The personalized recommendation feature provides more accurate item suggestions to your end users based on **automatically generated user profiles**. This feature applies to scenarios such as product recommendations on e\-commerce platforms, post distribution in content communities, and content stream recommendations in video apps, helping to improve user engagement and conversion rates. To ensure the quality of these personas and the effectiveness of the recommendations, the system requires a **sufficient volume of data that can be effectively interpreted by Large Language Models (LLMs)** . Before officially enabling recommendations, please complete the following data configuration and validation tasks **in the specified order**. :::tip **Overview of the Process** You must first activate at least one item dataset → then link the behavior dataset → finally activate the behavior dataset to start user profile extraction and creation. ::: # Prerequisites Before experiencing recommendations, you must ensure the following two conditions are met: 1. **At least one linked and successfully activated Image & Text Dataset** (such as graphics or products). This forms the basis for user behavior analysis and recommendations. 2. **Prepare a linkable user behavior dataset**. Please note that linking and activating the behavior dataset must occur only after the Image & Text Dataset is linked and activated. # Setup steps ## Step one: link and activate the item dataset 1. Go to your app → open **Datasets**. 2. In the "Item Dataset" section, click the **Linked Dataset** button. 3. Click the **Next: Configure fields** button and click **Done** to complete 2 step configurations. This triggers the system to validate the graphic data and build the index, which is necessary for subsequent user behavior analysis and recommendations.

4. Depending on the data size, activation may take some time. Please wait until the status shows **Configuration Active**.

## Step 2: Link User event dataset :::warning \*Note Before linking the User event dataset, ensure that the **items involved in the behaviors** have been uploaded to the Item dataset. If the matching rate between item IDs in behavior data and those in the item data is low, it will severely degrade the quality of the recommendations. ::: 1. Go to the ** Configuration → User Event Dataset** tab and click the **Linked Dataset** button. ![图片](https://p9-arcosite.byteimg.com/obj/tos-cn-i-goo7wpa0wc/866bf598c46b48149461fc57a6ea4c7f =2928x) 2. Select the target **User Event Dataset** and confirm. 1. The popup **only** shows the applicable **User Event** datasets. 2. Each app can link to **only 1** User Event dataset. ![图片](https://p9-arcosite.byteimg.com/obj/tos-cn-i-goo7wpa0wc/ddbec06dc3c54635a454064b0e7a2e26 =2844x) 3. Confirm the data field configuration to ensure the fields align with business definitions. Check and confirm that the platform’s standard fields on the right (such as user_id, item_id) are correctly mapped to corresponding fields in your dataset. ![图片](https://p9-arcosite.byteimg.com/obj/tos-cn-i-goo7wpa0wc/b20f8f1b19464bb19f1f4b4a7a680718 =900x) 4. Click "**Done**" ## Step 3: Historical behavior data backfill (optional) After confirming the data field configuration, the platform will provide one\-time historical data processing for the dataset. After configuration, the platform automatically processes behavior data within the User Event dataset and uses this historical behavior data to build user profiles for personalized recommendations. :::tip \*Explanation Processing historical behavior data incurs "Behavior data processing" fees. Confirm whether to use historical behavior data; processing fees for historical behavior data are charged per record. ::: 1. Through the popup, you can choose whether to "use" or "not use" historical behavior data.

2. After confirming, the platform will start (or skip) the data retrospection task based on your choice. ## Step four: activate the user event dataset and start the profiling task 1. Click **Confirm** to activate the User Event dataset, and the platform will begin the profiling and extraction tasks. You can view the linked Image and Text datasets through the interface. After the User Event dataset activates successfully, its status will automatically update to "Configuration active. Updates in real\-time".

--- # Related documentation After successfully activating the behavior dataset, you can 1. continue uploading User Event datasets via the platform interface and API: for details, please refer to [Write - real-time import/update data](/docs/85296/1544951) 2. Experience personalized recommendation capabilities: For details, refer to [Features Overview](/docs/85296/1791112)

Viking AI Search