modify

linucb/views/collect_data.py

@@ -9,6 +9,9 @@ import json
 from trans2es.models.tag import TopicTag
 import traceback
 from django.conf import settings
+from libs.es import ESPerform
+from search.utils.common import *
+
 
 class KafkaManager(object):
     consumser_obj = None
@@ -28,6 +31,8 @@ class CollectData(object):
     def __init__(self):
         self.linucb_matrix_redis_prefix = "physical:linucb:device_id:"
         self.linucb_recommend_redis_prefix = "physical:linucb:tag_recommend:device_id:"
+        self.linucb_recommend_topic_id_prefix = "physical:linucb:topic_recommend:device_id:"
+        self.tag_topic_id_redis_prefix = "physical:tag_id:topic_id_list:"
         # 默认
         self.user_feature = [0,1]
 
@@ -44,8 +49,38 @@ class CollectData(object):
             logging.error("catch exception,err_msg:%s" % traceback.format_exc())
             return dict()
 
+    def get_tag_topic_list(self,tag_id):
+        try:
+            q = {
+                "query":{
+                    "bool":{
+                        "must":[
+                            {"term":{"is_online": True}},
+                            {"term":{"is_deleted": False}},
+                            {"term":{"tag_list":tag_id}}
+                        ]
+                    }
+                },
+                "_source":{
+                    "include":["id"]
+                },
+                "sort":[
+                    {"create_time_val":{"order":"desc"}},
+                    {"language_type":{"order":"asc"}},
+                ]
+            }
+            result_dict = ESPerform.get_search_results(ESPerform.get_cli(), sub_index_name="topic-high-star", query_body=q,
+                                                       offset=0, size=5000)
+
+            topic_id_list = [item["_source"]["id"] for item in result_dict["hits"]]
+            return topic_id_list
+        except:
+            logging.error("catch exception,err_msg:%s" % traceback.format_exc())
+            return list()
+
     def update_recommend_tag_list(self, device_id,user_feature=None):
         try:
+            recommend_tag_set = set()
             recommend_tag_list = list()
             recommend_tag_dict = dict()
             redis_linucb_tag_data_dict = self._get_user_linucb_info(device_id)
@@ -54,15 +89,46 @@ class CollectData(object):
                 LinUCB.init_device_id_linucb_info(redis_client, self.linucb_matrix_redis_prefix,device_id,recommend_tag_list)
             else:
                 user_feature = user_feature if user_feature else self.user_feature
-                recommend_tag_dict = LinUCB.linucb_recommend_tag(device_id,redis_linucb_tag_data_dict,user_feature,list(redis_linucb_tag_data_dict.keys()))
+                (recommend_tag_dict,recommend_tag_set) = LinUCB.linucb_recommend_tag(device_id,redis_linucb_tag_data_dict,user_feature,list(redis_linucb_tag_data_dict.keys()))
 
-            logging.info("duan add,device_id:%s,recommend_tag_list:%s" % (str(device_id), str(recommend_tag_list)))
             if len(recommend_tag_dict) > 0:
+                recommend_tag_list = list(recommend_tag_set)
                 tag_recommend_redis_key = self.linucb_recommend_redis_prefix + str(device_id)
-                redis_client.set(tag_recommend_redis_key, json.dumps(list(recommend_tag_dict.keys())))
+                redis_client.set(tag_recommend_redis_key, json.dumps(recommend_tag_list))
                 # Todo:设置过期时间，调研set是否支持
                 redis_client.expire(tag_recommend_redis_key, 7*24*60*60)
 
+
+                redis_key = "physical:home_recommend" + ":device_id:" + device_id + ":query_type:" + str(TopicPageType.HOME_RECOMMEND)
+
+                have_read_topic_id_list = list()
+                redis_field_list = [b'have_read_topic_list']
+                redis_field_val_list = redis_client.hmget(redis_key, redis_field_list)
+                if redis_field_val_list[0]:
+                    have_read_topic_id_list = list(json.loads(redis_field_val_list[0]))
+
+
+                recommend_topic_id_list = list()
+                for index in range(0,1000):
+                    for tag_id in recommend_tag_list[0:5]:
+                        redis_tag_id_key = self.tag_topic_id_redis_prefix + str(tag_id)
+                        redis_tag_id_data = redis_client.get(redis_tag_id_key)
+                        tag_topic_id_list = json.loads(redis_tag_id_data) if redis_tag_id_data else []
+                        if not redis_tag_id_data:
+                            tag_topic_id_list = self.get_tag_topic_list(tag_id)
+                            redis_client.set(redis_tag_id_key,json.dumps(tag_topic_id_list))
+                            redis_client.expire(redis_tag_id_key,1*24*60*60)
+                        for topic_id in tag_topic_id_list:
+                            if topic_id not in have_read_topic_id_list:
+                                recommend_topic_id_list.append(topic_id)
+                                break
+
+                topic_recommend_redis_key = self.linucb_recommend_topic_id_prefix + str(device_id)
+                redis_data_dict = {
+                    "data": json.dumps(recommend_topic_id_list),
+                    "cursor":0
+                }
+                redis_client.hmset(topic_recommend_redis_key,redis_data_dict)
             return True
         except:
             logging.error("catch exception,err_msg:%s" % traceback.format_exc())

