TWI815400B - Emotion analysis system - Google Patents

Abstract

An emotion analysis system includes a voice receiver, a server and a processor. The server is configured to store a plurality of voice data. The processor is configured to perform the following steps according to a plurality of commands of the memory: storing a plurality of call record data in the server according to the plurality of voice data; marking an annotation data in the plural call record data according to the plural voice data; obtaining a customer emotion record parameter of the plural call record data according to the annotation data; obtaining a first customer emotion prediction parameter and a first emotion analysis model by applying the plural call record data through the deep learning algorithm; comparing whether the customer emotion record parameter is the same as the first customer emotion prediction parameter; and outputting the first emotion analysis model when the customer emotion record parameter is the same as the first customer emotion prediction parameter.

Description

sentiment analysis system

This case relates to an analysis system, and in particular to a sentiment analysis system.

Currently, customer complaints will bring losses to the company (such as the impact on goodwill, damage to the brand image, time and labor cost investment in the aftermath), and may even receive special attention from the competent authorities.

This summary is intended to provide a simplified summary of the disclosure to provide the reader with a basic understanding of the disclosure. This summary is not an extensive overview of the disclosure, and it is not intended to identify key/critical elements of the embodiments or to delineate the scope of the disclosure.

One of the technical aspects of this case is about an emotion analysis system. The sentiment analysis system includes a voice receiver, server and processor. The voice receiver is used to collect plural voice data. The server is coupled to the voice receiver and used to store plural voice data. The processor is electrically coupled to the server, and is used to perform the following steps according to the plurality of instructions in the memory: to store the plurality of call record data in the server according to the plurality of voice data; to mark the plurality of call record data according to the plurality of voice data; Obtain the customer emotion record parameters of multiple call record data based on the annotation data; use the multiple call record data through the deep learning algorithm to obtain the first customer emotion prediction parameters and the first emotion analysis model; compare the customer emotion record parameters with the first customer Whether the emotion prediction parameters are the same; and when the customer emotion record parameters are the same as the first customer emotion prediction parameters, the first emotion analysis model is output.

Therefore, according to the technical content of this case, the emotion analysis system shown in the embodiment of this case can detect customers' emotions in advance to achieve the effect of avoiding customer complaints.

After referring to the following embodiments, those with ordinary knowledge in the technical field to which this case belongs can easily understand the basic spirit and other purposes of the invention, as well as the technical means and implementation styles adopted in this case.

In order to make the description of this disclosure more detailed and complete, the following provides an illustrative description of the implementation aspects and specific embodiments of this case; but this is not the only form of implementing or using the specific embodiments of this case. The embodiments cover features of multiple specific embodiments as well as method steps and their sequences for constructing and operating these specific embodiments. However, other specific embodiments may also be used to achieve the same or equivalent functions and step sequences.

Unless otherwise defined in this specification, the scientific and technical terms used herein have the same meanings as commonly understood and customary by a person with ordinary knowledge in the technical field to which the subject matter belongs. In addition, unless there is conflict with the context, the singular noun used in this specification covers the plural form of the noun; and the plural noun used also covers the singular form of the noun.

In addition, "coupling" as used in this article can refer to two or more components that are in direct physical or electrical contact with each other, or that are in indirect physical or electrical contact with each other. It can also refer to two or more components that operate with each other. or action.

Figure 1 is a block diagram of an emotion analysis system according to an embodiment of the present invention. As shown in the figure, the sentiment analysis system 100 includes a voice receiver 110, a server 120 and a processor 130. In terms of connection relationship, the server 120 is electrically coupled to the voice receiver 110 , and the processor 130 is electrically coupled to the server 120 .

In order to detect customers' emotions in advance to avoid customer complaints, this case provides an emotion analysis system 100 as shown in Figure 1. The relevant operations of the emotion analysis system 100 are described in detail below.

Referring to Figure 1, in one embodiment, the voice receiver 110 is used to collect plural voice data. Then, the server 120 is used to store the plurality of voice data. For example, the voice receiver 110 can be a device driven by speech-to-text (Speech To Text, STT) software or algorithm. The voice receiver 110 can capture the words, audio, volume, and language spoken by the customer. The speed, call duration of the entire call, and the ID card number entered by the customer during Interactive Voice Response (IVR). When the customer talks to the customer service staff, the above plurality of voice-related data are instantly stored in the server. 120 for subsequent analysis, but this case is not limited to this.

