BruceSun
/
MaiBot
mirror of https://github.com/Mai-with-u/MaiBot.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

🤖 自动格式化代码 [skip ci]

pull/789/head
github-actions[bot] 2025-04-18 04:34:37 +00:00
parent bf7827a571
commit 3c95813c90
18 changed files with 119 additions and 236 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

25
import_openie.py
View File

@ -20,6 +20,7 @@ import sys
logger = get_module_logger("LPMM知识库-OpenIE导入")
def hash_deduplicate(
raw_paragraphs: Dict[str, str],
triple_list_data: Dict[str, List[List[str]]],
@ -43,14 +44,10 @@ def hash_deduplicate(
# 保存去重后的三元组
new_triple_list_data = dict()
for _, (raw_paragraph, triple_list) in enumerate(
zip(raw_paragraphs.values(), triple_list_data.values())
):
for _, (raw_paragraph, triple_list) in enumerate(zip(raw_paragraphs.values(), triple_list_data.values())):
# 段落hash
paragraph_hash = get_sha256(raw_paragraph)
if ((PG_NAMESPACE + "-" + paragraph_hash) in stored_pg_hashes) and (
paragraph_hash in stored_paragraph_hashes
):
if ((PG_NAMESPACE + "-" + paragraph_hash) in stored_pg_hashes) and (paragraph_hash in stored_paragraph_hashes):
continue
new_raw_paragraphs[paragraph_hash] = raw_paragraph
new_triple_list_data[paragraph_hash] = triple_list
@ -58,9 +55,7 @@ def hash_deduplicate(
return new_raw_paragraphs, new_triple_list_data
def handle_import_openie(
openie_data: OpenIE, embed_manager: EmbeddingManager, kg_manager: KGManager
) -> bool:
def handle_import_openie(openie_data: OpenIE, embed_manager: EmbeddingManager, kg_manager: KGManager) -> bool:
# 从OpenIE数据中提取段落原文与三元组列表
# 索引的段落原文
raw_paragraphs = openie_data.extract_raw_paragraph_dict()
@ -68,9 +63,7 @@ def handle_import_openie(
entity_list_data = openie_data.extract_entity_dict()
# 索引的三元组列表
triple_list_data = openie_data.extract_triple_dict()
if len(raw_paragraphs) != len(entity_list_data) or len(raw_paragraphs) != len(
triple_list_data
):
if len(raw_paragraphs) != len(entity_list_data) or len(raw_paragraphs) != len(triple_list_data):
logger.error("OpenIE数据存在异常")
return False
# 将索引换为对应段落的hash值
@ -112,11 +105,11 @@ def main():
print("知识导入时，会消耗大量系统资源，建议在较好配置电脑上运行")
print("同上样例，导入时10700K几乎跑满，14900HX占用80%，峰值内存占用约3G")
confirm = input("确认继续执行？(y/n): ").strip().lower()
if confirm != 'y':
if confirm != "y":
logger.info("用户取消操作")
print("操作已取消")
sys.exit(1)
print("\n" + "="*40 + "\n")
print("\n" + "=" * 40 + "\n")
logger.info("----开始导入openie数据----\n")
@ -129,9 +122,7 @@ def main():
)
# 初始化Embedding库
embed_manager = embed_manager = EmbeddingManager(
llm_client_list[global_config["embedding"]["provider"]]
)
embed_manager = embed_manager = EmbeddingManager(llm_client_list[global_config["embedding"]["provider"]])
logger.info("正在从文件加载Embedding库")
try:
embed_manager.load_from_file()

4
info_extraction.py
View File

@ -91,11 +91,11 @@ def main():
print("或者使用可以用赠金抵扣的Pro模型")
print("请确保账户余额充足，并且在执行前确认无误。")
confirm = input("确认继续执行？(y/n): ").strip().lower()
if confirm != 'y':
if confirm != "y":
logger.info("用户取消操作")
print("操作已取消")
sys.exit(1)
print("\n" + "="*40 + "\n")
print("\n" + "=" * 40 + "\n")
logger.info("--------进行信息提取--------\n")

