WO2016122294A1 - Evolutionary decision-making system and method operating according to criteria with automatic updates - Google Patents

Abstract

The invention relates to components of an evolutionary decision-making system which adjusts to the changes presented by the data feeding the components thereof over time. The advantage of this type of system is that, over time, they become more precise, such that the more cases evaluated, the greater the amount of information available for training classification or predictive models.

Description

 SYSTEM AND EVOLUTIVE METHOD OF DECISION MAKING THAT WORKS BY CRITERIA WITH

 AUTOMATIC UPDATES

BACKGROUND Although the first monograph on the expression of emotions in animals and humans was written by Charles Darwin in the last century, psychologists have gradually accumulated knowledge in the field of emotion detection and voice recognition, which It has attracted a new wave of interest recently by two psychologists and artificial intelligence specialists. There are several reasons for this renewed interest: the technological progress of recording, storage and processing of audio and visual information; the development of non-invasive sensors; the arrival of laptops; and the need to enrich the man-machine interface. Furthermore, a new field of research in artificial intelligence known as affective computing has recently been identified. As for research on the recognition of emotions in speech, on the one hand, psychologists have done many experiments and suggested theories. On the other hand, artificial intelligence researchers made contributions in the following areas: emotional voice synthesis, recognition of emotions and the use of agents for decoding and expressing emotions. Similar progress has been made with voice recognition.

Despite research on the recognition of emotions in speech, art has been devoid of methods and devices that use emotion recognition and voice recognition for commercial purposes.

Current needs demand the maximum use of resources and find competitive advantages that allow not only to remain in the market, but to highlight and surpass their competitors. Today's technologies offer a variety of solutions that can be used in various areas. High performance computing equipment and complex artificial intelligence algorithms allow to create dynamic solutions that precisely give the competitive advantage with intelligent systems capable of "learning" from "experience", increasingly reducing the need for human supervision.

Interpersonal relationships are constantly evolving, and we must consider that operators establish relationships with customers several times a day and therefore prove that the formula for achieving customer engagement varies depending on the customer himself, his mood and other unknown factors. This system evaluates the relationships between the client and the operator and decides which of the clients to call, according to the availability of the operators, using classification / prediction models. The models are updated every time the database grows by 10%. Growth verification, as well as re-training of models is done automatically, without human manual supervision.

DETAILED DESCRIPTION OF THE INVENTION Brief description of figures:

Figure 1 shows a flow chart of automatic verification when the Database (BD) has grown more than 10 percent. Figure 2 is a block diagram of call processing by a system that determines the emotions of the interlocutors (client or operator).

Figure 3 shows a decision diagram where the probability calculation processes and the classifications involving models.

Figure 4 illustrates the classification of customers based on payment response. Figure 5 shows the classification of users according to emotion / personality.

Assignment of the client-operator pair

 Traditional methods of assigning operators to customers (debtors) involve only the characteristics of the client and the needs of the company when making the call:

 - Customer availability (schedule, so that you can answer the call)

 - Amount

 - Due date

 - Status or classification (refers to customer compliance according to their history) and other criteria that vary according to the company's turn.

This invention complements the selection criteria of the customer to call with the information of the operators available at that time, so that the selection of the customer and the assignment to the operator is done so that the chances of success are greater, according to the result previously observed.

We have the characteristics of the client and a registry of operators who have succeeded in their efforts with him. Each operator has a history of successful clients with an indicator of their level of influence on the final result (outcome), this can be positive or negative. However, what should determine the call must be the relationship between the client and the operator, this is determined according to the emotions during the interaction at the time of the call.

 Example 1:

 Client 1 meets all the requirements according to the traditional criteria of the company to be required with a call. Operators A, B and C are available. The process that determines whether the customer is called or not is the following:

^"* -, ·· ·; '. ->

 Operator A Operator B Operator C |

NumLLam 15 25 0

Success 12 8 0

Yes 10 6 0

No 3 11 0

NumLLam: Number of times the operator has called the customer

 Success: Yes paid

 Yes: The client promised to pay

 No: The client did not agree to pay

The objective is to decide if the client is called or not according to. the operators available to assist you. In this case, looking at the numbers with the naked eye we could conclude that Operator A is a good option and that it has a good radius of success with the client. Operator 2 however has not been as successful, while Operator C has not had contact with Client 1 once. Getting carried away by the percentage of successful calls over total calls can be a mistake, as these numbers will not always give us a clear view of the client-operator relationship.

