feat(knowledge): Phase D 统一搜索 + 种子知识冷启动
Some checks failed
CI / Lint & TypeCheck (push) Has been cancelled
CI / Unit Tests (push) Has been cancelled
CI / Build Frontend (push) Has been cancelled
CI / Rust Check (push) Has been cancelled
CI / Security Scan (push) Has been cancelled
CI / E2E Tests (push) Has been cancelled

- search/recommend API 返回 UnifiedSearchResult (文档+结构化双通道)
- POST /api/v1/knowledge/seed 种子知识冷启动 (幂等, admin权限)
- seed_knowledge service: 按标题+行业查重, source=distillation
- SearchRequest 扩展: search_structured/search_documents/industry_id
This commit is contained in:
iven
2026-04-12 20:46:43 +08:00
parent f8c5a76ce6
commit 640df9937f
4 changed files with 167 additions and 16 deletions

View File

@@ -374,21 +374,17 @@ pub async fn rollback_version(
// === 检索 ===
/// POST /api/v1/knowledge/search — 语义搜索
/// POST /api/v1/knowledge/search — 统一搜索(双通道:文档 + 结构化)
pub async fn search(
State(state): State<AppState>,
Extension(ctx): Extension<AuthContext>,
Json(req): Json<SearchRequest>,
) -> SaasResult<Json<Vec<SearchResult>>> {
) -> SaasResult<Json<UnifiedSearchResult>> {
check_permission(&ctx, "knowledge:search")?;
let limit = req.limit.unwrap_or(5).min(10);
let min_score = req.min_score.unwrap_or(0.5);
let results = service::search(
let results = service::unified_search(
&state.db,
&req.query,
req.category_id.as_deref(),
limit,
min_score,
&req,
Some(&ctx.account_id),
).await?;
Ok(Json(results))
}
@@ -398,15 +394,15 @@ pub async fn recommend(
State(state): State<AppState>,
Extension(ctx): Extension<AuthContext>,
Json(req): Json<SearchRequest>,
) -> SaasResult<Json<Vec<SearchResult>>> {
) -> SaasResult<Json<UnifiedSearchResult>> {
check_permission(&ctx, "knowledge:search")?;
let limit = req.limit.unwrap_or(5).min(10);
let results = service::search(
let mut req = req;
req.min_score = Some(0.3);
req.search_structured = req.search_structured.or(Some(true));
let results = service::unified_search(
&state.db,
&req.query,
req.category_id.as_deref(),
limit,
0.3,
&req,
Some(&ctx.account_id),
).await?;
Ok(Json(results))
}
@@ -885,3 +881,34 @@ async fn handle_structured_upload(
_ => Err(SaasError::InvalidInput("意外的处理结果".into())),
}
}
// === 种子知识冷启动 ===
/// POST /api/v1/knowledge/seed — 触发种子知识冷启动
///
/// 需要 admin 权限,幂等(按标题+行业查重)
pub async fn seed_knowledge(
State(state): State<AppState>,
Extension(ctx): Extension<AuthContext>,
Json(req): Json<SeedKnowledgeRequest>,
) -> SaasResult<Json<serde_json::Value>> {
check_permission(&ctx, "knowledge:admin")?;
if req.items.len() > 100 {
return Err(SaasError::InvalidInput("单次种子不能超过 100 条".into()));
}
let created = service::seed_knowledge(
&state.db,
&req.industry_id,
req.category_id.as_deref().unwrap_or("seed"),
&req.items.iter().map(|i| (i.title.clone(), i.content.clone(), i.keywords.clone().unwrap_or_default())).collect::<Vec<_>>(),
&ctx.account_id,
).await?;
Ok(Json(serde_json::json!({
"industry_id": req.industry_id,
"created_count": created,
"total_submitted": req.items.len(),
})))
}

View File

@@ -32,6 +32,7 @@ pub fn routes() -> axum::Router<crate::state::AppState> {
// 检索
.route("/api/v1/knowledge/search", post(handlers::search))
.route("/api/v1/knowledge/recommend", post(handlers::recommend))
.route("/api/v1/knowledge/seed", post(handlers::seed_knowledge))
// 分析看板
.route("/api/v1/knowledge/analytics/overview", get(handlers::analytics_overview))
.route("/api/v1/knowledge/analytics/trends", get(handlers::analytics_trends))

View File

@@ -582,6 +582,113 @@ pub async fn search(
}).filter(|r| r.score >= min_score).collect())
}
// === 统一搜索(双通道合并) ===
/// 统一搜索:同时检索文档通道和结构化通道
pub async fn unified_search(
pool: &PgPool,
request: &SearchRequest,
viewer_account_id: Option<&str>,
) -> SaasResult<UnifiedSearchResult> {
let limit = request.limit.unwrap_or(5).min(10);
let search_docs = request.search_documents.unwrap_or(true);
let search_struct = request.search_structured.unwrap_or(true);
// 文档通道
let documents = if search_docs {
search(
pool,
&request.query,
request.category_id.as_deref(),
limit,
request.min_score.unwrap_or(0.5),
).await?
} else {
Vec::new()
};
// 结构化通道
let structured = if search_struct {
query_structured(
pool,
&StructuredQueryRequest {
query: request.query.clone(),
source_id: None,
industry_id: request.industry_id.clone(),
limit: Some(limit),
},
viewer_account_id,
).await?
} else {
Vec::new()
};
Ok(UnifiedSearchResult {
documents,
structured,
})
}
// === 种子知识冷启动 ===
/// 为指定行业插入种子知识(幂等)
pub async fn seed_knowledge(
pool: &PgPool,
industry_id: &str,
category_id: &str,
items: &[(String, String, Vec<String>)], // (title, content, keywords)
system_account_id: &str,
) -> SaasResult<usize> {
let mut created = 0;
for (title, content, keywords) in items {
if content.trim().is_empty() {
continue;
}
// 幂等:按标题 + source='distillation' + tags 含行业ID 查重
let exists: (i64,) = sqlx::query_as(
"SELECT COUNT(*) FROM knowledge_items \
WHERE title = $1 AND source = 'distillation' \
AND $2 = ANY(tags)"
)
.bind(title)
.bind(format!("industry:{}", industry_id))
.fetch_one(pool)
.await?;
if exists.0 > 0 {
continue;
}
let id = uuid::Uuid::new_v4().to_string();
let now = chrono::Utc::now();
let kw_json = serde_json::to_value(keywords).unwrap_or(serde_json::json!([]));
let tags = vec![
format!("industry:{}", industry_id),
"source:distillation".to_string(),
];
sqlx::query(
"INSERT INTO knowledge_items \
(id, category_id, title, content, keywords, status, priority, visibility, account_id, source, tags, version, created_by, created_at, updated_at) \
VALUES ($1, $8, $2, $3, $4, 'active', 5, 'public', NULL, \
'distillation', $5, 1, $6, $7, $7)"
)
.bind(&id)
.bind(title)
.bind(content)
.bind(&kw_json)
.bind(&tags)
.bind(system_account_id)
.bind(&now)
.bind(category_id)
.execute(pool)
.await?;
created += 1;
}
Ok(created)
}
// === 分析 ===
/// 分析总览

View File

@@ -331,3 +331,19 @@ pub struct StructuredQueryResult {
pub total_matched: i64,
pub generated_sql: Option<String>,
}
// === 种子知识 ===
#[derive(Debug, Deserialize)]
pub struct SeedKnowledgeRequest {
pub industry_id: String,
pub category_id: Option<String>,
pub items: Vec<SeedKnowledgeItem>,
}
#[derive(Debug, Deserialize)]
pub struct SeedKnowledgeItem {
pub title: String,
pub content: String,
pub keywords: Option<Vec<String>>,
}