CN116702789A

CN116702789A - Semantic analysis method, semantic analysis device, computer equipment and storage medium

Info

Publication number: CN116702789A
Application number: CN202310673714.8A
Authority: CN
Inventors: 葛琪超; 林子涯; 杜亚东
Original assignee: Industrial and Commercial Bank of China Ltd ICBC
Current assignee: Industrial and Commercial Bank of China Ltd ICBC
Priority date: 2023-06-08
Filing date: 2023-06-08
Publication date: 2023-09-05

Abstract

The application relates to a semantic analysis method, a semantic analysis device, computer equipment and a storage medium, and relates to the technical field of artificial intelligence. The method comprises the following steps: acquiring at least one section of call record to be analyzed; determining a predicted service label and a predicted behavior label corresponding to each call record to be analyzed through a target analysis model; judging whether error labels with wrong prediction exist in the prediction business labels and/or the prediction behavior labels of the call records to be analyzed; and determining the semantic analysis result of each call record to be analyzed based on the judgment result and the prediction service label and the prediction behavior label of each call record to be analyzed. The application reduces the manpower and time consumed when the call of the customer is counted, improves the semantic analysis accuracy of the call record to be analyzed, ensures that the semantic analysis result can accurately and effectively reflect the actual call intention of the customer in the call record to be analyzed, and thereby accurately acquires the service requirement of the customer.

Description

Semantic analysis method, semantic analysis device, computer equipment and storage medium

Technical Field

The present application relates to the field of artificial intelligence, and in particular, to a semantic analysis method, apparatus, computer device, and storage medium.

Background

With the continuous increase of the enterprise scale and the continuous increase of business related enterprises, the daily received clients call more and more; to better provide services to customers, personnel are typically required to manually count all customer calls received, thereby determining the customer's service needs.

However, since the number of incoming calls of the clients is large, a large amount of time and labor are consumed each time the staff member counts all received incoming calls of the clients, and thus excessive unnecessary resource waste is caused.

Disclosure of Invention

In view of the above, it is necessary to provide a semantic analysis method, apparatus, computer device, and storage medium capable of reducing the manpower and time consumption.

In a first aspect, the present application provides a semantic analysis method. The method comprises the following steps:

acquiring at least one section of call record to be analyzed;

determining a predicted service label and a predicted behavior label corresponding to each call record to be analyzed through a target analysis model;

judging whether error labels with wrong prediction exist in the prediction business labels and/or the prediction behavior labels of the call records to be analyzed;

and determining the semantic analysis result of each call record to be analyzed based on the judgment result and the prediction service label and the prediction behavior label of each call record to be analyzed.

In one embodiment, determining the semantic analysis result of each call record to be analyzed based on the determination result and the predicted business label and the predicted behavior label of each call record to be analyzed includes:

if the prediction business label and/or the prediction behavior label of each call record to be analyzed have a wrong prediction label, determining an abnormal call record corresponding to the wrong label from each call record to be analyzed, and carrying out semantic verification on the abnormal call record based on at least one section of candidate call record to obtain a verification result; each candidate call record comprises a call record to be analyzed which does not belong to an abnormal call record, and/or a historical call record with a known historical analysis result;

and determining the semantic analysis result of each call record to be analyzed based on the verification result, the predicted business label and the predicted behavior label of each call record to be analyzed.

In one embodiment, performing semantic verification on the abnormal call record based on at least one segment of the candidate call record to obtain a verification result, including:

determining candidate call vectors corresponding to the candidate call records and abnormal call vectors corresponding to the abnormal call records;

obtaining the similarity between the abnormal call vector and each candidate call vector by carrying out similarity operation on the abnormal call vector and each candidate call vector;

And determining a verification result based on the similarity between the abnormal call vector and each candidate call vector and a preset similarity threshold.

