CrawlFeaturizer

The goal of this sample is to show how to use Personalization API with the crawl featurizer pipeline

To try this sample

• Clone the Azure Personalization Service

git clone https://github.com/Azure/personalization-rl.git

• Navigate to quickstart\CrawlFeaturizer
• Open CrawlFeaturizer.sln

Prerequisites

Set up Personalization Service

• Navigate to Getting Started with Personalization

• Follow the instructions to get the endpoint and key to your Personalization Loop

• Replace ApiKey variable's value in Programs.cs with the loop Key

• Replace ServiceEndpoint variable's value in Programs.cs with the Endpoint url

Set up Cognitive Services Text Analytics

• Navigate to Microsoft Azure Cognitive Services.

• Click on Language APIs tab.

• Click on Get API Key button and select the account you want to sign in with.

• Replace cognitiveTextAnalyticsEndpoint variable's value in Programs.cs with one of the Endpoints

• Replace CognitiveTextAnalyticsAPIKey variable's value in Programs.cs with one of the Keys

Visual Studio

• Set CrawlFeaturizer as the Start Up project in Visual Studio

• Run the project (press F5 key)

Crawl Pipeline

The Crawl pipeline consists of 2 stages

1. Crawl a feed url and get all the items listed in the feed. These items are the Actions that will be ranked by Personalization API. This is exposed through the IActionsProvider interface.
2. Each Action is decorated with Features by using some ActionFeaturizer e.g Cognitive Services Text Analytics, Cognitive Services Vision. This functionality is exposed through the IActionFeaturizer interface. Once we have a set of actions with features, those actions can be ranked using the Personalization API.

Sample Walkthrough

• 6 News Topics and their RSS feed urls are hardcoded in the program

• When the program starts, the RSSFeedActionProvider accesses each RSS Feed and creates a collection of CrawlActions for each of the news articles listed in the feed.

• Next, each CrawlAction is sent to the Cognitive Services Text Analytics endpoint to extract key word phrases and get a sentiment score. These along with the articles news topic and title are used as features for the article

• In the user interaction loop, the user is asked the time of day and location where he/she would be reading the article. All the articles then passed to the Personalization API endpoint for ranking along with the given user context.

• Once ranking is done the top ranked article is displayed to the user asked if he/she would read it.

• The user provided reward value (yes/no) is then passed to the Personalization API endpoint as reward.

• Backend OnlineTrainer learns user preferences by analysing the reward values for the user context and recommended article.

• Overtime the system learns the user's preferences and starts returning very accurate news article recommendations.