Next, the processor 130 is used to perform the following steps according to the plurality of instructions in the memory 140: storing the plurality of call record data in the server 120 according to the plurality of voice data. For example, the memory 140 can be coupled to the processor 130. The processor 130 can use data analysis software (such as Python) to organize the plurality of speech data and segment the customer's words. Actually, the processor 130 can segment the words through stuttering. Methods (such as the Jieba word segmentation system) use built-in stop words and new words to organize word data in plural voice data into call record data, and store them in the server 120, but this case is not limited to this. In addition, the customer's audio, volume, speech speed, and the duration of the entire phone call can be digitized into call record data, and the call record data can include the customer's ID card number and be stored together in the server 120. This digitization The process is to convert unstructured data into structured data, but this case is not limited to this.

Then, the processor 130 is used to mark annotation data in the plurality of call record data according to the plurality of voice data. Subsequently, the processor 130 is used to obtain customer emotion record parameters of the plurality of call record data according to the annotation data. For example, the customer service staff can use the processor 130 to mark annotation data in the call record data based on the plurality of voice data. The annotation data can be when the customer is unhappy, strictly requests to be transferred to a supervisor, or directly makes a customer complaint call. Relevant content specifically noted by customer service staff in the call record data. In addition, the processor 130 can obtain the customer emotion record parameter from the marked call record data. At this time, the customer emotion record parameter is 1, which means that the customer has negative emotions. If the call record data is not marked, the customer emotion record parameter is 0, but this case is not limited to this.

Then, the processor 130 is used to apply the plurality of call record data through a deep learning algorithm to obtain the first customer emotion prediction parameters and the first emotion analysis model. For example, the deep learning algorithm can be a deep learning model, and the framework of the deep learning model can include Embedding, LSTM, and Softmax, and the complex call record data can be split into a training set and a verification set at a ratio of 8:2, and Multiple call record data are put into the deep learning model to obtain the first customer emotion prediction parameters and the first emotion analysis model, but this case is not limited to this.

Subsequently, the processor 130 is used to compare whether the customer emotion record parameter is the same as the first customer emotion prediction parameter. Then, when the customer emotion record parameter is the same as the first customer emotion prediction parameter, the processor 130 is used to output the first emotion analysis model. For example, the customer emotion record parameter may range from 0 to 1, and the first customer emotion prediction parameter may range from 0 to 1. When the customer emotion record parameter is the same as the first customer emotion prediction parameter, the first emotion analysis model is used for prediction. The accuracy of customer emotions is high. The processor 130 can output and package the first emotion analysis model into an application programming interface (Application Programming Interface, API), and this API can be used to output customer emotion prediction parameters, but this case does not use this as an example. limit.

In one embodiment, the processor 130 is further configured to perform the following steps according to the plurality of instructions in the memory 140: obtain the second customer emotion prediction parameter through the first emotion analysis model stored in the server 120 according to the plurality of voice data. For example, the plurality of voice data may be the voice content of the customer's call, and the second customer emotion prediction parameter is obtained through the first emotion analysis model stored in the server 120 based on the voice content of the customer's call.

Figure 2 is a usage scenario diagram of an emotion analysis system according to an embodiment of this case. Please refer to Figures 1 and 2 together. In one embodiment, the processor 130 is further configured to perform the following steps according to the plurality of instructions of the memory 140: when the second customer emotion prediction parameter exceeds the emotion parameter threshold, control The display 150 displays warning information A1. For example, the emotion analysis system 100 may further include a display 150, the emotion parameter threshold may be 0.8, and the second customer emotion prediction parameter may be 0.9. At this time, because the second customer emotion prediction parameter exceeds the emotion parameter threshold, the warning information A1 is displayed. on display 150. In addition, the second customer sentiment prediction parameter can be called a negative index, and the content of warning information A1 can be "Customer negativity is high, please pay attention", but this case is not limited to this.

In other words, the purpose of displaying the warning information A1 on the display 150 is to remind customer service personnel and their direct supervisors to pay special attention and take immediate actions to appease customers to avoid potential customer complaints.

In one embodiment, the processor 130 is further configured to perform the following steps according to the plurality of instructions in the memory 140: when the customer emotion record parameter is different from the first customer emotion prediction parameter, adjust the hyperparameters in the deep learning algorithm. Then, the processor 130 is used to compare whether the customer emotion record parameter and the third customer emotion prediction parameter are the same. Subsequently, when the customer emotion record parameter is the same as the third customer emotion prediction parameter, the processor 130 is used to output the second emotion analysis model. For example, when the customer emotion record parameters are different from the first customer emotion prediction parameters, that is, the accuracy of the first emotion analysis model is too low, so the hyperparameters in the deep learning algorithm are adjusted until the deep learning algorithm generates The accuracy of the sentiment analysis model is in line with expectations, but this case is not limited to this.