33
raw_data_preprocessor.py
View File

@ -3,18 +3,17 @@ import os
from pathlib import Path
import sys # 新增系统模块导入
def check_and_create_dirs():
"""检查并创建必要的目录"""
required_dirs = [
"data/lpmm_raw_data",
"data/imported_lpmm_data"
]
required_dirs = ["data/lpmm_raw_data", "data/imported_lpmm_data"]
for dir_path in required_dirs:
if not os.path.exists(dir_path):
os.makedirs(dir_path)
print(f"已创建目录: {dir_path}")
def process_text_file(file_path):
"""处理单个文本文件，返回段落列表"""
with open(file_path, "r", encoding="utf-8") as f:
@ -29,12 +28,13 @@ def process_text_file(file_path):
paragraph = ""
else:
paragraph += line + "\n"
if paragraph != "":
paragraphs.append(paragraph.strip())
return paragraphs
def main():
# 新增用户确认提示
print("=== 重要操作确认 ===")
@ -43,42 +43,43 @@ def main():
print("在进行知识库导入之前")
print("请修改config/lpmm_config.toml中的配置项")
confirm = input("确认继续执行？(y/n): ").strip().lower()
if confirm != 'y':
if confirm != "y":
print("操作已取消")
sys.exit(1)
print("\n" + "="*40 + "\n")
print("\n" + "=" * 40 + "\n")
# 检查并创建必要的目录
check_and_create_dirs()
# 检查输出文件是否存在
if os.path.exists("data/import.json"):
print("错误: data/import.json 已存在，请先处理或删除该文件")
sys.exit(1)
if os.path.exists("data/openie.json"):
print("错误: data/openie.json 已存在，请先处理或删除该文件")
sys.exit(1)
# 获取所有原始文本文件
raw_files = list(Path("data/lpmm_raw_data").glob("*.txt"))
if not raw_files:
print("警告: data/lpmm_raw_data 中没有找到任何 .txt 文件")
sys.exit(1)
# 处理所有文件
all_paragraphs = []
for file in raw_files:
print(f"正在处理文件: {file.name}")
paragraphs = process_text_file(file)
all_paragraphs.extend(paragraphs)
# 保存合并后的结果
output_path = "data/import.json"
with open(output_path, "w", encoding="utf-8") as f:
json.dump(all_paragraphs, f, ensure_ascii=False, indent=4)
print(f"处理完成，结果已保存到: {output_path}")
if __name__ == "__main__":
main()
main()

1
src/do_tool/tool_can_use/lpmm_get_knowledge.py
View File

@ -1,5 +1,6 @@
from src.do_tool.tool_can_use.base_tool import BaseTool
from src.plugins.chat.utils import get_embedding
# from src.common.database import db
from src.common.logger import get_module_logger
from typing import Dict, Any

5
src/plugins/knowledge/knowledge_lib.py
View File

@ -22,9 +22,7 @@ for key in global_config["llm_providers"]:
)
# 初始化Embedding库
embed_manager = EmbeddingManager(
llm_client_list[global_config["embedding"]["provider"]]
)
embed_manager = EmbeddingManager(llm_client_list[global_config["embedding"]["provider"]])
logger.info("正在从文件加载Embedding库")
try:
embed_manager.load_from_file()
@ -63,4 +61,3 @@ inspire_manager = MemoryActiveManager(
embed_manager,
llm_client_list[global_config["embedding"]["provider"]],
)

4
src/plugins/knowledge/lib/quick_algo/__init__.py
View File

@ -23,9 +23,7 @@ def _nx_graph_to_lists(
A tuple containing the list of edges and the list of nodes.
"""
nodes = [node for node in graph.nodes()]
edges = [
(u, v, graph.get_edge_data(u, v).get("weight", 0.0)) for u, v in graph.edges()
]
edges = [(u, v, graph.get_edge_data(u, v).get("weight", 0.0)) for u, v in graph.edges()]
return edges, nodes

