CN111865926A

CN111865926A - Call channel construction method and device based on double models and computer equipment

Info

Publication number: CN111865926A
Application number: CN202010592337.1A
Authority: CN
Inventors: 赵桂花; 叶松
Original assignee: OneConnect Financial Technology Co Ltd Shanghai
Current assignee: OneConnect Smart Technology Co Ltd; OneConnect Financial Technology Co Ltd Shanghai
Priority date: 2020-06-24
Filing date: 2020-06-24
Publication date: 2020-10-30

Abstract

The application relates to the technical field of artificial intelligence, and discloses a call channel construction method and device based on dual models, computer equipment and a storage medium, wherein the call channel construction method comprises the following steps: acquiring a call request sent by a first terminal; acquiring first personal feature information to obtain a first voice feature cluster; calculating a first clustering center; acquiring second voice characteristic data, and mapping the second voice characteristic data into a first central vector and a second voice vector; calculating a first distance value; acquiring second person characteristic information to obtain a second voice characteristic cluster; calculating a second cluster center; acquiring first voice feature data, a second central vector and a first voice vector; calculating a second distance value; and if the distance values are smaller than the first distance threshold value and the sum of the first distance value and the second distance value is smaller than a preset second distance threshold value, establishing a communication channel between the first terminal and the second terminal. Thereby improving the call efficiency. In addition, the application also relates to a block chain technology, and the dual model can be stored in the block chain.

Description

Call channel construction method and device based on double models and computer equipment

Technical Field

The application relates to the technical field of artificial intelligence, in particular to a call channel construction method and device based on double models, computer equipment and a storage medium.

Background

Remote call technology has become popular, and in the conventional remote call technology, as long as a call initiator initiates a call request and a call receiver agrees with the call request, remote call can be performed. Because the conventional remote communication technology cannot determine whether the two parties can effectively communicate with each other, the communication efficiency cannot be guaranteed. Specifically, in a communication process, there is a requirement for a batch communication, for example, a communication requirement between a class i person and a class ii person, and at this time, a specific natural person is not called another natural person, but a communication between a natural person group and another natural person group (for example, a communication between a customer service department of a company and a corresponding customer group) is performed, and at this time, if a conventional remote communication technology is adopted, there is a defect that the communication efficiency is low. Therefore, the call efficiency of the conventional remote call technology cannot be guaranteed.

Disclosure of Invention

The application mainly aims to provide a call channel construction method and device based on a dual model, computer equipment and a storage medium, and aims to improve call efficiency.

In order to achieve the above object, the present application provides a method for constructing a call channel based on dual models, including the following steps:

the method comprises the steps that a call request sent by a first terminal is obtained, and the call request is used for requesting remote call with a second terminal; the first terminal belongs to a preset first cluster, and the second terminal belongs to a preset second cluster;

acquiring first personal characteristic information of a first user corresponding to the first terminal, and inputting the first personal characteristic information into a preset forward prediction model for calculation, so as to obtain a first voice characteristic cluster output by the forward prediction model; the first voice feature cluster is composed of a plurality of voice feature data, the forward prediction model is based on a preset neural network model and is obtained by training with training data, and the training data comprises a call record which is collected in advance and marked as an efficient call by an artificial marker, personal feature information of a call initiator in the call record and voice feature information of a call receiver in the call record;

calculating a first clustering center of the first voice feature cluster according to a preset clustering center calculation method;

Acquiring second voice characteristic data of a user corresponding to the second terminal, and mapping the first clustering center and the second voice characteristic data into a first center vector and a second voice vector of a high-dimensional space respectively;

calculating a first distance value of the first center vector and a second voice vector according to a preset distance calculation method;

acquiring second personal characteristic information of a second user corresponding to the second terminal, and inputting the second personal characteristic information into a preset reverse prediction model for calculation so as to obtain a second voice characteristic cluster output by the reverse prediction model;

calculating a second clustering center of the second voice feature cluster according to a preset clustering center calculation method;

acquiring first voice characteristic data of a user corresponding to the first terminal, and mapping the second clustering center and the first voice characteristic data into a second center vector and a first voice vector of a high-dimensional space respectively;

calculating a second distance value between the second center vector and the first voice vector according to a preset distance calculation method;

judging whether the first distance value and the second distance value are both smaller than a preset first distance threshold value, and judging whether the sum of the first distance value and the second distance value is smaller than a preset second distance threshold value, wherein the value range of the second distance threshold value is (the first distance threshold value, p is multiplied by the first distance threshold value), and p is a positive number which is larger than 1 and smaller than 2;

And if the first distance value and the second distance value are both smaller than a preset first distance threshold value, and the sum of the first distance value and the second distance value is smaller than a preset second distance threshold value, establishing a communication channel between the first terminal and the second terminal.

Further, the first personal feature information of the first user corresponding to the first terminal is obtained and is input into a preset forward prediction model for calculation, so that a first voice feature cluster output by the forward prediction model is obtained; the method comprises the following steps of obtaining a forward prediction model based on a preset neural network model, obtaining training data by utilizing the training data, wherein the first voice feature cluster is composed of a plurality of voice feature data, the forward prediction model is obtained by training the forward prediction model based on the preset neural network model, and the training data comprises a call record which is collected in advance and is marked as an efficient call by an artificial marker, personal feature information of a call initiator in the call record and voice feature information of a call receiver in the call record, and comprises the following steps:

calling a specified amount of sample data from a preset sample database, and dividing the sample data into training data and verification data according to a preset proportion; the sample data comprises call records which are manually marked as efficient calls and collected in advance, personal characteristic information of a call initiator in the call records and voice characteristic information of a call receiver in the call records;

Inputting the training data into a preset neural network model for training so as to obtain an intermediate model;

verifying the intermediate model by using the verification data to obtain a verification result, and judging whether the verification result is passed;

and if the verification result is that the verification is passed, marking the intermediate model as a forward prediction model.

