botpress / v12

Botpress OSS – v12
https://v12.botpress.com
GNU Affero General Public License v3.0
71 stars 83 forks source link

NLU does not work correctly with the Russian language #1661

Open SergeyKhodusov opened 2 years ago

SergeyKhodusov commented 2 years ago

Describe the bug NLU does not work correctly with the Russian language.

To Reproduce Steps to reproduce the behavior:

  1. Created a bot
  2. Added 1 intent
  3. Added 1 question Q&A
  4. I am writing a message that should not fall into either the intent or the Q&A at all.

Gives match percentages for intent. Neither in the intent nor in the QnA is there a match even for one word.

Expected behavior none equal to approximately 0%

Screenshots

false-positive-1 false-positive-2

Environment (please complete the following information):

Additional context If I create more intents and QnA questions, I always get false positives of 50%+.

SergeyKhodusov commented 2 years ago

Prepared a better example. Created QnA on the topic of what the bot can do. Questions like: "Tell me about your abilities?" "Tell me what you can do?" "How else can you help me?" Wrote a text that has no relation in meaning to the questions asked (Like "Lorem Ipsum"). Gives a 75% match with QnA. It is wrong. 2022-06-28_18-28-17

SergeyKhodusov commented 2 years ago

There are indeed matches for several words such as "What", "You" and "How". But these are words that do not make much sense in the question. They occur in almost every question or intent and should not affect the outcome.