通用信息提取数据预处理

数据预处理才是最重要的，至于训练评估部署，是个人都会，别小看预处理

train_data='./datasets/duuie'

output_folder='./datasets/duuie_pre'
ignore_datasets=["DUEE", "DUEE_FIN_LITE"]
schema_folder='./datasets/seen_schema'

import shutil
import os

# shutil.rmtree(output_folder) if os.path.exists(
#         output_folder) else None
# shutil.copytree(train_data,output_folder)

life_folder = os.path.join(output_folder, "DUIE_LIFE_SPO") #人生
org_folder = os.path.join(output_folder, "DUIE_ORG_SPO") # 组织

print(life_folder,org_folder)

import json

def load_jsonlines_file(filename):
    return [json.loads(line) for line in open(filename, encoding="utf8")]

life_train_instances = load_jsonlines_file(f"{life_folder}/train.json")
org_train_instances = load_jsonlines_file(f"{org_folder}/train.json")

print(type(life_train_instances),len(life_train_instances),len(org_train_instances))

for i in range(27695,27696):
    print(life_train_instances[i],'|',org_train_instances[i])

class RecordSchema: # 格式
    def __init__(self, type_list, role_list, type_role_dict):
        self.type_list = type_list # 类型
        self.role_list = role_list # 论元角色
        self.type_role_dict = type_role_dict #类型角色字典
    def __repr__(self) -> str:
        repr_list = [f"Type: {self.type_list}\n", f"Role: {self.role_list}\n", f"Map: {self.type_role_dict}"]
        return "\n".join(repr_list)
    @staticmethod
    def get_empty_schema(): # 初始化
        return RecordSchema(type_list=list(), role_list=list(), type_role_dict=dict())
    @staticmethod
    def read_from_file(filename):
        lines = open(filename, encoding="utf8").readlines()
        type_list = json.loads(lines[0])# 类型
        role_list = json.loads(lines[1]) # 角色
        type_role_dict = json.loads(lines[2])#类型－角色
        return RecordSchema(type_list, role_list, type_role_dict)
    def write_to_file(self, filename):
        with open(filename, "w", encoding="utf8") as output:
            # 用于将Python对象编码（序列化）为JSON格式的字符串。设置ensure_ascii=False参数
            # 会告诉json.dumps()函数不要转义非ASCII字符，用dumps把json对象转换成json字符串
            output.write(json.dumps(self.type_list, ensure_ascii=False) + "\n")
            output.write(json.dumps(self.role_list, ensure_ascii=False) + "\n")
            output.write(json.dumps(self.type_role_dict, ensure_ascii=False) + "\n")

RecordSchema.read_from_file(f"{life_folder}/record.schema")

life_relation = RecordSchema.read_from_file(f"{life_folder}/record.schema").role_list #人生关系

org_relation = RecordSchema.read_from_file(f"{org_folder}/record.schema").role_list

from collections import defaultdict
instance_dict = defaultdict(list) # 543个重复的,会被替换

for instance in life_train_instances + org_train_instances:
    instance_dict[instance["text"]] += [instance]

a=[i for i in life_train_instances for j in org_train_instances if i['text']==j['text']]
b=[i for i in org_train_instances for j in a if i['text']==j['text']]
for i in range(3):
    print(a[i]['relation'],'|',b[i]['relation'])

dict_1={1:2,3:4}
for i in dict_1:#相当于字典的keys()
    print(i)

from typing import Tuple, List, Dict

def merge_instance(instance_list):
    def all_equal(_x):#判断是否全相同
        for __x in _x:
            if __x != _x[0]:
                return False
        return True
    def entity_key(_x): # 实体键
        return (tuple(_x["offset"]), _x["type"])
    def relation_key(_x): #关系键
        return (
            tuple(_x["type"]),
            tuple(_x["args"][0]["offset"]),
            _x["args"][0]["type"],
            tuple(_x["args"][1]["offset"]),
            _x["args"][1]["type"],
        )