Further, the step of calculating a first distance value between the first center vector and the second speech vector according to a preset distance calculation method includes:

obtaining values of u components of the first center vector and values of u components of the second speech vector, the first center vector and the second speech vector each having u components;

according to the formula:

and calculating a first distance value T of the first central vector and the second voice vector, wherein Ei is the numerical value of the ith component vector of the first central vector, and Ri is the numerical value of the ith component vector of the second voice vector.

Further, after the step of determining whether both the first distance value and the second distance value are smaller than a preset first distance threshold and determining whether a sum of the first distance value and the second distance value is smaller than a preset second distance threshold, where a value range of the second distance threshold is (the first distance threshold, p × the first distance threshold), and p is a positive number greater than 1 and smaller than 2, the method includes:

If the first distance value is not smaller than a preset first distance threshold, or the second distance value is not smaller than a preset first distance threshold, or the sum of the first distance value and the second distance value is not smaller than a preset second distance threshold, screening a third terminal from the second cluster according to a preset terminal screening method;

acquiring third voice characteristic data corresponding to the third terminal, and mapping the third voice characteristic data into a third voice vector of a high-dimensional space;

calculating a third distance value of the first center vector and a third voice vector according to a preset distance calculation method;

acquiring third personal characteristic information of a third user corresponding to the third terminal, and inputting the third personal characteristic information into a preset reverse prediction model for calculation, so as to obtain a third voice characteristic cluster output by the reverse prediction model;

calculating a third clustering center of the third voice feature cluster according to a preset clustering center calculation method;

mapping the third cluster center to a third center vector of a high-dimensional space;

calculating a fourth distance value between the third center vector and the first voice vector according to a preset distance calculation method;

Judging whether the third distance value and the fourth distance value are both smaller than the first distance threshold value, and judging whether the sum of the third distance value and the fourth distance value is smaller than a preset third distance threshold value, wherein the value range of the third distance threshold value is (the first distance threshold value, q is multiplied by the first distance threshold value), and q is a positive number which is larger than 1 and smaller than p;

and if the third distance value and the fourth distance value are both smaller than the first distance threshold value, and the sum of the third distance value and the fourth distance value is smaller than a preset third distance threshold value, establishing a communication channel between the first terminal and the third terminal.

Further, the step of establishing a communication channel between the first terminal and the second terminal includes:

acquiring the call request times A1 and successful call times B1 sent by the terminal I in the preset time, and acquiring the requested call times A2 and successful call times B2 of the terminal II in the preset time;

according to the formula: y1 ═ B1 × a2)/(a1 × B2), Y2 ═ B1/a1, and Y3 ═ B2/a2, and a first factor Y1, a second factor Y2, and a third factor Y3 were calculated, respectively;

judging whether the value of max (Y1, Y2, Y3,1) is equal to Y1 or not, and judging whether the value of min (Y1, Y2, Y3, v) is equal to v or not, wherein v is a preset parameter which is less than 1 and more than 0;

And if the value of max (Y1, Y2, Y3,1) is equal to Y1, and the value of min (Y1, Y2, Y3, v) is equal to v, constructing a call channel between the first terminal and the second terminal.

The application provides a conversation passageway founds device based on bimodulus, includes:

the first call request acquisition unit is used for acquiring a call request sent by a first terminal, and the call request is used for requesting remote call with a second terminal; the first terminal belongs to a preset first cluster, and the second terminal belongs to a preset second cluster;

the first voice feature cluster obtaining unit is used for obtaining first personal feature information of a first user corresponding to the first terminal, and inputting the first personal feature information into a preset forward prediction model for calculation, so that a first voice feature cluster output by the forward prediction model is obtained; the first voice feature cluster is composed of a plurality of voice feature data, the forward prediction model is based on a preset neural network model and is obtained by training with training data, and the training data comprises a call record which is collected in advance and marked as an efficient call by an artificial marker, personal feature information of a call initiator in the call record and voice feature information of a call receiver in the call record;

The first clustering center calculating unit is used for calculating a first clustering center of the first voice feature cluster according to a preset clustering center calculating method;

a second voice feature data obtaining unit, configured to obtain second voice feature data of a user corresponding to the second terminal, and map the first clustering center and the second voice feature data into a first center vector and a second voice vector of a high-dimensional space, respectively;

the first distance value calculation unit is used for calculating first distance values of the first center vector and the second voice vector according to a preset distance calculation method;

a second speech feature cluster obtaining unit, configured to obtain second personal feature information of a second user corresponding to the second terminal, and input the second personal feature information into a preset reverse prediction model for calculation, so as to obtain a second speech feature cluster output by the reverse prediction model;

the second clustering center calculating unit is used for calculating a second clustering center of the second voice feature cluster according to a preset clustering center calculating method;

the first voice characteristic data acquisition unit is used for acquiring first voice characteristic data of a user corresponding to the first terminal, and mapping the second clustering center and the first voice characteristic data into a second center vector and a first voice vector of a high-dimensional space respectively;

The second distance value calculation unit is used for calculating a second distance value between the second center vector and the first voice vector according to a preset distance calculation method;

a distance threshold determination unit, configured to determine whether both the first distance value and the second distance value are smaller than a preset first distance threshold, and determine whether a sum of the first distance value and the second distance value is smaller than a preset second distance threshold, where a value range of the second distance threshold is (the first distance threshold, p × the first distance threshold), and p is a positive number greater than 1 and smaller than 2;

and the call channel construction unit is used for constructing a call channel between the first terminal and the second terminal if the first distance value and the second distance value are both smaller than a preset first distance threshold value and the sum of the first distance value and the second distance value is smaller than a preset second distance threshold value.

