| CPC G06Q 30/02 (2013.01) [G10L 15/26 (2013.01)] | 13 Claims |
|
1. A method for analyzing a sales conversation based on voice recognition, the method performed by a processor executing instructions stored in a memory and comprising:
obtaining, via a communication interface unit, voice information about a sales conversation between a sales representative and a customer;
converting, by using a deep neural network (DNN), the voice information into text,
wherein the converting of the voice information into the text comprises: extracting, using a sales conversation analysis apparatus, a Mel-frequency cepstral coefficient (MFCC) feature vector from voices of the sales representative and the customer, and separating and extracting the voice of the sales representative and the voice of the customer based on the extracted MFCC vector through K-mean clustering;
extracting at least one of a keyword and a sentence corresponding to each of a plurality of business items from the text;
extracting analysis information for each of the plurality of business items based on at least one of the keyword and the sentence;
calculating an evaluation score for each of the plurality of business items based on the analysis information for each of the plurality of business items,
wherein the plurality of business items includes items about a budget of a customer, an authority of the customer, needs of the customer, a purchase time of the customer, and a competitor of a sales entity,
wherein first information about the budget of the customer, second information about the authority of the customer, third information about the needs of the customer, fourth information about the purchase time of the customer, and fifth information about the competitor of the sales entity are extracted, and
wherein a first score for the budget of the customer is calculated based on the first information, a second score for the authority of the customer is calculated based on the second information, a third score for the needs of the customer is calculated based on the third information, a fourth score for the purchase time of the customer is calculated based on the fourth information, and a fifth score for the competitor of the sales entity is calculated based on the fifth information;
automatically outputting, by involving the DNN, a recommendation query for at least one business item based on at least one of analysis information for each of the plurality of business items and the evaluation score for each of the plurality of business items,
wherein at least one business item corresponding to a score smaller than a reference score is selected from among the first to fifth scores, a reference sentence identical to or similar to the sentence that is extracted from the text in relation to the selected business item is extracted from a reference database (DB), and the recommendation query is automatically generated and output based on a query list corresponding to the reference text,
wherein the plurality of business items further includes an item for a customer question,
sixth information for the customer question is extracted and a sixth score for the customer question is calculated based on the sixth information, and
the sixth information includes information about the number of customer questions, and
wherein the sixth information includes information about a pending customer question, and
the method further comprises:
automatically generating and outputting schedule information for the sales representative based on alarm information for the pending customer question;
automatically categorizing, by involving the DNN, the customer question based on the plurality of business items;
calculating an evaluation score for the customer question based on analysis of the categorized customer question; and
correcting the evaluation score for each of the business items based on the evaluation score for the customer question.
|