linucb/views/linucb.py

@@ -14,7 +14,7 @@ from django.conf import settings
 
 class LinUCB:
     d = 2
-    alpha = 0.1
+    alpha = 0.01
     r1 = 10
     r0 = -0.1
     default_tag_list = list()
@@ -70,6 +70,7 @@ class LinUCB:
             #art_max = tag_list[np.argmax(np.dot(xaT, theta_tmp) + cls.alpha * np.sqrt(np.dot(np.dot(xaT, AaI_tmp), xa)))]
 
 
+            top_tag_set = set()
             top_tag_dict = dict()
             np_score_list = list()
             np_score_dict = dict()
@@ -87,17 +88,18 @@ class LinUCB:
                 for top_score_index in np_score_dict[top_score]:
                     tag_id = str(tag_list[top_score_index], encoding="utf-8")
                     top_tag_dict[tag_id] = top_score
-                    if len(top_tag_dict) >= 10:
+                    top_tag_set.add(tag_id)
+                    if len(top_tag_dict) >= 20:
                         break
 
-                if len(top_tag_dict) >= 10:
+                if len(top_tag_dict) >= 20:
                     break
 
             logging.info("duan add,device_id:%s,sorted_np_score_list:%s,np_score_dict:%s" % (str(device_id), str(sorted_np_score_list), str(np_score_dict)))
-            return top_tag_dict
+            return (top_tag_dict,top_tag_set)
         except:
             logging.error("catch exception,err_msg:%s" % traceback.format_exc())
-            return {}
+            return ({},())
 
     @classmethod
     def init_device_id_linucb_info(cls, redis_cli,redis_prefix, device_id, tag_list):

search/utils/topic.py

@@ -165,7 +165,7 @@ class TopicUtils(object):
                             "language_type": 1
                         }
                     },
-                    "weight": 4
+                    "weight": 6
                 },
                 {
                     "gauss": {
@@ -224,25 +224,25 @@ class TopicUtils(object):
                         "weight": 1
                     }
                 )
-            if len(recommend_tag_list)>0:
-                if len(recommend_tag_list)>1:
-                    functions_list += [
-                        {
-                            "filter": {"term": {"tag_list": recommend_tag_list[0]}},
-                            "weight": 4
-                        },
-                        {
-                            "filter": {"terms": {"tag_list": recommend_tag_list[1:]}},
-                            "weight": 3
-                        }
-                    ]
-                else:
-                    functions_list.append(
-                        {
-                            "filter": {"terms": {"tag_list": recommend_tag_list}},
-                            "weight": 3
-                        }
-                    )
+            # if len(recommend_tag_list)>0:
+            #     if len(recommend_tag_list)>1:
+            #         functions_list += [
+            #             {
+            #                 "filter": {"term": {"tag_list": recommend_tag_list[0]}},
+            #                 "weight": 4
+            #             },
+            #             {
+            #                 "filter": {"terms": {"tag_list": recommend_tag_list[1:]}},
+            #                 "weight": 3
+            #             }
+            #         ]
+            #     else:
+            #         functions_list.append(
+            #             {
+            #                 "filter": {"terms": {"tag_list": recommend_tag_list}},
+            #                 "weight": 3
+            #             }
+            #         )
 
             # for tag_id in recommend_tag_dict:
             #     functions_list.append(
@@ -252,13 +252,13 @@ class TopicUtils(object):
             #         }
             #     )
 
