Computing l-diversity for a dataset

L-diversity is a property of a dataset and an extension of k-anonymity that measures the diversity of sensitive values for each column in which they occur. A dataset has l-diversity if, for every set of rows with identical quasi-identifiers, there are at least l distinct values for each sensitive attribute.

You can compute the l-diversity value based on one or more columns, or fields, of a dataset. This topic demonstrates how to compute l-diversity values for a dataset using Sensitive Data Protection. For more information about l-diversity or risk analysis in general, see the risk analysis concept topic before continuing on.

Before you begin

Before continuing, be sure you've done the following:

  1. Sign in to your Google Account.
  2. In the Google Cloud console, on the project selector page, select or create a Google Cloud project.
  3. Go to the project selector
  4. Make sure that billing is enabled for your Google Cloud project. Learn how to confirm billing is enabled for your project.
  5. Enable Sensitive Data Protection.
  6. Enable Sensitive Data Protection

  7. Select a BigQuery dataset to analyze. Sensitive Data Protection calculates the l-diversity metric by scanning a BigQuery table.
  8. Determine a sensitive field identifier (if applicable) and at least one quasi-identifier in the dataset. For more information, see Risk analysis terms and techniques.

Compute l-diversity

Sensitive Data Protection performs risk analysis whenever a risk analysis job runs. You must create the job first, either by using the Google Cloud console, sending a DLP API request, or using a Sensitive Data Protection client library.

Console

  1. In the Google Cloud console, go to the Create risk analysis page.

    Go to Create risk analysis

  2. In the Choose input data section, specify the BigQuery table to scan by entering the project ID of the project containing the table, the dataset ID of the table, and the name of the table.

  3. Under Privacy metric to compute, select l-diversity.

  4. In the Job ID section, you can optionally give the job a custom identifier and select a resource location in which Sensitive Data Protection will process your data. When you're done, click Continue.

  5. In the Define fields section, you specify sensitive fields and quasi-identifiers for the l-diversity risk job. Sensitive Data Protection accesses the metadata of the BigQuery table you specified in the previous step and attempts to populate the list of fields.

    1. Select the appropriate checkbox to specify a field as either a sensitive field (S) or quasi-identifier (QI). You must select 1 sensitive field and at least 1 quasi-identifier.
    2. If Sensitive Data Protection isn't able to populate the fields, click Enter field name to manually enter one or more fields and set each one as sensitive field or quasi-identifier. When you're done, click Continue.
  6. In the Add actions section, you can add optional actions to perform when the risk job is complete. The available options are:

    • Save to BigQuery: Saves the results of the risk analysis scan to a BigQuery table.
    • Publish to Pub/Sub: Publishes a notification to a Pub/Sub topic.

    • Notify by email: Sends you an email with results. When you're done, click Create.

The l-diversity risk analysis job starts immediately.

C#

To learn how to install and use the client library for Sensitive Data Protection, see Sensitive Data Protection client libraries.

To authenticate to Sensitive Data Protection, set up Application Default Credentials. For more information, see Set up authentication for a local development environment.


using Google.Api.Gax.ResourceNames;
using Google.Cloud.Dlp.V2;
using Google.Cloud.PubSub.V1;
using Newtonsoft.Json;
using System;
using System.Collections.Generic;
using System.Linq;
using System.Threading;
using System.Threading.Tasks;
using static Google.Cloud.Dlp.V2.Action.Types;
using static Google.Cloud.Dlp.V2.PrivacyMetric.Types;

public class RiskAnalysisCreateLDiversity
{
    public static object LDiversity(
        string callingProjectId,
        string tableProjectId,
        string datasetId,
        string tableId,
        string topicId,
        string subscriptionId,
        IEnumerable<FieldId> quasiIds,
        string sensitiveAttribute)
    {
        var dlp = DlpServiceClient.Create();

        // Construct + submit the job
        var ldiversityConfig = new LDiversityConfig
        {
            SensitiveAttribute = new FieldId { Name = sensitiveAttribute },
            QuasiIds = { quasiIds }
        };

        var config = new RiskAnalysisJobConfig
        {
            PrivacyMetric = new PrivacyMetric
            {
                LDiversityConfig = ldiversityConfig
            },
            SourceTable = new BigQueryTable
            {
                ProjectId = tableProjectId,
                DatasetId = datasetId,
                TableId = tableId
            },
            Actions =
            {
                new Google.Cloud.Dlp.V2.Action
                {
                    PubSub = new PublishToPubSub
                    {
                        Topic = $"projects/{callingProjectId}/topics/{topicId}"
                    }
                }
            }
        };

        var submittedJob = dlp.CreateDlpJob(
            new CreateDlpJobRequest
            {
                ParentAsProjectName = new ProjectName(callingProjectId),
                RiskJob = config
            });