    def event_key(_x): # 事件键
        return (tuple(_x["offset"]), _x["type"])
    assert all_equal([x["text"] for x in instance_list]) #只有实体文本text都相同才合并
    element_dict = {
        "entity": dict(),
        "relation": dict(),
        "event": dict(),
    }
    instance_id_list = list()
    for x in instance_list:
        instance_id_list += [x["id"]] #同样的文本，实体id的集合
        for entity in x.get("entity", list()):
            element_dict["entity"][entity_key(entity)] = entity #建立键到实体的映射
        for relation in x.get("relation", list()):
            element_dict["relation"][relation_key(relation)] = relation
        for event in x.get("event", list()):
            element_dict["event"][event_key(event)] = event

    return {#合并后的实体是从原始数据集的不同id的样本组合起来的
        "id": "-".join(instance_id_list),
        "text": instance_list[0]["text"],
        "tokens": instance_list[0]["tokens"],
        "entity": list(element_dict["entity"].values()),
        "relation": list(element_dict["relation"].values()),
        "event": list(element_dict["event"].values()),
    }

for text in instance_dict:
    instance_dict[text] = merge_instance(instance_dict[text])

for i in range(800,802):
    print(list(instance_dict.values())[i])

import copy

with open(f"{life_folder}/train.json", "w") as output: #写人生训练集到文件
    for instance in instance_dict.values():
        new_instance = copy.deepcopy(instance) # 深拷贝
        # 过滤当前样本的["relation"]在life_relation列表内的
        new_instance["relation"] = list(filter(lambda x: x["type"] in life_relation, instance["relation"]))
        output.write(json.dumps(new_instance) + "\n")

f"{org_folder}/train.json"

with open(f"{org_folder}/train.json", "w") as output:
    for instance in instance_dict.values():
        new_instance = copy.deepcopy(instance)
        #过滤当前样本的["relation"]在组织关系列表内的
        new_instance["relation"] = list(filter(lambda x: x["type"] in org_relation, instance["relation"]))
        output.write(json.dumps(new_instance) + "\n")

a_instances = load_jsonlines_file(f"{life_folder}/train.json")
b_instances = load_jsonlines_file(f"{org_folder}/train.json")
print(len(a_instances),len(b_instances))

import yaml
def load_definition_schema_file(filename):#加载yaml文件
    return yaml.load(open(filename, encoding="utf8"), Loader=yaml.FullLoader)

aa = load_definition_schema_file(os.path.join(schema_folder,'体育竞赛.yaml'))

aa['事件'].values()

mm=[]
for i in aa['事件'].values():
    mm+=i["参数"] # 有重复
print(mm)
mm=list(set(mm)) #用set是为了去重复

[x for x in aa['事件']]

aa['事件']['退役']["参数"].keys()

aaa={1:2,3:4}
for k,v in aaa.items():
    print(k,v)

def dump_schema(output_folder, schema_dict):
    if not os.path.exists(output_folder):
        os.makedirs(output_folder)
    for schema_name, schema in schema_dict.items(): #写"entity"，"relation",...到文件entity.schema
        schema_file = f"{output_folder}/{schema_name}.schema"
        with open(schema_file, "w", encoding="utf8") as output:
            for element in schema:
                output.write(json.dumps(element, ensure_ascii=False) + "\n")

def dump_event_schema(event_map, output_folder):
    role_list = list()
    for roles in event_map.values():
        role_list += roles["参数"]
    rols_list = list(set(role_list)) #论元角色列表
    type_list = list(event_map.keys()) #事件类型
    #事件类型到论元角色列表的映射
    type_role_map = {event_type: list(event_map[event_type]["参数"].keys()) for event_type in event_map}
    dump_schema(
        output_folder=output_folder,
        schema_dict={
            "entity": [[], [], {}],
            "relation": [[], [], {}],
            "event": [type_list, rols_list, type_role_map],
            "record": [type_list, rols_list, type_role_map],
        },
    )

def filter_event_in_instance(instances,required_event_types):
    # 过滤实例中的事件，只保留需要的事件类别的事件标注
    new_instances = list()
    for instance in instances:
        new_instance = copy.deepcopy(instance)
        # 重新设定样本["event"]，只保留当前需要的事件
        new_instance["event"] = list(filter(lambda x: x["type"] in required_event_types, new_instance["event"]))
        new_instances += [new_instance]
    return new_instances

def dump_instances(instances, output_filename): #写新的样本到目标文件夹
    with open(output_filename, "w", encoding="utf8") as output:
        for instance in instances:
            output.write(json.dumps(instance, ensure_ascii=False) + "\n")

def filter_event(data_folder, event_types, output_folder):
    dump_event_schema(event_types, output_folder) #写schema文件到目标文件夹
    for split in ["train", "val"]:
        filename = os.path.join(data_folder, f"{split}.json")
        instances = [json.loads(line.strip()) for line in open(filename, encoding="utf8")]
        new_instances = filter_event_in_instance(instances, required_event_types=event_types)
        dump_instances(new_instances, os.path.join(output_folder, f"{split}.json"))

os.path.join(output_folder, "灾害意外")

load_definition_schema_file(os.path.join(schema_folder, f"灾害意外.yaml"))

