Upload 3 files

app.py

@@ -0,0 +1,85 @@
+from fastapi import FastAPI, Request
+from sentence_transformers import SentenceTransformer, util
+import json
+import torch
+import os
+
+app = FastAPI()
+
+# ================= 配置区域 =================
+# 设定匹配阈值 (0-1)。
+# 建议 0.3 - 0.4。太高会导致匹配不到，太低会导致乱匹配。
+THRESHOLD = 0.35 
+
+# 加载轻量级模型 (80MB)
+# 第一次启动时会自动下载
+print("正在加载模型...")
+model = SentenceTransformer('all-MiniLM-L6-v2')
+print("模型加载完成")
+
+# ================= 数据预处理 =================
+# 读取 JSON 文件并预计算向量
+def load_and_encode_data():
+    if not os.path.exists('emoji_labels.json'):
+        print("错误: 找不到 emoji_labels.json")
+        return [], None
+        
+    with open('emoji_labels.json', 'r', encoding='utf-8') as f:
+        data = json.load(f)
+    
+    # 提取描述文本用于计算
+    texts = [item['text'] for item in data]
+    # 计算向量并转为 Tensor
+    embeddings = model.encode(texts, convert_to_tensor=True)
+    
+    return data, embeddings
+
+# 初始化数据
+emoji_data, emoji_embeddings = load_and_encode_data()
+
+# ================= API 接口 =================
+
+@app.get("/")
+def home():
+    return {"status": "Kouri Emotion API is running"}
+
+@app.post("/match")
+async def match_emoji(request: Request):
+    """
+    接收 {"text": "我想吃汉堡"}
+    返回 {"label": "burger", "score": 0.85}
+    """
+    try:
+        body = await request.json()
+        user_text = body.get("text", "")
+        
+        if not user_text or emoji_embeddings is None:
+            return {"label": None, "reason": "empty_input_or_data"}
+
+        # 1. 计算用户输入的向量
+        query_emb = model.encode(user_text, convert_to_tensor=True)
+        
+        # 2. 计算与库中所有描述的余弦相似度
+        scores = util.cos_sim(query_emb, emoji_embeddings)[0]
+        
+        # 3. 找到得分最高的那个
+        best_score = float(torch.max(scores))
+        best_idx = int(torch.argmax(scores))
+        
+        # 4. 判断是否超过阈值
+        if best_score > THRESHOLD:
+            matched_item = emoji_data[best_idx]
+            return {
+                "label": matched_item['label'],
+                "score": best_score,
+                "matched_text": matched_item['text'] # 方便调试看它匹配到了哪一条
+            }
+        else:
+            return {
+                "label": None, 
+                "score": best_score,
+                "reason": "low_confidence"
+            }
+            
+    except Exception as e:
+        return {"error": str(e)}

emoji_labels.json

@@ -0,0 +1,10 @@
+[
+  {"text": "开心快乐高兴哈哈笑嘻嘻好耶", "label": "happy"},
+  {"text": "难过哭泣悲伤痛苦呜呜呜", "label": "sad"},
+  {"text": "生气愤怒发火暴躁", "label": "angry"},
+  {"text": "爱你喜欢你笔芯么么哒", "label": "love"},
+  {"text": "无语发呆不知道说什么", "label": "neutral"},
+  {"text": "想吃汉堡炸鸡快餐麦当劳肯德基", "label": "burger"},
+  {"text": "好喝的奶茶饮料咖啡", "label": "drink"},
+  {"text": "看到小猫咪喵喵叫好可爱", "label": "cat"}
+]

requirements.txt

@@ -0,0 +1,5 @@
+sentence-transformers
+torch
+fastapi
+uvicorn
+pandas