[英]logic determining dialog in Watson assistant
I want to improve ibm's Watson assistant results. 我想改善ibm的Watson助手结果。 So, I want to know the algorithm to determine a dialog in Watson assistant's conversations.
因此,我想知道在Watson Assistant的对话中确定对话的算法。 Is it a svm algorithm?
它是svm算法吗? A paper is welcome.
欢迎论文。
There are a number of ML/NLP technologies under the covers of Watson Assistant. Watson Assistant涵盖了许多ML / NLP技术。 So it's not just a single algorithm.
因此,这不仅仅是一个算法。 Knowing them is not going to help you improve your results.
知道它们不会帮助您改善结果。
I want to improve ibm's Watson assistant results.
我想改善ibm的Watson助手结果。
There are a number of ways. 有很多方法。
Focus on getting true representative questions from the end users. 专注于从最终用户那里获得真正的代表性问题。 Not only in the language that they use, but if possible from the same medium you plan to use WA on (eg. Mobile device, Web, Audio).
不仅使用他们使用的语言,而且还尽可能使用计划在其上使用WA的媒体(例如,移动设备,Web,音频)。
This is the first factor that reduces accuracy. 这是降低准确性的第一个因素。 Manufacturing an intent can mean you build an intent that a customer may never ask (even if you think they do).
制造意图可能意味着您建立了一个客户可能从未问过的意图(即使您认为他们这样做了)。 Second you will use language/terms with similar patterns.
其次,您将使用类似模式的语言/术语。 This makes it harder for WA to train.
这使得WA难以训练。
It's possible to train an intent with one question, but for best results 10-20 example questions. 可以用一个问题来训练意图,但是为了获得最佳结果,可以回答10-20个示例问题。 Where intents are close together then more examples are needed.
如果意图紧密相关,则需要更多示例。
The current process is to create what is called a K-Fold Cross validation ( sample script ). 当前过程是创建所谓的K折交叉验证( 示例脚本 )。 If your questions are representative then the results should give you an accurate indicator of how well it is performing.
如果您的问题具有代表性,那么结果应为您提供一个准确的指标,表明其效果如何。
However, it is possible to overfit the training. 但是,可能会过度适合训练。 So you should use a blind set.
因此,您应该使用盲目设置。 This is a 10-20% of all questions (Random sample).
这是所有问题的10-20%(随机样本)。 They should never be used to train WA.
绝对不要将它们用于训练WA。 Then run them against the system.
然后针对系统运行它们。 Both your Blind + K-Fold should fall within 5% of each other.
盲注+ K折应都落在5%之内。
You can look at the results of the K-Fold to fix issues, but blind set you should not. 您可以查看K形折叠的结果以解决问题,但盲目设置则不然。 Blinds can go stale as well.
百叶窗也可能过时。 So try to create a new blind set after 2-3 training cycles.
因此,尝试在2-3个训练周期后创建一个新的盲注集。
No matter how well your system is trained, I can guarantee you that new things will pop up when put in front of end users. 无论您的系统培训得如何好,我都可以保证当在最终用户面前出现新事物时会弹出。 So you should plan to have users test before you put it into production.
因此,您应该计划让用户进行测试,然后再将其投入生产。
When getting users to test, ensure they understand the general areas it has been trained on. 在让用户进行测试时,请确保他们了解接受培训的一般领域。 You can do this with user stories, but try not to prime the user into asking a narrow scoped question.
您可以使用用户故事来做到这一点,但请尽量避免使用户提出范围狭窄的问题。
Example: 例:
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.