vlambda博客
学习文章列表

Task 5 Neo4j 图数据库查询


Datawhale分享地址:https://github.com/datawhalechina/team-learning-nlp/blob/master/KnowledgeGraph_Basic/task04.md原项目地址:https://github.com/zhihao-chen/QASystemOnMedicalGraph

1.Cypher查询
Cypher作为Neo4j的查询语言,是一个描述性的图形查询语言,允许不必编写图形结构的遍历代码对图形存储有表现力和效率的查询。
在前面我们构建好的图数据库中,查询症状可以输入语句:
MATCH (d:Disease)-[:HAS_SYMPTOM]->(s) WHERE d.name='糖尿病' RETURN d,s
返回结果如下:


2.代码结构
这部分功能主要是主体类 AnswerSearching实现的。其框架结构如下:
class AnswerSearching: def __init__(self): pass # 主要是根据不同的实体和意图构造cypher查询语句 def question_parser(self, data): """ 主要是根据不同的实体和意图构造cypher查询语句 :param data: {"Disease":[], "Alias":[], "Symptom":[], "Complication":[]} :return: """ pass # 将问题转变为cypher查询语句 def transfor_to_sql(self, label, entities, intent): """ 将问题转变为cypher查询语句 :param label:实体标签 :param entities:实体列表 :param intent:查询意图 :return:cypher查询语句 """ pass # 执行cypher查询,返回结果 def searching(self, sqls): """ 执行cypher查询,返回结果 :param sqls: :return:str """ pass # 根据不同意图,返回不同模板的答案 def answer_template(self, intent, answers): """ 根据不同意图,返回不同模板的答案 :param intent: 查询意图 :param answers: 知识图谱查询结果 :return: str """ Pass



3.源码介绍

首先安装py2neo,pip install py2neo
连接上neo4j数据库
from py2neo import Graph
graph=Graph("http://localhost:7474",username="neo4j", password="neo4j")
3.1.根据不同的实体和意图构造cypher查询语句

   

 def question_parser(data): """ 主要是根据不同的实体和意图构造cypher查询语句 :param data: {"Disease":[], "Alias":[], "Symptom":[], "Complication":[]} :return: """ sqls = [] if data: for intent in data["intentions"]: sql_ = {} sql_["intention"] = intent sql = [] if data.get("Disease"): sql = transfor_to_sql("Disease", data["Disease"], intent) elif data.get("Alias"): sql = transfor_to_sql("Alias", data["Alias"], intent) elif data.get("Symptom"): sql = transfor_to_sql("Symptom", data["Symptom"], intent) elif data.get("Complication"): sql = transfor_to_sql("Complication", data["Complication"], intent)

if sql: sql_['sql'] = sql sqls.append(sql_) return sql


3.2将问题转变为cypher查询语句

  

 def transfor_to_sql(label, entities, intent): """ 将问题转变为cypher查询语句 :param label:实体标签 :param entities:实体列表 :param intent:查询意图 :return:cypher查询语句 """ if not entities: return [] sql = []

# 查询症状 if intent == "query_symptom" and label == "Disease": sql = ["MATCH (d:Disease)-[:HAS_SYMPTOM]->(s) WHERE d.name='{0}' RETURN d.name,s.name".format(e) for e in entities] # 查询治疗方法 if intent == "query_cureway" and label == "Disease": sql = ["MATCH (d:Disease)-[:HAS_DRUG]->(n) WHERE d.name='{0}' return d.name,d.treatment," \ "n.name".format(e) for e in entities] # 查询治疗周期 if intent == "query_period" and label == "Disease": sql = ["MATCH (d:Disease) WHERE d.name='{0}' return d.name,d.period".format(e) for e in entities ...


6.执行cypher查询,返回结果

  

 def searching(sqls): """ 执行cypher查询,返回结果 :param sqls: :return:str """ final_answers = [] for sql_ in sqls: intent = sql_['intention'] queries = sql_['sql'] answers = [] for query in queries: ress = graph.run(query).data() answers += ress final_answer = answer_template(intent, answers) if final_answer: final_answers.append(final_answer) return final_answers


7.根据不同意图,返回不同模板的答案

   

 def answer_template(intent, answers): """ 根据不同意图,返回不同模板的答案 :param intent: 查询意图 :param answers: 知识图谱查询结果 :return: str """ final_answer = "" if not answers: return "" # 查询症状 if intent == "query_symptom": disease_dic = {} for data in answers: d = data['d.name'] s = data['s.name'] if d not in disease_dic: disease_dic[d] = [s] else: disease_dic[d].append(s) i = 0 for k, v in disease_dic.items(): if i >= 10: break final_answer += "疾病 {0} 的症状有:{1}\n".format(k, ','.join(list(set(v)))) i += 1 ...


知识图谱系列到这里就完结了,我觉得还是给了我不少启发的,大家可以想下这个在公安工作中的落地应用,应该有非常多值得研究的地方,欢迎交流!