Further, the apparatus comprises:

the system comprises a sample data dividing unit, a verification unit and a data processing unit, wherein the sample data dividing unit is used for calling sample data with specified quantity from a preset sample database and dividing the sample data into training data and verification data according to a preset proportion; the sample data comprises call records which are manually marked as efficient calls and collected in advance, personal characteristic information of a call initiator in the call records and voice characteristic information of a call receiver in the call records;

The training unit is used for inputting the training data into a preset neural network model for training so as to obtain an intermediate model;

the verification unit is used for verifying the intermediate model by using the verification data to obtain a verification result and judging whether the verification result is passed;

and the forward prediction model marking unit is used for marking the intermediate model as a forward prediction model if the verification result is that the verification is passed.

Further, the first distance value calculating unit includes:

a component value obtaining subunit, configured to obtain values of u component vectors of the first center vector and obtain values of u component vectors of the second speech vector, where the first center vector and the second speech vector both have u component vectors;

a first distance value T calculation subunit configured to:

The present application provides a computer device comprising a memory storing a computer program and a processor implementing the steps of any of the above methods when the processor executes the computer program.

The present application provides a computer-readable storage medium having stored thereon a computer program which, when being executed by a processor, carries out the steps of the method of any of the above.

The call channel construction method and device based on the dual model, the computer equipment and the storage medium acquire a call request sent by a first terminal; acquiring first personal feature information to obtain a first voice feature cluster; calculating a first clustering center; acquiring second voice characteristic data, and mapping the first clustering center and the second voice characteristic data into a first center vector and a second voice vector respectively; calculating a first distance value of the first center vector and the second voice vector; acquiring second person characteristic information of the second user to obtain a second voice characteristic cluster; calculating a second cluster center; acquiring first voice feature data, a second central vector and a first voice vector; calculating a second distance value between the second center vector and the first voice vector; and if the first distance value and the second distance value are both smaller than a preset first distance threshold value, and the sum of the first distance value and the second distance value is smaller than a preset second distance threshold value, establishing a communication channel between the first terminal and the second terminal. Thereby improving the call efficiency.

Drawings

Fig. 1 is a schematic flowchart of a call channel construction method based on dual models according to an embodiment of the present application;

fig. 2 is a schematic block diagram of a call channel constructing apparatus based on dual models according to an embodiment of the present application;

fig. 3 is a block diagram illustrating a structure of a computer device according to an embodiment of the present application.

The implementation, functional features and advantages of the objectives of the present application will be further explained with reference to the accompanying drawings.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

Referring to fig. 1, an embodiment of the present application provides a method for constructing a call channel based on a dual model, including the following steps:

s1, acquiring a call request sent by the first terminal, wherein the call request is used for requesting remote call with the second terminal; the first terminal belongs to a preset first cluster, and the second terminal belongs to a preset second cluster;

s2, acquiring first personal feature information of a first user corresponding to the first terminal, and inputting the first personal feature information into a preset forward prediction model for calculation, so as to obtain a first voice feature cluster output by the forward prediction model; the first voice feature cluster is composed of a plurality of voice feature data, the forward prediction model is based on a preset neural network model and is obtained by training with training data, and the training data comprises a call record which is collected in advance and marked as an efficient call by an artificial marker, personal feature information of a call initiator in the call record and voice feature information of a call receiver in the call record;

S3, calculating a first clustering center of the first voice feature cluster according to a preset clustering center calculation method;

s4, acquiring second voice characteristic data of a user corresponding to the second terminal, and mapping the first clustering center and the second voice characteristic data into a first center vector and a second voice vector of a high-dimensional space respectively;

s5, calculating a first distance value of the first center vector and the second voice vector according to a preset distance calculation method;

s6, obtaining second personal characteristic information of a second user corresponding to the second terminal, and inputting the second personal characteristic information into a preset reverse prediction model for calculation, so as to obtain a second voice characteristic cluster output by the reverse prediction model;

s7, calculating a second clustering center of the second voice feature cluster according to a preset clustering center calculation method;

s8, acquiring first voice characteristic data of a user corresponding to the first terminal, and mapping the second clustering center and the first voice characteristic data into a second center vector and a first voice vector of a high-dimensional space respectively;

s9, calculating a second distance value between the second center vector and the first voice vector according to a preset distance calculation method;

S10, determining whether the first distance value and the second distance value are both smaller than a preset first distance threshold, and determining whether the sum of the first distance value and the second distance value is smaller than a preset second distance threshold, where the value range of the second distance threshold is (the first distance threshold, p × the first distance threshold), and p is a positive number greater than 1 and smaller than 2;

s11, if the first distance value and the second distance value are both smaller than a preset first distance threshold value, and the sum of the first distance value and the second distance value is smaller than a preset second distance threshold value, a communication channel between the first terminal and the second terminal is constructed.

According to the method and the device, the suitable conversation opposite side of the two parties is comprehensively considered through comprehensive setting of the forward prediction model and the reverse prediction model, and therefore conversation efficiency is improved. The execution subject of the present application may be any feasible subject, such as a server. In addition, the method and the device can be applied to any feasible scene, such as conversation between a business terminal corresponding to business personnel and a client terminal corresponding to a client, wherein the business personnel are borrow collection urging personnel. Wherein the dual models refer to a forward prediction model and a reverse prediction model.

As described in step S1, a call request sent by the terminal one is obtained, where the call request is used to request a remote call with the terminal two; the first terminal belongs to a preset first cluster, and the second terminal belongs to a preset second cluster. The first cluster comprises a plurality of terminals with the same attribute as the first terminal, for example, the terminals are all corresponding to service personnel. The second cluster comprises a plurality of terminals having the same attributes as terminal two, for example, terminals corresponding to clients. In the conventional communication technology, as long as a terminal one requests to communicate with a terminal two, a communication channel between the two terminals is established. The method is different from the method, whether the first terminal and the second terminal can carry out efficient conversation or not is predicted through special design, and if the first terminal and the second terminal can carry out efficient conversation, a conversation channel is established, so that the efficiency is improved, and the time is saved.