# 对事件数据进行预处理，过滤除 `灾害意外` 和 `体育竞赛` 外的事件标注
for schema in ["灾害意外", "体育竞赛"]:
    print(f"Building {schema} dataset ...")
    duee_folder = os.path.join(output_folder, "DUEE") #原始duee数据集
    schema_file = os.path.join(schema_folder, f"{schema}.yaml")
    output_folder2 = os.path.join(output_folder, schema)
    schema = load_definition_schema_file(schema_file)
    filter_event(
        duee_folder,# 原数据集的路径
        schema["事件"],#获取事件里面的具体事件组
        output_folder2, #要生成数据集的路径
    )

ty_instances = load_jsonlines_file(f"{output_folder}/体育竞赛/train.json")
zh_instances = load_jsonlines_file(f"{output_folder}/灾害意外/train.json")

print(len(ty_instances),len(zh_instances))

for i in range(11508,11510):
    print(ty_instances[i],'|',zh_instances[i])

bb=load_definition_schema_file(os.path.join(schema_folder, "金融信息.yaml"))

for i in bb['事件'].keys():
    print(i)

mm=list()
mm+=bb['事件']['中标']["参数"]   
mm=list(set(mm))

bb["事件"]['中标']["参数"] .keys()

os.path.join(output_folder, "DUEE_FIN_LITE")

for schema in ["金融信息"]:
    print(f"Building {schema} dataset ...")
    duee_fin_folder = os.path.join(output_folder, "DUEE_FIN_LITE") # 原金融数据集
    schema_file = os.path.join(schema_folder, f"{schema}.yaml") # 金融信息schema
    output_folder2 = os.path.join(output_folder, schema) #要构建的目标数据集存放路径
    schema = load_definition_schema_file(schema_file) #加载schema
    # 根据不同事件类型将多事件抽取分割成多个单事件类型抽取
    # 把多事件分割成单事件提取
    for event_type in schema["事件"]:
        filter_event(
           duee_fin_folder,
           {event_type: schema["事件"][event_type]},# 具体的事件
            output_folder2 + "_" + event_type, # 具体事件对应的输出文件夹
        )

vv=load_jsonlines_file(f"{output_folder}/DUEE_FIN_LITE/train.json")

f"{output_folder}/DUEE_FIN_LITE/train.json"

zb_instances = load_jsonlines_file(f"{output_folder}/金融信息_中标/train.json")
zy_instances = load_jsonlines_file(f"{output_folder}/金融信息_质押/train.json")
print(len(zb_instances),len(zy_instances))

for i in range(6985,6986):
    print(zb_instances[i],'|',zy_instances[i])