In one embodiment, determining the semantic analysis result of each call record to be analyzed based on the verification result, the predicted business label and the predicted behavior label of each call record to be analyzed includes:

determining an abnormal service label and an abnormal behavior label corresponding to the abnormal call record according to the verification result;

and determining semantic analysis results of the call records to be analyzed based on the predicted service tags and the predicted behavior tags of the call records to be analyzed and the abnormal service tags and the abnormal behavior tags corresponding to the abnormal call records.

In one embodiment, determining an abnormal service tag and an abnormal behavior tag corresponding to the abnormal call record according to the verification result includes:

if the verification result is that the similarity of each candidate call vector and the abnormal call vector is smaller than the similarity threshold value, acquiring an abnormal service label and an abnormal behavior label corresponding to the abnormal call record.

In one embodiment, according to the verification result, determining an abnormal service tag and an abnormal behavior tag corresponding to the abnormal call record, and further includes:

If the verification result is that the similarity between the candidate call vectors and the abnormal call vector is greater than the similarity threshold, the similar service label and the similar behavior label corresponding to the similar call vector are used as the abnormal service label and the abnormal behavior label corresponding to the abnormal call record.

In one embodiment, the method further comprises:

if the predicted business label and/or the predicted behavior label of each call record to be analyzed do not have the error label of the prediction error, the predicted business label and the predicted behavior label of each call record to be analyzed are used as the target business label and the target behavior label of each call record to be analyzed;

and determining the semantic analysis result of each call record to be analyzed based on the target service label and the target behavior label of each call record to be analyzed.

In one embodiment, the training process of the target analytic model includes:

based on a professional phrase sample of the service field to which the call record to be analyzed belongs, performing professional phrase training on the initial analysis model to obtain an intermediate analysis model;

training a business analysis layer of the intermediate analysis model based on the business label training sample in the conversation training sample, and training a behavior analysis layer of the intermediate analysis model based on the behavior label training sample in the conversation training sample to obtain a trained target analysis model.

In a second aspect, the application further provides a semantic analysis device. The device comprises:

the acquisition module is used for acquiring at least one section of call record to be analyzed;

the first determining module is used for determining a predicted service label and a predicted behavior label corresponding to each call record to be analyzed through the target analysis model;

the judging module is used for judging whether error labels with wrong prediction exist in the prediction business labels and/or the prediction behavior labels of the call records to be analyzed;

the second determining module is used for determining semantic analysis results of all call records to be analyzed based on the judgment results, and the prediction business labels and the prediction behavior labels of all call records to be analyzed.

In a third aspect, the present application also provides a computer device. The computer device comprises a memory storing a computer program and a processor implementing the semantic analysis method according to any of the embodiments of the first aspect described above when the processor executes the computer program.

In a fourth aspect, the present application also provides a computer-readable storage medium. A computer readable storage medium having stored thereon a computer program which when executed by a processor implements a semantic analysis method as in any of the embodiments of the first aspect described above.

In a fifth aspect, the present application also provides a computer program product. A computer program product comprising a computer program which, when executed by a processor, implements a semantic analysis method as in any of the embodiments of the first aspect described above.

According to the semantic analysis method, the semantic analysis device, the computer equipment and the storage medium, after the call record to be analyzed of the client is obtained, the predicted business label and the predicted behavior label corresponding to the call record to be analyzed are determined through the target analysis model, so that the predicted business label and the predicted behavior label corresponding to the call record to be analyzed can be obtained without participation of staff in the process, labor and time consumed in statistics of incoming calls of the client are reduced, resource consumption in statistics of the incoming calls of the client is greatly reduced, semantic analysis accuracy of the call record to be analyzed is improved by judging whether the error label exists or not, and real incoming call intention of the client in the call record to be analyzed is guaranteed to be accurately and effectively reflected by semantic analysis results, so that service requirements of the user are accurately obtained.