As described in step S2, obtaining first personal feature information of the first user corresponding to the first terminal, and inputting the first personal feature information into a preset forward prediction model for calculation, so as to obtain a first speech feature cluster output by the forward prediction model; the first voice feature cluster is composed of a plurality of voice feature data, the forward prediction model is based on a preset neural network model and is obtained by training with training data, and the training data comprises call records which are collected in advance and marked as efficient calls manually, personal feature information of a call initiator in the call records and voice feature information of a call receiver in the call records. The human being is a sensing creature, and the reaction is that the intention of the call and the efficiency of the call are different for different people (with different voice characteristics) in the call process. The method and the device utilize the natural attribute of the person to improve the conversation efficiency. The personal characteristic information includes, for example: educational background, work position, gender, age, recent calls, and the like. Due to the differences of the education background, the working position and the gender, the communication objects with the preferences of the users can be different, for example, the users who do not have high education tend to have popular spoken language expressions, the users who are male tend to communicate with the users with soft tone. Therefore, the method and the device perform prediction through a forward prediction model, wherein the forward prediction model is based on a preset neural network model and is obtained by training through training data, and the training data comprises call records which are collected in advance and marked as efficient calls manually, personal characteristic information of a call initiator in the call records and voice characteristic information of a call receiver in the call records. Therefore, the forward prediction model is sufficient to predict the speech features that another party capable of performing efficient communication with an object should have according to the personal feature information of the object, and then output all the speech features, so as to obtain a first speech feature cluster (i.e. a set of all call objects capable of performing efficient communication). The first speech feature cluster thus actually includes speech feature data respectively possessed by a plurality of call partners, for example, the first speech feature cluster includes n speech feature data, where each speech feature data represents one or a class of people. The voice features include, for example: speech rate, timbre, spoken language, frequency of use of the term of expertise, and the like. In addition, it should be mentioned here that the forward prediction model of the present application is trained based on special training data, that is, the training data includes a call record collected in advance and manually marked as an efficient call, personal feature information of a call initiator in the call record, and voice feature information of a call recipient in the call record. The training data only comprises call records which are manually marked as efficient calls, but not comprises call records of inefficient calls, so that the training data is a semi-supervised training mode and can effectively improve the training speed of the model. The efficiency call is manually marked, but may also be assisted by a computer, for example, the computer marks the call time less than a preset time threshold, marks the call that achieves the purpose of the call as a tentative call, and manually verifies whether the tentative call is an efficiency call (or may be directly marked as an efficiency call).

As described in step S3, the first cluster center of the first speech feature cluster is calculated according to a preset cluster center calculation method. Wherein the first speech feature cluster is composed of a plurality of speech feature data, each representing a population of people. If the second user corresponding to the second terminal can perform an efficient call with the first user corresponding to the first terminal only by using whether the first voice feature cluster falls into the first voice feature cluster or not as a basis, the judgment accuracy is low. The reason is that: if the speech feature of the second user is at the edge of the first speech feature cluster, the probability that the second user will effectively communicate with the first user (the user corresponding to the first terminal) is low (relative to). In order to overcome the defect, the method comprehensively utilizes the forward prediction model, the reverse prediction model, the first clustering center, the second clustering center, the first distance value and the second distance value to measure the possibility of efficiency conversation, thereby improving the prediction accuracy. Therefore, clustering calculation is required at this time. The cluster calculation may be performed in any feasible manner, such as KNN clustering, and will not be described herein again.

As described in the above steps S4-S5, second voice feature data of the user corresponding to the terminal ii is obtained, and the first clustering center and the second voice feature data are mapped into a first center vector and a second voice vector of a high-dimensional space, respectively; and calculating a first distance value of the first center vector and the second voice vector according to a preset distance calculation method. The mapping method may be any feasible method, and the simplest method is to map the speech feature data into a dimensional value a if the speech speed is a word/second, for example. The distance calculation method is used for calculating the distance between vectors, and any feasible algorithm can be adopted, such as an Euclidean distance algorithm and the like. The first distance value obtained at this time reflects the difference between the second user and the ideal speech feature.

As described in step S6, second personal feature information of the second user corresponding to the second terminal is obtained, and the second personal feature information is input into a preset reverse prediction model for calculation, so as to obtain a second speech feature cluster output by the reverse prediction model. The reverse prediction model is opposite to the forward prediction model, namely, the forward prediction model predicts the voice characteristics of the suitable call receiver by the personal characteristic information of the call initiator, and the reverse prediction model predicts the voice characteristics of the suitable call initiator by the personal characteristic information of the call receiver. These two models look the same but are in fact distinct because the master-slave roles of the call initiator and the call receiver are different and irreversible, reflected on the training data, i.e. the training data of the two models are not common, and the training data of the inverse predictive model needs to include the call record that is manually marked as an efficient call, the personal characteristics information of the call receiver in the call record, and the voice characteristics information of the call initiator in the call record. In addition, the inverse prediction model may also be trained by using a neural network model, and no additional limitation is imposed here.

Calculating a second cluster center of the second speech feature cluster according to a preset cluster center calculation method as described in the above steps S7-S9; acquiring first voice characteristic data of a user corresponding to the first terminal, and mapping the second clustering center and the first voice characteristic data into a second center vector and a first voice vector of a high-dimensional space respectively; and calculating a second distance value between the second center vector and the first voice vector according to a preset distance calculation method. The method for calculating the second clustering center, the method for vector mapping between the second clustering center and the first speech feature data, and the method for calculating the second distance value may be any feasible method, and may also be the same as the method for calculating the first clustering center, the method for vector mapping between the first clustering center and the second speech feature data, and the method for calculating the first distance value.

