# Capability Introduction The platform provides a flexible **Exposure De\-duplication** capability. Through two independent configurations in the console, you can fine\-tune how items are filtered to prevent repetitive content. You can establish de\-duplication rules based on **actual user exposure** while simultaneously configuring a short\-term "fallback" rule based on **delivered items**. This enables a robust, **no\-code** de\-duplication system that effectively filters items a user has truly seen, while providing a safety net for the brief window of "exposure behavior data latency." This feature is ideal for homepage recommendations, news feeds, and any scenario where repetitive content must be avoided. This guide explains how to configure and utilize these de\-duplication capabilities via the console. ## Core value 1. **True Exposure De\-duplication:** Ensure that high\-quality items—which were cached or retrieved but never actually displayed to the user—are not incorrectly filtered out. The platform only filters items that the user has **actually seen**. 2. **Flexible Cross\-Scenario Logic:** Clients can implement sophisticated logic to ensure that items a user has interacted with in other channels (e.g., **already purchased items**) are not repeatedly recommended across different recommendation scenarios. --- # Operation guide ## Exposure deduplication configuration :::tip \*Note Recommendation configurations created after January 6, 2026, automatically use exposure deduplication capability. Previously created configurations will retain the existing delivery deduplication capability. If you want to use the exposure deduplication capability, you can **upgrade** the deduplication configuration. Please refer to [升级历史场景](/docs/85296/2163565#de053e22). ::: 1. Go to **Experience Center** \> **Recommendation Experience,** and locate the “Strategy” configuration entry on the left. Expand “Related Event Scenes” to enter the configuration area for the actual behavior occurrence scene.

2. In the "Related Event Scenes" setting, select the corresponding scene from the dropdown list. This configuration affects the automatic extraction task of exposure data and ultimately impacts the update of exposed items. Ensure the configuration is accurate and meets expectations. :::warning \*Attention 1. Changing the "Related Event Scenes" will not trigger exposure data backtracking; the configuration will only take effect incrementally. 2. Removing the "Related Event Scenes" will not trigger exposure data cleanup; configurations will only take effect incrementally. ::: 3. In **Experience Center** ＞ ** Recommendation Experience,** check the "Strategy" configuration entry on the left. Expand "Deduplication Rules" to access the exposure deduplication configuration area.

4. Configure the "Exposure deduplication" and "Remove recently recommended" rules based on business requirements. The exposure deduplication capability is managed by two settings simultaneously. Ensure both are correctly configured to achieve the desired deduplication effect.

|**rules** |**configuration item** |**input restriction** |**capability explanation** | |---|---|---|---| |Exposure deduplication |time window |minimum: 1 hour|Define the deduplication logic for exposed items; the platform will automatically obtain **items exposed most recently within the time window.** The platform ensures that the recommended results exclude these item IDs.|\ | | |maximum: unlimited |For the source of exposure data, see [曝光物品数据来源](/docs/85296/2163565#c67c3441).|\ | | | |:::tip \*description|\ | | | ||\ | | | |1. If the number of exposed items \> Maximum number of results within the time window|\ | | | | 1. Deduplicated items = **items exposed within the time window, with exposure time from newest to oldest** (up to the item limit)|\ | | | |2. If the number of exposed items < Maximum number of results within the time window|\ | | | | 1. Exposure deduplication items = **full exposure items within the time frame**|\ | | | |3. When the "Maximum number of results" exceeds the number of recommendable items, recommendations will return empty.|\ | | | ||\ | | | |:::| |^^|Item quantity limit |Minimum: 1 item|^^|\ | | |Maximum: 10,000 items | | |Remove recently recommended |Time window |Minimum: 1 minute|Due to brief delays in exposure behavior receiving and processing, deduplication of recommended items prevents users from receiving duplicate recommendations of the same item.|\ | | |Maximum: 120 minutes |:::tip \*Explanation|\ | | | |It is recommended to set the time window to be greater than or equal to the delay in behavior data upload.|\ | | | ||\ | | | |:::| |^^|Item quantity limit |Minimum: 1 item|^^|\ | | |Maximum: 30,000 items | | 5. The new deduplication rules configuration takes effect in the experience interface with the next recommendation request. 6. After confirming the configuration meets expectations, click the "**Sync Changes to API**" button at the top right of the page. The online API interface will also utilize the new deduplication rules. --- ## Exposure Item Data Sources Each user's exposure data supports dual sources, which can be automatically incrementally extracted from User Event datasets or manually uploaded via the deduplication API. The platform merges two exposure item lists based on the exposure behavior occurrence time when processing requests. ## Automatic extraction: User Event dataset (recommended) If you are already uploading exposure behavior data, automatic extraction allows you to use the "exposure deduplication" feature with zero code and zero modification. The platform automatically extracts exposure behavior from the **past 7 days** after receiving new behavior data. The platform writes exposure items into the exposure item list in real time based on the "Related Event Scenes" configured for each recommendation configuration. For the next request, the platform ensures that the new recommendation results do not include already exposed items. :::tip \*Note There is a millisecond\-level delay in behavior data reception, data processing, exposure behavior extraction, and updating. To ensure reliable exposure deduplication, each recommendation configuration must configure at least a one\-minute "Remove recently recommended" rule. ::: ### Manual upload: deduplication API :::warning \*Attention Before calling the API, you must configure the corresponding recommendation strategy’s "Related Event Scenes". For configuration details, see [Remove items already exposed to users](/docs/85296/2163565). **If the recommendation strategy is not configured, the API will return an error.** ::: If you do not upload exposure behavior or need to implement more complex deduplication logic across multiple "Related Event Scenes", the platform supports writing the list of items to be deduplicated using the deduplication interface. To learn more, please check [Deduplicate - Deduplication interface](/docs/393/92269). --- ## Upgrade historical scenarios Previously created configurations will retain the existing "Remove recently recommended" capability. Users can upgrade to exposure deduplication through the platform's provided "Upgrade Capability". 1. Go to **Experience Center** ＞ **Recommendation Experience,** and check the "Strategy" configuration entry on the left. In scenes that have not been upgraded, Deduplication Rules will show an "**Upgradeable Capabilities"** prompt.

2. Clicking "**Upgradeable Capabilities**" opens a pop\-up where the platform explains the differences between "Remove recently recommended" and "Exposure deduplication" and requires users to configure the "Related Event Scenes" for the recommendation scene.

|**Configuration item** |**Description** | |---|---| |Related Event Scenes |**Corresponding "behavior occurrence interface / interaction interface" for the recommendation configuration**|\ | |This setting affects the automatic extraction task of exposure data, ultimately impacting the update of exposed items. Ensure the configuration is accurate and meets expectations. | 3. After clicking **"Upgrade"** , the platform immediately upgrades to exposure deduplication. The historical "Remove recently recommended" rules will be retained and used for the "deduplication of delivered items" rule. **This feature upgrade is irreversible.**

Viking AI Search