# Background Personalized recommendation capability helps you deliver customized item suggestions, ensuring recommendations closely match users' interests and preferences, ultimately improving your business metrics. When the recommended item pool contains highly similar items, the recommendation system may tend to show these similar items repeatedly, reducing diversity and affecting the user experience. # Feature introduction With the "Item Diversity Rule" feature, you can define rules to reorder the recommendation list before returning results to users, ensuring diversity among recommended items. This feature is primarily designed for operation teams and product managers, aiming to enhance end\-user browsing experience and content discovery efficiency. This article explains how to configure item diversity rules within a single recommendation configuration. ## **Diversity rule list** |**Diversity rules** |**Description** | |---|---| |Disperse by consecutive item dimensions |Prevent clustering of similar content to improve recommendation diversity and user experience. The rules here are divided into two categories:|\ | ||\ | |1. Within a specified sliding window, ensure **uniqueness of specific dimensions** (e.g., author ID, content category)|\ | |2. Within a specified sliding window, **disperse only items that meet certain conditions** (e.g., is lottery = yes) | # Typical use cases * **Content platforms avoid consecutive display of homogeneous content**: In news, short video, or live streaming recommendations, prevent consecutive appearances of content from the same author, same category (e.g., technology, entertainment), or same topic. For example, short video platforms configure dispersion rules based on the "author ID" dimension to avoid users seeing videos from the same creator consecutively, enhancing content discovery diversity. * **E\-commerce platforms ensure diversity of categories and brands**: In product recommendation scenarios, avoid concentrated display of products from the same brand, same category (e.g., dresses, sports shoes), or same price range. For example, clothing e\-commerce configures dispersion rules based on the "category ID" dimension to ensure alternating appearances of different clothing categories in the recommendation list, reducing user aesthetic fatigue. * **Music/podcast platforms diversify content styles**: In audio content recommendations, avoid consecutive playback of content from the same singer, same music style (e.g., rock, classical), or same podcast series. By configuring dispersion rules based on the "style tag" dimension, users can discover more diverse audio content, improving listening experience and content exploration motivation. --- # Operational steps All rule changes are released in two phases: your changes first take effect in the "experience center" for effect verification. After confirmation, update the configuration to apply the rules to the live production environment. This design aims to provide a secure testing environment to prevent unverified rules from affecting live users. ## Add new rule Follow the steps below to create and activate a diversity rule within the recommendation scenario. 1. Go to **“Experience Center” — “Recommendation Experience” — “Strategy”** , then click “Diversity Rules” to access the diversity rule section.

2. If there is no existing configuration in the recommendation scenario, the interface will guide you to configure a new item diversity rule.

3. Click the “Add Rule” button in the interface to open the new rule configuration window. 4. Add the required rule configuration according to business needs. |**Configuration items** |**Explanation** | |---|---| |Rule name |Enter the rule name to identify different diversity rules. Set a simple and clear name. | |Consecutively recommended items |Define the window size for dispersed sliding items.|\ | |:::tip \*Example|\ | |When setting "consecutively recommended items" = 10 and "items that fulfill these conditions" = 5; within 10 consecutive recommended items, at most 5 items share the same item field value.|\ | ||\ | |:::| |Item condition |Configure the effective logic of diversity rules for item fields|\ | ||\ | |1. **Disperse dimension:** Select the field to be dispersed. This field must exist in the item dataset.|\ | |2. **Filter some items:** If the dispersion condition applies only to certain items, add a filter condition for effective items (e.g., category = lottery).|\ | ||\ | |:::warning \*Note|\ | ||\ | |If the item field is an array type, items are considered identical and dispersed only when the array **content and order are exactly the same**.|\ | |To prevent logic redundancy and system conflicts, the platform enforces a strict uniqueness check when saving rules: You cannot save a new diversity rule if its **"consecutively recommended items"** AND **"items that fulfill these conditions"** are identical to an existing rule.|\ | ||\ | |:::| |Return limit |Return up to how many items with the same **item field value**. "Return limit" cannot be greater than "**consecutively recommended items**". | 5. Click “Confirm,” and the platform will create the new diversity rule. The new rule is set to the lowest priority by default; to update the priority, please refer to [管理规则优先级](/docs/85296/1956414#db920a59). 6. The new diversity rule configuration will take effect in the experience interface from the next recommendation request. 6. When the configuration meets expectations, click "Sync Changes to API" at the top right of the rule list page. The online API interface will also apply the new diversity rules. ## View and edit rules Follow these steps to view and edit active diversity rules: 1. View diversity rules at **"Experience Center" — "Recommendation Experience" — ** **"Strategy"** ** — "Diversity Rules"** .

