Improve the return on your marketing investments with intelligent user segmentation in Amazon Personalize

Today, were excited to reveal smart user segmentation powered by machine learning (ML) in Amazon Personalize, a brand-new way to deliver customized experiences to your users and run more effective campaigns through your marketing channels.
Traditionally, user segmentation depends on group or psychographic info to arrange users into predefined audiences. These methods rely on assumptions about the users choices and objectives that limit their scalability, do not automatically find out from altering user habits, and do not provide user experiences individualized for each user. User division in Amazon Personalize utilizes ML strategies, improved and developed at Amazon, to learn what is appropriate to users.
Amazon Personalize allows developers to build personalized user experiences with the very same ML innovation utilized by Amazon with no ML expertise needed. You can begin developing user sectors quickly with the Amazon Personalize API or AWS Management Console and only pay for what you utilize, with no minimum costs or upfront dedications.
This post walks you through how to use Amazon Personalize to segment your users based on choices for grocery products utilizing an Amazon Prime Pantry dataset.
Summary of solution
Were presenting 2 brand-new recipes that section your users based on their interest in various item categories, brands, and more. Our item affinity dish (aws-item-affinity) recognizes users based on their interest in the individual items in your brochure, such as a motion picture, item, or song. The product quality affinity recipe (aws-item-attribute) determines users based upon the characteristics of items in your brochure, such as genre or brand. This allows you to much better engage users with your marketing campaigns and enhance retention through targeted messaging.
The notebook that accompanies this post demonstrates how to use the aws-item-affinity and aws-item-attribute dish to produce user segments based upon their choices for grocery products in an Amazon Prime Pantry dataset. We use one dataset group which contains user-item interaction data and product metadata. We utilize these datasets to train solutions utilizing the 2 dishes and produce user segments in batch.
To evaluate the performance of the option, we split the interactions information into a training set and test set. The Amazon Prime Pantry dataset has approximately 18 years of interaction data from August 9, 2000, to October 5, 2018, with roughly 1.7 million interactions. We hold out 5% of the most current interactions and train on the staying 95%. This results in a split where we use interactions from August 9, 2000, through February 1, 2018, to train the option and use the remaining 8 months of interactions to simulate future activity as ground reality.
Outcomes
When reproducing these tests in the notebook, your outcomes may vary somewhat. This is since when training, the solution the specifications of the underlying designs are arbitrarily initialized.
Lets very first evaluation the outcomes by taking a look at a couple of examples. We ran queries on 3 products, and recognized 10 users that have a high tendency to engage with the products. We then look at the users shopping histories to assess if they would likely be interested in the queried item.
The following table reveals the results of a segmentation inquiry on gingerbread coffee, an item we might desire to promote for the holiday season. Each row in the table reveals the last three purchases of the 10 users returned from the question. The majority of the users we identified are plainly coffee drinkers, having recently acquired coffee and coffee creamers. Surprisingly, the item we queried on is a whole bean coffee, not a ground coffee. We see in the product histories that, where the details is available, the users have actually recently acquired entire bean coffee.

Gingerbread Coffee, 1 pound Whole Bean FlavorSeal Vacuum Bag: Bite into a freshly baked Gingerbread Coffee

A2XERDJ6I2K38U.
Egyptian Gold Luster Dust.
Kelloggs Rice Krispies Treats.
Wilton Decorator Preferred Green Fondant.

A2NLJJVA0IEK2S.
Coffee Masters Flavored Coffee.
Lays 15pk Hickory Sticks Original (47g/ 1.6 oz per.
Albanese Confectionery Sugar Free Gummy Bears.

A385P0YAW6U5J3.
Tinksky Wedding Cake Topper God Gave Me You Sparkl.
Sweet Sixteen Cake Topper 16th Birthday Cake Toppe.
Capturing the Big One DecoSet Cake DecorationReel i.

A1474SH2RB49MP.
Various Snowflake Sugar Decorations Disney Movie.
Darice VL3L Mirror Acrylic Initial Letter Cake Top.
Edible Snowflakes Sugar Decorations (15 pc).

AOZ0D3AGVROT5.
Sea Green Disco Glitter Dust.
Christmas Green Disco Glitter Dust.
Child Green Disco Glitter Dust.

A3RLEN577P4E3M.
The Republic Of Tea.
Alyssas Gluten Free Oatmeal Cookies– Pack of 4.
Double Honey Filled Candies.