def annonote_graph(
    entities: List[Dict] = [],
    relations: List[Dict] = [],
    events: List[Dict] = []):
    spot_dict = dict() # 信息点字典
    asoc_dict = defaultdict(list)
    # 将实体,关系,事件转换为点关联图
    def add_spot(spot):
        spot_key = (tuple(spot["offset"]), spot["type"])
        spot_dict[spot_key] = spot
    def add_asoc(spot, asoc, tail):
        spot_key = (tuple(spot["offset"]), spot["type"])
        asoc_dict[spot_key] += [(tuple(tail["offset"]), tail["text"], asoc)]
    for entity in entities: # 为实体添加键值映射，其实就是为实体添加了索引
        add_spot(spot=entity)
    for relation in relations: # 关系
        add_spot(spot=relation["args"][0])
        # 建立relation["args"][0]的索引到组合关系的映射
        add_asoc(spot=relation["args"][0], asoc=relation["type"], tail=relation["args"][1])
    for event in events: #整个就是建立一个索引到一堆东西的映射，以便访问
        add_spot(spot=event)
        for argument in event["args"]:
            add_asoc(spot=event, asoc=argument["type"], tail=argument)
    spot_asoc_instance = list()
    for spot_key in sorted(spot_dict.keys()): # 遍历索引
        offset, label = spot_key
        if len(spot_dict[spot_key]["offset"]) == 0: #这个对应具体的位置，如果没有，就过滤掉
            continue
        spot_instance = {
            "span": spot_dict[spot_key]["text"],
            "label": label,
            "asoc": list(),
        }
        for tail_offset, tail_text, asoc in sorted(asoc_dict.get(spot_key, [])):
            if len(tail_offset) == 0:
                continue
            spot_instance["asoc"] += [(asoc, tail_text)]
        spot_asoc_instance += [spot_instance]
    spot_labels = set([label for _, label in spot_dict.keys()])
    asoc_labels = set()
    for _, asoc_list in asoc_dict.items():
        for _, _, asoc in asoc_list:
            asoc_labels.add(asoc)
    return spot_labels, asoc_labels, spot_asoc_instance

def add_spot_asoc_to_single_file(filename):
    instances = [json.loads(line) for line in open(filename, encoding="utf8")]
    print(f"Add spot asoc to {filename} ...")
    with open(filename, "w", encoding="utf8") as output:
        for instance in instances:
            spots, asocs, spot_asoc_instance = annonote_graph(
                entities=instance["entity"],# 实体
                relations=instance["relation"],# 关系
                events=instance["event"],# 事件
            )
            # 为对象添加spot_asoc
            instance["spot_asoc"] = spot_asoc_instance
            # 为对象添加spot
            instance["spot"] = list(spots)
            # 为对象添加asoc
            instance["asoc"] = list(asocs)
            output.write(json.dumps(instance, ensure_ascii=False) + "\n")

ff = os.path.join(output_folder,'金融信息_企业破产',"train.json")
ff_instances = [json.loads(line) for line in open(ff, encoding="utf8")]

for i in range(1046,1048):
    print(ff_instances[i])

a,b,yyj=annonote_graph(entities=ff_instances[5000]["entity"],
                relations=ff_instances[5000]["relation"],
                events=ff_instances[5000]["event"],)

data_folder=output_folder

def merge_schema(schema_list: List[RecordSchema]):
    type_set = set()
    role_set = set()
    type_role_dict = defaultdict(list) #字典，字典的值是列表类型
    for schema in schema_list:
        for type_name in schema.type_list:
            type_set.add(type_name)
        for role_name in schema.role_list:
            role_set.add(role_name)
        for type_name in schema.type_role_dict:
            # type_name为索引
            type_role_dict[type_name] += schema.type_role_dict[type_name]
    for type_name in type_role_dict:
        type_role_dict[type_name] = list(set(type_role_dict[type_name])) # 去重复
    return RecordSchema(
        type_list=list(type_set),
        role_list=list(role_set),
        type_role_dict=type_role_dict,
    )

def convert_duuie_to_spotasoc(data_folder, ignore_datasets):
    #
    schema_list = list() # 用来保存所有任务的schema
    for task_folder in os.listdir(data_folder):#遍历下面的子文件夹
        if task_folder in ignore_datasets: #过滤掉不处理的，这两个是原始数据
            continue
        if not os.path.isdir(os.path.join(data_folder, task_folder)):# 过滤掉非文件夹，就是文件
            continue
        print(f"Add spot asoc to {task_folder} ...")#为其他的添加spot asoc
        # 读取单任务的 Schema
        task_schema_file = os.path.join(data_folder, task_folder, "record.schema")
        # 向单任务数据中添加 Spot Asoc 标注
        add_spot_asoc_to_single_file(os.path.join(data_folder, task_folder, "train.json"))
        add_spot_asoc_to_single_file(os.path.join(data_folder, task_folder, "val.json"))
        record_schema = RecordSchema.read_from_file(task_schema_file) #读取单任务schema
        schema_list += [record_schema]
    # 融合不同任务的 Schema
    multi_schema = merge_schema(schema_list)
    multi_schema.write_to_file(os.path.join(data_folder, "record.schema"))

convert_duuie_to_spotasoc(output_folder,ignore_datasets)