| name | video-classification |
| description | Establishes a multi-tiered review system for classifying, escalating, and moderating user-generated video content. |
| required_tools | ["validateVideo","checkUserHistory","assignReviewer","getReview","submitContentModeration","detectHateSpeech","detectExplicitContent","implementModeration","assessAgeRating","generateContentWarnings"] |
| output_fields | ["final_decision"] |
Step 1: Validate Video Content and Uploader History
Execute the Video Validation Protocol (VVP) to validate video format compliance against supported codec specifications (MP4, HEVC/H.264) and ensure a minimum resolution of 720p. Account for typos in specifications and do not discard based on them. Perform metadata extraction and validation according to the platform's Metadata Extraction Protocol (MEP).
Call the validateVideo tool with video_id and video_path.
Perform an analysis to ensure the uploader's history is clean; otherwise, identify red flags and exercise caution.
Call the checkUserHistory tool with uploader_id.
If video validation and uploader history analysis are positive, proceed to the next stage of the moderation process.
Step 2: Assign Initial Reviewer
Run the Reviewer Assignment Algorithm (RAA) incorporating reviewer expertise metrics, language proficiency scores, and current workload distribution factors. Assign an initial_reviewer_id based on optimal reviewer selection criteria.
Call the assignReviewer tool with video_id, video_language, and region.
Based on the initial reviewer's detected categories and their confidences, decide if the content needs to be escalated to a moderation expert.
Step 3: Conduct Initial Content Review
Execute the Review Session Protocol (RSP), including mandatory full-length video review and documentation requirements.
Call the getReview tool with video_id and initial_reviewer_id to fetch the review details.
Apply the Content Classification Taxonomy (CCT) to identify and classify content violations.
Calculate confidence_scores for each identified violation category.
Return empty fields if no categories and confidence scores are detected.
Step 4: Process Content Classification and Generate Report
Apply classification algorithms to detected_categories based on review findings.
Call the detectHateSpeech tool with video_id and transcript (assuming transcript is available).
Call the detectExplicitContent tool with video_id.
Calculate a composite violation severity score using weighted category metrics.
Generate a preliminary classification report including violation details and confidence scores.
Call the submitContentModeration tool with video_id and initial_reviewer_id.
[APPROVAL REQUIRED]
Step 5: Determine Escalation Requirements
Calculate the Escalation Threshold Metric (ETM) based on the violation severity and confidence scores from the preliminary classification report.
Compare the ETM against established escalation thresholds.
If the ETM exceeds the threshold, initiate the escalation protocol and assign a moderator_id.
Document the escalation justification and associated metrics.
Step 6: Implement Moderation Actions
Generate comprehensive moderator_notes documenting timestamps and descriptions of potentially objectionable content.
Apply the Moderation Action Matrix (MAM) to determine appropriate actions and embed them in your notes.
Upload the data to the database as per data upload guidelines within 24 hours of the case being assigned.
Call the implementModeration tool with video_id and moderator_id (if assigned in Step 5).
[APPROVAL REQUIRED]
Step 7: Final Review, Documentation, and Decision
Generate comprehensive review documentation including all decision points and justifications.
If the content was escalated, go through the moderator's detailed notes to identify potential moderation actions: Age Restrict, Remove, Strike Issued, Warning. You may issue one or more actions as per the case or None.
Assign an age rating based on the reviewers' input and moderation actions.
Call the assessAgeRating tool with video_id and content_flags (derived from detected issues and moderation actions) to determine the age rating ('18+', '13+', or None).
Assign a content warning (True or False) based on all signals.
Call the generateContentWarnings tool with video_id and detected_issues (derived from detected issues and moderation actions).
Based on all signals, including technical issues with the video or inappropriate content flagged by reviewers and moderators, assign a final_decision from these options: Remove, Warning, Allow, Age Restrict. Note that allowed content may or may not have a content warning status as True, and technical hiccups in content may result in a Remove decision.
Archive review session data according to data retention policies.
[APPROVAL REQUIRED]
Complete the step by providing the final decision.
<final_decision>value</final_decision>