我发布了一个开源项目,实现中文语义理解,包括训练和识别两部分。
已经发布Docker用于运行服务,通过客户端的SDK进行访问,SDK支持Python,Java,Node等语言。
API,实现原理,运行服务,下载示例程序或观看教学视频,参考:https://github.com/chatopera/clause
数据标注基于Intent, Entity, Slots 和 词典,举例:
{
"chatbotID": "bot007",
"dicts": [
{
"name": "food",
"dictwords": [
{
"word": "汉堡",
"synonyms": "汉堡包;漢堡;漢堡包"
}
]
}
],
"intents": [
{
"name": "take_out",
"description": "下外卖订单",
"slots": [
{
"name": "time",
"dictname": "@TIME",
"requires": true,
"question": "您想什么时候送到?"
},
{
"name": "loc",
"dictname": "@LOC",
"requires": true,
"question": "您希望该订单送到哪里?"
},
{
"name": "food",
"dictname": "food",
"requires": true,
"question": "您需要什么食物?"
}
],
"utters": [
{
"utterance": "我想订一份{food}"
}
]
}
]
}