feat: add compare, trends routes, embedder service, and phase5 tests

app/services/searcher.py

@@ -1,15 +1,19 @@
-"""FTS5 全文搜索服务 — 关键词 + 标签筛选，命中片段高亮，分页。"""
+"""搜索服务 — FTS5 关键词搜索 + ChromaDB 语义搜索，命中片段高亮，分页。"""
 
 from __future__ import annotations
 
+import logging
 import math
 import re
 
 from sqlalchemy import text
 from sqlalchemy.orm import Session, joinedload
 
+from app.config import settings
 from app.models import Paper
 
+logger = logging.getLogger(__name__)
+
 # ── 输入清洗 ──────────────────────────────────────────────────────────
 
 # FTS5 查询语法中的特殊字符，用户输入时需要移除
@@ -41,8 +45,9 @@ def search_papers(
     sort: str = "relevance",
     page: int = 1,
     page_size: int = 20,
+    mode: str = "keyword",
 ) -> dict:
-    """FTS5 搜索论文。
+    """搜索论文，支持 keyword (FTS5) 和 semantic (ChromaDB) 两种模式。
 
     返回::
         {
@@ -51,8 +56,14 @@ def search_papers(
             "total": int,
             "page": int,
             "total_pages": int,
+            "distances": dict[str, float],  # arxiv_id → distance (仅 semantic)
         }
     """
+    # ── semantic 模式 ──
+    if mode == "semantic" and settings.CHROMA_ENABLED and query:
+        return _search_semantic(db, query, tag, sort, page, page_size)
+
+    # ── keyword 模式（默认）──
     match_expr = _sanitize_query(query) if query else None
 
     # ── 无关键词 + 无标签 → 空结果 ──
@@ -63,6 +74,7 @@ def search_papers(
             "total": 0,
             "page": page,
             "total_pages": 0,
+            "distances": {},
         }
 
     # ── 构建条件性 JOIN 和 WHERE 片段 ──
@@ -146,6 +158,75 @@ def _search_with_fts(
         "total": total,
         "page": page,
         "total_pages": math.ceil(total / page_size) if total else 0,
+        "distances": {},
+    }
+
+
+def _search_semantic(
+    db: Session,
+    query: str,
+    tag: str | None,
+    sort: str,
+    page: int,
+    page_size: int,
+) -> dict:
+    """ChromaDB 语义搜索，失败时回退到 FTS5。"""
+    try:
+        from app.services.embedder import search_similar
+
+        top_k = page_size * 3  # 多取一些用于 tag 过滤
+        candidates = search_similar(query, top_k=top_k)
+    except Exception:
+        logger.exception("Semantic search failed, falling back to keyword")
+        candidates = []
+
+    if not candidates:
+        # 回退到 FTS5
+        return _search_with_fts(
+            db,
+            _sanitize_query(query) or query,
+            "JOIN paper_tags pt ON pt.paper_id = p.id" if tag else "",
+            "AND pt.tag = :tag" if tag else "",
+            {"tag": tag} if tag else {},
+            sort, page, page_size, (page - 1) * page_size,
+        )
+
+    # 按 arxiv_id 从 DB 加载完整数据
+    arxiv_ids = [c["arxiv_id"] for c in candidates]
+    distance_map = {c["arxiv_id"]: c["distance"] for c in candidates}
+
+    papers_query = (
+        db.query(Paper)
+        .filter(Paper.arxiv_id.in_(arxiv_ids))
+        .options(
+            joinedload(Paper.authors),
+            joinedload(Paper.tags),
+            joinedload(Paper.summary_status),
+            joinedload(Paper.bookmark),
+            joinedload(Paper.reading_status),
+        )
+    )
+    if tag:
+        papers_query = papers_query.filter(Paper.tags.any(tag=tag))
+
+    papers = papers_query.all()
+
+    # 按语义距离排序
+    id_order = {aid: idx for idx, aid in enumerate(arxiv_ids)}
+    papers.sort(key=lambda p: id_order.get(p.arxiv_id, 999))
+
+    # 分页
+    total = len(papers)
+    start = (page - 1) * page_size
+    page_papers = papers[start:start + page_size]
+
+    return {
+        "results": page_papers,
+        "snippets": {},
+        "total": total,
+        "page": page,
+        "total_pages": math.ceil(total / page_size) if total else 0,
+        "distances": distance_map,
     }
 
 
@@ -187,6 +268,7 @@ def _search_tag_only(
         "total": total,
         "page": page,
         "total_pages": math.ceil(total / page_size) if total else 0,
+        "distances": {},
     }