66
src/plugins/knowledge/lib/quick_algo/pagerank_py.py
View File

@ -9,7 +9,7 @@ def pagerank_py(
personalization: Optional[Dict[str, float]] = None,
alpha: float = 0.85,
max_iter: int = 100,
tol: float = 1e-6
tol: float = 1e-6,
) -> Dict[str, float]:
"""使用 Python、NumPy 和 SciPy 计算个性化 PageRank。
@ -44,14 +44,14 @@ def pagerank_py(
raw_values = np.maximum(raw_values, 0)
norm_sum = np.sum(raw_values)
if norm_sum > 1e-9: # 避免除以零
if norm_sum > 1e-9: # 避免除以零
personalization_vec = raw_values / norm_sum
else:
# 如果所有提供的个性化值都为零或负数，则回退到均匀分布
print("警告：个性化值总和为零或所有值均为非正数。回退到均匀个性化设置。")
personalization_vec.fill(1.0 / num_nodes)
# --- 构建稀疏邻接矩阵 ---
# --- 构建稀疏邻接矩阵 ---
# 标准 PageRank 需要基于出度的归一化
row_ind = []
col_ind = []
@ -66,9 +66,9 @@ def pagerank_py(
col_ind.append(src_idx)
# 暂存原始权重，如果需要加权 PageRank，可以在此使用 w
# 对于标准 PageRank，我们只需要知道连接存在
data.append(1.0) # 初始数据设为 1，之后归一化
data.append(1.0) # 初始数据设为 1，之后归一化
# 标准 PageRank 的出度是边的数量，加权 PageRank 可以用 w
out_degree[src_idx] += 1
out_degree[src_idx] += 1
# 归一化权重（构建转移矩阵 M 的转置 M.T）
# M[j, i] 是从 i 到 j 的概率
@ -82,17 +82,16 @@ def pagerank_py(
if out_degree[c] > 0:
# 标准 PageRank: 1.0 / out_degree[c]
# 如果要用原始权重 w 作为转移概率（需确保它们已归一化），则用 w / sum(w for edges from c)
normalized_data.append(d / out_degree[c])
new_row_ind.append(c) # M.T 的行索引是 src_idx
new_col_ind.append(r) # M.T 的列索引是 dst_idx
normalized_data.append(d / out_degree[c])
new_row_ind.append(c) # M.T 的行索引是 src_idx
new_col_ind.append(r) # M.T 的列索引是 dst_idx
# 创建稀疏矩阵 (M.T)
# 注意：scipy.sparse 期望 (data, (row_ind, col_ind)) 格式
# 这里构建的是 M 的转置，方便后续计算 scores = alpha * M.T @ scores + ...
if len(normalized_data) > 0:
# 使用 csc_matrix 以便高效地进行列操作（矩阵向量乘法）
M_T = sp.csc_matrix((normalized_data, (new_row_ind, new_col_ind)),
shape=(num_nodes, num_nodes))
M_T = sp.csc_matrix((normalized_data, (new_row_ind, new_col_ind)), shape=(num_nodes, num_nodes))
else:
M_T = sp.csc_matrix((num_nodes, num_nodes))
@ -109,44 +108,45 @@ def pagerank_py(
# 还有一种做法是仅分配给个性化向量中非零的节点
# --- PageRank 迭代 ---
scores = personalization_vec.copy() # 从个性化向量开始
scores = personalization_vec.copy() # 从个性化向量开始
for iteration in range(max_iter):
prev_scores = scores.copy()
# 计算来自链接的贡献
linked_scores = M_T @ scores
# 计算来自悬挂节点的贡献
# 悬挂节点的总分数 * 悬挂权重向量
dangling_sum = np.sum(scores[is_dangling])
dangling_contribution = dangling_sum * dangling_weights
# 结合瞬移、链接贡献和悬挂节点贡献
scores = alpha * (linked_scores + dangling_contribution) + (1 - alpha) * personalization_vec
# 检查收敛性 (L1 范数)
diff = np.sum(np.abs(scores - prev_scores))
if diff < tol:
print(f"在 {iteration + 1} 次迭代后收敛。")
break
else: # 循环完成但未中断
else: # 循环完成但未中断
print(f"达到最大迭代次数 ({max_iter}) 但未收敛。")
# --- 格式化输出 ---
result_dict = {index_to_node[i]: scores[i] for i in range(num_nodes)}
return result_dict
# --- 示例用法（类似于 pr.c 中的 main）---
if __name__ == "__main__":
nodes_test = ["0", "1", "2", "3", "4"]
edges_test = [
("0", "1", 0.5), # 权重在此实现中仅用于确定出度
("0", "1", 0.5), # 权重在此实现中仅用于确定出度
("1", "2", 0.3),
("2", "0", 0.2),
("1", "3", 0.4),
("3", "4", 0.6),
("4", "1", 0.7)
("4", "1", 0.7),
]
# 添加一个悬挂节点示例
nodes_test.append("5")
@ -161,37 +161,33 @@ if __name__ == "__main__":
print("运行优化的 Python PageRank 实现...")
result = pagerank_py(
nodes_test,
edges_test,
personalization_test,
alpha=alpha_test,
max_iter=max_iter_test,
tol=tol_test
nodes_test, edges_test, personalization_test, alpha=alpha_test, max_iter=max_iter_test, tol=tol_test
)
print("\nPageRank 分数:")
# 按节点索引排序以获得一致的输出
sorted_nodes = sorted(result.keys(), key=lambda x: int(x))
for node_id in sorted_nodes:
print(f"节点 {node_id}: {result[node_id]:.6f}")
print(f"节点 {node_id}: {result[node_id]:.6f}")
print("\n使用默认个性化设置运行...")
result_default_pers = pagerank_py(
nodes_test,
edges_test,
personalization=None, # 使用默认的统一性化设置
personalization=None, # 使用默认的统一性化设置
alpha=alpha_test,
max_iter=max_iter_test,
tol=tol_test
tol=tol_test,
)
print("\nPageRank 分数（默认个性化）:")
sorted_nodes_default = sorted(result_default_pers.keys(), key=lambda x: int(x))
for node_id in sorted_nodes_default:
print(f"节点 {node_id}: {result_default_pers[node_id]:.6f}")
print(f"节点 {node_id}: {result_default_pers[node_id]:.6f}")
# 与 NetworkX 对比 (如果安装了)
try:
import networkx as nx
print("\n与 NetworkX PageRank 对比 (个性化)...")
G = nx.DiGraph()
G.add_nodes_from(nodes_test)
@ -200,25 +196,29 @@ if __name__ == "__main__":
# 为了更接近我们的实现，我们不传递权重给 add_edges_from
edges_for_nx = [(u, v) for u, v, w in edges_test]
G.add_edges_from(edges_for_nx)
# 归一化 NetworkX 的个性化向量
nx_pers = {node: personalization_test.get(node, 0.0) for node in nodes_test}
pers_sum = sum(nx_pers.values())
if pers_sum > 0:
nx_pers = {k: v / pers_sum for k, v in nx_pers.items()}
else: # 如果全为0，NetworkX 会报错或行为未定义，我们设为 None
else: # 如果全为0，NetworkX 会报错或行为未定义，我们设为 None
nx_pers = None
nx_result = nx.pagerank(G, alpha=alpha_test, personalization=nx_pers, max_iter=max_iter_test, tol=tol_test, weight=None) # weight=None 强制标准 PageRank
nx_result = nx.pagerank(
G, alpha=alpha_test, personalization=nx_pers, max_iter=max_iter_test, tol=tol_test, weight=None
) # weight=None 强制标准 PageRank
for node_id in sorted_nodes:
print(f"节点 {node_id}: {nx_result.get(node_id, 0.0):.6f}")
print("\n与 NetworkX PageRank 对比 (默认)...")
nx_result_default = nx.pagerank(G, alpha=alpha_test, personalization=None, max_iter=max_iter_test, tol=tol_test, weight=None)
nx_result_default = nx.pagerank(
G, alpha=alpha_test, personalization=None, max_iter=max_iter_test, tol=tol_test, weight=None
)
for node_id in sorted_nodes_default:
print(f"节点 {node_id}: {nx_result_default.get(node_id, 0.0):.6f}")
except ImportError:
print("\n未安装 NetworkX，跳过对比。")
except Exception as e:
print(f"\n运行 NetworkX PageRank 时出错: {e}")
print(f"\n运行 NetworkX PageRank 时出错: {e}")

