finish

.gitignore

@@ -17,6 +17,7 @@ app/model/model_save/*
 #data
 app/services/graph_service/data/indices/paper_ivf.index
 app/services/graph_service/data/indices
+app/services/papers_service/data
 
 app/indices
 app/routers/previous/output/summaries_dir
@@ -25,11 +26,17 @@ app/data
 app/routers/output/summaries_dir/*
 
 
+
+
 #logs
 app/scripts/logs
 app/scripts/pyrouge_root
 app/scripts/running_logs
 
+app/services/old
+
+
+
 #etc
 app/etc/*
 
@@ -58,6 +65,11 @@ share/python-wheels/
 *.egg
 MANIFEST
 
+
+app/data/
+app/model
+app/tools/
+
 # PyInstaller
 #  Usually these files are written by a python script from a template
 #  before PyInstaller builds the exe, so as to inject date/other infos into it.
@@ -206,3 +218,4 @@ cython_debug/
 .pypirc
 app/data/inductive/train.jsonl
 app/routers/output/summaries_dir/
+app/services/papers_service/data/paper_db.json

app/runtime/Dockerfile

@@ -5,21 +5,23 @@ ENV DEBIAN_FRONTEND=noninteractive \
     PYTHONUNBUFFERED=1 \
     NVIDIA_VISIBLE_DEVICES=all
 
+WORKDIR /app
+
 # 필수 패키지
 RUN apt-get update && apt-get install -y \
         python3-pip python3-dev git && \
     rm -rf /var/lib/apt/lists/*
 
-# 프로젝트 소스 복사
-WORKDIR /workspace
-COPY .. /workspace      
+# 1) requirements만 먼저 복사 → 레이어 캐시
+COPY runtime/requirements.txt ./requirements.txt
+RUN pip install --no-cache-dir --upgrade pip \
+ && pip install --no-cache-dir -r requirements.txt
 
-# 의존성(인퍼런스 전용)
-RUN pip3 install --upgrade pip \
- && pip3 install -r app/runtime/requirements.txt
+# 2) 실제 런타임 코드 복사
+COPY runtime/ ./runtime/
 
 # 지표·인덱스(volume 으로 붙여도 OK)
 # COPY indices/ /workspace/indices
 
 EXPOSE 8004
-CMD ["uvicorn", "app.runtime.api:app", "--host", "0.0.0.0", "--port", "8004"]
+CMD ["uvicorn", "runtime.api:app", "--host", "0.0.0.0", "--port", "8004"]

app/runtime/api.py

@@ -6,6 +6,7 @@
 from runtime.cluster_searcher import search_clusters, cluster2pids, meta
 from runtime.graph_builder    import build_tree
 
+
 app = FastAPI(title="SearchForest-AI Recommend API")
 
 
@@ -32,7 +33,7 @@ class RecResponse(BaseModel):
 @app.get("/inference", response_model=RecResponse)
 def recommend(
     query: str = Query(..., description="검색 쿼리"),
-    top_k: int = Query(5, gt=1, le=10)          # default 5
+    top_k: int = Query(10, gt=1, le=10)          # default 10
 ):
     # 1) 쿼리 기준 top-k 클러스터
     hits = search_clusters(query, top_k)
@@ -46,11 +47,6 @@ def recommend(
         root["children"].append(cluster_node)
     return {"results": root }
 
-
-
-
-
-
 # 로컬 실행용
 if __name__ == "__main__":
     uvicorn.run(app, host="0.0.0.0", port=8004)

app/runtime/graph_builder.py

@@ -9,7 +9,7 @@
 import torch
 from runtime.cluster_searcher import meta, cluster2pids   # ← meta 와 함께 추가로 import
 
-
+kw2pids: dict[str, list[str]] = {}     # keyword → [paper_ids]
 
 
 # ── 전역 설정 ───────────────────────────────────────────
@@ -121,6 +121,19 @@ def contains_kw(abs_txt: str, kw: str) -> bool:
     max_features=40_000,
 )
 