        // Listen to pub/sub for the job
        var subscriptionName = new SubscriptionName(callingProjectId, subscriptionId);
        var subscriber = SubscriberClient.CreateAsync(subscriptionName).Result;

        // SimpleSubscriber runs your message handle function on multiple
        // threads to maximize throughput.
        var done = new ManualResetEventSlim(false);
        subscriber.StartAsync((PubsubMessage message, CancellationToken cancel) =>
        {
            if (message.Attributes["DlpJobName"] == submittedJob.Name)
            {
                Thread.Sleep(500); // Wait for DLP API results to become consistent
                done.Set();
                return Task.FromResult(SubscriberClient.Reply.Ack);
            }
            else
            {
                return Task.FromResult(SubscriberClient.Reply.Nack);
            }
        });

        done.Wait(TimeSpan.FromMinutes(10)); // 10 minute timeout; may not work for large jobs
        subscriber.StopAsync(CancellationToken.None).Wait();

        // Process results
        var resultJob = dlp.GetDlpJob(
            new GetDlpJobRequest
            {
                DlpJobName = DlpJobName.Parse(submittedJob.Name)
            });

        var result = resultJob.RiskDetails.LDiversityResult;

        for (var bucketIdx = 0; bucketIdx < result.SensitiveValueFrequencyHistogramBuckets.Count; bucketIdx++)
        {
            var bucket = result.SensitiveValueFrequencyHistogramBuckets[bucketIdx];
            Console.WriteLine($"Bucket {bucketIdx}");
            Console.WriteLine($"  Bucket size range: [{bucket.SensitiveValueFrequencyLowerBound}, {bucket.SensitiveValueFrequencyUpperBound}].");
            Console.WriteLine($"  {bucket.BucketSize} unique value(s) total.");

            foreach (var bucketValue in bucket.BucketValues)
            {
                // 'UnpackValue(x)' is a prettier version of 'x.toString()'
                Console.WriteLine($"    Quasi-ID values: [{String.Join(',', bucketValue.QuasiIdsValues.Select(x => UnpackValue(x)))}]");
                Console.WriteLine($"    Class size: {bucketValue.EquivalenceClassSize}");

                foreach (var topValue in bucketValue.TopSensitiveValues)
                {
                    Console.WriteLine($"    Sensitive value {UnpackValue(topValue.Value)} occurs {topValue.Count} time(s).");
                }
            }
        }

        return result;
    }

    public static string UnpackValue(Value protoValue)
    {
        var jsonValue = JsonConvert.DeserializeObject<Dictionary<string, object>>(protoValue.ToString());
        return jsonValue.Values.ElementAt(0).ToString();
    }
}

Go

To learn how to install and use the client library for Sensitive Data Protection, see Sensitive Data Protection client libraries.

To authenticate to Sensitive Data Protection, set up Application Default Credentials. For more information, see Set up authentication for a local development environment.

import (
	"context"
	"fmt"
	"io"
	"strings"
	"time"

	dlp "cloud.google.com/go/dlp/apiv2"
	"cloud.google.com/go/dlp/apiv2/dlppb"
	"cloud.google.com/go/pubsub"
)

// riskLDiversity computes the L Diversity of the given columns.
func riskLDiversity(w io.Writer, projectID, dataProject, pubSubTopic, pubSubSub, datasetID, tableID, sensitiveAttribute string, columnNames ...string) error {
	// projectID := "my-project-id"
	// dataProject := "bigquery-public-data"
	// pubSubTopic := "dlp-risk-sample-topic"
	// pubSubSub := "dlp-risk-sample-sub"
	// datasetID := "nhtsa_traffic_fatalities"
	// tableID := "accident_2015"
	// sensitiveAttribute := "city"
	// columnNames := "state_number", "county"
	ctx := context.Background()
	client, err := dlp.NewClient(ctx)
	if err != nil {
		return fmt.Errorf("dlp.NewClient: %w", err)
	}
	defer client.Close()

