內容簡介
總結：SpamJudge利用AI大型語言模型自動偵測和過濾垃圾評論。
問題與答案：
1. SpamJudge使用什麼來自動偵測和過濾垃圾評論？
- 使用AI大型語言模型。
2. 該外掛的特點有哪些？
- 支援符合OpenAI格式的任何API。
- 可自訂AI提示以根據網站特性調整評分標準。
- 可配置評分閾值以靈活控制過濾強度。
- 詳細記錄以追踪每個評論的處理過程。
3. 描述SpamJudge的工作流程。
- 訪客提交評論。
- 外掛攔截評論並將其發送給AI進行評分。
- AI返回介於0和100之間的分數（0=垃圾，100=高質量)。
- 根據分數和閾值自動處理評論：如果分數>閾值，則批准；如果分數<閾值，根據設置將其移至垃圾或移至審核；如果超時/出錯，根據設置將其轉移到審核或直接批准。
- 記錄詳細的評論處理情況供管理員查看。
外掛標籤
開發者團隊
原文外掛簡介
SpamJudge uses AI large language models to automatically detect and filter spam comments.
Features
Supports any API compatible with the OpenAI format
Supports both /v1/chat/completions and /v1/response endpoints
Customizable AI prompts to adjust scoring criteria based on the characteristics of the website
Configurable score thresholds for flexible control over filtering intensity
Detailed logging to track the processing of each comment
Workflow
Visitor submits a comment
The plugin intercepts the comment and sends it to the AI for scoring
The AI returns a score between 0 and 100 (0 = spam, 100 = high quality)
The comment is automatically processed based on the score and threshold:
Score >= threshold: approved
Score < threshold: moved to spam or moved to moderation based on settings
Timeout/error: moved to moderation or directly approved based on settings
Detailed logs are recorded for administrators to review
Default system prompt in the current version
You are a spam comment detection system. Your ONLY task is to output a single number between 0 and 100.
SCORING RULES:
- 0-20: Obvious spam (ads, malicious links, gibberish)
- 21-40: Likely spam (suspicious links, bot-like comments)
- 41-60: Uncertain (short comments, borderline content)
- 61-80: Likely legitimate (relevant, thoughtful)
- 81-100: Clearly legitimate (detailed, helpful, on-topic)
CRITICAL INSTRUCTIONS:
1. Output ONLY a number (0-100)
2. NO explanations
3. NO additional text
4. NO punctuation
5. Just the number
Example valid outputs: 85
Example INVALID outputs: "Score: 85", "85 points", "I think it's 85"
If you output anything other than a single number, the system will fail.