+# from sklearn.feature_extraction.text import TfidfVectorizer
+
+# # 전처리한 phrase 문서를 그대로 feeding
+# vectorizer = TfidfVectorizer(
+#     tokenizer=lambda s: s, preprocessor=lambda s: s, lowercase=False,
+#     ngram_range=(1,3), min_df=5, max_df=0.8
+# )
+# tfidf_mat = vectorizer.fit_transform(docs)           # shape (N_docs, N_terms)
+# idf   = vectorizer.idf_
+# vocab = vectorizer.vocabulary_                       # dict{phrase: idx}
+
+
+
 def top_keywords(pids, n=8):
     docs = [_as_text(G.nodes[p].get("abstract", "")).lower()
             for p in pids if G.has_node(p)]
@@ -156,7 +169,17 @@ def build_tree(root_kw: str, cid: int, depth: int = 1):
     ))
     pids_lvl0 = cluster2pids[cid]
 
-    tree = {"id": root_kw, "value": 1.0, "children": []}
+
+    # ① root 에도 pids 부여
+    tree = {
+            "id":    root_kw,
+            "value": 1.0,
+            "pids":  pids_lvl0,     # ★ 추가
+            "children": []
+        }
+
+    # kw2pids 전역 캐시에도 root 등록
+    kw2pids[root_kw] = pids_lvl0
 
     # ── depth-1  (최대 3개) ──────────────────────
     for kw1, sc1 in select_kw_scored(root_kw, cand, tfidf_dict, k=3):
@@ -172,16 +195,36 @@ def build_tree(root_kw: str, cid: int, depth: int = 1):
             "id":      kw1,
             "value":   round(sc1, 4),
             "pids":    hop1,          # 필요 없으면 제거
+            "children": []
         }
 
+        kw2pids[kw1] = hop1
+
         # ── depth-2 : parent=kw1, 최대 3개 ───────
         if depth > 1:
             for kw2, sc2 in select_kw_scored(kw1, cand, tfidf_dict, k=3):
+                # node1["children"].append({
+                #     "id":    kw2,
+                #     "value": round(sc2, 4),
+                # })
+                # hop-2 pids (root → kw1 → kw2)
+                kw2_emb = model.encode([kw2],
+                                        normalize_embeddings=True)[0]
+                hop2 = [
+                    p for p in hop1
+                    if (emb := get_abs_emb(p)) is not None
+                    and util.cos_sim(kw2_emb, emb).item() > COS_TH2
+                ]
+
                 node1["children"].append({
                     "id":    kw2,
                     "value": round(sc2, 4),
+                    "pids":  hop2,
                 })
 
+                # kw2pids에 2-depth 저장
+                kw2pids[kw2] = hop2
+
         tree["children"].append(node1)
 
     return tree

app/runtime/requirements.txt

@@ -4,4 +4,8 @@ torch
 sentence-transformers
 faiss-gpu
 networkx
-tqdm
+tqdm
+fastapi
+uvicorn
+pydantic
+spacy

app/runtime/run_uvicorn.sh

@@ -1,3 +1,3 @@
 #!/bin/bash
 
-uvicorn runtime.api:app --reload --port 8004
+uvicorn runtime.api:app --host 0.0.0.0 --port 8004 --reload

app/services/get_meta.py

@@ -0,0 +1,17 @@
+import aiohttp, asyncio
+
+S2_URL = "https://api.semanticscholar.org/graph/v1/paper/{}"
+FIELDS = "title,abstract,year,venue,referenceCount,citationCount," \
+         "influentialCitationCount,fieldsOfStudy,authors,tldr"
+
+async def fetch_paper(session, pid):
+    url = S2_URL.format(pid)
+    params = {"fields": FIELDS}
+    async with session.get(url, params=params, timeout=10) as r:
+        r.raise_for_status()
+        return await r.json()
+
+async def fetch_many(pids):
+    async with aiohttp.ClientSession() as session:
+        tasks = [fetch_paper(session, pid) for pid in pids]
+        return await asyncio.gather(*tasks)

app/services/graph_service/Dockerfile

@@ -2,10 +2,9 @@
 FROM python:3.9-slim
 
 WORKDIR /app
-COPY requirements.txt .
+COPY . .
 RUN pip install --no-cache-dir -r requirements.txt
 
-COPY . .
 
 
 # FastAPI Uvicorn 실행