In one embodiment, the plurality of call record data voice data may be word data, data on the number of incoming calls on the day, audio data, volume data, speech rate data or call time data. For example, the word data can be the word data of the customer's voice after being segmented into sentences. The data of the number of incoming calls on the day can be the number of times calculated based on the customer's ID number as a unit. For example, every time a customer contacts customer service You need to report your ID card number once to calculate the number of incoming calls that day. The audio data can be the audio value when the customer contacts the customer service. For example, when the customer is so angry that the pitch of the voice breaks, the audio data can be 10. The voice volume data can be the voice volume value when the customer is in contact with the customer service. For example, when the customer is so angry that he roars, the voice volume data can be 9. The speech speed data can be the speech speed value when the customer is in contact with the customer service. For example, When a customer is so anxious that he speaks so fast that he can hardly hear clearly, the speaking speed data can be 8, and the call time data can be the total call time when the customer is in contact with customer service, but this case is not limited to this.

In one embodiment, the data on the number of incoming calls on the day is related to the customer's ID card data. For example, the data on the number of incoming calls can be the number of incoming calls. The processor 130 can calculate the data on the number of incoming calls on that day from the number of customer ID data recorded in the server 120 in a day. For details, It is said that if the number of customer ID data in the server 120 is 3 in a day, it can be calculated that the number of incoming calls that day is 3, but this case is not limited to this.

It can be seen from the above embodiments that the application of this case has the following advantages. The emotion analysis system shown in the embodiment of this case can detect customers' emotions in advance to avoid customer complaints.

Although the above implementation mode discloses specific examples of the present case, it is not intended to limit the present case. Persons with ordinary knowledge in the technical field to which the present case belongs can, without departing from the principles and spirit of the present case, proceed with it. Various changes and modifications, therefore the scope of protection in this case shall be subject to the scope of the accompanying patent application.

100: Sentiment Analysis System 110:Voice receiver 120:Server 130: Processor 140:Memory 150:Display A1: Warning information

In order to make the above and other purposes, features, advantages and embodiments of this case more obvious and understandable, the attached drawings are described as follows: Figure 1 is a block diagram of an emotion analysis system according to an embodiment of the present invention. Figure 2 is a usage scenario diagram of an emotion analysis system according to an embodiment of this case. In accordance with common practice, the various features and components in the drawings are not drawn to scale, but are drawn in such a way as to best present the specific features and components relevant to this case. In addition, the same or similar reference symbols are used to refer to similar elements/components in different drawings.

Domestic storage information (please note in order of storage institution, date and number) without Overseas storage information (please note in order of storage country, institution, date, and number) without

100: Sentiment Analysis System

110:Voice receiver

120:Server

130: Processor

140:Memory

Claims (10)

A sentiment analysis system that includes: a voice receiver for collecting plural voice data; a server telecommunications coupled to the voice receiver and used to store plural voice data; and A processor, telecommunications coupled to the server, and configured to perform the following steps according to a plurality of instructions from a memory: Store multiple call record data in the server based on the voice data; marking the call record data with annotation data based on the voice data; Obtain a customer sentiment record parameter for the call record data based on the annotation data; Apply the call record data through a deep learning algorithm to obtain a first customer emotion prediction parameter and a first emotion analysis model; Compare whether the customer sentiment record parameter is the same as the first customer sentiment prediction parameter; and When the customer emotion record parameter is the same as the first customer emotion prediction parameter, the first emotion analysis model is output. The sentiment analysis system of claim 1, wherein the processor is further used to perform the following steps according to the instructions of the memory: A second customer emotion prediction parameter is obtained through the first emotion analysis model stored in the server based on the voice data. The sentiment analysis system of claim 2, wherein the processor is further used to perform the following steps according to the instructions of the memory: When the second customer emotion prediction parameter exceeds an emotion parameter threshold, a display is controlled to display a warning message. The sentiment analysis system of claim 2, wherein the processor is further used to perform the following steps according to the instructions of the memory: When the customer emotion record parameter is different from the first customer emotion prediction parameter, a hyperparameter in the deep learning algorithm is adjusted. The sentiment analysis system of claim 4, wherein the processor is further used to perform the following steps according to the instructions of the memory: The voice data is passed through the deep learning algorithm to obtain a third customer emotion prediction parameter and a second emotion analysis model. The sentiment analysis system of claim 5, wherein the processor is further used to perform the following steps according to the instructions of the memory: Compare whether the customer emotion record parameter is the same as the third customer emotion prediction parameter. The sentiment analysis system of claim 6, wherein the processor is further used to perform the following steps according to the instructions of the memory: When the customer emotion record parameter is the same as the third customer new emotion prediction parameter, the second emotion analysis model is output. The sentiment analysis system as described in request item 1, wherein the call record data includes one word data, one day's incoming call number data, one audio data, one sound volume data, one speech speed data and one call time data. At least one of them. The sentiment analysis system as described in claim 1, wherein the call record data includes customer ID card data. For the sentiment analysis system described in request item 9, one of the data on the number of incoming calls on that day is related to the customer's ID card data.

Images