28
src/plugins/knowledge/src/embedding_store.py
View File

@ -47,9 +47,7 @@ class EmbeddingStore:
self.idx2hash = None
def _get_embedding(self, s: str) -> List[float]:
return self.llm_client.send_embedding_request(
global_config["embedding"]["model"], s
)
return self.llm_client.send_embedding_request(global_config["embedding"]["model"], s)
def batch_insert_strs(self, strs: List[str]) -> None:
"""向库中存入字符串"""
@ -83,14 +81,10 @@ class EmbeddingStore:
logger.info(f"{self.namespace}嵌入库保存成功")
if self.faiss_index is not None and self.idx2hash is not None:
logger.info(
f"正在保存{self.namespace}嵌入库的FaissIndex到文件{self.index_file_path}"
)
logger.info(f"正在保存{self.namespace}嵌入库的FaissIndex到文件{self.index_file_path}")
faiss.write_index(self.faiss_index, self.index_file_path)
logger.info(f"{self.namespace}嵌入库的FaissIndex保存成功")
logger.info(
f"正在保存{self.namespace}嵌入库的idx2hash映射到文件{self.idx2hash_file_path}"
)
logger.info(f"正在保存{self.namespace}嵌入库的idx2hash映射到文件{self.idx2hash_file_path}")
with open(self.idx2hash_file_path, "w", encoding="utf-8") as f:
f.write(json.dumps(self.idx2hash, ensure_ascii=False, indent=4))
logger.info(f"{self.namespace}嵌入库的idx2hash映射保存成功")
@ -103,24 +97,18 @@ class EmbeddingStore:
logger.info(f"正在从文件{self.embedding_file_path}中加载{self.namespace}嵌入库")
data_frame = pd.read_parquet(self.embedding_file_path, engine="pyarrow")
for _, row in tqdm.tqdm(data_frame.iterrows(), total=len(data_frame)):
self.store[row["hash"]] = EmbeddingStoreItem(
row["hash"], row["embedding"], row["str"]
)
self.store[row["hash"]] = EmbeddingStoreItem(row["hash"], row["embedding"], row["str"])
logger.info(f"{self.namespace}嵌入库加载成功")
try:
if os.path.exists(self.index_file_path):
logger.info(
f"正在从文件{self.index_file_path}中加载{self.namespace}嵌入库的FaissIndex"
)
logger.info(f"正在从文件{self.index_file_path}中加载{self.namespace}嵌入库的FaissIndex")
self.faiss_index = faiss.read_index(self.index_file_path)
logger.info(f"{self.namespace}嵌入库的FaissIndex加载成功")
else:
raise Exception(f"文件{self.index_file_path}不存在")
if os.path.exists(self.idx2hash_file_path):
logger.info(
f"正在从文件{self.idx2hash_file_path}中加载{self.namespace}嵌入库的idx2hash映射"
)
logger.info(f"正在从文件{self.idx2hash_file_path}中加载{self.namespace}嵌入库的idx2hash映射")
with open(self.idx2hash_file_path, "r") as f:
self.idx2hash = json.load(f)
logger.info(f"{self.namespace}嵌入库的idx2hash映射加载成功")
@ -215,9 +203,7 @@ class EmbeddingManager:
for triples in triple_list_data.values():
graph_triples.extend([tuple(t) for t in triples])
graph_triples = list(set(graph_triples))
self.relation_embedding_store.batch_insert_strs(
[str(triple) for triple in graph_triples]
)
self.relation_embedding_store.batch_insert_strs([str(triple) for triple in graph_triples])
def load_from_file(self):
"""从文件加载"""