2. In the rule list, find the rule you want to view or modify and click its "Edit" button to access the rule details page.

3. On the rule details page, you can view the current configuration or modify the following information as needed:

|**Configuration item** |**Description** | |---|---| |Rule name |Enter the rule name to identify different diversity rules. Set a simple and clear name. | |Consecutively recommended items |Define the window size for dispersed sliding items.|\ | |:::tip \*Example|\ | |When setting "consecutively recommended items" = 10 and "items that fulfill these conditions" = 5; within 10 consecutive recommended items, at most 5 items share the same item field value.|\ | ||\ | |:::| |Item condition |Configure the effective logic of diversity rules for item fields|\ | ||\ | |1. **Disperse dimension:** Select the field to be dispersed. This field must exist in the item dataset.|\ | |2. **Filter some items:** If the dispersion condition applies only to certain items, add a filter condition for effective items (e.g., category = lottery).|\ | ||\ | |:::warning \*Note|\ | ||\ | |If the item field is an array type, items are considered identical and dispersed only when the array **content and order are exactly the same**.|\ | |To prevent logic redundancy and system conflicts, the platform enforces a strict uniqueness check when saving rules: You cannot save a new diversity rule if its **"consecutively recommended items"** AND **"items that fulfill these conditions"** are identical to an existing rule.|\ | ||\ | |:::| |Return limit |Return up to how many items with the same **item field value**. "Return limit" cannot be greater than "**consecutively recommended items**". | 4. Click "Confirm" to update the rule on the platform. The rule priority will not change. To update the priority, please refer to [管理规则优先级](/docs/85296/1956414#db920a59). 5. Updated diversity rules take effect in the experience interface on the next recommendation request. 6. When the configuration meets expectations, click "Sync Changes to API" at the top right of the rules list page. The updated diversity rules also apply to the online API interface. --- ## Manage rule priority If multiple diversity rules **take effect simultaneously** in a recommendation scenario, the platform might **fail to satisfy all diversity rules** during item reordering. Therefore, the platform allows users to control the execution order by setting priorities. In the rules list, rules higher in the list have higher priority and are executed first. Follow these steps to view and manage diversity rule priorities: :::warning \*Note the platform applies diversity rules based on priority when higher\-priority rules take effect, lower\-priority rules **may not apply** ::: 1. You can view diversity rules in **"Experience Center" — "Recommendation Experience" — "Strategy" — "Diversity Rules"** .

2. In the **"Diversity Rules"** list, you can adjust rule priorities by dragging and dropping, or remove rules by clicking the delete icon.

3. Hold the drag icon on the left of a rule entry and move it to a new position to adjust its priority. The platform will immediately save the rule priority. 4. The updated diversity rule priority takes effect in the experience interface at the next recommendation request. 5. When the configuration meets expectations, click "Sync Changes to API" at the top right of the rule list page. The online API interface will also use the updated diversity rule priority. --- ## Delete rule Follow the steps below to delete an active diversity rule configuration: 1. You can view diversity rules under **"Experience Center" — "Recommendation Experience" — "Strategy" — "Diversity Rules"** .

2. Click the "Delete" button of the target rule and then click "Confirm" again in the pop\-up confirmation dialog.

3. After the deletion, the rule immediately becomes invalid and will no longer be used in the experience interface at the next recommendation request. 4. When the configuration meets expectations, click "Sync Changes to API" at the top right of the rule list page. The online API interface will also update the diversity rule configuration and disable deleted rules. --- # FAQ **Q: Why does it prompt "** **The number of items shown in “maximum display” cannot exceed the number of items in “continuous recommendation”** **" when saving?** **A:** This happens because the "items that fulfill these conditions" configured within your rule is greater than the "consecutively recommended items". Increase the size of the "consecutively recommended items" or decrease the "items that fulfill these conditions". **Q: Why does it prompt a rule conflict when saving?** **A:** The platform does not allow saving a new rule with "consecutively recommended items" and "item conditions" that completely match an existing diversity rule. Check your current configuration and ensure the new rule's "consecutively recommended items" and "item conditions" do not conflict with existing rules. **Q: How does the diversity rule handle array\-type fields?** **A:** Items are considered identical and diversified only when the array's **content and order match exactly**. Here is an example of the processing logic. |**item a** |**item b** |**Disperse or not** | |---|---|---| |[1, 2, 3] |[1, 2, 3] |Dispersed | |[1, 2] |[1, 2, 3] |No | |[1, 2, 3] |[1, 3, 2] |No | The platform will later support disperse functions based on array content granularity. Stay tuned.

Viking AI Search