Amazon Personalize is a Machine Learning (ML) service by AWS designed to simplify the process of building in personalization capabilities to recommend products and content on eCommerce platforms and websites. Customizing customer experiences based on browsing behavior, purchase history, demographics, psychographics and other data can deliver more engaging experiences to customers, exposing them selectively to content and products that fall in their scope of interest. Not only does this increase the value the customer derives from their browsing experience, it can also result in higher conversions, sales and value generation for companies.
A plethora of different data sets and personalization characteristics help in developing a comprehensive identity for customers which can inform personalization algorithms. Developing a rich machine learning model with as many such characteristics as possible, can prove to be incredibly difficult. Amazon Personalize completely simplifies this process, allowing developers with no ML experience to utilize the same level of technology as is utilized at Amazon.com to enhance their eCommerce platforms.
Why do online platforms need personalization capabilities?
Personalization helps customers discover products and content that can meet their needs and interests. It motivates engagement on digital properties, as customers see more of what they want and like. This growth in engagement can be observed through measurable improvements in metrics such as Click-Through-Rates (CTR), watch durations on videos, dwell times on articles and bounce rates. On top of all this, personalization is proven to improve conversions, resulting in growth in revenues or other desired outcomes.
The massive business potential observed in personalization motivated foresighted companies such as Amazon to invest and capitalize on the technology very early on. Amazon uses personalization intensively across their suite of platforms ranging from the Amazon.com website to Amazon Prime Video, Amazon Kindle, Amazon Music and more. Today, the algorithms have become significantly advanced, seen in widgets that recommend based on browsing history or from categories that are of high interest to the customer.
A common approach to starting personalization is with a rule based system that is designed with pre-existing ideas or trends in mind, for eg. a retailer recommending boots to women who visit their website from New York at the start of winter. This system makes sense, however, it misses out on capturing the diversity of needs across people as user numbers and catalog sizes grow. Hence using machine learning should prove a better solution, producing a comprehensive algorithm more catered to individual level recommendations. Working with ML however, comes along with some challenges:
- Recommendation systems should respond to the actions and intentions of a user in real time and they should be able to effectively handle new users and new items in the catalog – a common challenge for ML systems.
- Recommendations should not have biases for popular goods, pushing the most relevant product or content based on the needs of the customer.
- Algorithms are not one size fits all, requiring customizations for different use cases.
- Building good personalization models is very hard, requiring a high level of ML expertise.
These complexities push people away from using ML, choosing to go with rule based systems or no personalization at all. Using a rule based system may come with lesser complications, however they present poor performance, poor scalability and a high cost and effort of maintenance. Those that choose to tackle machine learning tools for better results can also find them hard to build and manage, in addition to their limitations in matching recommendations with customer intent and managing real-time personalization for new customers.
Reducing the barriers to entry – Amazon Personalize
Amazon Personalize is a solution for individualization woes in the form of a fully managed service for generating personalized recommendations. The service uses ML & deep learning technologies to generate accurate & relevant recommendations in addition to reducing the time it takes to get started from a typical period of 6 months or more to just a few weeks. It builds custom and private ML models using your accumulated data and best of all, it requires no ML expertise to use.
How it Works
Amazon Personalize consists of three components:
- Amazon Personalize – used to create, manage and deploy solution versions.
- Amazon Personalize events – used to record user events for training data.
- Amazon Personalize Runtime – used to get recommendations from a campaign.
To make recommendations, Amazon Personalize uses a machine learning model that is trained with your data which is stored in related datasets in a dataset group. Each model is trained using a recipe containing an algorithm for a specific use case. A solution version, the name for a trained model in Amazon Personalize, is deployed for use as a campaign. People using your applications receive recommendations based on the deployed solution versions.
Data such as a user’s activity stream on your platform, inventory metadata and user metadata are input into Amazon Personalize, which then processes and outputs a customized personalization & recommendation API. Fully managed by AWS, Amazon Personalize inspects the data, identifies unique features, selects hyperparameters, trains and optimizes models and performs a host of functions on the backend before you receive recommendations.