Drawings

FIG. 1 is an application environment diagram of a semantic analysis method according to an embodiment of the present application;

FIG. 2 is a flow chart of a semantic analysis method according to an embodiment of the present application;

FIG. 3 is a flowchart for determining a semantic analysis result according to an embodiment of the present application;

FIG. 4 is a flowchart of another method for determining semantic analysis results according to an embodiment of the present application;

FIG. 5 is a flowchart for training a target analytical model according to an embodiment of the present application;

FIG. 6 is a flowchart of another semantic analysis method according to an embodiment of the present application;

FIG. 7 is a block diagram of a first semantic analysis device according to an embodiment of the present application;

FIG. 8 is a block diagram of a second semantic analysis device according to an embodiment of the present application;

FIG. 9 is a block diagram of a third semantic analysis device according to an embodiment of the present application;

fig. 10 is a block diagram of a fourth semantic analysis device according to an embodiment of the present application;

FIG. 11 is a block diagram of a fifth semantic analysis device according to an embodiment of the present application;

FIG. 12 is a block diagram of a sixth semantic analysis device according to an embodiment of the present application;

fig. 13 is an internal structural view of a computer device in one embodiment.

Detailed Description

The present application will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present application more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.

It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application. In the description of the present application, a description of the terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present application. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.

Based on the above situation, the semantic analysis method provided by the embodiment of the application can be applied to an application environment as shown in fig. 1. In one embodiment, a computer device is provided, which may be a server, the internal structure of which may be as shown in FIG. 1. The computer device includes a processor, a memory, and a network interface connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the computer device is used for storing acquired data of the semantic analysis method. 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 semantic analysis method.

The application discloses a semantic analysis method, a semantic analysis device, computer equipment and a storage medium, wherein the computer equipment determines a predicted business label and a predicted behavior label corresponding to each call record to be analyzed through a target analysis model, judges whether error labels with prediction errors exist in the predicted business labels and/or the predicted behavior labels of the call records to be analyzed, and further determines semantic analysis results of the call records to be analyzed.

In one embodiment, as shown in fig. 2, fig. 2 is a flowchart of a semantic analysis method provided by an embodiment of the present application, where the semantic analysis method performed by the computer device in fig. 1 may include the following steps:

step 201, at least one section of call record to be analyzed is obtained.

The call record to be analyzed may include call content of the client and call content of the agent, or only include call content of the client.

When the call records to be analyzed need to be obtained, validity verification can be performed on all call records in advance, and the call records to be analyzed can be obtained from all call records according to the validity verification results of all call records.

Further illustratively, the validity verification for all call records may be a plurality of validity verification processes, for example, the plurality of validity verification processes may include, but are not limited to: and verifying whether the call content of the client in the call records is clear, verifying whether the call records contain the call content of the client, verifying whether the call duration of the client in the call records is greater than a duration threshold value, and the like, so that screening out call records in which the call content of the client in all call records is not clear, the call content of the client is not contained, and the effectiveness verification that the call duration of the client is less than the duration threshold value is not passed is realized by performing effectiveness verification on all call records.

The time length threshold may be set according to the historical experience of the staff, and the setting of the time length threshold is not limited here.

In one embodiment of the present application, when at least one section of call records to be analyzed needs to be obtained, validity verification can be performed on all call records respectively, whether each call record in all call records passes through all validity verification processes is determined, and if a certain call record passes through all validity verification processes, the call record is the call record to be analyzed; if at least one validity verification process fails in the validity verification process of a certain call record, the call record is not the call record to be analyzed.

Step 202, determining a predicted service label and a predicted behavior label corresponding to each call record to be analyzed through a target analysis model.

The predicted service label refers to the service type which is predicted by the target analysis model and is required to be transacted by a customer in the call record to be analyzed; the predicted behavior label refers to a specific behavior which is predicted by the target analysis model and is expected to be executed by a client aiming at the service type in the predicted service label in the call record to be analyzed; for example, if the predicted business label is "financial transaction", the predicted behavior label may be "credit card application"; alternatively, if the predicted traffic label is "information query," the predicted behavior label may be "balance query.

