Artificial Intelligence Recommendation:Content industry

Data tables

AIRec requires three data tables for content industry scenes.

Before uploading:

• Items and users each have a quota limit. Deduplicate both tables before upload.

• Provide behavior data from the last one to two weeks. If historical behavior data is unavailable (new scene or technical constraints), use the AIRec test data — expect up to two weeks of suboptimal results while the model accumulates real data.

• Fill in as many optional fields as possible. The more complete your data, the better the recommendation results.

Item table

Each row represents one piece of content (article, video, audio, image, and so on).

Sample data row (item table):

item_id,item_type,status,scene_id,pub_time,expire_time,title,tags,channel,weight,category_level,category_path
34513,article,1,"a101,b102",1520327038,1620000000,"The Digital Era Provides the Greatest Opportunity","digitalization,AI",economy,1,3,12_1024_56

User table

Each row represents one user.

Identifying users:

• Logged-on users: use user_id

• Anonymous users: use imei

• Mixed scenarios: use user_id for logged-on users and imei for anonymous users

Make sure all user records are unique. When requesting recommendation results, specify the correct user identifier — otherwise personalized recommendations cannot be generated.

Behavior table

Each row represents one user interaction with an item.

CREATE TABLE statements

Use these MaxCompute statements to create the tables required for starting an AIRec instance. Optional fields can be left blank, but all fields must be present in the schema.

--- Create a behavior table in the content industry.
DROP TABLE IF EXISTS behavior_table;
CREATE TABLE IF NOT EXISTS `behavior_table`
(
    trace_id STRING COMMENT "Request tracking ID"
    ,trace_info STRING COMMENT "Request tracking information"
    ,platform STRING COMMENT "Client platform"
    ,device_model STRING COMMENT "Device model"
    ,imei STRING COMMENT "Device ID"
    ,app_version STRING COMMENT "App version number"
    ,net_type STRING COMMENT "Network type"
    ,longitude STRING COMMENT "Longitude"
    ,latitude STRING COMMENT "Latitude"
    ,ip STRING COMMENT "Client IP address"
    ,login STRING COMMENT "Whether the user has logged on"
    ,report_src STRING COMMENT "Source of the report"
    ,scene_id STRING COMMENT "Scene ID"
    ,user_id STRING COMMENT "User ID"
    ,item_id STRING COMMENT "Item ID"
    ,item_type STRING COMMENT "Type of the item"
    ,module_id STRING COMMENT "Module ID"
    ,page_id STRING COMMENT "Page ID"
    ,position STRING COMMENT "Position of the item"
    ,bhv_type STRING COMMENT "Behavior type"
    ,bhv_value STRING COMMENT "Behavior details"
    ,bhv_time STRING COMMENT "Time at which the behavior occurs"
)
PARTITIONED BY
(
    ds STRING
)
LIFECYCLE 30
;


--- Create a user table in the content industry.
DROP TABLE IF EXISTS user_table;
CREATE TABLE IF NOT EXISTS `user_table`
(
    user_id STRING COMMENT "Unique user ID"
    ,user_id_type STRING COMMENT "Registration type of the user"
    ,third_user_name STRING COMMENT "Name of the third-party user"
    ,third_user_type STRING COMMENT "Name of the third-party platform"
    ,phone_md5 STRING COMMENT "MD5 hash value of the mobile phone number of the user."
    ,imei STRING COMMENT "Device ID of the user"
    ,content STRING COMMENT "User content"
    ,gender STRING COMMENT "Gender"
    ,age STRING COMMENT "Age"
    ,age_group STRING COMMENT "Age group"
    ,country STRING COMMENT "Country"
    ,city STRING COMMENT "City"
    ,ip STRING COMMENT "Last logon IP address"
    ,device_model STRING COMMENT "Device model"
    ,register_time STRING COMMENT "Registration time"
    ,last_login_time STRING COMMENT "Last logon time"
    ,last_modify_time STRING COMMENT "Last modification time of user information"
    ,tags STRING COMMENT "User tags"
    ,source STRING COMMENT "Source of the user"
    ,features STRING COMMENT "Additional user characteristics, which are strings"
    ,num_features STRING COMMENT "Additional user characteristics, which are numerical values"
)
PARTITIONED BY
(
    ds STRING
)
LIFECYCLE 30
;


--- Create an item table in the content industry.
DROP TABLE IF EXISTS item_table;
CREATE TABLE IF NOT EXISTS `item_table`
(
    item_id STRING COMMENT "Unique ID of the item"
    ,item_type STRING COMMENT "Type of the item"
    ,title STRING COMMENT "Item title"
    ,content STRING COMMENT "Body part of the item"
    ,pub_time STRING COMMENT "Release time"
    ,status STRING COMMENT "Whether the item can be recommended"
    ,expire_time STRING COMMENT "Time at which the item expires"
    ,last_modify_time STRING COMMENT "Last modification time of the item information"
    ,scene_id STRING COMMENT "Scene ID"
    ,duration STRING COMMENT "Duration, in seconds"
    ,category_level STRING COMMENT "Category level"
    ,category_path STRING COMMENT "Category path"
    ,tags STRING COMMENT "Tags"
    ,channel STRING COMMENT "Channels"
    ,organization STRING COMMENT "Organizations"
    ,author STRING COMMENT "Authors"
    ,pv_cnt STRING COMMENT "Number of exposures"
    ,click_cnt STRING COMMENT "Number of clicks"
    ,like_cnt STRING COMMENT "Number of likes"
    ,unlike_cnt STRING COMMENT "Number of dislikes"
    ,comment_cnt STRING COMMENT "Number of comments"
    ,collect_cnt STRING COMMENT "Number of favorites"
    ,share_cnt STRING COMMENT "Number of shares"
    ,download_cnt STRING COMMENT "Number of downloads"
    ,tip_cnt STRING COMMENT "Number of rewards"
    ,subscribe_cnt STRING COMMENT "Number of follows"
    ,source_id STRING COMMENT "Item source"
    ,country STRING COMMENT "Country"
    ,city STRING COMMENT "City"
    ,features STRING COMMENT "Additional characteristics"
    ,num_features STRING COMMENT "Additional characteristics, which are numerical values"
    ,weight STRING COMMENT "Weight of the item, default value: 1"
)
PARTITIONED BY
(
    ds STRING
)
LIFECYCLE 30
;

What's next

• Use scene IDs — learn how scene IDs map to pages and recommendation placements

• Improve recommendation diversity using instance operation rules — configure discretization policies for category-based diversity

• Negative feedback — use the dislike behavior type to suppress unwanted content