README
¶
GLM Content Reviewer Plugin
GLM Content Reviewer is an Apache Answer plugin that uses ZhipuAI's content moderation API to automatically review forum content for inappropriate material.
Features
- Automatic content moderation using GLM-4 API
- Configurable review for questions, answers, and comments independently
- Flexible handling: review queue or auto-delete flagged content
- Admin bypass: administrators are not subject to review
- Smart caching: reduces API calls by caching review results (1-hour TTL)
- Rate limiting: prevents API abuse with token bucket algorithm (10 req/sec)
- Retry logic: automatic retry with exponential backoff on API failures
- Content length validation: handles long content gracefully (8000 char limit)
- Configurable timeout: adjust API request timeout to your needs
- Metrics tracking: monitors review statistics and performance
- Multi-language support: English and Chinese translations
Installation
- Build the plugin:
cd reviewer-glm
go build
-
Copy the compiled plugin to your Answer plugins directory
-
Restart Answer server
Configuration
-
Get your API key from ZhipuAI Console
-
Configure the plugin in Answer admin panel:
- API Key: Your ZhipuAI API key (required)
- API Timeout: Request timeout in seconds (default: 30)
- Max Content Length: Maximum content length in characters (default: 8000)
- Cache TTL: Cache time-to-live in minutes (default: 60)
- Cache Max Size: Maximum cached entries (default: 1000)
- Rate Limit RPS: Maximum requests per second (default: 10)
- Max Retries: Maximum retry attempts for failed API calls (default: 2)
- Review Questions: Enable/disable question review
- Review Answers: Enable/disable answer review
- Review Comments: Enable/disable comment review
- Spam Filtering: Choose "review queue" or "auto-delete"
How It Works
When enabled, the plugin:
- Intercepts new content submissions (questions/answers/comments)
- Checks cache for previously reviewed identical content
- Applies rate limiting to prevent API abuse
- Sends content to GLM-4 API for moderation
- Parses response for safety indicators
- Approved content is published immediately
- Flagged content is queued for review or deleted based on configuration
Performance Features
Caching
- Content is hashed and cached with configurable TTL (default: 1 hour)
- Configurable cache size (default: 1000 entries)
- Reduces redundant API calls for duplicate content
- Automatic cleanup of expired cache entries
Rate Limiting
- Token bucket algorithm with configurable rate (default: 10 req/s)
- Prevents API quota exhaustion
- Graceful degradation under high load
- Adjustable to match your API plan
Retry Logic
- Configurable retry attempts (default: 2 retries)
- Exponential backoff between retries
- Detailed error logging
Content Validation
- Configurable content length limit (default: 8000 chars)
- Automatic truncation with warning logs
- Prevents token limit issues
Metrics
- Tracks total reviews, approvals, rejections
- Monitors cache hit rate and API errors
- Logs statistics every 5 minutes
API Integration
The plugin uses GLM-4 chat completion API with a specialized prompt for content moderation. Responses are parsed for safety keywords in both English and Chinese:
- Safe: "safe", "Safe", "SAFE", "安全"
- Unsafe: "unsafe", "违规", "不安全", "不合规"
Troubleshooting
High API errors: Increase the API timeout value in configuration Rate limit exceeded: Content is automatically rejected for review; increase rate limit RPS if needed Long content truncated: Content over the configured max length is truncated before review; adjust max content length setting Cache not working: Check cache TTL and max size settings; ensure values are positive Performance issues: Adjust cache size and rate limit to balance performance and API costs
License
Apache License 2.0
Documentation
¶
Index ¶
Constants ¶
This section is empty.
Variables ¶
var Info embed.FS
Functions ¶
This section is empty.
Types ¶
type ContentFilterError ¶
ContentFilterError indicates that GLM detected unsafe content in the input
func (*ContentFilterError) Error ¶
func (e *ContentFilterError) Error() string
type GLMRequest ¶
type GLMResponse ¶
type GLMResponse struct {
Choices []struct {
Message struct {
Content string `json:"content"`
} `json:"message"`
} `json:"choices"`
Error *struct {
Message string `json:"message"`
Code string `json:"code"`
} `json:"error,omitempty"`
ContentFilter []struct {
Level int `json:"level"`
Role string `json:"role"`
} `json:"contentFilter,omitempty"`
}
type Reviewer ¶
type Reviewer struct {
Config *ReviewerConfig
}
func (*Reviewer) ConfigFields ¶
func (r *Reviewer) ConfigFields() []plugin.ConfigField
func (*Reviewer) ConfigReceiver ¶
func (*Reviewer) Review ¶
func (r *Reviewer) Review(content *plugin.ReviewContent) (result *plugin.ReviewResult)
func (*Reviewer) TestConnection ¶
TestConnection tests the GLM API connection
type ReviewerConfig ¶
type ReviewerConfig struct {
APIKey string `json:"api_key"`
ReviewQuestion bool `json:"review_question"`
ReviewAnswer bool `json:"review_answer"`
ReviewComment bool `json:"review_comment"`
SpamFiltering string `json:"spam_filtering"`
APITimeout int `json:"api_timeout"` // Timeout in seconds
MaxContentLength int `json:"max_content_length"` // Maximum content length in characters
CacheTTL int `json:"cache_ttl"` // Cache TTL in minutes
CacheMaxSize int `json:"cache_max_size"` // Maximum cache entries
RateLimitRPS int `json:"rate_limit_rps"` // Rate limit requests per second
MaxRetries int `json:"max_retries"` // Maximum retry attempts
}
Source Files
Directories