Enabling autonomous agents to discriminate between questions and requests

• US 10,796,099 B2
• Filed: 09/28/2018
• Issued: 10/06/2020
• Est. Priority Date: 09/28/2017
• Status: Active Grant

• Current Assignee: Oracle International Corporation (Oracle Corporation)
• Original Assignee: Oracle International Corporation (Oracle Corporation)
• Inventors: Galitsky, Boris, Vishnoi, Vishal, Xu, Anfernee
• Primary Examiner(s): Koeth, Michelle M
• Application Number:
  US16/145,702
• Publication Number:
  US 20190095425A1
• Time in Patent Office: 739 Days
• CPC Class Codes: G06F 40/205 Parsing
• CPC Class Codes: G06F 40/253 Grammatical analysis; Style critique
• CPC Class Codes: G06F 40/295 Named entity recognition
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G10L 15/18 using natural language modelling

Abstract

Systems, devices, and methods of the present invention relate to text classification. A text classification system accesses an utterance of text. The utterance includes at least one word. The text classification system generates a parse tree for the utterance. The parse tree includes at least one terminal node with a word type. The terminal node represents a word of the utterance. The text classification system applies one or more rules to the text. The text classification system then classifies the utterance as a question or a request for an autonomous agent to perform an action.

Claims

1. A system for classifying text as a question or a request, comprising:

a linguistic engine configured to;

access an utterance comprising at least one word;

generate a parse tree for the utterance, wherein the parse tree comprises at least one terminal node comprising a word type, wherein each terminal node represents a word or phrase of the utterance and wherein the parse tree identifies syntactic information of the utterance; and

apply, to the parse tree, one or more linguistic templates comprising one or more word types; and

a rule engine configured to generate one or more match results obtained from the linguistic engine and determine a classification of the text by applying rules comprising;

(i) responsive to determining, from the parse tree, that the utterance comprises a first person pronoun followed by a noun, classifying the utterance as a request;

(ii) responsive to determining, from the parse tree, that the utterance comprises an imperative verb as a first word of the utterance by identifying a verb that is (a) in a present tense or in an active voice and (b) neither modal, mental, or a form of the verb “

to be,”

classifying the utterance as a request;

(iii) responsive to identifying, in the utterance, one or more predefined request keywords, classifying the utterance as a request;

(iv) responsive to identifying, in the utterance, a mental verb, classifying the utterance as a question; and

(v) responsive to determining that the utterance comprises a question intent prefix in a first word position, classifying the utterance as a question,

wherein the system is further configured, based on the classification as the question or the request, to send a message to a user device or adjust a configuration of an external device.

2. The system of claim 1, wherein the rules further comprise (vi) responsive to failing to identify the utterance as a question or a request, classifying the utterance as unknown.

3. The system of claim 1, wherein the rules further comprise (vi) responsive to determining that the utterance comprises a request suffix at in a last word position of the utterance, classifying the utterance as a request.

4. The system of claim 1, wherein the rules further comprise (vi) responsive to determining that the utterance comprises a transaction verb, classifying the utterance as a request.

5. The system of claim 1, wherein the linguistic engine is configured to, prior to determining the parse tree, removing one or more stop words from the utterance.

15. The system of claim 1, wherein the rules further comprise (vi) responsive to determining, from the parse tree, that the utterance comprises a first person pronoun and a mental verb, classifying the utterance as a request.

17. The system of claim 1, wherein the rule engine is implemented by a classification model.

6. A method of using a classification model to classify utterances as questions or requests, the method comprising:

accessing an utterance, wherein the utterance comprises words;

generating a parse tree for the utterance, wherein the parse tree comprises at least one terminal node comprising a word type representing a word or phrase of the utterance, and wherein the parse tree identifies syntactic information of the utterance;

applying, to the parse tree, a classification model, that is trained to apply rules comprising;

(i) responsive to determining, from the parse tree, that the utterance comprises a first person pronoun followed by a noun, classifying the utterance as a request;

(ii) responsive to determining, from the parse tree, that the utterance comprises an imperative verb as a first word of the utterance by identifying a verb that is (a) in a present tense or in an active voice and (b) neither modal, mental, or a form of the verb “

to be,”

classifying the utterance as a request;

(iii) responsive to identifying, in the utterance, one or more predefined request keywords, classifying the utterance as a request;

(iv) responsive to identifying, in the utterance, a mental verb, classifying the utterance as a question; and

(v) responsive to determining that the utterance comprises a question intent prefix in a first word position, classifying the utterance as a question; and

receiving, from the classification model, a classification of the utterance, wherein the classification of the utterance identifies the utterance as (i) a question, (ii) a request, or (iii) unknown.

7. The method of claim 6, further comprising:

accessing a set of training data comprising a set of training data pairs, wherein each training data pair comprises text and an expected classification, and wherein the set of training data includes both (i) a first training data pair that comprises a question and (ii) a second training data pair comprises a request; and

training the classification model by iteratively;

providing one of the training data pairs to the classification model;

receiving, from the classification model, a determined classification;

calculating a loss function by calculating a difference between the

determined classification and the expected classification; and

adjusting internal parameters of the classification model to minimize the loss function.

8. A method for classifying text as a question or a request, the method comprising:

accessing an utterance comprising at least one word;

generating a parse tree for the utterance, wherein the parse tree comprises at least one terminal node comprising a word type, wherein each terminal node represents a word or phrase of the utterance and wherein the parse tree identifies syntactic information of the utterance; and

applying, to the parse tree, one or more linguistic templates comprising one or more word types; and

applying rules comprising;

(i) responsive to determining, from the parse tree, that the utterance comprises a first person pronoun followed by a noun, classifying the utterance as a request;

(ii) responsive to determining, from the parse tree, that the utterance comprises an imperative verb as a first word of the utterance, classifying the utterance as a request;

(iii) responsive to identifying, in the utterance, one or more predefined request keywords, classifying the utterance as a request;

(iv) responsive to identifying, in the utterance, a mental verb, classifying the utterance as a question;

(v) responsive to determining that the utterance comprises a question intent prefix in a first word position, classifying the utterance as a question; and

(vi) responsive to determining, from the parse tree, that the utterance comprises a first person pronoun and a mental verb, classifying the utterance as a request; and

sending, based on the classification as the question or the request, a message to a user device or adjust a configuration of an external device.

9. The method of claim 8, wherein the rules further comprise:

(vii) responsive to determining, a linguistic template match, classifying the utterance as a request.

10. The method of claim 8, wherein the rules further comprise (vii) responsive to failing to classify the utterance as the question or the request, classifying the utterance as unknown.

11. The method of claim 8, wherein the rules further comprise (vii) responsive to identifying, in the utterance, a transaction suffix or a transaction verb, classifying the utterance as a request.

12. The method of claim 8, wherein the rules further comprise (vii) responsive to determining that the utterance comprises a question prefix, classifying the utterance as a question.

13. The method of claim 8 further comprising removing one or more stop words from the utterance.

14. The method of claim 9, wherein the classification model is a tree kernel or a support vector machine.

16. The method of claim 8, wherein determining that the utterance comprises an imperative verb comprises determining that the utterance comprises a verb that is (a) in a present tense or in an active voice and (b) neither modal, mental or a form of the verb “

to be.”

18. The method of claim 8, wherein applying the rules comprises inputting the parse tree to a classification model.