A1MHK19QSCV8SY.
Hoosier Hill Farm Prague Powder No. 1 Pink Curing S.
APPLE CIDER VINEGAR.
Fleischmanns Instant Dry Yeast 1lb bagDry Yeast.M.

A2IPDJISO5T6AX.
Angel Brand Oyster Sauce.
Bullhead Barbecue Sauce.
ONE ORGANIC Sushi Nori Premium Roasted Organic Sea.

USER_ID.
Last Three Purchases.

Wrights Natural Hickory Seasoning Liquid Smoke, 128 Ounce This seasoning is produced by burning fresh cut hickory chips, then condensing the smoke into a liquid kind.

Active User Baseline.
0.0720.
0.0320.

A1MDO8RZCZ40B0.
Master Chef Ground Coffee.
New England Ground Coffee.
Maxwell House Wake Up Roast Medium Coffee.

APHFL4MDJRGWB.
Dunkin Donuts Original Blend Ground Coffee.
Coffee-Mate Coffee Mix.
Folgers Gourmet Selections Coconut Cream Pie Flavo.

A3DCP979LU7CTE.
DecoPac Heading for The Green DecoSet Cake TopperL.
Rhinestne Cake Topper Number 90This delicate and h.
Rhinestone Cake Topper Letter KThis fragile and h.

A1GDEQIGFPRBNO.
Christopher Bean Coffee Flavored Ground Coffee.
Camerons French Vanilla Almond Whole Bean Coffee.
Camerons Coffee Roasted Whole Bean Coffee.

A2LK2DENORQI8S.
The Bean Coffee Company Organic Holiday Bean (Vani.
Lola Savannah Angel Dust Ground.
New England Coffee Blueberry Cobbler.

ALXKY9T83C4Z6.
Heart Language of Love Bride and Groom White Weddi.
Happiness Cake Topper by Lenox (836473 )Its a present tha.
First Dance Bride and Groom Wedding Cake TopperRom.

ANX42D33MNOVP.
The Coffee Fool Fools House American.
Don Franciscos Hawaiian Hazelnut.
Don Franciscos French Roast Coffee.

AC7O52PQ4HPYR.
Rhinestone Cake Topper Number 7 by otherThis delic.
Rhinestone Cake Topper Number 5This fragile and h.
Rhinestone Cake Topper Number 8 by otherThis delic.

Although these outcomes are informative, theyre not an ideal reflection of the efficiency of the dish because the user segmentation wasnt utilized to promote the items which users later on communicated with. The finest method to measure efficiency is an online A/B test– running a marketing campaign on a list of users stemmed from the aws-item-affinity option along with a set of the most active users to measure the distinction in engagement.
Conclusion.
Amazon Personalize now makes it easy to run more smart user division at scale, without needing to keep intricate sets of rules or depending on broad presumptions about the preferences of your users. This allows you to better engage users with your marketing campaigns and enhance retention through targeted messaging.
For more information about Amazon Personalize, check out the item page.

To do this broader assessment, we run the aws-item-affinity option on 500 randomly picked items that appear in the test set to query a list of 2,262 users (1% of the users in the dataset). We then use the test set to evaluate how frequently the 2,262 users bought the items during the test period.

A3NDGGX7CWV8RT.
Frontier Mustard Seed.
Da Bomb Ghost Pepper HOT SaucesWe infused our hot.
Starwest Botanicals Organic Rosemary Leaf Whole.

ANEDXRFDZDL18.
Pepperidge Farm Goldfish Crackers.
Boston Baked Beans (1) 5.3 Oz Theater Box Sizecont.
Increase Simply Complete Nutritional Drink.

A1H3ATRIQ098I7
Brew La Red Velvet Cupcake Coffee
Olas Exotic Super Premium Coffee Organic Uganda B.
Coffee Masters Gourmet Coffee.

A2WW9T8EEI8NU4.
Hidden Valley Dips Mix Creamy Dill.9 oz Packets (.
Frontier Garlic Powder.
Wolf Chili Without Beans.

Test Metrics.

Letter C– Swarovski Crystal Monogram Wedding Cake Topper Letter, Jazz up your cakes with a shimmering monogram from our Sparkling collection! These single letter monograms are silver plated covered in crystal rhinestones and can be found in a number of sizes for your benefit.

A3QW120I2BY1MU.
Goldas Kitchen Acetate Cake Collars– 4.
Twinings of London English Breakfast Tea K-Cups fo.
Chefmaster by United States Cake Supply 9-Ounce Airbrush Clea.