	// Create a PubSub Client used to listen for when the inspect job finishes.
	pubsubClient, err := pubsub.NewClient(ctx, projectID)
	if err != nil {
		return err
	}
	defer pubsubClient.Close()

	// Create a PubSub subscription we can use to listen for messages.
	// Create the Topic if it doesn't exist.
	t := pubsubClient.Topic(pubSubTopic)
	topicExists, err := t.Exists(ctx)
	if err != nil {
		return err
	}
	if !topicExists {
		if t, err = pubsubClient.CreateTopic(ctx, pubSubTopic); err != nil {
			return err
		}
	}

	// Create the Subscription if it doesn't exist.
	s := pubsubClient.Subscription(pubSubSub)
	subExists, err := s.Exists(ctx)
	if err != nil {
		return err
	}
	if !subExists {
		if s, err = pubsubClient.CreateSubscription(ctx, pubSubSub, pubsub.SubscriptionConfig{Topic: t}); err != nil {
			return err
		}
	}

	// topic is the PubSub topic string where messages should be sent.
	topic := "projects/" + projectID + "/topics/" + pubSubTopic

	// Build the QuasiID slice.
	var q []*dlppb.FieldId
	for _, c := range columnNames {
		q = append(q, &dlppb.FieldId{Name: c})
	}

	// Create a configured request.
	req := &dlppb.CreateDlpJobRequest{
		Parent: fmt.Sprintf("projects/%s/locations/global", projectID),
		Job: &dlppb.CreateDlpJobRequest_RiskJob{
			RiskJob: &dlppb.RiskAnalysisJobConfig{
				// PrivacyMetric configures what to compute.
				PrivacyMetric: &dlppb.PrivacyMetric{
					Type: &dlppb.PrivacyMetric_LDiversityConfig_{
						LDiversityConfig: &dlppb.PrivacyMetric_LDiversityConfig{
							QuasiIds: q,
							SensitiveAttribute: &dlppb.FieldId{
								Name: sensitiveAttribute,
							},
						},
					},
				},
				// SourceTable describes where to find the data.
				SourceTable: &dlppb.BigQueryTable{
					ProjectId: dataProject,
					DatasetId: datasetID,
					TableId:   tableID,
				},
				// Send a message to PubSub using Actions.
				Actions: []*dlppb.Action{
					{
						Action: &dlppb.Action_PubSub{
							PubSub: &dlppb.Action_PublishToPubSub{
								Topic: topic,
							},
						},
					},
				},
			},
		},
	}
	// Create the risk job.
	j, err := client.CreateDlpJob(ctx, req)
	if err != nil {
		return fmt.Errorf("CreateDlpJob: %w", err)
	}
	fmt.Fprintf(w, "Created job: %v\n", j.GetName())
	// Wait for the risk job to finish by waiting for a PubSub message.
	// This only waits for 10 minutes. For long jobs, consider using a truly
	// asynchronous execution model such as Cloud Functions.
	ctx, cancel := context.WithTimeout(ctx, 10*time.Minute)
	defer cancel()
	err = s.Receive(ctx, func(ctx context.Context, msg *pubsub.Message) {
		// If this is the wrong job, do not process the result.
		if msg.Attributes["DlpJobName"] != j.GetName() {
			msg.Nack()
			return
		}
		msg.Ack()
		time.Sleep(500 * time.Millisecond)
		j, err := client.GetDlpJob(ctx, &dlppb.GetDlpJobRequest{
			Name: j.GetName(),
		})
		if err != nil {
			fmt.Fprintf(w, "GetDlpJob: %v", err)
			return
		}
		h := j.GetRiskDetails().GetLDiversityResult().GetSensitiveValueFrequencyHistogramBuckets()
		for i, b := range h {
			fmt.Fprintf(w, "Histogram bucket %v\n", i)
			fmt.Fprintf(w, "  Size range: [%v,%v]\n", b.GetSensitiveValueFrequencyLowerBound(), b.GetSensitiveValueFrequencyUpperBound())
			fmt.Fprintf(w, "  %v unique values total\n", b.GetBucketSize())
			for _, v := range b.GetBucketValues() {
				var qvs []string
				for _, qv := range v.GetQuasiIdsValues() {
					qvs = append(qvs, qv.String())
				}
				fmt.Fprintf(w, "    QuasiID values: %s\n", strings.Join(qvs, ", "))
				fmt.Fprintf(w, "    Class size: %v\n", v.GetEquivalenceClassSize())
				for _, sv := range v.GetTopSensitiveValues() {
					fmt.Fprintf(w, "    Sensitive value %v occurs %v times\n", sv.GetValue(), sv.GetCount())
				}
			}
		}
		// Stop listening for more messages.
		cancel()
	})
	if err != nil {
		return fmt.Errorf("Recieve: %w", err)
	}
	return nil
}

Java

To learn how to install and use the client library for Sensitive Data Protection, see Sensitive Data Protection client libraries.

