Question

我的wit.ai应用程序中有三个故事：

- order_pizza: typical pizza ordering with size and toppings
- reserve_table: book a table for a specific date and time
- fav_topping: query favourite pizza topping from historical data by date

第三个，fav_topping有点特别。用户可以问一下6月份我最喜欢的打算是什么？＆＃39;。在得到答案之后，可以通过更改日期和“7月”以及7月份来再次提出问题吗？＆＃39;。

更正式地说：让我说我有一个有多个实体（X和Y）的故事。用户询问一个带有几个参数的问题（一个用于X，一个用于Y），并得到答案。再问（c代表X），我的机器人通过将当前的参数与前面的参数混合来给出答案（c代表X，b代表Y）。

为了解决这个问题，我从头到尾回放了我最喜欢的故事： fav_topping story with a loop-back

工作完美，但我遇到了问题。它永远不会结束。因此无法阻止它，用户无法更改主题。这是来自我的node.js客户端的日志，使用交互模式：

> What was my favourite topping in June?
fav_topping
context in: {}
entities in: {"contact":[{"confidence":0.9965486088458536,"type":"value","value":"my","suggested":true}],"datetime":[{"confidence":0.9954243915206602,"type":"value","value":"2017-06-01T00:00:00.000-07:00","grain":"month","values":[{"type":"value","value":"2017-06-01T00:00:00.000-07:00","grain":"month"},{"type":"value","value":"2018-06-01T00:00:00.000-07:00","grain":"month"},{"type":"value","value":"2019-06-01T00:00:00.000-07:00","grain":"month"}]}],"intent":[{"confidence":0.9969200405818317,"value":"fav_topping"}]}
context out is: {"ft_answer":"Your favourite topping was tomato for sure :) *"}
sending... {"text":"Your favourite topping was tomato for sure :) "}
> and in June?
fav_topping
context in: {"ft_answer":"Your favourite topping was tomato for sure :) "}
entities in: {"datetime":[{"confidence":0.994470434703695,"type":"value","value":"2017-06-01T00:00:00.000-07:00","grain":"month","values":[{"type":"value","value":"2017-06-01T00:00:00.000-07:00","grain":"month"},{"type":"value","value":"2018-06-01T00:00:00.000-07:00","grain":"month"},{"type":"value","value":"2019-06-01T00:00:00.000-07:00","grain":"month"}]}]}
context out is: {"ft_answer":"Your favourite topping was tomato for sure :) "}
sending... {"text":"Your favourite topping was tomato for sure :) "}
> OK, I want a normal pizza with tomato and pepperoni
fav_topping
context in: {"ft_answer":"Your favourite topping was tomato for sure :) "}
entities in: {"contact":[{"confidence":0.5452832324099028,"type":"value","value":"I","suggested":true}],"pizza_size":[{"confidence":0.9769454518025678,"type":"value","value":"middle"}],"pizza_topping":[{"confidence":0.9934795348369287,"type":"value","value":"tomato"},{"confidence":0.9505825366331593,"type":"value","value":"pepperoni"}],"intent":[{"confidence":0.6669035510798399,"value":"order_pizza"}]}
context out is: {"ft_answer":"Your favourite topping was tomato for sure :) "}
sending... {"text":"Your favourite topping was tomato for sure :) "}
> Book me a table for 5 persons at 9pm
fav_topping
context in: {"ft_answer":"Your favourite topping was tomato for sure :) "}
entities in: {"contact":[{"confidence":0.8387365394318869,"type":"value","value":"me","suggested":true}],"number":[{"confidence":1,"type":"value","value":5}],"datetime":[{"confidence":1,"type":"value","value":"2016-07-07T21:00:00.000-07:00","grain":"hour","values":[{"type":"value","value":"2016-07-07T21:00:00.000-07:00","grain":"hour"},{"type":"value","value":"2016-07-08T21:00:00.000-07:00","grain":"hour"},{"type":"value","value":"2016-07-09T21:00:00.000-07:00","grain":"hour"}]}],"intent":[{"confidence":0.9997353858626356,"value":"reserve_table"}]}
context out is: {"ft_answer":"Your favourite topping was tomato for sure :) "}
sending... {"text":"Your favourite topping was tomato for sure :) *"}

问题是：如何手动/编程地告诉BE完成故事？或者，我如何设置故事来处理切换？

还有一点需要注意：正如您在我的日志中可以看到BE理解得很好，用户开始了一个新主题（认识到意图改为＆＃39; order_pizza＆＃39;＆＃39; reserve_table＆＃39;）。如果我能明确定义故事的意图，也许可以帮助很多。你怎么看？（对我们现在所拥有的东西没有太大的改变：定义故事的第一句话）

Answer 1

感谢您的详细解释。这是目前Bot Engine的限制，我们正在以更优雅的方式来解决这个问题。与此同时，当意图不是“fav_topping”时，您可能希望删除“ft_answer”上下文密钥。

有关上下文的更多信息

Wit读取上下文以预测下一步。我们永远不会更新我们这方面的背景。您可以使用旁边的操作来更新上下文中的键。这些键可以反过来影响基于故事的预测，但也可以通过变量建立Bot答案。 Wit使用当前上下文（仅用于现在的键），上下文的状态（在交互N-1，N-2 ......）以及用户消息（特征实体）来预测下一个动作。

如何退出从最后回归的wit.ai故事？

1 个答案: