Hybrid natural language understanding
Hybrid natural language understanding (NLU) systems and methods are provided that capitalize on the strengths of the rule-based models and the statistical models, lowering the cost of development and increasing the speed of construction, without sacrificing control and accuracy. Two models are used for intent recognition, one statistical and one rule-based. Both models define the same set of intents, but the rule-based model is devoid of any grammars or patterns initially. Each model may or may not be hierarchical in that it may be composed of a set of specialized models that are in a tree form or it may be just a singular model.
1. A hybrid natural language understanding (NLU) system, comprising:
a processor; and
a chatbot comprising:
a rule-based NLU module configured to determine a first intent based on a user input during a conversation between a client device of the user and the chatbot, and a confidence value associated with the first intent;
a statistical NLU module configured to determine a second intent based on the user input, and a confidence value associated with the second intent; and
a decider configured to determine and output a third intent using the first intent, the confidence value associated with the first intent, the second intent, and the confidence value associated with the second intent, wherein the third intent is further determined by the decider:
(1) determining whether both the first intent and the second intent are correct,
(2) determining whether only the first intent is correct,
(3) determining whether only the second intent is correct, and
(4) determining whether neither the first intent nor the second intent is correct,
wherein the chatbot is configured to generate and provide a processed language output based on the third intent.
2. The system of claim 1 , wherein the rule-based NLU module, the statistical NLU module, and the decider are comprised within a computing device.
3. The system of claim 1 , wherein the rule-based NLU module comprises an intent determiner and at least one of patterns or grammars to determine the first intent.
4. The system of claim 1 , wherein the statistical NLU module comprises an intent determiner and a trained model of input text over intentions to determine the second intent.
5. The system of claim 1 , wherein the rule-based NLU module comprises a first confidence value determiner configured to determine a first intent confidence value, and the statistical NLU module comprises a second confidence value determiner configured to determine a second intent confidence value.
6. The system of claim 5 , wherein the decider is further configured to determine the third intent further based on the first intent confidence value and the second intent confidence value.
7. The system of claim 1 , further comprising a training module configured to train the statistical NLU module.
8. The system of claim 1 , further comprising a trainer configured to train the decider.
9. A method of hybrid natural language understanding (NLU), the method comprising:
receiving an input data at a rule-based NLU module comprised within a chatbot, wherein the input data is received during a conversation between a client device of a user and the chatbot;
receiving the input data at a statistical NLU module comprised within the chatbot;
determining a first intent, and a confidence value associated with the first intent, at the rule-based NLU module using the input data;
determining a second intent, and a confidence value associated with the second intent, at the statistical NLU module using the input data;
determining a third intent at a decider using the first intent, the confidence value associated with the first intent, the second intent, the confidence value associated with the second intent, and:
(1) determining whether both the first intent and the second intent are correct,
(2) determining whether only the first intent is correct,
(3) determining whether only the second intent is correct, and
(4) determining whether neither the first intent nor the second intent is correct;
outputting the third intent; and
generating and providing a processed language output from the chatbot based on the third intent.
10. The method of claim 9 , wherein determining the first intent comprises using an intent determiner and at least one of patterns or grammars to determine the first intent.
11. The method of claim 9 , wherein determining the second intent comprises using an intent determiner and a trained model of input text over intentions to determine the second intent.
12. The method of claim 9 , further comprising determining a first intent confidence value and a second intent confidence value.
13. The method of claim 12 , wherein determining the third intent is further based on the first intent confidence value and the second intent confidence value.
14. The method of claim 9 , further comprising training the statistical NLU module and the decider.
15. A system comprising:
a processor; and
a chatbot comprising:
a natural language understanding (NLU) component configured to determine a rule-based NLU intent using an input data, wherein the input data is received during a conversation between a client device of a user and the chatbot, determine a statistical NLU intent using the input data, determine an intent using the rule-based NLU intent and the statistical NLU intent and using a confidence value associated with the rule-based NLU intent and a confidence value associated with the statistical NLU intent, based on: (1) determining whether both the rule-based NLU intent and the statistical NLU intent are correct, (2) determining whether only the rule-based NLU intent is correct, (3) determining whether only the statistical NLU intent is correct, and (4) determining whether neither the rule-based NLU intent nor the statistical NLU intent is correct, and generate and output a processed language output based on the intent; and
a memory that stores training data for the NLU component.
16. The system of claim 15 , wherein the input data comprises chat data.
17. The system of claim 15 , wherein the NLU component is further configured to determine a first intent confidence value for the rule-based NLU intent and a second intent confidence value for the statistical NLU intent, and to determine the intent further based on the first intent confidence value and the second intent confidence value.