As described in step 10 above, it is determined whether both the first distance value and the second distance value are smaller than a preset first distance threshold, and it is determined whether the sum of the first distance value and the second distance value is smaller than a preset second distance threshold, where a value range of the second distance threshold is (the first distance threshold, p × the first distance threshold), and p is a positive number greater than 1 and smaller than 2. The first distance value and the second distance value represent the difference between the voice characteristics of both parties of the call and the ideal state, respectively, and if the first distance value and the second distance value are both small enough, for example, 0, both parties of the call can be regarded as an ideal call target of the opposite party. However, such ideal situation occurs less frequently, and therefore, the present application determines whether the sum of the first distance value and the second distance value is smaller than the preset second distance threshold by determining whether both the first distance value and the second distance value are smaller than the preset first distance threshold, so as to comprehensively determine whether the efficient call between the first terminal and the second terminal can be achieved.

As described in step S11, if the first distance value and the second distance value are both smaller than a preset first distance threshold, and the sum of the first distance value and the second distance value is smaller than a preset second distance threshold, a call channel between the first terminal and the second terminal is established. As described above, it is understood that when the first distance value is larger but still smaller than the first distance threshold, the reaction is that both parties of the call can realize the efficient call during the actual call, but when the second distance is larger and smaller than the first distance threshold, the possibility that both parties of the call can realize the efficient call is greatly reduced. Therefore, the present application considers both parties of a call comprehensively under the condition that both the first distance value and the second distance value are smaller than a preset first distance threshold and the sum of the first distance value and the second distance value is smaller than a preset second distance threshold, so as to prevent the occurrence of a phenomenon that an erroneous judgment is made that an efficient call can be realized in the case that only one voice feature of the other party is adapted to the other party. If the first distance value and the second distance value are both smaller than a preset first distance threshold value, and the sum of the first distance value and the second distance value is smaller than a preset second distance threshold value, it can be ensured that both the first distance value and the second distance value are smaller, and therefore the possibility of efficient conversation is improved after a conversation channel is established.

In one embodiment, the first personal feature information of the first user corresponding to the first terminal is obtained and is input into a preset forward prediction model for calculation, so that a first speech feature cluster output by the forward prediction model is obtained; before step S2, where the first speech feature cluster is composed of a plurality of speech feature data, the forward prediction model is based on a preset neural network model and is obtained by training using training data, and the training data includes a call record that is collected in advance and is manually marked as an efficient call, personal feature information of a call initiator in the call record, and speech feature information of a call recipient in the call record, the method includes:

s101, calling sample data of a specified quantity from a preset sample database, and dividing the sample data into training data and verification data according to a preset proportion; the sample data comprises call records which are manually marked as efficient calls and collected in advance, personal characteristic information of a call initiator in the call records and voice characteristic information of a call receiver in the call records;

S102, inputting the training data into a preset neural network model for training so as to obtain an intermediate model;

s103, verifying the intermediate model by using the verification data to obtain a verification result, and judging whether the verification result is passed;

and S104, if the verification result is that the verification is passed, marking the intermediate model as a forward prediction model.

As described above, obtaining a forward prediction model is achieved. Wherein the predetermined ratio is, for example, 7:3 to 0.99: 0.01. The neural network model may be any feasible model, such as a recurrent neural network model, and so on. Inputting the training data into a preset neural network model for training so as to obtain an intermediate model; verifying the intermediate model by using the verification data to obtain a verification result; if the verification result is that the verification is passed, the intermediate model is proved to be competent for forward prediction, and accordingly the intermediate model is marked as a forward prediction model. Further, the method for dividing the sample data into training data and verification data according to the preset proportion adopts the following method: dividing the sample data into a plurality of batches to obtain a plurality of batches of training data and verification data, wherein according to the increase of the number of the batches, the numerical value of the preset proportion is reduced, so that the proportion of the obtained plurality of batches of training data is more and more, the proportion of the verification data is smaller and smaller, and the plurality of batches of training data and the verification data are used for training and verifying in sequence, so that the sample data is fully used for training. Further, the forward prediction model may be stored in a preset block chain. For example, the forward prediction model is sent to other auditing nodes in the blockchain through one blockchain node in the blockchain. And after the other auditing nodes pass the auditing, storing the forward prediction model into a public account book of the block chain in a new block form. The block chain is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, a consensus mechanism, an encryption algorithm and the like.

In one embodiment, the step S5 of calculating the first distance value between the first center vector and the second speech vector according to a preset distance calculation method includes:

s501, obtaining numerical values of u component vectors of the first center vector and obtaining numerical values of u component vectors of the second voice vector, wherein the first center vector and the second voice vector are provided with u component vectors;

s502, according to a formula:

As described above, it is realized that the first distance values of the first center vector and the second speech vector are calculated according to a preset distance calculation method. The first distance value of the first center vector and the second voice vector is calculated in a special mode, namely a formula is adopted:

the minimum value of the first distance value T is-1, namely when the value of the first distance value T is-1, the first center vector and the second voice vector are completely equal; the larger the value of the first distance value T is, the larger the distance between the first center vector and the second voice vector is, the more dissimilar the first center vector and the second voice vector is. The above formula not only considers the numerical difference between the first center vector and the second speech vector, but also considers the angular difference between the first center vector and the second speech vector, thereby improving the accuracy of distance calculation.

In one embodiment, after the step S10 of determining whether the first distance value and the second distance value are both smaller than a preset first distance threshold and determining whether the sum of the first distance value and the second distance value is smaller than a preset second distance threshold, where the range of the second distance threshold is (the first distance threshold, p × the first distance threshold), and p is a positive number greater than 1 and smaller than 2, the method includes:

s1011, if the first distance value is not less than a preset first distance threshold, or the second distance value is not less than a preset first distance threshold, or the sum of the first distance value and the second distance value is not less than a preset second distance threshold, screening a third terminal from the second cluster according to a preset terminal screening method;

s1012, acquiring third voice characteristic data corresponding to the third terminal, and mapping the third voice characteristic data into a third voice vector of a high-dimensional space;

s1013, calculating a third distance value of the first center vector and a third voice vector according to a preset distance calculation method;

s1014, acquiring third person characteristic information of a third user corresponding to the third terminal, and inputting the third person characteristic information into a preset reverse prediction model for calculation, so as to obtain a third voice characteristic cluster output by the reverse prediction model;

