1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
| class RAGRetriever:
"""RAG检索器 - 结合向量搜索和关键词搜索"""
def __init__(self, vector_store: VectorStore,
memory_manager: MemoryManager):
self.vector_store = vector_store
self.memory_manager = memory_manager
self.bm25_scorer = BM25Scorer()
async def retrieve_relevant(self, query: str,
top_k: int = 5,
retrieval_mode: str = "hybrid") -> List[Dict]:
"""
检索相关内容
retrieval_mode:
- "vector": 仅向量搜索
- "keyword": 仅关键词搜索
- "hybrid": 混合搜索
"""
results = []
if retrieval_mode in ["vector", "hybrid"]:
query_embedding = await self.memory_manager._generate_embedding(query)
vector_results = await self.vector_store.search(
query_embedding,
top_k=top_k * 2 if retrieval_mode == "hybrid" else top_k
)
for memory in vector_results:
results.append({
'memory': memory,
'score': 0.8,
'source': 'vector'
})
if retrieval_mode in ["keyword", "hybrid"]:
all_memories = self.memory_manager.short_term_memory + \
self.memory_manager.long_term_memory
bm25_results = self.bm25_scorer.search(query, all_memories, top_k=top_k)
for score, memory in bm25_results:
existing_ids = [r['memory'].id for r in results]
if memory.id not in existing_ids:
results.append({
'memory': memory,
'score': score,
'source': 'keyword'
})
if retrieval_mode == "hybrid":
results = self._hybrid_fusion(results, top_k)
else:
results = sorted(results, key=lambda x: x['score'], reverse=True)[:top_k]
return results
def _hybrid_fusion(self, results: List[Dict], top_k: int) -> List[Dict]:
"""混合搜索结果融合"""
k = 60
fused_scores: Dict[str, float] = {}
for result in results:
memory_id = result['memory'].id
source = result['source']
rank = results.index(result) + 1
if source == 'vector':
base_score = 0.6
else:
base_score = 0.4
rrf_score = base_score / (k + rank)
fused_scores[memory_id] = fused_scores.get(memory_id, 0) + rrf_score
fused_results = []
for result in results:
memory_id = result['memory'].id
if memory_id in fused_scores:
result['fused_score'] = fused_scores[memory_id]
result['score'] = fused_scores[memory_id]
fused_results.append(result)
return sorted(fused_results, key=lambda x: x['fused_score'], reverse=True)[:top_k]
class BM25Scorer:
"""BM25关键词评分器"""
def __init__(self, k1: float = 1.5, b: float = 0.75):
self.k1 = k1
self.b = b
self.corpus_size = 0
self.avgdl = 0
self.doc_freqs: Dict[str, int] = {}
self.idf: Dict[str, float] = {}
def fit(self, documents: List[str]):
"""构建BM25索引"""
tokenized_docs = [self._tokenize(doc) for doc in documents]
self.corpus_size = len(documents)
nd = {}
for tokens in tokenized_docs:
for token in set(tokens):
nd[token] = nd.get(token, 0) + 1
for token, freq in nd.items():
self.idf[token] = math.log(self.corpus_size - freq + 0.5) - \
math.log(freq + 0.5)
self.avgdl = sum(len(tokens) for tokens in tokenized_docs) / self.corpus_size
def _tokenize(self, text: str) -> List[str]:
"""简单分词"""
import re
text = text.lower()
tokens = re.findall(r'\b\w+\b', text)
return tokens
def search(self, query: str, memories: List[Memory], top_k: int) -> List[Tuple[float, Memory]]:
"""搜索相关记忆"""
if not self.idf:
self.fit([m.content for m in memories])
query_tokens = self._tokenize(query)
scores = []
for memory in memories:
score = self._calc_bm25(memory.content, query_tokens)
scores.append((score, memory))
return sorted(scores, key=lambda x: x[0], reverse=True)[:top_k]
def _calc_bm25(self, document: str, query_tokens: List[str]) -> float:
"""计算单文档的BM25分数"""
doc_tokens = self._tokenize(document)
doc_len = len(doc_tokens)
score = 0.0
for token in query_tokens:
if token not in self.idf:
continue
tf = doc_tokens.count(token)
if tf == 0:
continue
idf = self.idf[token]
numerator = tf * (self.k1 + 1)
denominator = tf + self.k1 * (1 - self.b + self.b * doc_len / self.avgdl)
score += idf * numerator / denominator
return score
|
