Pobierz otwarty kod źródłowy bota Libre ogólny wydanie i zainstaluj bota Libre na własny serwer
Bot Libre Dev

Natural language understanding in Botlibre bots

w eavinash1983 wysłany Feb 7 2019, 9:20

Hi,

I just created a bot that provides information on motivation speakers,videos and categories.

It was a bot with no knowledge. As I started interacting with bot I corrected its response. I wan to understand whether Botlibre bots 

1. converstational (they can remember the context during a conversation)

2. possess NLU so that it can respond as a human does.

 

for e.g in my bot

 I asked the bot who is Eric Thomas

it said he is French actor

Then I corrected it saying

He is a motivational speaker.

When I asked a question to the bot after corrects like

Tell me something about Eric Thomas, Bot was unable to respond with the answer I taught him.

So wasnted to know how NLU is possessed by Botlibre bots.


by admin posted Feb 7 2019, 14:11
When you correct a bot, it will then use your correction the next time you ask the same question, or a similar question.

Using correction is a very easy way to train a bot, but not the best way. It is best to training your bot from its Training & Chat Logs page in its Admin Console.

See,
https://www.botlibre.com/forum-post?id=483549

or using a response list, or AIML

Looking at your bot you have learning and comprehension enabled. These are not recommend, most certainly not for a bot for a business. You should create your bot from the "service_template" not the "ai_template", or disable learning, comprehension and delete the Understanding script. Review your bot's responses from its Training & Chat Logs page and delete the invalid responses that you trained it with through learning.

The Understanding script does let the bot understand simple language and learn and remember new things. i.e. if you say "the sky is blue", then ask "is the sky blue" it will answer "yes, the sky is blue". But this is meant for an experimental AI bot, not a bot for a business.

Note that with learning enabled "every" response you enter when you chat with your bot it will remember and use for its previous response. This makes it very easy to incorrectly train a bot with bad responses, we recommend to disable learning and comprehension.

To train a bot correctly, again see,
https://www.botlibre.com/forum-post?id=483549

We can also train a bot for you, or review and cleanup your bot, or implement advanced scripts through our development services. Contact [email protected]

Thumbs up: 0, thumbs down: 0, stars: 0.0
Views: 2685, today: 1, week: 6, month: 40

by eavinash1983 posted Feb 15 2019, 9:00

Thanks for the reply.

I went through the process of NLP in Botlibre but still not able to understand two things

1. How is intend classification done in Botlibre NLP

2. How is entitiy extraction done in Botlibre NLP.


Thumbs up: 0, thumbs down: 0, stars: 0.0
Views: 2513, today: 1, week: 6, month: 34

by admin posted Feb 15 2019, 10:04
From your bot's Training & Chat Logs page you can enter questions and responses. The question is indexed and use by Bot Libre's NLP to match similar user questions to the response.

For an intent, or for having many question refer to the same response you can use intent labels. Just label the primary question/response and use the label for other question that you want to use the same response.

https://www.botlibre.com/forum-post?id=12117061

To extract information from the user's question there are a few mechanisms.

1 - You can use a pattern for the question, then you automatically get access to the * star variables in the response template.

2 - You can use a Template response, or a response "think" tag to process information. You can use Regex for parse the sentence to extract data.
https://www.botlibre.com/forum-post?id=23543115
https://www.botlibre.com/manual-regex.jsp

3 - You can use a Self script to parse a question, or do processing on a response. Self lets you do advanced things like parse and evaluate a mathematical expression, or call a web service,
https://www.botlibre.com/forum-post?id=699077

See also,
https://www.botlibre.com/doc.jsp

https://www.botlibre.com/browse?browse-type=Script&browse=true

Thumbs up: 0, thumbs down: 0, stars: 0.0
Views: 2823, today: 0, week: 5, month: 54

Identyfikator: 25826612
Wysłany: Feb 7 2019, 9:20
Aktualizacja: Feb 7 2019, 14:01
Odpowiedzi: 3
Widok: 2845, dziś: 1, tydzień: 8, miesiąc: 49
0 0 0.0/5