4
src/plugins/knowledge/src/global_logger.py
View File

@ -7,8 +7,6 @@ logger = logging.getLogger(__name__)
logger.setLevel(logging.DEBUG)
console_logging_handler = logging.StreamHandler()
console_logging_handler.setFormatter(
logging.Formatter("%(asctime)s - %(levelname)s - %(message)s")
)
console_logging_handler.setFormatter(logging.Formatter("%(asctime)s - %(levelname)s - %(message)s"))
console_logging_handler.setLevel(logging.DEBUG)
logger.addHandler(console_logging_handler)

18
src/plugins/knowledge/src/ie_process.py
View File

@ -38,16 +38,12 @@ def _entity_extract(llm_client: LLMClient, paragraph: str) -> List[str]:
return entity_extract_result
def _rdf_triple_extract(
llm_client: LLMClient, paragraph: str, entities: list
) -> List[List[str]]:
def _rdf_triple_extract(llm_client: LLMClient, paragraph: str, entities: list) -> List[List[str]]:
"""对段落进行实体提取，返回提取出的实体列表（JSON格式）"""
entity_extract_context = prompt_template.build_rdf_triple_extract_context(
paragraph, entities=json.dumps(entities, ensure_ascii=False)
)
_, request_result = llm_client.send_chat_request(
global_config["rdf_build"]["llm"]["model"], entity_extract_context
)
_, request_result = llm_client.send_chat_request(global_config["rdf_build"]["llm"]["model"], entity_extract_context)
# 去除‘{’前的内容（结果中可能有多个‘{’）
if "[" in request_result:
@ -60,11 +56,7 @@ def _rdf_triple_extract(
entity_extract_result = json.loads(fix_broken_generated_json(request_result))
for triple in entity_extract_result:
if (
len(triple) != 3
or (triple[0] is None or triple[1] is None or triple[2] is None)
or "" in triple
):
if len(triple) != 3 or (triple[0] is None or triple[1] is None or triple[2] is None) or "" in triple:
raise Exception("RDF提取结果格式错误")
return entity_extract_result
@ -91,9 +83,7 @@ def info_extract_from_str(
try_count = 0
while True:
try:
rdf_triple_extract_result = _rdf_triple_extract(
llm_client_for_rdf, paragraph, entity_extract_result
)
rdf_triple_extract_result = _rdf_triple_extract(llm_client_for_rdf, paragraph, entity_extract_result)
break
except Exception as e:
logger.warning(f"实体提取失败，错误信息：{e}")

78
src/plugins/knowledge/src/kg_manager.py
View File

@ -22,6 +22,7 @@ from .lpmmconfig import (
from .global_logger import logger
class KGManager:
def __init__(self):
# 会被保存的字段
@ -35,9 +36,7 @@ class KGManager:
# 持久化相关
self.dir_path = global_config["persistence"]["rag_data_dir"]
self.graph_data_path = self.dir_path + "/" + RAG_GRAPH_NAMESPACE + ".graphmlz"
self.ent_cnt_data_path = (
self.dir_path + "/" + RAG_ENT_CNT_NAMESPACE + ".parquet"
)
self.ent_cnt_data_path = self.dir_path + "/" + RAG_ENT_CNT_NAMESPACE + ".parquet"
self.pg_hash_file_path = self.dir_path + "/" + RAG_PG_HASH_NAMESPACE + ".json"
def save_to_file(self):
@ -50,9 +49,7 @@ class KGManager:
nx.write_graphml(self.graph, path=self.graph_data_path, encoding="utf-8")
# 保存实体计数到文件
ent_cnt_df = pd.DataFrame(
[{"hash_key": k, "appear_cnt": v} for k, v in self.ent_appear_cnt.items()]
)
ent_cnt_df = pd.DataFrame([{"hash_key": k, "appear_cnt": v} for k, v in self.ent_appear_cnt.items()])
ent_cnt_df.to_parquet(self.ent_cnt_data_path, engine="pyarrow", index=False)
# 保存段落hash到文件
@ -77,9 +74,7 @@ class KGManager:
# 加载实体计数
ent_cnt_df = pd.read_parquet(self.ent_cnt_data_path, engine="pyarrow")
self.ent_appear_cnt = dict(
{row["hash_key"]: row["appear_cnt"] for _, row in ent_cnt_df.iterrows()}
)
self.ent_appear_cnt = dict({row["hash_key"]: row["appear_cnt"] for _, row in ent_cnt_df.iterrows()})
# 加载KG
self.graph = nx.read_graphml(self.graph_data_path)
@ -101,20 +96,14 @@ class KGManager:
# 一个triple就是一条边（同时构建双向联系）
hash_key1 = ENT_NAMESPACE + "-" + get_sha256(triple[0])
hash_key2 = ENT_NAMESPACE + "-" + get_sha256(triple[2])
node_to_node[(hash_key1, hash_key2)] = (
node_to_node.get((hash_key1, hash_key2), 0) + 1.0
)
node_to_node[(hash_key2, hash_key1)] = (
node_to_node.get((hash_key2, hash_key1), 0) + 1.0
)
node_to_node[(hash_key1, hash_key2)] = node_to_node.get((hash_key1, hash_key2), 0) + 1.0
node_to_node[(hash_key2, hash_key1)] = node_to_node.get((hash_key2, hash_key1), 0) + 1.0
entity_set.add(hash_key1)
entity_set.add(hash_key2)
# 实体出现次数统计
for hash_key in entity_set:
self.ent_appear_cnt[hash_key] = (
self.ent_appear_cnt.get(hash_key, 0) + 1.0
)
self.ent_appear_cnt[hash_key] = self.ent_appear_cnt.get(hash_key, 0) + 1.0
@staticmethod
def _build_edges_between_ent_pg(
@ -126,9 +115,7 @@ class KGManager:
for triple in triple_list_data[idx]:
ent_hash_key = ENT_NAMESPACE + "-" + get_sha256(triple[0])
pg_hash_key = PG_NAMESPACE + "-" + str(idx)
node_to_node[(ent_hash_key, pg_hash_key)] = (
node_to_node.get((ent_hash_key, pg_hash_key), 0) + 1.0
)
node_to_node[(ent_hash_key, pg_hash_key)] = node_to_node.get((ent_hash_key, pg_hash_key), 0) + 1.0
@staticmethod
def _synonym_connect(
@ -177,9 +164,7 @@ class KGManager:
new_edge_cnt += 1
res_ent.append(
(
embedding_manager.entities_embedding_store.store[
res_ent_hash
].str,
embedding_manager.entities_embedding_store.store[res_ent_hash].str,
similarity,
)
) # Debug
@ -236,22 +221,16 @@ class KGManager:
if node_hash not in existed_nodes:
if node_hash.startswith(ENT_NAMESPACE):
# 新增实体节点
node = embedding_manager.entities_embedding_store.store[
node_hash
]
node = embedding_manager.entities_embedding_store.store[node_hash]
assert isinstance(node, EmbeddingStoreItem)
self.graph.nodes[node_hash]["content"] = node.str
self.graph.nodes[node_hash]["type"] = "ent"
elif node_hash.startswith(PG_NAMESPACE):
# 新增文段节点
node = embedding_manager.paragraphs_embedding_store.store[
node_hash
]
node = embedding_manager.paragraphs_embedding_store.store[node_hash]
assert isinstance(node, EmbeddingStoreItem)
content = node.str.replace("\n", " ")
self.graph.nodes[node_hash]["content"] = (
content if len(content) < 8 else content[:8] + "..."
)
self.graph.nodes[node_hash]["content"] = content if len(content) < 8 else content[:8] + "..."
self.graph.nodes[node_hash]["type"] = "pg"
def build_kg(
@ -319,9 +298,7 @@ class KGManager:
ent_sim_scores = {}
for relation_hash, similarity, _ in relation_search_result:
# 提取主宾短语
relation = embed_manager.relation_embedding_store.store.get(
relation_hash
).str
relation = embed_manager.relation_embedding_store.store.get(relation_hash).str
assert relation is not None # 断言：relation不为空
# 关系三元组
triple = relation[2:-2].split("', '")
@ -335,9 +312,7 @@ class KGManager:
ent_mean_scores = {} # 记录实体的平均相似度
for ent_hash, scores in ent_sim_scores.items():
# 先对相似度进行累加，然后与实体计数相除获取最终权重
ent_weights[ent_hash] = (
float(np.sum(scores)) / self.ent_appear_cnt[ent_hash]
)
ent_weights[ent_hash] = float(np.sum(scores)) / self.ent_appear_cnt[ent_hash]
# 记录实体的平均相似度，用于后续的top_k筛选
ent_mean_scores[ent_hash] = float(np.mean(scores))
del ent_sim_scores
@ -354,21 +329,14 @@ class KGManager:
for ent_hash, score in ent_weights.items():
# 缩放相似度
ent_weights[ent_hash] = (
(score - ent_weights_min)
* (1 - down_edge)
/ (ent_weights_max - ent_weights_min)
(score - ent_weights_min) * (1 - down_edge) / (ent_weights_max - ent_weights_min)
) + down_edge
# 取平均相似度的top_k实体
top_k = global_config["qa"]["params"]["ent_filter_top_k"]
if len(ent_mean_scores) > top_k:
# 从大到小排序，取后len - k个
ent_mean_scores = {
k: v
for k, v in sorted(
ent_mean_scores.items(), key=lambda item: item[1], reverse=True
)
}
ent_mean_scores = {k: v for k, v in sorted(ent_mean_scores.items(), key=lambda item: item[1], reverse=True)}
for ent_hash, _ in ent_mean_scores.items():
# 删除被淘汰的实体节点权重设置
del ent_weights[ent_hash]
@ -389,9 +357,7 @@ class KGManager:
# 归一化
for pg_hash, similarity in pg_sim_scores.items():
# 归一化相似度
pg_sim_scores[pg_hash] = (similarity - pg_sim_score_min) / (
pg_sim_score_max - pg_sim_score_min
)
pg_sim_scores[pg_hash] = (similarity - pg_sim_score_min) / (pg_sim_score_max - pg_sim_score_min)
del pg_sim_score_max, pg_sim_score_min
for pg_hash, score in pg_sim_scores.items():
@ -401,9 +367,7 @@ class KGManager:
del pg_sim_scores
# 最终权重数据 = 实体权重 + 文段权重
ppr_node_weights = {
k: v for d in [ent_weights, pg_weights] for k, v in d.items()
}
ppr_node_weights = {k: v for d in [ent_weights, pg_weights] for k, v in d.items()}
del ent_weights, pg_weights
# PersonalizedPageRank
@ -418,14 +382,12 @@ class KGManager:
# 从搜索结果中提取文段节点的结果
passage_node_res = [
(node_key, score)
for node_key, score in ppr_res.items() # Iterate over dictionary items
for node_key, score in ppr_res.items() # Iterate over dictionary items
if node_key.startswith(PG_NAMESPACE)
]
del ppr_res
# 排序：按照分数从大到小
passage_node_res = sorted(
passage_node_res, key=lambda item: item[1], reverse=True
)
passage_node_res = sorted(passage_node_res, key=lambda item: item[1], reverse=True)
return passage_node_res, ppr_node_weights