To authenticate to Sensitive Data Protection, set up Application Default Credentials. For more information, see Set up authentication for a local development environment.


import com.google.api.core.SettableApiFuture;
import com.google.cloud.dlp.v2.DlpServiceClient;
import com.google.cloud.dlp.v2.DlpServiceSettings;
import com.google.cloud.pubsub.v1.AckReplyConsumer;
import com.google.cloud.pubsub.v1.MessageReceiver;
import com.google.cloud.pubsub.v1.Subscriber;
import com.google.privacy.dlp.v2.Action;
import com.google.privacy.dlp.v2.Action.PublishToPubSub;
import com.google.privacy.dlp.v2.AnalyzeDataSourceRiskDetails.LDiversityResult;
import com.google.privacy.dlp.v2.AnalyzeDataSourceRiskDetails.LDiversityResult.LDiversityEquivalenceClass;
import com.google.privacy.dlp.v2.AnalyzeDataSourceRiskDetails.LDiversityResult.LDiversityHistogramBucket;
import com.google.privacy.dlp.v2.BigQueryTable;
import com.google.privacy.dlp.v2.CreateDlpJobRequest;
import com.google.privacy.dlp.v2.DlpJob;
import com.google.privacy.dlp.v2.FieldId;
import com.google.privacy.dlp.v2.GetDlpJobRequest;
import com.google.privacy.dlp.v2.LocationName;
import com.google.privacy.dlp.v2.PrivacyMetric;
import com.google.privacy.dlp.v2.PrivacyMetric.LDiversityConfig;
import com.google.privacy.dlp.v2.RiskAnalysisJobConfig;
import com.google.privacy.dlp.v2.Value;
import com.google.privacy.dlp.v2.ValueFrequency;
import com.google.pubsub.v1.ProjectSubscriptionName;
import com.google.pubsub.v1.ProjectTopicName;
import com.google.pubsub.v1.PubsubMessage;
import java.io.IOException;
import java.util.Arrays;
import java.util.List;
import java.util.concurrent.ExecutionException;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.TimeoutException;
import java.util.stream.Collectors;
import org.threeten.bp.Duration;

@SuppressWarnings("checkstyle:AbbreviationAsWordInName")
class RiskAnalysisLDiversity {

  public static void main(String[] args) throws Exception {
    // TODO(developer): Replace these variables before running the sample.
    String projectId = "your-project-id";
    String datasetId = "your-bigquery-dataset-id";
    String tableId = "your-bigquery-table-id";
    String topicId = "pub-sub-topic";
    String subscriptionId = "pub-sub-subscription";
    calculateLDiversity(projectId, datasetId, tableId, topicId, subscriptionId);
  }

  public static void calculateLDiversity(
      String projectId, String datasetId, String tableId, String topicId, String subscriptionId)
      throws ExecutionException, InterruptedException, IOException {

    // Initialize client that will be used to send requests. This client only needs to be created
    // once, and can be reused for multiple requests. After completing all of your requests, call
    // the "close" method on the client to safely clean up any remaining background resources.
    DlpServiceSettings.Builder dlpServiceSettingsBuilder = DlpServiceSettings.newBuilder();
    dlpServiceSettingsBuilder
        .getDlpJobSettings()
        .setRetrySettings(
            dlpServiceSettingsBuilder
                .getDlpJobSettings()
                .getRetrySettings()
                .toBuilder()
                .setTotalTimeout(Duration.ofSeconds(600))
                .build());
    try (DlpServiceClient dlpServiceClient =
        DlpServiceClient.create(dlpServiceSettingsBuilder.build())) {
      // Specify the BigQuery table to analyze
      BigQueryTable bigQueryTable =
          BigQueryTable.newBuilder()
              .setProjectId(projectId)
              .setDatasetId(datasetId)
              .setTableId(tableId)
              .build();

      // These values represent the column names of quasi-identifiers to analyze
      List<String> quasiIds = Arrays.asList("Age", "Mystery");

