Walmart shoppers can link their customer accounts to Google and get personalized recommendations. ... " 

Evolution of Amazon Recommender Systems

Building a store for each consumer.  And how those consumers evolve over time?  Instructive piece. What is the driving force for change?  In IEEE Computer Society:

Two Decades of Recommender Systems at Amazon.com
Brent Smith, Amazon.com,  Greg Linden, Microsoft

Pages: 12–18

Abstract—Amazon is well-known for personalization and recommendations, which help customers discover items they might otherwise not have found. In this update to our original paper, we discuss some of the changes as Amazon has grown.

Keywords—Internet/Web technologies; recommender systems; intelligent systems; artificial intelligence; software engineering; item-based collaborative filtering; personalization  ..... "

Recommender Engine in Cortana

Good instructive example for Cortana.  How easy is this for maintaining the recommendations?   I like to show these kinds of examples because they include simple business process.  Which is forgotten these days when mentioning AI.

Cortana Intelligence and Machine Learning Blog
Using a recommendation system in an application
by Ankit Asthana, Principal PM VS.NET at Microsoft.

Recommendation systems are extremely popular today and are used everywhere, to predict music you’d like, products to buy, and movies to see! In this post, we would like to show you how you can build a movie recommendation engine. The post will describe how to build this model in Azure Machine Learning Studio. It comes with an end-end sample that walks you through the process of building a movie recommendation service that you can incorporate into your application today. .... "

AI Driven Clothes Shopping

I think I have seen at least three recommender-engine driven apparel selection systems pitched to me, none have survived.  This is a tough sell.  Touch and 3D would help.  Another mirror like idea utilizing machine vision.   Will AI help this?  I do hope so, the fashion industry could also use the data involved, have seen that from students in my classes.  In Fastcompany, with considerable detail:

The Robots Are Coming For Your Wardrobe

The husband and wife duo behind Chennai-born AI startup Mad Street Den are determined to change the way you shop for clothes.    By Jackie Snow.

Artificial intelligence is rapidly changing just about every industry, with billions being poured into algorithms for everything from self-driving cars to detecting cancer to chatbots. Not so much fashion.

Mad Street Den, a three-year-old artificial intelligence startup founded by a husband and wife team in Chennai, India, is boldly going where few startups have gone before. The couple’s complementary career paths have helped. CEO Ashwini Asokan was previously with Intel Labs working on product design; her husband and CTO, Anand Chandrasekaran, is a neuroscientist who switched over to AI when he realized the technology wasn’t quite there yet to build silicon brains. ... " 

Product Recommender in Python

More Python example than recommender.  But also a useful view of programming in this domain. Via O'Reilly.

Recommender Systems Beyond Matrix Completion

Recommender Systems: Beyond Matrix Completion/Abstract
By Dietmar Jannach, Paul Resnick, Alexander Tuzhilin, Markus Zanker
Communications of the ACM, Vol. 59 No. 11, Pages 94-102
10.1145/2891406

Abstract: The use of recommender systems has exploded over the last decade, making personalized recommendations ubiquitous online. Most of the major companies, including Google, Facebook, Twitter, LinkedIn, Netflix, Amazon, Microsoft, Yahoo!, eBay, Pandora, Spotify, and many others use recommender systems (RS) within their services.... 

These systems are used to recommend a whole range of items, including consumer products, movies, songs, friends, news articles, restaurants and various others. Recommender systems constitute a mission-critical technology in several companies. For example, Netflix reports that at least 75% of its downloads and rentals come from their RS, thus making it of strategic importance to the company. ... " 

Requires registration to see full aricle.

Building a Reco Engine with a Graph Database

Building a recommendation Engine ... Powering recommendations with a Graph Database.  I like the idea of decision process starting with a visualization.   " You May also like …" ... Product recommendations help businesses maximize their online revenue. It requires advanced technology, but this is now available off-the-shelf, and is already being used by Wal-Mart and other market leaders.

“You may also like” is a deceptively simple phrase that encapsulates a new era in customer relationship management. In offering the advice, businesses maximize the value they deliver by providing highly targeted, real-time product recommendations to their online consumers.

Big Data in Recommendation Systems

A good overview in KDNuggets.  Example with Amazon books. When you think of it, much of our purchase interaction online is influenced by such systems.  It is important to make the premise for the reco clear to the user.  Our early pump recommendation expert system was a form of this method, but with much less data, and a clear reason for selection.  Also a way to map markets of need.  Look also at Procurement systems.