Automated systems and methods for natural language processing with speaker intention inference

1. A method comprising:

calling, by an automated robotic telemarketing system, a device for a customer selected from a customer list;

parsing, by a real-time speech recognition module of the automated robotic telemarketing system, a first audio encoding of a customer statement spoken by the customer;

determining, by a language intention determining module, a likely customer purchase intention according to the parsed customer statement;

selecting a sales pitch response corresponding to the determined likely customer purchase intention;

providing a second audio signal that encodes the selected sales pitch response to the device for the customer; and

training an optimal sales pitch model to select the sales pitch response corresponding to the determined likely customer purchase intention by:

obtaining multiple historical sales calls;

separating the obtained calls into multiple datasets according to attributes of the obtained calls;

for each dataset:

generating a word-vector matrix based on word frequency in the obtained calls, and

assigning a label to each word or cluster of words; and

training a machine-learning model for each dataset using at least one of the generated word-vector matrix, the word or cluster labels, or determined response success rates of the obtained calls.

2. The method of claim 1 , wherein calling includes obtaining, by a customer list management module, a phone number of the device for the selected customer from the customer list managed by the customer list management module.

3. The method of claim 1 , wherein determining the likely customer purchase intention includes determining the likely customer purchase intention using keyword matching.

4. The method of claim 3 , wherein the keyword matching includes:

mapping each parsed word in the customer statement to a label; and

applying one or more business rules to the labeled words to determine the likely customer purchase intention and an accuracy of the likely customer purchase intention.

5. The method of claim 1 , wherein determining the likely customer purchase intention includes determining the likely customer purchase intention using natural language processing.

6. The method of claim 1 , wherein providing the second audio signal to the device of the customer includes at least one of:

obtaining a pre-recorded audio signal of the selected response; and

converting the selected response into the second audio signal using computer voice synthesis.

7. The method of claim 1 , wherein determining the likely customer purchase intention includes classifying the likely customer purchase intention as a positive response or a negative response.

8. The method of claim 7 , wherein selecting the sales pitch response includes selecting a sales pitch response that corresponds to the classified positive response or selecting a sales pitch response that corresponds to the classified negative response.

9. The method of claim 1 , further comprising determining, based on the first audio encoding of the parsed customer statement, whether to end the call or transfer the call to a second device for a salesperson.

10. A computer system comprising:

memory configured to store a customer list and computer-executable instructions, wherein the customer list includes phone numbers for multiple potential telemarketing customers and

at least one processor configured to execute the instructions, wherein the instructions include:

calling a device for one of the customers selected from the customer list;

parsing, by a real-time speech recognition module, a first audio encoding of a customer statement spoken by the customer;

determining, by a language intention determining module, a likely customer purchase intention according to the parsed customer statement;

selecting a sales pitch response corresponding to the determined likely customer purchase intention;

providing a second audio signal that encodes the selected sales pitch response to the device for the customer; and

training an optimal sales pitch model to select the sales pitch response corresponding to the determined likely customer purchase intention by:

obtaining multiple historical sales calls;

separating the obtained calls into multiple datasets according to attributes of the obtained calls:

for each dataset:

generating a word-vector matrix based on word frequency in the obtained calls, and

assigning a label to each word or cluster of words; and

training a machine-learning model for each dataset using at least one of the generated word-vector matrix, the word or cluster labels, or determined response success rates of the obtained calls.

11. The computer system of claim 10 , wherein calling includes obtaining, by a customer list management module, the phone number of the device for the selected customer from the customer list managed by the customer list management module.

12. The computer system of claim 10 , wherein determining the likely customer purchase intention includes determining the likely customer purchase intention using keyword matching.

13. The computer system of claim 12 , wherein the keyword matching includes:

mapping each parsed word in the customer statement to a label; and

applying one or more business rules to the labeled words to determine the likely customer purchase intention and an accuracy of the likely customer purchase intention.

14. The computer system of claim 10 , wherein determining the customer purchase intention includes determining the likely customer purchase intention using natural language processing.

15. The computer system of claim 10 , wherein providing the second audio signal to the device of the customer includes at least one of:

obtaining a pre-recorded audio signal of the selected response; and

converting the selected response into the second audio signal using computer voice synthesis.

16. The computer system of claim 10 , wherein determining the likely customer purchase intention includes classifying the likely customer purchase intention as a positive response or a negative response.

17. The computer system of claim 16 , wherein selecting the sales pitch response includes selecting a sales pitch response that corresponds to the classified positive response or selecting a sales pitch response that corresponds to the classified negative response.

18. The computer system of claim 10 , wherein the instructions include determining, based on the parsed customer statement, whether to end the call or transfer the call to a salesperson.