Social listening workflow fails to trigger automation on multilingual keywords

[analyst_carmen]

We’re running AEC 2021 with social listening workflows configured to trigger sentiment analysis automation when specific keywords are detected. The workflow works perfectly for English keywords but completely fails to trigger when monitoring multilingual content - specifically Arabic, Chinese, and Japanese keywords.

Our language detection is configured to support these languages, and we’ve verified the Unicode keyword patterns are correctly stored in the workflow trigger settings. However, the automation simply doesn’t fire when social posts containing these keywords appear. English keywords in the same workflow trigger immediately.

The workflow trigger logic shows the multilingual keywords as active, but the execution logs show zero matches despite confirmed posts containing exact keyword matches. Has anyone encountered similar issues with Unicode keyword support in workflow automation triggers?

I’ve resolved this exact issue multiple times in AEC 2021 deployments. The problem is a three-layer configuration gap that affects multilingual workflow triggers.

Language Detection Configuration: First, verify your language detection is configured at the data source level, not just the workflow level. Navigate to Social Listening > Data Sources > [Your Source] > Advanced Settings and ensure “Pre-ingestion Language Detection” is enabled. This ensures language metadata is attached to each social post before it enters the workflow evaluation pipeline. Without this, the workflow trigger engine doesn’t know which language context to use for keyword matching.

Unicode Keyword Support: The keyword configuration in AEC 2021 has a critical limitation - it doesn’t properly handle Unicode normalization variants. You need to configure BOTH NFC and NFD normalized forms of your keywords. For example, if you’re monitoring “café” you need to add both the composed form (U+00E9) and decomposed form (e + U+0301). For CJK languages, ensure you’re using the correct Unicode blocks - Japanese keywords should explicitly specify Hiragana (U+3040-U+309F), Katakana (U+30A0-U+30FF), or Kanji ranges as needed. Add each variant as a separate keyword in your trigger configuration.

Workflow Trigger Logic: The trigger evaluation engine in AEC 2021 uses a legacy string matching algorithm that isn’t fully Unicode-aware by default. You need to enable the enhanced Unicode matching mode by adding a custom property to your workflow definition. In the workflow configuration XML (accessible via Export Workflow), add this property block:

<triggerProperties>
  <property name="unicode.matching.mode" value="enhanced"/>
  <property name="charset.normalization" value="NFC"/>
  <property name="language.context.aware" value="true"/>
</triggerProperties>

After making these changes, reimport the workflow and clear the trigger evaluation cache (Admin Console > System > Clear Workflow Cache). Test with a known social post containing your multilingual keywords - you should see the language detection metadata in the execution logs and the trigger should fire correctly.

One additional consideration: if you’re using regex patterns in your keywords, ensure they’re Unicode-aware regex. Standard ASCII regex patterns won’t match Unicode character classes properly. Use \p{Script=Arabic} or \p{Script=Han} for script-specific matching instead of character ranges.

This configuration addresses all three focus areas and should resolve your multilingual trigger failures. Monitor the workflow execution logs after implementation to confirm language detection is occurring before trigger evaluation and that Unicode normalization is being applied consistently.

---

This draft is based on general Adobe Experience Cloud knowledge. It has not been verified against your specific version and environment. Practitioners: verify the steps and share your experience below.

[srinivasdatta]

I’ve seen this before - it’s typically a character encoding mismatch between the social listening data ingestion layer and the workflow trigger evaluation engine. Check if your social data source is being normalized to UTF-8 during ingestion. AEC 2021 had known issues with certain Unicode normalization forms (NFC vs NFD) causing trigger mismatches.

[patriciaad]

Are you using the native AEC social listening connector or a custom integration? We had a similar problem where our custom API integration wasn’t preserving Unicode properly. The fix involved explicitly setting charset=UTF-8 in our HTTP headers and ensuring the database collation supported full Unicode ranges. Also verify your keyword matching is case-insensitive where appropriate - some languages don’t have case distinctions but the workflow engine might be applying case-sensitive matching by default.

[jackg]

Check your workflow trigger configuration XML or JSON - there might be escape sequence issues with the Unicode characters. When we configured Japanese keywords, we had to use Unicode escape sequences (\uXXXX format) instead of literal characters in some configuration fields. The AEC UI might display them correctly but the underlying trigger engine could be receiving malformed strings. Export your workflow definition and inspect the raw XML/JSON to see how the multilingual keywords are actually stored.

[adob_larsson]

This sounds like the language detection is running AFTER the trigger evaluation instead of before. The workflow should detect language first, then match keywords in the appropriate language context. In AEC 2021, you need to explicitly enable pre-trigger language detection in the workflow settings - it’s not enabled by default. There’s a checkbox in the advanced trigger settings called “Enable language-aware keyword matching” that needs to be checked for multilingual workflows.

[mwhite]

Have you verified the actual byte representation of your keywords in the database? We discovered our Arabic keywords were being stored with incorrect byte sequences due to a charset conversion during the initial workflow setup. Use a database query tool to examine the raw bytes of your keyword configuration - they should match the UTF-8 byte sequences of your intended keywords. Also check if there are any character normalization settings in your social listening data source configuration that might be stripping diacritical marks or modifying character compositions.

[victoria_powerbiar]

I need to check the sentiment analysis logs as well. Sometimes the issue isn’t with trigger evaluation but with how the workflow engine handles Unicode in the execution context variables passed to downstream automation steps.

We discovered our Arabic keywords were being stored with incorrect byte sequences due to a charset conversion during the initial workflow setup.