 The explanation is simple: relationships between people tend to change, it evolves. So the Operator 1 that we believe is the one indicated for the call, could have succeeded in the first 12 calls and had a terrible interaction during the remaining 3, which disqualifies him to attend to the client. On the other hand, if successful calls and bad interactions have been interspersed then Operator A may not be discarded at all. Finally, if the success calls have been all at the end, having the negatives at the beginning, definitely the Operator A is an excellent option and the call is made to Client 1.

 The key is then in the relationship between the operator and the customer. This is determined with a series of factors based mainly on the interaction during calls by means of software that determines the emotions in audio for each interlocutor. The representation of each call for these purposes is then a chain of labels that indicate emotion (joy, sadness, anger and neutral) and the interlocutor (operator, client). A time series algorithm then classifies the transition of the call from the beginning to the end and determines whether the call had a good, bad or neutral interaction. Each tag has a probability. Example 2:

 Call: 13

 Client: 1

 Operator: A Good Bad Neutral

Ratio: 0.78 0.12 0.10

In this case the probability that the call has had a good interaction is 78% and is qualified as such, without discarding the numbers for the other labels, which are stored in a table for future processes.

Figure 2. The call is processed by a system that determines the emotions of the partners (client or operator). A time series algorithm classifies the transition of the call from the beginning to the end and determines whether the call had a good, bad or neutral interaction. Classes of operators and Clients

 The selection of clients and operators is preferably made in a personalized way, that is, there is a list of specific operators for each client. However, the staff is constantly rotated and it is convenient to have a relationship between the characteristics of the operators that are successful with a specific customer or, the characteristics of the clients for a particular operator. We can generalize based on the knowledge obtained through observations and define levels of success or failure between types of customers and operators according to their characteristics.

Figure 3 shows a diagram of decisions where the probability calculation processes and the classifications that involve models (in blue background) are updated automatically when the growth of the records that intervene equals or exceeds 10%. Customer history is evaluated against operator history and a list of customers with a higher probability of committing to pay is obtained. The blocks are modified according to the growth of the operator-client records in the database. Figure 4. Classification of customers based on payment response. Classification / prediction models are re-trained or updated automatically when customer history increases by 10%. It shows the general structure of the customer classification process based on payment response. In this case, when automatic monitoring detects a 10% or greater growth in customer history, the prediction models are trained, so that the new information becomes part of the system's knowledge.

Figure 5. Classification of users according to emotion / personality. In this case, the class assignment according to the emotion is updated when the history growth meets the established criteria (growth of at least 10%). In the case of the classification of users according to emotion / personality, the update consists in recalculating the emotion label when the number of instances or calls exceeds or equals 10%. In this process the models of emotion recognition and voice recognition are not retrained. The established criteria for re-training can be modified without altering the functioning of the system components. Although the update is a measure that requires constant monitoring that consumes resources, in the long run it will result in greater accuracy in predicting the response of customers, which will allow assigning operators according to each case. The classification models of the operators and all the indicators that are calculated based on histories are kept in this form constantly updated.

Claims

CLAIMS that are claimed is: 1. An evolutionary system and method of decision making that adjusts to the changes presented by the data that feed its components over time and that works as follows: a. The system automatically verifies when the Database (BD) has grown by more than 10 percent, by the time this condition is met, a retraining process of the classification models is launched. b. A call is processed by a system that determines the emotions of the partners (client or operator). A time series algorithm classifies the transition of the call from the beginning to the end and determines whether the call had a good, bad or neutral interaction. C. Decision making where the probability calculation processes and the classifications that involve models are updated automatically when the growth of the records that intervene equals or exceeds 10%. Customer history is evaluated against operator history and a list of customers with a higher probability of committing to pay is obtained. The blocks are modified according to the growth of the operator-client records in the database. d. Classification of customers based on payment response. Classification / prediction models are re-trained or updated automatically when customer history increases by 10%. It shows the general structure of the customer classification process based on payment response. and. User classification according to emotion / personality. In this case, the class assignment according to the emotion is updated when the history growth meets the established criteria (growth of at least 10%). 2. The system and method of the previous claim that in the case of the classification of users according to the emotion / personality, the update consists in recalculating the emotion label when the number of instances or calls exceeds or equals the 10% In this process the models of emotion recognition and voice recognition are not retrained. The system and method of claim 1 that when automatic monitoring detects a 10% or greater growth in customer history, the prediction models are re-trained, so that the new information becomes part of the knowledge of the system.