14
src/plugins/knowledge/src/llm_client.py
View File

@ -21,20 +21,14 @@ class LLMClient:
def send_chat_request(self, model, messages):
"""发送对话请求，等待返回结果"""
response = self.client.chat.completions.create(
model=model, messages=messages, stream=False
)
response = self.client.chat.completions.create(model=model, messages=messages, stream=False)
if hasattr(response.choices[0].message, "reasoning_content"):
# 有单独的推理内容块
reasoning_content = response.choices[0].message.reasoning_content
content = response.choices[0].message.content
else:
# 无单独的推理内容块
response = (
response.choices[0]
.message.content.split("<think>")[-1]
.split("</think>")
)
response = response.choices[0].message.content.split("<think>")[-1].split("</think>")
# 如果有推理内容，则分割推理内容和内容
if len(response) == 2:
reasoning_content = response[0]
@ -48,6 +42,4 @@ class LLMClient:
def send_embedding_request(self, model, text):
"""发送嵌入请求，等待返回结果"""
text = text.replace("\n", " ")
return (
self.client.embeddings.create(input=[text], model=model).data[0].embedding
)
return self.client.embeddings.create(input=[text], model=model).data[0].embedding

8
src/plugins/knowledge/src/mem_active_manager.py
View File

@ -16,13 +16,9 @@ class MemoryActiveManager:
def get_activation(self, question: str) -> float:
"""获取记忆激活度"""
# 生成问题的Embedding
question_embedding = self.embedding_client.send_embedding_request(
"text-embedding", question
)
question_embedding = self.embedding_client.send_embedding_request("text-embedding", question)
# 查询关系库中的相似度
rel_search_res = self.embed_manager.relation_embedding_store.search_top_k(
question_embedding, 10
)
rel_search_res = self.embed_manager.relation_embedding_store.search_top_k(question_embedding, 10)
# 动态过滤阈值
rel_scores = dyn_select_top_k(rel_search_res, 0.5, 1.0)

23
src/plugins/knowledge/src/open_ie.py
View File

@ -9,12 +9,7 @@ def _filter_invalid_entities(entities: List[str]) -> List[str]:
"""过滤无效的实体"""
valid_entities = set()
for entity in entities:
if (
not isinstance(entity, str)
or entity.strip() == ""
or entity in INVALID_ENTITY
or entity in valid_entities
):
if not isinstance(entity, str) or entity.strip() == "" or entity in INVALID_ENTITY or entity in valid_entities:
# 非字符串/空字符串/在无效实体列表中/重复
continue
valid_entities.add(entity)
@ -74,9 +69,7 @@ class OpenIE:
for doc in self.docs:
# 过滤实体列表
doc["extracted_entities"] = _filter_invalid_entities(
doc["extracted_entities"]
)
doc["extracted_entities"] = _filter_invalid_entities(doc["extracted_entities"])
# 过滤无效的三元组
doc["extracted_triples"] = _filter_invalid_triples(doc["extracted_triples"])
@ -100,9 +93,7 @@ class OpenIE:
@staticmethod
def load() -> "OpenIE":
"""从文件中加载OpenIE数据"""
with open(
global_config["persistence"]["openie_data_path"], "r", encoding="utf-8"
) as f:
with open(global_config["persistence"]["openie_data_path"], "r", encoding="utf-8") as f:
data = json.loads(f.read())
openie_data = OpenIE._from_dict(data)
@ -112,9 +103,7 @@ class OpenIE:
@staticmethod
def save(openie_data: "OpenIE"):
"""保存OpenIE数据到文件"""
with open(
global_config["persistence"]["openie_data_path"], "w", encoding="utf-8"
) as f:
with open(global_config["persistence"]["openie_data_path"], "w", encoding="utf-8") as f:
f.write(json.dumps(openie_data._to_dict(), ensure_ascii=False, indent=4))
def extract_entity_dict(self):
@ -141,7 +130,5 @@ class OpenIE:
def extract_raw_paragraph_dict(self):
"""提取原始段落"""
raw_paragraph_dict = dict(
{doc_item["idx"]: doc_item["passage"] for doc_item in self.docs}
)
raw_paragraph_dict = dict({doc_item["idx"]: doc_item["passage"] for doc_item in self.docs})
return raw_paragraph_dict