S1015, calculating a third clustering center of the third speech feature clustering according to a preset clustering center calculation method;

s1016, mapping the third cluster center to be a third center vector of a high-dimensional space;

s1017, calculating a fourth distance value between the third center vector and the first voice vector according to a preset distance calculation method;

s1018, determining whether both the third distance value and the fourth distance value are smaller than the first distance threshold, and determining whether a sum of the third distance value and the fourth distance value is smaller than a preset third distance threshold, where a value range of the third distance threshold is (the first distance threshold, q × the first distance threshold), and q is a positive number greater than 1 and smaller than p;

s1019, if the third distance value and the fourth distance value are both smaller than the first distance threshold value, and the sum of the third distance value and the fourth distance value is smaller than a preset third distance threshold value, constructing a communication channel between the first terminal and the third terminal.

As mentioned above, the construction of the call channel between the first terminal and the third terminal is realized. Since the first terminal intends to establish a call channel with the second terminal, which is substantially in that a call is implemented with one terminal in the second cluster to which the second terminal belongs, when it is determined that the second terminal is not suitable for performing an efficient call (for example, the first distance value is not less than the preset first distance threshold), a suitable other terminal is searched from the second cluster to establish an efficient call channel. And then, according to the same processing mode as the second terminal, calculating a third distance value and a fourth distance value, judging whether the third distance value and the fourth distance value are both smaller than a preset first distance threshold value, and judging whether the sum of the third distance value and the fourth distance value is smaller than a preset third distance threshold value, thereby determining whether the third terminal is suitable for carrying out efficient call with the first terminal. If the third distance value and the fourth distance value are both smaller than a preset first distance threshold value, and the sum of the third distance value and the fourth distance value is smaller than a preset third distance threshold value, it is indicated that efficient communication can be performed between the first terminal and the third terminal, and therefore a communication channel between the first terminal and the third terminal is established. It should be noted that, the value range of the third distance threshold is (the first distance threshold, qx the first distance threshold), and q is a positive number greater than 1 and smaller than p, which is different from the second distance threshold involved in determining the first terminal and the second terminal (the value range of the second distance threshold is (the first distance threshold, p x the first distance threshold), and p is a positive number greater than 1 and smaller than 2), that is, as the mismatch number increases, the value of the parameter (e.g., p, q) becomes smaller and smaller, that is, the value of the distance threshold corresponding to the sum value becomes smaller and smaller, so that the unqualified first terminal can be found in time (screened out by querying the mismatch number), and the calculation amount of the whole system is reduced in time. Further, the inverse prediction model may be stored in a preset block chain. For example, the reverse prediction model is sent to other auditing nodes in the blockchain through one blockchain node in the blockchain. And after the other auditing nodes pass the auditing, storing the reverse prediction model in a new block form into a public account book of the block chain.

In one embodiment, the step S11 of constructing a call channel between the terminal one and the terminal two includes:

s1101, acquiring the call request times A1 and the successful call times B1 sent by the first terminal in the preset time, and acquiring the requested call times A2 and the successful call times B2 of the second terminal in the preset time;

s1102, according to a formula: y1 ═ B1 × a2)/(a1 × B2), Y2 ═ B1/a1, and Y3 ═ B2/a2, and a first factor Y1, a second factor Y2, and a third factor Y3 were calculated, respectively;

s1103, judging whether the value of max (Y1, Y2, Y3,1) is equal to Y1, and judging whether the value of min (Y1, Y2, Y3, v) is equal to v, wherein v is a preset parameter which is less than 1 and more than 0;

and S1104, if the value of max (Y1, Y2, Y3,1) is equal to Y1, and the value of min (Y1, Y2, Y3, v) is equal to v, constructing a call channel between the first terminal and the second terminal.

As mentioned above, the construction of the call channel between the first terminal and the second terminal is realized. The method also comprises the steps of obtaining the call request times A1 and the successful call times B1 sent by the terminal I in the preset time, and obtaining the requested call times A2 and the successful call times B2 of the terminal II in the preset time; according to the formula: y1 ═ B1 × a2)/(a1 × B2), Y2 ═ B1/a1, Y3 ═ B2/a2, and whether the value of max (Y1, Y2, Y3,1) is equal to Y1 and the value of min (Y1, Y2, Y3, v) is equal to v are determined. The method for determining whether the first terminal and the second terminal are suitable for efficient call by using the forward prediction model, the reverse prediction model and the like is based on personal feature information and voice features to judge whether the state of a person is fluctuating, so that fine adjustment is further performed to improve call efficiency. If the value of max (Y1, Y2, Y3,1) is equal to Y1, and the value of min (Y1, Y2, Y3, v) is equal to v, it indicates that the current successful call duty ratio of the first terminal is higher than that of the second terminal, and the successful call duty ratios of the first terminal and the second terminal are both higher than the ratio v, so that it indicates that the first terminal and the second terminal are suitable for making a call in a call data plane, and thus a call channel between the first terminal and the second terminal is constructed.

The call channel construction method based on the dual model obtains a call request sent by a first terminal; acquiring first personal feature information to obtain a first voice feature cluster; calculating a first clustering center; acquiring second voice characteristic data, and mapping the first clustering center and the second voice characteristic data into a first center vector and a second voice vector respectively; calculating a first distance value of the first center vector and the second voice vector; acquiring second person characteristic information of the second user to obtain a second voice characteristic cluster; calculating a second cluster center; acquiring first voice feature data, a second central vector and a first voice vector; calculating a second distance value between the second center vector and the first voice vector; and if the first distance value and the second distance value are both smaller than a preset first distance threshold value, and the sum of the first distance value and the second distance value is smaller than a preset second distance threshold value, establishing a communication channel between the first terminal and the second terminal. Thereby improving the call efficiency.

