Configure Snowplow Signals behavioral attributes

Users express intent through their browsing behavior - page views, filter interactions, content engagement - long before they type a message to an agent. In this step, you'll define behavioral attributes using Snowplow Signals, publish them as a service, and validate that they compute correctly from raw events.

These attributes form a behavioral profile that the agent's get_signals tool fetches at runtime, enabling personalized responses without requiring the user to state their preferences manually.

Set up your credentials

You need your Signals connection credentials. If you haven't set these up yet, see connecting to Signals. These are the same values you configured in the notebook's credentials cell:

CDI

python

API_URL = 'example.signals.snowplowanalytics.com'
API_KEY = ''
API_KEY_ID = ''
ORG_ID = ''

Sandbox

python

API_URL = 'you.signals.snowplowanalytics.com'
ACCESS_TOKEN = ''

Define attributes

Run the notebook cell that defines the attributes. The notebook creates a series of attributes that represent different user preferences and behaviors based on these content tags:

python

cultural_explorer_tags = ["culture", "history", "heritage", "ancient", "temples", "art", "traditional"]
modern_urbanite_tags = ["urban", "nightlife", "shopping", "modern", "architecture"]
tranquil_seeker_tags = ["nature", "peaceful", "wellness", "beaches", "mountains", "river", "wellness"]
family_fun_tags = ["family-friendly", "beaches", "nature", "food", "mountains", "culture"]
culinary_tourist_tags = ["food", "street food", "multicultural", "traditional", "urban", "shopping"]

The notebook defines these counter attributes:

• page_view_count - overall engagement level
• destination_page_view_count - interest in destination content
• family_destination_count - interactions with family destination content
• cultural_explorer - affinity for cultural/historical experiences
• modern_urbanite - affinity for urban/nightlife experiences
• tranquil_seeker - preference for nature and wellness
• family_fun - interest in family-friendly activities
• culinary_tourist - interest in food-related experiences
• budget_conscious_count - frequency of budget filtering
• luxury_inclined_count - frequency of luxury filtering

And these attributes with the last aggregation:

• preferred_experience_length - most recently selected experience duration
• latest_schedule - last itinerary schedule update

Running this cell defines the attributes locally but does not publish them yet.

Snowplow schemas

These attributes are designed around specific Snowplow events and entities (for example, filter_tag_applied, destination_filter, and the content entity). To use them on your own site, your tracking implementation must send the same events with matching schemas. You can adapt the attribute definitions to match your own event data instead.

Create an attribute group

Run the next notebook cell to define an attribute group. This creates a StreamAttributeGroup with domain_sessionid as the attribute key, meaning attributes are computed per browser session:

python

session_attributes_group = StreamAttributeGroup(
    name="travel_view",
    version=1,
    attribute_key=domain_sessionid,
    attributes=[page_view_count, dest_page_view_count, family_destination_count,
                cultural_explorer, modern_urbanite, tranquil_seeker, family_fun,
                culinary_tourist, preferred_experience_length, budget_conscious,
                luxury_inclined, latest_schedule],
    owner='you@email.com',
)

Create a service

Run the notebook cell that defines a service to manage the attribute group:

python

travel_service = Service(
    name="travel_service",
    description="Behavioral profile service for agent personalization.",
    attribute_groups=[session_attributes_group],
    owner='you@email.com'
)

The service name travel_service is what the agent's get_signals tool uses when querying the Profiles API.

Publish to Signals

Run the cell that publishes the attribute group and service:

python

response = sp_signals.publish([session_attributes_group, travel_service])

Signals will start processing attributes from your real-time event stream.

Validate attribute computation

Run the notebook cells that send synthetic test events and retrieve the computed attributes. The Snowplow tracker in the notebook sends events directly to your event Collector - no demo site or web application is needed. The page_url field is metadata used by the attribute criteria to match against URL patterns.

After sending the events, retrieve the results:

python

response = sp_signals.get_service_attributes(
    name="travel_service",
    attribute_key="domain_sessionid",
    identifier=user_id,
)

print(response)

You should see a response similar to the following (exact counts depend on the test events sent):

json

{
  "page_view_count": 10,
  "destination_page_view_count": 6,
  "family_destination_count": 3,
  "cultural_explorer": 5,
  "modern_urbanite": 1,
  "tranquil_seeker": 2,
  "family_fun": 3,
  "culinary_tourist": 2,
  "preferred_experience_length": "half-day",
  "budget_conscious_count": 2,
  "luxury_inclined_count": 1,
  "latest_schedule": null
}

tip

The get_signals tool uses the same Signals credentials you configured earlier in the notebook.