27
src/plugins/knowledge/src/qa_manager.py
View File

@ -30,7 +30,7 @@ class QAManager:
"""处理查询"""
# 生成问题的Embedding
part_start_time =time.perf_counter()
part_start_time = time.perf_counter()
question_embedding = self.llm_client_list["embedding"].send_embedding_request(
global_config["embedding"]["model"], question
)
@ -38,7 +38,7 @@ class QAManager:
logger.debug(f"Embedding用时：{part_end_time - part_start_time:.5f}s")
# 根据问题Embedding查询Relation Embedding库
part_start_time =time.perf_counter()
part_start_time = time.perf_counter()
relation_search_res = self.embed_manager.relation_embedding_store.search_top_k(
question_embedding,
global_config["qa"]["params"]["relation_search_top_k"],
@ -46,10 +46,7 @@ class QAManager:
# 过滤阈值
# 考虑动态阈值：当存在显著数值差异的结果时，保留显著结果；否则，保留所有结果
relation_search_res = dyn_select_top_k(relation_search_res, 0.5, 1.0)
if (
relation_search_res[0][1]
< global_config["qa"]["params"]["relation_threshold"]
):
if relation_search_res[0][1] < global_config["qa"]["params"]["relation_threshold"]:
# 未找到相关关系
relation_search_res = []
@ -65,12 +62,10 @@ class QAManager:
# part_start_time = time.time()
# 根据问题Embedding查询Paragraph Embedding库
part_start_time =time.perf_counter()
paragraph_search_res = (
self.embed_manager.paragraphs_embedding_store.search_top_k(
question_embedding,
global_config["qa"]["params"]["paragraph_search_top_k"],
)
part_start_time = time.perf_counter()
paragraph_search_res = self.embed_manager.paragraphs_embedding_store.search_top_k(
question_embedding,
global_config["qa"]["params"]["paragraph_search_top_k"],
)
part_end_time = time.perf_counter()
logger.debug(f"文段检索用时：{part_end_time - part_start_time:.5f}s")
@ -78,7 +73,7 @@ class QAManager:
if len(relation_search_res) != 0:
logger.info("找到相关关系，将使用RAG进行检索")
# 使用KG检索
part_start_time =time.perf_counter()
part_start_time = time.perf_counter()
result, ppr_node_weights = self.kg_manager.kg_search(
relation_search_res, paragraph_search_res, self.embed_manager
)
@ -93,9 +88,7 @@ class QAManager:
result = dyn_select_top_k(result, 0.5, 1.0)
for res in result:
raw_paragraph = self.embed_manager.paragraphs_embedding_store.store[
res[0]
].str
raw_paragraph = self.embed_manager.paragraphs_embedding_store.store[res[0]].str
print(f"找到相关文段，相关系数：{res[1]:.8f}\n{raw_paragraph}\n\n")
return result, ppr_node_weights
@ -113,4 +106,4 @@ class QAManager:
for res in query_res
]
found_knowledge = "\n".join([f"{i + 1}. 相关性：{k[0]}\n{k[1]}" for i, k in enumerate(knowledge)])
return found_knowledge
return found_knowledge

4
src/plugins/knowledge/src/raw_processing.py
View File

@ -17,9 +17,7 @@ def load_raw_data() -> tuple[list[str], list[str]]:
"""
# 读取import.json文件
if os.path.exists(global_config["persistence"]["raw_data_path"]) is True:
with open(
global_config["persistence"]["raw_data_path"], "r", encoding="utf-8"
) as f:
with open(global_config["persistence"]["raw_data_path"], "r", encoding="utf-8") as f:
import_json = json.loads(f.read())
else:
raise Exception("原始数据文件读取失败")

8
src/plugins/knowledge/src/utils/dyn_topk.py
View File

@ -36,14 +36,10 @@ def dyn_select_top_k(
# 计算均值
mean_score = sum([s[2] for s in normalized_score]) / len(normalized_score)
# 计算方差
var_score = sum([(s[2] - mean_score) ** 2 for s in normalized_score]) / len(
normalized_score
)
var_score = sum([(s[2] - mean_score) ** 2 for s in normalized_score]) / len(normalized_score)
# 动态阈值
threshold = jmp_factor * jump_threshold + (1 - jmp_factor) * (
mean_score + var_factor * var_score
)
threshold = jmp_factor * jump_threshold + (1 - jmp_factor) * (mean_score + var_factor * var_score)
# 重新过滤
res = [s for s in normalized_score if s[2] > threshold]

5
src/plugins/knowledge/src/utils/json_fix.py
View File

@ -29,10 +29,7 @@ def _find_unclosed(json_str):
elif char in "{[":
unclosed.append(char)
elif char in "}]":
if unclosed and (
(char == "}" and unclosed[-1] == "{")
or (char == "]" and unclosed[-1] == "[")
):
if unclosed and ((char == "}" and unclosed[-1] == "{") or (char == "]" and unclosed[-1] == "[")):
unclosed.pop()
return unclosed