Setting Up an Amazon Personalize Workflow
The process of working with Amazon Personalize starts with creating related datasets and a data set group. Training data consisting of historical data and live recorded event data is then input into the data set group. The next step is to create a solution version using a recipe or AutoML and evaluate it using metrics. AutoML is the process of allowing Amazon Personalize to automatically choose a recipe that it feels is best for your use case. Finally, the campaign is deployed and users on the platform begin receiving recommendations.
Preparing & Importing Data
To train models, Amazon Personalize uses data provided from source files (historical) or live recorded data such as activity on a website. To provide a source file to import data in Amazon Personalize, you must:
- Format in a comma-separated values (CSV) file
- Provide a schema to guide Amazon Personalize to import the data correctly
- Upload the file into an Amazon S3 bucket that Amazon Personalize can access
Three types of historical datasets are recognized by Amazon Personalize. Each type has an associated schema with a name key that matches value with the dataset type. These include:
- Users – contains metadata about your users
- Items – contains item metadata
- Interactions – contains information of historical interaction data between users and items, metadata on browsing context, location, device, etc.
Only certain recipes use the Users and Items datasets which are also known as metadata types.
Prior to adding datasets to Amazon Personalize, a schema must be defined for that dataset. Schemas in AWS are defined in the Avro format. There are precise guidelines to defining dataset schemas and sample resources to reference for creating each schema type can be found in the AWS console. An example of a schema for User data can be seen below:
For a more detailed look at specific versions of schemas for interactions or items, watch our webinar.
Amazon Personalize can utilize real time user event data and process it individually or combined with historical data to produce more accurate and relevant recommendations. Unlike historical data, new recorded data is used automatically when getting recommendations. Minimum requirements for new user data are:
- 1,000 records of combined interaction data
- 25 unique users with a minimum of 2 interactions each
Solutions & Recipes
A solutions version is a trained machine learning model that is ready to make recommendations to customers. Historical and live data available in interactions datasets are processed to train the model which is done with the help of a recipe. Recipes consist of algorithms and data processing steps that optimize a solution for the type of recommendation being developed. You can choose what recipe is used to train the model. If utilizing AutoML, Amazon Personalize can automatically choose appropriate recipes based on an analysis of training data.
Hyperparameters & Hyperparameter Optimization
Hyperparameters are utilized to optimize training models before training begins. Different recipes utilize different hyperparameters and it is essential to perform Hyperparameter optimization (HPO) or tuning, to ensure the right hyperparameter is picked for a specific learning objective. Optimal hyperparameters are found by running training jobs using different values from the available specified range. HPO is not a default Amazon Personalize process.
Evaluating a Solution Version and Creating Campaigns
Amazon Personalize generates a slew of metrics when creating solution versions, which allow you to evaluate the performance prior to creating campaigns and providing recommendations. These metrics can be compared between solutions that use the same training data with different recipes or solutions with modified hyperparameters, to determine the best option. The solution versions producing the best results in metrics are typically used for deployment as campaigns which are either created in the console or by calling the CreateCampaign API. Once deployed, these campaigns can finally be used to make recommendations.
Campaigns can be used to produce two types of recommendations in Amazon Personalize.
Real-time recommendations are used to enrich individual user experiences. These can be obtained by calling the GetRecommendations API and supplying the user ID or item ID, depending on the recipe.
Batch recommendations are useful for large datasets that do not need real time updates. For example, getting product recommendations for all the users on an email list. These can be obtained by calling the CreateBatchInferenceJob API.
Both the above types of recommendation results are returned as JSON files to an S3 bucket.
Enriching Customer Experiences
Using the steps detailed above, you can utilize Amazon Personalize to deliver high quality recommendations to your customers for almost any product or content and adapt to their evolving needs and intents in real time. Training new models require just a few clicks and can be deployed with the specificity you desire for recommendation algorithms.
For a demo and a more detailed look at Amazon Personalize, watch our webinar on Customizing the Customer Journey with Amazon Personalize. If you’re interested in utilizing Amazon Personalize for your eCommerce or online platform, get in touch with our experts today!