      // This value represents the column name to compare the quasi-identifiers against
      String sensitiveAttribute = "Name";

      // Configure the privacy metric for the job
      FieldId sensitiveAttributeField = FieldId.newBuilder().setName(sensitiveAttribute).build();
      List<FieldId> quasiIdFields =
          quasiIds.stream()
              .map(columnName -> FieldId.newBuilder().setName(columnName).build())
              .collect(Collectors.toList());
      LDiversityConfig ldiversityConfig =
          LDiversityConfig.newBuilder()
              .addAllQuasiIds(quasiIdFields)
              .setSensitiveAttribute(sensitiveAttributeField)
              .build();
      PrivacyMetric privacyMetric =
          PrivacyMetric.newBuilder().setLDiversityConfig(ldiversityConfig).build();

      // Create action to publish job status notifications over Google Cloud Pub/
      ProjectTopicName topicName = ProjectTopicName.of(projectId, topicId);
      PublishToPubSub publishToPubSub =
          PublishToPubSub.newBuilder().setTopic(topicName.toString()).build();
      Action action = Action.newBuilder().setPubSub(publishToPubSub).build();

      // Configure the risk analysis job to perform
      RiskAnalysisJobConfig riskAnalysisJobConfig =
          RiskAnalysisJobConfig.newBuilder()
              .setSourceTable(bigQueryTable)
              .setPrivacyMetric(privacyMetric)
              .addActions(action)
              .build();

      // Build the request to be sent by the client
      CreateDlpJobRequest createDlpJobRequest =
          CreateDlpJobRequest.newBuilder()
              .setParent(LocationName.of(projectId, "global").toString())
              .setRiskJob(riskAnalysisJobConfig)
              .build();

      // Send the request to the API using the client
      DlpJob dlpJob = dlpServiceClient.createDlpJob(createDlpJobRequest);

      // Set up a Pub/Sub subscriber to listen on the job completion status
      final SettableApiFuture<Boolean> done = SettableApiFuture.create();

      ProjectSubscriptionName subscriptionName =
          ProjectSubscriptionName.of(projectId, subscriptionId);

      MessageReceiver messageHandler =
          (PubsubMessage pubsubMessage, AckReplyConsumer ackReplyConsumer) -> {
            handleMessage(dlpJob, done, pubsubMessage, ackReplyConsumer);
          };
      Subscriber subscriber = Subscriber.newBuilder(subscriptionName, messageHandler).build();
      subscriber.startAsync();

      // Wait for job completion semi-synchronously
      // For long jobs, consider using a truly asynchronous execution model such as Cloud Functions
      try {
        done.get(15, TimeUnit.MINUTES);
      } catch (TimeoutException e) {
        System.out.println("Job was not completed after 15 minutes.");
        return;
      } finally {
        subscriber.stopAsync();
        subscriber.awaitTerminated();
      }

      // Build a request to get the completed job
      GetDlpJobRequest getDlpJobRequest =
          GetDlpJobRequest.newBuilder().setName(dlpJob.getName()).build();

      // Retrieve completed job status
      DlpJob completedJob = dlpServiceClient.getDlpJob(getDlpJobRequest);
      System.out.println("Job status: " + completedJob.getState());
      System.out.println("Job name: " + dlpJob.getName());

      // Get the result and parse through and process the information
      LDiversityResult ldiversityResult = completedJob.getRiskDetails().getLDiversityResult();
      List<LDiversityHistogramBucket> histogramBucketList =
          ldiversityResult.getSensitiveValueFrequencyHistogramBucketsList();
      for (LDiversityHistogramBucket result : histogramBucketList) {
        for (LDiversityEquivalenceClass bucket : result.getBucketValuesList()) {
          List<String> quasiIdValues =
              bucket.getQuasiIdsValuesList().stream()
                  .map(Value::toString)
                  .collect(Collectors.toList());

          System.out.println("\tQuasi-ID values: " + String.join(", ", quasiIdValues));
          System.out.println("\tClass size: " + bucket.getEquivalenceClassSize());

          for (ValueFrequency valueFrequency : bucket.getTopSensitiveValuesList()) {
            System.out.printf(
                "\t\tSensitive value %s occurs %d time(s).\n",
                valueFrequency.getValue().toString(), valueFrequency.getCount());
          }
        }
      }
    }
  }

  // handleMessage injects the job and settableFuture into the message reciever interface
  private static void handleMessage(
      DlpJob job