It should be noted that, when determining the predicted service label and the predicted behavior label corresponding to each call record to be analyzed through the target analysis model, the following may be specifically included; performing text conversion on the call record to be analyzed in advance to obtain a text to be analyzed corresponding to the call record to be analyzed; performing text error correction on the text to be analyzed, and modifying words with conversion errors in the text conversion process in the text to be analyzed; word segmentation processing is carried out on the text to be analyzed after text error correction, word segmentation results of the text to be analyzed are obtained, keyword extraction is carried out on the word segmentation results of the text to be analyzed, and keywords in call records to be analyzed are obtained; carrying out vectorization processing on keywords in the call records to be analyzed to obtain keyword vectors corresponding to the keywords in the call records to be analyzed; and inputting the keyword vector into the target analysis model to obtain a predicted service label and a predicted behavior label corresponding to each call record to be analyzed, which are output by the target analysis model.

The keywords refer to words in the call record to be analyzed, which can reflect the actual intention of the customer, and for example, the keywords may be: loans, repayment, car credits, credit cards, etc. Further, keyword extraction may be achieved through TF-IDF (term frequency-inverse document frequency, a commonly used weighting technique for information retrieval and data mining).

When the text to be analyzed needs to be subjected to word segmentation, the word segmentation of the text to be analyzed can be realized by using the barker word segmentation.

The method can be used for vectorizing the keywords in the call records to be analyzed based on mean-pooling to obtain keyword vectors corresponding to the keywords in the call records to be analyzed.

In one embodiment of the application, a service tag training sample and a behavior tag training sample can be predetermined, and the initial analytical model is trained according to the service tag training sample and the behavior tag training sample to obtain a target analytical model, wherein the service tag training sample is a sample call record marked with a service tag, and the behavior tag training sample is a sample call record marked with a behavior tag.

The initial analytical model may be a Mengzi (a model) model.

In another embodiment of the present application, before training the target analysis model according to the service tag training sample and the behavior tag training sample, a professional phrase sample in the service domain to which the call record to be analyzed belongs may be obtained, and the target analysis model is trained according to the professional phrase sample, so that the target analysis model can adapt to the service domain to which the call record to be analyzed belongs, and accuracy of the target analysis model in outputting the predicted service tags and the predicted behavior tags corresponding to the call records to be analyzed is improved.

And 203, judging whether error labels with wrong prediction exist in the prediction business labels and/or the prediction behavior labels of the call records to be analyzed.

It should be noted that, there are many methods for judging whether there are error labels of prediction errors in the predicted service labels and/or the predicted behavior labels of the call records to be analyzed, for example: transmitting each call record to be analyzed, and a predicted service label and a predicted behavior label corresponding to each call record to be analyzed to service personnel, acquiring feedback information transmitted by the service personnel, and further determining a judgment result according to the content recorded in the feedback information; or after determining the predicted service label and the predicted behavior label corresponding to each call record to be analyzed, displaying each call record to be analyzed, the predicted service label and the predicted behavior label corresponding to each call record to be analyzed on a display interface of the computer equipment executing the semantic analysis method, acquiring operation selection of service personnel on the computer equipment, and determining a judgment result.

As an implementation manner, when it is required to determine whether there is an error label with a prediction error in the predicted service label and/or the predicted behavior label of each call record to be analyzed, the method may include the following steps: after receiving and determining the prediction service label and the prediction behavior label corresponding to each call record to be analyzed, transmitting each call record to be analyzed and the prediction service label and the prediction behavior label corresponding to each call record to be analyzed to service personnel, providing an information feedback interface, receiving feedback information transmitted by the service personnel through the information feedback interface, determining whether error labels with prediction errors exist in the prediction service label and/or the prediction behavior label of each call record to be analyzed according to experience by the service personnel, editing a judging result into feedback information, transmitting the feedback information to the information feedback interface, and further determining the judging result according to contents recorded in the feedback information.