Referring to fig. 2, an embodiment of the present application provides a call path constructing apparatus based on dual models, including:

A first call request obtaining unit 10, configured to obtain a call request sent by a first terminal, where the call request is used to request a remote call with a second terminal; the first terminal belongs to a preset first cluster, and the second terminal belongs to a preset second cluster;

a first speech feature cluster obtaining unit 20, configured to obtain first personal feature information of a first user corresponding to the first terminal, and input the first personal feature information into a preset forward prediction model for calculation, so as to obtain a first speech feature cluster output by the forward prediction model; the first voice feature cluster is composed of a plurality of voice feature data, the forward prediction model is based on a preset neural network model and is obtained by training with training data, and the training data comprises a call record which is collected in advance and marked as an efficient call by an artificial marker, personal feature information of a call initiator in the call record and voice feature information of a call receiver in the call record;

a first clustering center calculating unit 30, configured to calculate a first clustering center of the first speech feature cluster according to a preset clustering center calculating method;

A second voice feature data obtaining unit 40, configured to obtain second voice feature data of a user corresponding to the second terminal, and map the first clustering center and the second voice feature data into a first center vector and a second voice vector of a high-dimensional space, respectively;

a first distance value calculating unit 50, configured to calculate a first distance value between the first center vector and the second speech vector according to a preset distance calculating method;

a second speech feature cluster obtaining unit 60, configured to obtain second personal feature information of the second user corresponding to the second terminal, and input the second personal feature information into a preset reverse prediction model for calculation, so as to obtain a second speech feature cluster output by the reverse prediction model;

a second clustering center calculating unit 70, configured to calculate a second clustering center of the second speech feature cluster according to a preset clustering center calculating method;

a first voice feature data obtaining unit 80, configured to obtain first voice feature data of a user corresponding to the first terminal, and map the second clustering center and the first voice feature data into a second center vector and a first voice vector of a high-dimensional space, respectively;

A second distance value calculating unit 90, configured to calculate a second distance value between the second center vector and the first speech vector according to a preset distance calculating method;

a distance threshold determining unit 100, configured to determine whether both the first distance value and the second distance value are smaller than a preset first distance threshold, and determine whether a sum of the first distance value and the second distance value is smaller than a preset second distance threshold, where a value range of the second distance threshold is (the first distance threshold, p × the first distance threshold), and p is a positive number greater than 1 and smaller than 2;

a call channel constructing unit 110, configured to construct a call channel between the first terminal and the second terminal if the first distance value and the second distance value are both smaller than a preset first distance threshold, and a sum of the first distance value and the second distance value is smaller than a preset second distance threshold.

The operations respectively executed by the above units or sub-units correspond to the steps of the call channel construction method based on dual models in the foregoing embodiment one by one, and are not described herein again.

In one embodiment, the apparatus comprises:

In one embodiment, the first distance value calculating unit includes:

a first distance value T calculation subunit configured to:

In one embodiment, the apparatus comprises:

a third terminal screening unit, configured to screen, if the first distance value is not less than a preset first distance threshold, or the second distance value is not less than the preset first distance threshold, or a sum of the first distance value and the second distance value is not less than a preset second distance threshold, a third terminal from the second cluster according to a preset terminal screening method;

a third voice vector mapping unit, configured to obtain third voice feature data corresponding to the third terminal, and map the third voice feature data into a third voice vector in a high-dimensional space;

a third distance value calculating unit, configured to calculate a third distance value between the first center vector and a third speech vector according to a preset distance calculating method;

a third person feature information obtaining unit, configured to obtain third person feature information of a user number three corresponding to the terminal number three, and input the third person feature information into a preset reverse prediction model for calculation, so as to obtain a third speech feature cluster output by the reverse prediction model;

The third clustering center calculating unit is used for calculating a third clustering center of the third voice feature cluster according to a preset clustering center calculating method;

a third center vector mapping unit, configured to map the third cluster center into a third center vector of a high-dimensional space;

a fourth distance value calculating unit, configured to calculate a fourth distance value between the third center vector and the first speech vector according to a preset distance calculating method;

a third distance threshold determination unit, configured to determine whether both the third distance value and the fourth distance value are smaller than the first distance threshold, and determine whether a sum of the third distance value and the fourth distance value is smaller than a preset third distance threshold, where a value range of the third distance threshold is (the first distance threshold, qx the first distance threshold), and q is a positive number greater than 1 and smaller than p;

and the third terminal communication channel establishing unit is used for establishing a communication channel between the first terminal and the third terminal if the third distance value and the fourth distance value are both smaller than the first distance threshold value, and the sum of the third distance value and the fourth distance value is smaller than a preset third distance threshold value.

In one embodiment, the call channel constructing unit 110 includes:

a call number obtaining subunit, configured to obtain a call request number a1 and a successful call number B1 sent by the terminal one in a preset time, and obtain a requested call number a2 and a successful call number B2 of the terminal two in a preset time;

a factor calculation subunit, configured to: y1 ═ B1 × a2)/(a1 × B2), Y2 ═ B1/a1, and Y3 ═ B2/a2, and a first factor Y1, a second factor Y2, and a third factor Y3 were calculated, respectively;

a value judgment subunit, configured to judge whether a value of max (Y1, Y2, Y3,1) is equal to Y1, and whether a value of min (Y1, Y2, Y3, v) is equal to v, where v is a preset parameter that is less than 1 and greater than 0;

and the call channel construction subunit is used for constructing a call channel between the first terminal and the second terminal if the value of max (Y1, Y2, Y3,1) is equal to Y1 and the value of min (Y1, Y2, Y3, v) is equal to v.