A3F7NO1Q3RQ9Y0.
Yankee Traders Brand Whole Allspice.
Aji No Moto Ajinomoto Monosodium Glutamate Umami S.
Hoosier Hill Farm Prague Powder No. 1 Pink Curing S.

Customize– Item Affinity.
0.2880.
0.1297.

Hits.
Remember.

USER_ID.
Last Three Purchases.

A2TEJ1S0SK7ZT.
Black Tai Salt Cos– (Food Grade) Himalayan Cryst.
Marukan Genuine Brewed Rice Vinegar Unseasoned.
Cheddar Cheese Powder.

A24E9YGY3V94N8.
TOOGOO( R) Double-Heart Cake Topper Decoration for.
Custom-made Personalized Mr Mrs Wedding Cake Topper Wit.
Jacobs Twiglets 6 Pack Jacobs Twiglets are one of.

The next table shows a segmentation inquiry on hickory liquid smoke, a seasoning used for grilling and treating bacon. We see a number of different grocery items that might accompany barbecue in the users recent purchases: barbecue sauces, spices, and hot sauce. Two of the users recently bought Prague Powder No. 1 Pink Curing Salt, a product likewise utilized for treating bacon. We might have missed these 2 users if we had relied on rules to identify people interested in grilling.

Our third example shows a segmentation inquiry on a decor used to leading cakes. We see that the users identified are not just bakers, but are also clearly interested in decorating their baked products. We see current purchases like other cake toppers, edible decorations, and fondant (an icing used to sculpt cakes).

A2U77Z3Z7DC9T9.
Food to Live Yellow Mustard Seeds (Kosher) 5 Pound.
100 Sheets (6.7 oz) Dried Kelp Seaweed Nori Raw Uns.
SB Oriental Hot Mustard Powder.

A3MPY3AGRMPCZL.
Wrights Natural Hickory Seasoning Liquid Smoke.
San Francisco Bay OneCup Fog Chaser (120 Count) Si.
Kikkoman Soy Sauce.

A13YHYM6FA6VJO.
Lola Savannah Triple Vanilla Whole Bean.
Lola Savannah Vanilla Cinnamon Pecan Whole Bean.
Pecan Maple Nut.

A3JKI7AWYSTILO.
Lalahs Heated Indian Curry Powder 3 Lb LargeCurry.
Ducal Beans Black Beans with Cheese.
Emerald Nuts Whole Cashews.

Traditionally, user division depends on group or psychographic details to sort users into predefined audiences. These strategies rely on presumptions about the users choices and intents that restrict their scalability, dont instantly find out from changing user habits, and dont provide user experiences individualized for each user. User division in Amazon Personalize utilizes ML strategies, developed and refined at Amazon, to learn what is appropriate to users. Most of the users we identified are plainly coffee drinkers, having just recently acquired coffee and coffee creamers. To do this broader assessment, we run the aws-item-affinity option on 500 randomly picked products that appear in the test set to query a list of 2,262 users (1% of the users in the dataset).

USER_ID
Last Three Purchases

About the Authors.
Daniel Foley is a Senior Product Manager for Amazon Personalize. He is concentrated on building applications that take advantage of synthetic intelligence to resolve our customers largest challenges. Beyond work, Dan is an avid skier and hiker.
Debarshi Raha is a Senior Software Engineer for Amazon Personalize. He is enthusiastic about developing AI-based customization systems at scale. In his extra time, he enjoys taking a trip and photography.
Ge Liu is an Applied Scientist at AWS AI Labs dealing with developing next generation recommender system for Amazon Personalize. Her research study interests consist of Recommender System, Deep Learning, and Reinforcement Learning.
Haizhou Fu is a senior software application engineer on the Amazon Personalize group working on building and developing suggestion systems and solutions for various markets. Beyond his work, he enjoys playing soccer, basketball and enjoying films, reading and learning more about physics, especially theories related to time and area.
Yifei Ma is a Senior Applied Scientist at AWS AI Labs working on recommender systems. His research study interests depend on modeling and choice making in large-scale temporal domains, utilizing tools in causal analysis, support knowing, distributed deep learning, approximate inference, and uncertainty-driven expedition.

AGW1F5N8HV3AS.
New England Coffee Colombian.
Kirkland Signature chicken breast.
Lola Savannah Banana Nut Whole Bean.

A3G5P0SU1AW2DO.
Wrights Natural Hickory Seasoning Liquid Smoke.
8 OClock Whole Bean Coffee.
Cooking Area Bouquet Browning and Seasoning Sauce.

Leave a Reply

Your email address will not be published.