For example, if the content recorded in the feedback information sent by the information feedback interface receiving service personnel is the prediction error of the prediction service label of the call record a to be analyzed, it can be determined that the prediction service label of the call record a to be analyzed is the error label.

As another implementation manner, when it is required to determine whether there is an error label with a prediction error in the predicted service label and/or the predicted behavior label of each call record to be analyzed, the method may include the following steps: after the prediction service label and the prediction behavior label corresponding to each call record to be analyzed are determined, each call record to be analyzed, the prediction service label and the prediction behavior label corresponding to each call record to be analyzed are displayed on a display interface of a computer device executing a semantic analysis method, and a trigger control key of an error label is arranged in the display interface of the computer device, when a business person judges that the prediction service label and/or the prediction behavior label of a certain call record to be analyzed have the error label of the prediction error, the judgment result can be determined by triggering the trigger control key of the error label according to the operation selection of the business person in the computer device.

For example, when the display interface of the computer device displays the call record B to be analyzed and the predicted service label and the predicted behavior label of the call record B to be analyzed, the service personnel triggers the "error label" beside the predicted service label to trigger the control key, and then determines that the predicted service label of the call record B to be analyzed is the error label.

Step 204, determining the semantic analysis result of each call record to be analyzed based on the judgment result and the prediction service label and the prediction behavior label of each call record to be analyzed.

It should be noted that, the judgment result includes two cases, the first case is: error labels with prediction errors exist in the prediction business labels and/or the prediction behavior labels of the call records to be analyzed; the second case is: and if the predicted business labels and/or the predicted behavior labels of the call records to be analyzed do not have error labels with wrong prediction. Further, the corresponding methods for determining the semantic analysis results of the call records to be analyzed are different from the judgment results of different conditions.

In one implementation manner of the present application, if the judgment result is the first case, it is: when the semantic analysis result of each call record to be analyzed is determined, determining an abnormal call record corresponding to the error label from each call record to be analyzed, and determining an abnormal service label and an abnormal behavior label corresponding to the abnormal call record; and determining the semantic analysis result of each call record to be analyzed according to the abnormal service label and the abnormal behavior label corresponding to the abnormal call record and the predicted service label and the predicted behavior label of each call record to be analyzed.

In another implementation manner of the present application, if the determination result is the second case, it is: if the predicted business labels and/or the predicted behavior labels of the call records to be analyzed do not have error labels with prediction errors; the predicted service label and the predicted behavior label of each call record to be analyzed can be used as the target service label and the target behavior label of each call record to be analyzed; and determining the semantic analysis result of each call record to be analyzed based on the target service label and the target behavior label of each call record to be analyzed.

According to the semantic analysis method, after the call record to be analyzed of the client is obtained, the prediction service label and the prediction behavior label corresponding to the call record to be analyzed are determined through the target analysis model, so that the prediction service label and the prediction behavior label corresponding to the call record to be analyzed can be obtained without participation of staff in the process, labor and time consumed in statistics of incoming calls of the client are reduced, resource consumption in statistics of the incoming calls of the client is greatly reduced, semantic analysis accuracy of the call record to be analyzed is improved by judging whether the error label exists or not, and real incoming call intention of the client in the call record to be analyzed is guaranteed to be accurately and effectively reflected by a semantic analysis result, so that service requirements of the user are accurately obtained.

With the increasing of enterprise scale and the increasing of enterprise related business, the number of customer calls received every day is increased, so that the number of label types corresponding to the customer calls is increased, the difficulty of semantic analysis on the customer calls is increased, and the semantic analysis result of each call record to be analyzed is accurately analyzed, and the computer device in this embodiment can determine the semantic analysis result of each call record to be analyzed based on the judgment result, the predicted business label and the predicted behavior label of each call record to be analyzed by a manner shown in fig. 3, and specifically includes the following steps:

step 301, if a prediction service label and/or a prediction behavior label of each call record to be analyzed has a wrong label of a prediction error, determining an abnormal call record corresponding to the wrong label from each call record to be analyzed, and performing semantic verification on the abnormal call record based on at least one segment of candidate call record to obtain a verification result.