The call channel construction device based on the dual model obtains a call request sent by a first terminal; acquiring first personal feature information to obtain a first voice feature cluster; calculating a first clustering center; acquiring second voice characteristic data, and mapping the first clustering center and the second voice characteristic data into a first center vector and a second voice vector respectively; calculating a first distance value of the first center vector and the second voice vector; acquiring second person characteristic information of the second user to obtain a second voice characteristic cluster; calculating a second cluster center; acquiring first voice feature data, a second central vector and a first voice vector; calculating a second distance value between the second center vector and the first voice vector; and if the first distance value and the second distance value are both smaller than a preset first distance threshold value, and the sum of the first distance value and the second distance value is smaller than a preset second distance threshold value, establishing a communication channel between the first terminal and the second terminal. Thereby improving the call efficiency.

Referring to fig. 3, an embodiment of the present invention further provides a computer device, where the computer device may be a server, and an internal structure of the computer device may be as shown in the figure. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the computer designed processor is used to provide computational and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer equipment is used for storing data used by the call channel construction method based on the dual model. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a call path construction method based on dual models.

The processor executes the above dual-model-based call channel construction method, wherein the steps included in the method correspond to the steps of executing the dual-model-based call channel construction method of the foregoing embodiment one to one, and are not described herein again.

It will be understood by those skilled in the art that the structures shown in the drawings are only block diagrams of some of the structures associated with the embodiments of the present application and do not constitute a limitation on the computer apparatus to which the embodiments of the present application may be applied.

The computer equipment acquires a call request sent by a first terminal; acquiring first personal feature information to obtain a first voice feature cluster; calculating a first clustering center; acquiring second voice characteristic data, and mapping the first clustering center and the second voice characteristic data into a first center vector and a second voice vector respectively; calculating a first distance value of the first center vector and the second voice vector; acquiring second person characteristic information of the second user to obtain a second voice characteristic cluster; calculating a second cluster center; acquiring first voice feature data, a second central vector and a first voice vector; calculating a second distance value between the second center vector and the first voice vector; and if the first distance value and the second distance value are both smaller than a preset first distance threshold value, and the sum of the first distance value and the second distance value is smaller than a preset second distance threshold value, establishing a communication channel between the first terminal and the second terminal. Thereby improving the call efficiency.

An embodiment of the present application further provides a computer-readable storage medium, where a computer program is stored thereon, and when the computer program is executed by a processor, the method for constructing a dual-model-based communication channel is implemented, where steps included in the method correspond to steps of the method for constructing a dual-model-based communication channel in the foregoing embodiment one to one, and are not described herein again.

The computer-readable storage medium of the application acquires a call request sent by a first terminal; acquiring first personal feature information to obtain a first voice feature cluster; calculating a first clustering center; acquiring second voice characteristic data, and mapping the first clustering center and the second voice characteristic data into a first center vector and a second voice vector respectively; calculating a first distance value of the first center vector and the second voice vector; acquiring second person characteristic information of the second user to obtain a second voice characteristic cluster; calculating a second cluster center; acquiring first voice feature data, a second central vector and a first voice vector; calculating a second distance value between the second center vector and the first voice vector; and if the first distance value and the second distance value are both smaller than a preset first distance threshold value, and the sum of the first distance value and the second distance value is smaller than a preset second distance threshold value, establishing a communication channel between the first terminal and the second terminal. Thereby improving the call efficiency.

Further, the computer-readable storage medium may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function, and the like; the storage data area may store data created according to the use of the blockchain node, and the like.

The block chain is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, a consensus mechanism, an encryption algorithm and the like. A block chain (Blockchain), which is essentially a decentralized database, is a series of data blocks associated by using a cryptographic method, and each data block contains information of a batch of network transactions, so as to verify the validity (anti-counterfeiting) of the information and generate a next block. The blockchain may include a blockchain underlying platform, a platform product service layer, an application service layer, and the like.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium provided herein and used in the examples may include non-volatile and/or volatile memory. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), double-rate SDRAM (SSRSDRAM), Enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and bus dynamic RAM (RDRAM).

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, apparatus, article, or method. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, apparatus, article, or method that includes the element.

The above description is only a preferred embodiment of the present application, and not intended to limit the scope of the present application, and all modifications of equivalent structures and equivalent processes, which are made by the contents of the specification and the drawings of the present application, or which are directly or indirectly applied to other related technical fields, are also included in the scope of the present application.

Claims

1. A call channel construction method based on dual models is characterized by comprising the following steps:

2. The dual-model-based call channel construction method according to claim 1, wherein the first personal feature information of the first user corresponding to the first terminal is obtained and input into a preset forward prediction model for calculation, so as to obtain a first speech feature cluster output by the forward prediction model; the method comprises the following steps of obtaining a forward prediction model based on a preset neural network model, obtaining training data by utilizing the training data, wherein the first voice feature cluster is composed of a plurality of voice feature data, the forward prediction model is obtained by training the forward prediction model based on the preset neural network model, and the training data comprises a call record which is collected in advance and is marked as an efficient call by an artificial marker, personal feature information of a call initiator in the call record and voice feature information of a call receiver in the call record, and comprises the following steps:

3. The method for constructing a call channel according to claim 1, wherein the step of calculating the first distance value between the first center vector and the second speech vector according to a predetermined distance calculation method comprises:

according to the formula:

4. The method according to claim 1, wherein the steps of determining whether the first distance value and the second distance value are both smaller than a preset first distance threshold, and determining whether the sum of the first distance value and the second distance value is smaller than a preset second distance threshold, where the value range of the second distance threshold is (the first distance threshold, p × the first distance threshold), and p is a positive number greater than 1 and smaller than 2, include:

5. The method for constructing a call channel according to claim 1, wherein the step of constructing a call channel between a first terminal and a second terminal comprises:

6. A conversation channel constructing device based on dual models is characterized by comprising:

7. The dual model based call channel construction apparatus of claim 6, wherein the apparatus comprises:

8. The dual-model-based call channel construction apparatus according to claim 6, wherein the first distance value calculation unit includes:

a first distance value T calculation subunit configured to:

9. A computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor implements the steps of the method of any one of claims 1 to 5 when executing the computer program.

10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 5.