Tags
Overview
Content tags are an important data processing feature of the Omni Insights Platform, providing LLM-based custom data processing. By filling in a form, you can define information processing and extraction tasks for content such as posts, videos, and comments, enabling LLM processing scenarios such as content classification, key information extraction, and quantitative scoring. This tag data can be used for in-depth analysis and detail display.
In the Omni Insights Platform, Content Tags refers broadly to the output of LLM processing of community content. For example, if you want to extract the hero names discussed in a batch of content, quantify players' satisfaction with a new event, or sort player comments into several major categories, the resulting Hero Name, Satisfaction Score, and Comment Type can all be called tags. These tags supplement the related content, much like the sentiment scores, keywords, and other information the system processes by default, and can be displayed in multiple feature modules or used for filtering and analysis.
Tag list page
When you open the Content Tags module, you see the existing tags of the current game and their basic information. Click a tag name to open its details page and view its rules and tag distribution. You can also edit, run manually (that is, run LLM processing on community content within a specified period), copy, and delete existing tags. To create a new tag task, click Create Tag in the upper-right corner.
Create a tag
Click Create Tag in the upper-right corner of the tag list to open the Create Tag page.
On the Create Tag page, set the tag's name, data scope, and tag type, and write the tag rules according to the type. Tags are generated by the LLM processing the content, and the tag configuration is applied to the LLM prompt, so provide as much detail as possible when you configure it.
- Tag Name
The tag name is the main identifier of a tag and also key information that describes the LLM processing task, so include as much business meaning as possible
The tag name cannot be changed after the tag is created, so choose it carefully
When you write a tag name, follow these guidelines:
- Use a name that describes the tag content, such as User Satisfaction or Hero Names Discussed
- Add qualifiers where useful, such as New Event User Satisfaction
- Add suffixes such as Category or Score to reflect the tag type, such as New Event User Satisfaction Score
Tag names to avoid:
- Meaningless numbers or letters, such as tag_1 or 123
- Names that are too general, such as Gift Pack Name or Level Name
- Data Scope
The data scope defines which data the tag processes. Post, video, and comment data from each platform can currently be processed. You can choose to process data from specific channels, or include all channels of the game, including channels added later.
- Tag Type
The tag type determines the type of the LLM processing task. Different tag types have different processing logic and generate tags of different data types. Currently, only Category is supported; more types will be added later.
Note that the tag type cannot be changed after the tag is created, so choose it carefully
Classification tags
Classification tags correspond to a common text processing task: classification. For community data, classification tags are useful in the following scenarios:
- Classify what players discuss, for example, into gameplay, cards, story, payment and gift packs, character art and skins, fan creations, and so on.
- Classify player attitudes. This is similar to sentiment scores but can focus on a specific theme, for example, classifying player attitudes toward the gameplay of a new version as positive, negative, or neutral.
- Classify players into audience profiles based on what they post, for example, new players, paying players, and expert players.
- Classify noteworthy content for more precise targeting, for example, content that mentions a core gameplay feature, mentions buying a gift pack, or mentions a character.
- Tag Configuration
The configuration of a classification tag is shown in the following figure. At its core, you define several classification types: the type names become the generated tag values, and the classification rules are the logic for identifying content as each type.
The following figure shows the classification tag rules for Player Attitude Toward Game Graphics. Positive, Neutral, and Negative are the final tag values, and the figure shows the rule for Positive. When the LLM determines that content matches the Positive rule, it marks the content as Positive in the Player Attitude Toward Game Graphics tag.
- Additional settings
In addition to the core configuration, you can configure the following additional settings for a classification tag.
- Task Requirements: Limit the scope of content to include and exclude irrelevant content
- Matches: The maximum number of types a piece of content can be assigned to when it matches multiple rules. If content can be assigned to multiple types, the LLM tags it with the best-fitting types, and the tag value is a list
- Default Value if No Match: Whether to provide a default value when content is not classified into any category. If you configure a default value, it is also treated as a tag value; if you select Null, such content has no tag value
- Sample Data: Data examples provided to the LLM. Build data examples for the tag task using the following template to help the LLM complete the task better
The following are input and output examples for {tag name}:
Input: [Enter content]
Output: ['Category 1']
Input: [Enter content]
Output:['Category 1', 'Category 2']
- Update Settings
The update settings determine whether the tag runs automatically and when. If automatic updates are enabled, the system tags the content of the previous period on a schedule. Currently, daily scheduled runs are supported, which tag the previous day's content; more frequency options will be available later.
Set the scheduled update time according to your data collection frequency, so that scheduled tasks do not run before data collection finishes and miss data.
More tag types
In later versions, we will keep adding more tag types, such as scoring tags that quantify content and extraction tags that extract key information from content, to meet more diverse and complex content processing needs.
Tag details page
Click a tag name on the tag list page to open the tag's details page, where you can view the tag's configuration and results. As shown in the following figure, you can view the tag's coverage, whether it updates automatically, and its logic. You can also copy, run, or edit the tag from the action panel in the upper-right corner.
Switch to the Tag Overview tab to view the distribution of tag values over a period of time. In later versions, we will provide more ways to view the overview.
Run a tag manually
After a tag is created, it is not executed automatically, which means you need to run the tagging task manually. The prompt that confirms the tag was created has a Run Now button; click it to start a manual run. You can also click Run Manually on the list page or details page to do the same.
The manual run dialog is shown below. Select the start time and end time for the manual run, that is, the publish time range of the content to process, and then click Run to execute the LLM tagging task. Because LLM tasks may take a long time, you can leave the page after you start the manual run; the tagging task keeps running in the background.
Note that tag computation tasks consume LLM tokens. To avoid waste, do not run tags manually too often unless necessary.