add api

personas_vector/api.py

+import json
+import os
+import traceback
+
+import faiss
+import numpy as np
+import pandas as pd
+from gm_rpcd.all import bind
+from utils.cache import redis_client_db
+from utils.message import send_msg_to_dingtalk
+from utils.personas import get_user_portrait_tag3_from_redis
+
+MODEL_PATH = os.path.join(os.getcwd(), "_models")
+INDEX_PATH = os.path.join(MODEL_PATH, "faiss_personas_vector.index")
+FAISS_TAGS_INDEX = faiss.read_index(INDEX_PATH)
+
+TAG_EMBEDDING_DICT = redis_client_db.hgetall("personas_tags_embedding")
+
+
+@bind("strategy_embedding/personas_vector/match")
+def match_tractate_by_device(device_id, n=10):
+    try:
+        business_tags = get_user_portrait_tag3_from_redis(device_id, tags_num=5).get("business_tags", [])
+        business_tags = ["假体下巴", "你好", "假体隆胸"]
+        res = []
+        vectors = []
+        for tag in business_tags:
+            lst = json.loads(TAG_EMBEDDING_DICT.get(bytes(tag, "utf-8"), b"[]"))
+            if lst:
+                vectors.append(np.array(lst).astype("float32"))
+        if vectors:
+            D, I = FAISS_TAGS_INDEX.search(np.array([np.average(vectors, axis=0)]).astype("float32"), n)
+            res = I.tolist()[0]
+        return res
+    except Exception as e:
+        send_msg_to_dingtalk(str(traceback.format_exc()))
+        return []

personas_vector/to_vector.py

 import json
 import os
-import random
 import sys
 
 sys.path.append(os.path.realpath("."))

requirements.txt

@@ -33,3 +33,4 @@ keras==2.4.3
 protobuf==3.13.0
 
 ipython
+prompt-toolkit==2.0.10

utils/personas.py

+import json
+
+from utils.cache import redis_client2
+
+
+def get_user_portrait_tag3_redis_key(device_id):
+    return "doris:user_portrait:tag3:device_id:" + str(device_id)
+
+
+def get_user_portrait_tag3_from_redis(device_id, limit_score=0, tags_num=5):
+    def items_gt_score(d, tags_num=tags_num):
+        new_d = dict(sorted(d.items(), key=lambda x: x[1], reverse=True)[:tags_num])
+        res = {tag: float(score) for tag, score in new_d.items() if float(score) >= limit_score}
+        return list(res.keys())
+
+    portrait_key = get_user_portrait_tag3_redis_key(device_id)
+    if redis_client2.exists(portrait_key):
+        user_portrait = json.loads(redis_client2.get(portrait_key))
+        first_demands = items_gt_score(user_portrait.get("first_demands", {}))
+        second_demands = items_gt_score(user_portrait.get("second_demands", {}))
+        first_solutions = items_gt_score(user_portrait.get("first_solutions", {}))
+        second_solutions = items_gt_score(user_portrait.get("second_solutions", {}))
+        first_positions = items_gt_score(user_portrait.get("first_positions", {}))
+        second_positions = items_gt_score(user_portrait.get("second_positions", {}))
+        projects = items_gt_score(user_portrait.get("projects", {}))
+        anecdote_tags = items_gt_score(user_portrait.get("anecdote_tags", {}))
+        business_tags = items_gt_score(user_portrait.get("business_tags", {}))
+        return {
+            "first_demands": first_demands,
+            "second_demands": second_demands,
+            "first_solutions": first_solutions,
+            "second_solutions": second_solutions,
+            "first_positions": first_positions,
+            "second_positions": second_positions,
+            "projects": projects,
+            "anecdote_tags": anecdote_tags,
+            "business_tags": business_tags
+        }
+    return {}

word_vector/word_to_vec.py

@@ -112,7 +112,7 @@ def clicked_tractate_ids_item2vec_model(id, n=5):
     try:
         return TRACTATE_CLICK_IDS_MODEL.wv.most_similar(id, topn=n)
     except Exception as e:
-        send_msg_to_dingtalk(str(traceback.format_exc()))
+        # send_msg_to_dingtalk(str(traceback.format_exc()))
         return []