-            low_content_level = 4 if query_type == TopicPageType.FIND_PAGE else 3
+            # low_content_level = 4 if query_type == TopicPageType.FIND_PAGE else 3
             query_function_score = {
                 "query": {
                     "bool": {
                         "filter": [
                             # {"range": {"content_level": {"gte": low_content_level, "lte": 5}}},
-                            {"term": {"has_image":True}},
+                            # {"term": {"has_image":True}},
                             {"term": {"is_online": True}},
                             {"term": {"is_deleted": False}}
                         ],

search/views/topic.py

@@ -46,11 +46,14 @@ def get_home_recommend_topic_ids(user_id, device_id, tag_id, offset, size, query
                                  query_type=TopicPageType.HOME_RECOMMEND):
     try:
         if query is None:
-            redis_key = "physical:home_recommend" + ":user_id:" + str(
-                user_id) + ":device_id:" + device_id + ":query_type:" + str(query_type)
+            # redis_key = "physical:home_recommend" + ":user_id:" + str(
+            #     user_id) + ":device_id:" + device_id + ":query_type:" + str(query_type)
+            redis_key = "physical:home_recommend" + ":device_id:" + device_id + ":query_type:" + str(query_type)
         else:
-            redis_key = "physical:home_query" + ":user_id:" + str(
-                user_id) + ":device_id:" + device_id + ":query:" + str(query) + ":query_type:" + str(query_type)
+            # redis_key = "physical:home_query" + ":user_id:" + str(
+            #     user_id) + ":device_id:" + device_id + ":query:" + str(query) + ":query_type:" + str(query_type)
+            redis_key = "physical:home_query" + ":device_id:" + device_id + ":query:" + str(query) + ":query_type:" + str(query_type)
+
 
         redis_field_list = [b'have_read_topic_list']
         redis_field_val_list = redis_client.hmget(redis_key, redis_field_list)
@@ -63,12 +66,14 @@ def get_home_recommend_topic_ids(user_id, device_id, tag_id, offset, size, query
         # if tag_recommend_val:
         #     recommend_tag_dict = json.loads(str(tag_recommend_val, encoding="utf-8"))
 
-        recommend_tag_list = list()
-        tag_recommend_val = redis_client.get(tag_recommend_redis_key)
-        if tag_recommend_val:
-            recommend_tag_list = json.loads(str(tag_recommend_val, encoding="utf-8"))
+        recommend_topic_list=list()
+        recommend_tag_dict = redis_client.hgetall(tag_recommend_redis_key)
+        if "data" in recommend_tag_dict:
+            recommend_topic_id_list = json.loads(recommend_tag_dict["data"])
+            cursor = recommend_tag_dict["cursor"]
+            if len(recommend_topic_id_list) > (cursor+5):
+                recommend_topic_list = recommend_topic_id_list[cursor:cursor+5]
 
-        recommend_topic_ids = []
         have_read_topic_id_list = list()
 
         if redis_field_val_list[0]:
@@ -83,10 +88,11 @@ def get_home_recommend_topic_ids(user_id, device_id, tag_id, offset, size, query
         user_similar_score_redis_list = json.loads(
             redis_user_similar_score_redis_val) if redis_user_similar_score_redis_val else []
 
+        size = size-len(recommend_topic_list)
         topic_id_list = TopicUtils.get_recommend_topic_ids(user_id=user_id, tag_id=tag_id, offset=offset, size=size,
                                                            single_size=size,query=query, query_type=query_type,
                                                            filter_topic_id_list=have_read_topic_id_list,
-                                                           recommend_tag_list=recommend_tag_list,
+                                                           recommend_tag_list=recommend_topic_list,
                                                            user_similar_score_list=user_similar_score_redis_list,index_type="topic-high-star")
         have_read_group_id_set = set()
         have_read_user_id_set = set()
@@ -128,6 +134,7 @@ def get_home_recommend_topic_ids(user_id, device_id, tag_id, offset, size, query
         #         else:
         #             break
 
+        topic_id_list = recommend_topic_list.extend(topic_id_list)
         have_read_topic_id_list.extend(topic_id_list)
         if len(have_read_topic_id_list) > 30000:
             cut_len = len(have_read_topic_id_list)-30000

trans2es/mapping/topic.json

@@ -11,6 +11,7 @@
     "content":{"type":"text","analyzer":"gm_default_index","search_analyzer":"gm_default_index"},
     "content_level":{"type":"text"},
     "user_id":{"type":"long"},
+    "user_nick_name":{"type":"text","analyzer":"gm_default_index","search_analyzer":"gm_default_index"},//帖子用户名
     "group_id":{"type":"long"}, //所在组ID
     "tag_list":{"type":"long"},//标签属性
     "edit_tag_list":{"type":"long"},//编辑标签