Scoring profiles are a feature of Microsoft Azure Search that customize the calculation of search scores, influencing how items are ranked in a search results list. You can think of scoring profiles as a way to model relevance, by boosting items that meet predefined criteria. For example, suppose your application is an online hotel reservation site. By boosting the
location field, searches that include a term like Seattle will result in higher scores for items that have Seattle in the
location field. Note that you can have more than one scoring profile, or none at all, if the default scoring is sufficient for your application.
To help you experiment with scoring profiles, you can download a sample application that uses scoring profiles to change the rank order of search results. The sample is a console application – perhaps not very realistic for real-world application development – but useful nonetheless as a learning tool.
The sample application demonstrates scoring behaviors using fictional data, called the
musicstoreindex. The simplicity of the sample app makes it easy to modify scoring profiles and queries, and then see the immediate effects on rank order when the program is executed.
The sample application is written in C# using Visual Studio 2013. Try the free Visual Studio 2013 Express edition if you don't already have a copy of Visual Studio.
You will need an Azure subscription and an Azure Search service to complete the tutorial. See Create a Search service in the portal for help with setting up the service.
- You can open an Azure account for free: You get credits you can use to try out paid Azure services, and even after they're used up you can keep the account and use free Azure services, such as Websites. Your credit card will never be charged, unless you explicitly change your settings and ask to be charged.
- You can activate MSDN subscriber benefits: Your MSDN subscription gives you credits every month that you can use for paid Azure services.
Download the sample application
Go to Azure Search Scoring Profiles Demo on codeplex to download the sample application described in this tutorial.
On the Source Code tab, click Download to get a zip file of the solution.
- After you extract the files, open the solution in Visual Studio to edit the configuration file.
- In Solution Explorer, double-click app.config. This file specifies the service endpoint and an
api-keyused to authenticate the request. You can obtain these values from the Classic Portal.
- Sign in to the Azure Portal.
- Go to the service dashboard for Azure Search.
- Click the Properties tile to copy the service URL
- Click the Keys tile to copy the
When you are finished adding the URL and
api-key to app.config, application settings should look like this:
Explore the application
You're almost ready to build and run the app, but before you do, take a look at the JSON files used to create and populate the index.
Schema.json defines the index, including the scoring profiles that are emphasized in this demo. Notice that the schema defines all of the fields used in the index, including non-searchable fields, such as
margin, that you can use in a scoring profile. Scoring profile syntax is documented in Add a scoring profile to an Azure Search index.
Data1-3.json provides the data, 246 albums across a handful of genres. The data is a combination of actual album and artist information, with fictional fields like
margin used to illustrate search operations. The data files conform to the index and are uploaded to your Azure Search service. After the data is uploaded and indexed, you can issue queries against it.
Program.cs performs the following operations:
- Opens a console window.
- Connects to Azure Search using the service URL and
- Deletes the
musicstoreindexif it exists.
- Creates a new
musicstoreindexusing the schema.json file.
- Populates the index using the data files.
- Queries the index using four queries. Notice that the scoring profiles are specified as a query parameter. All of the queries search for the same term, 'best'. The first query demonstrates default scoring. The remaining three queries use a scoring profile.
Build and run the application
To rule out connectivity or assembly reference problems, build and run the application to ensure there are no issues to work out first. You should see a console application open in the background. All four queries execute in sequence without pausing. On many systems, the entire program executes in under 15 seconds. If the console application includes a message stating “Complete. Press enter to continue”, the program completed successfully.
To compare query runs, you can mark-copy-paste the query results from the console and paste them into an Excel file.
The following illustration shows results from the first three queries side-by-side. All of the queries use the same search term, 'best', which appears in numerous album titles.
The first query uses default scoring. Since the search term appears only in album titles, and no other criteria is specified, items having 'best' in the album title are returned in the order in which the search service finds them.
The second query uses a scoring profile, but notice that the profile had no effect. The results are identical to those of the first query. This is because the scoring profile boosts a field ('genre') that is not germane to the query. The search term 'best' does not exist in any 'genre' field of any document. When a scoring profile has no effect, the results are the same as default scoring.
The third query is the first evidence of scoring profile impact. The search term is still 'best' so we are working with the same set of albums, but because the scoring profile provides additional criteria that boosts 'rating' and 'last-updated', some items are propelled higher in the list.
The next illustration shows the fourth and final query, boosted by 'margin'. The 'margin' field is non-searchable and cannot be returned in search results. The 'margin' value was manually added to the spreadsheet to help illustrate the point that items with higher margins show up higher in the search results list.
Now that you have experimented with scoring profiles, try changing the program to use different query syntax, scoring profiles, or richer data. Links in the next section provide more information.
Learn more about scoring profiles. See Add a scoring profile to an Azure Search index for details.
Learn more about search syntax and query parameters. See Search Documents (Azure Search REST API) for details.
Need to step back and learn more about index creation? Watch this video to understand the basics.