Each candidate call record includes a call record to be analyzed that does not belong to an abnormal call record, and/or a history call record for which a history analysis result is known.

It should be noted that, when the abnormal call record needs to be semantically verified, the method specifically includes the following steps: determining candidate call vectors corresponding to the candidate call records and abnormal call vectors corresponding to the abnormal call records; obtaining the similarity between the abnormal call vector and each candidate call vector by carrying out similarity operation on the abnormal call vector and each candidate call vector; and determining a verification result based on the similarity between the abnormal call vector and each candidate call vector and a preset similarity threshold.

In one embodiment of the application, when the similarity operation of the abnormal call vector and each candidate call vector is needed, the vector distance between the abnormal call vector and each candidate call vector can be calculated, and if the vector distance between the abnormal call vector and a certain candidate call vector is larger, the similarity between the abnormal call vector and the candidate call vector is proved to be smaller; if the vector distance between the abnormal call vector and a certain candidate call vector is smaller, the similarity between the abnormal call vector and the candidate call vector is proved to be larger.

The verification result records the magnitude relation between the similarity between the abnormal call vector and each candidate call vector and the similarity threshold value, and if the similarity between the abnormal call vector and each candidate call vector is represented by the vector distance between the abnormal call vector and each candidate call vector, the magnitude relation between the vector distance between the abnormal call vector and each candidate call vector and the distance threshold value can be understood as the verification result records.

For example, when the verification result needs to be determined, the vector distance between the abnormal call vector and each candidate call vector may be calculated based on a vector-to-vector distance calculation formula, and the comparison result of the vector distance between the abnormal call vector and each candidate call vector and the distance threshold is determined, if the vector distance between the abnormal call vector and each candidate call vector is greater than the distance threshold, the verification result is: the similarity of each candidate call vector and the abnormal call vector is smaller than a similarity threshold value; if the vector distance between each candidate call vector and the abnormal call vector is smaller than the distance threshold value and the similar call vectors exist, the verification result is that: and each candidate call vector has a similar call vector with the similarity with the abnormal call vector being greater than a similarity threshold.

Step 302, determining a semantic analysis result of each call record to be analyzed based on the verification result, the predicted business label and the predicted behavior label of each call record to be analyzed.

In one embodiment of the present application, when it is required to determine the semantic analysis result of each call record to be analyzed, the method specifically includes the following steps: determining an abnormal service label and an abnormal behavior label corresponding to the abnormal call record according to the verification result; and determining semantic analysis results of the call records to be analyzed based on the predicted service tags and the predicted behavior tags of the call records to be analyzed and the abnormal service tags and the abnormal behavior tags corresponding to the abnormal call records.

When determining the abnormal service label and the abnormal behavior label corresponding to the abnormal call record according to the verification result, the abnormal service label and the abnormal behavior label can be divided into the following two cases according to the difference of the verification result, wherein one case is: if the verification result is that the similarity of each candidate call vector and the abnormal call vector is smaller than the similarity threshold value, acquiring an abnormal service label and an abnormal behavior label corresponding to the abnormal call record. And a second case: if the verification result is that the similarity between the candidate call vectors and the abnormal call vector is greater than the similarity threshold, the similar service label and the similar behavior label corresponding to the similar call vector are used as the abnormal service label and the abnormal behavior label corresponding to the abnormal call record.

In one embodiment of the present application, if the verification result is that the similarity between each candidate call vector and the abnormal call vector is smaller than the similarity threshold, the vector distances between each candidate call vector and the abnormal call vector are both greater than the distance threshold, which indicates that the abnormal call vector is dissimilar from each candidate call vector, so that the abnormal call record is sent to the computer equipment of the service personnel, so that the service personnel can determine the abnormal service label and the abnormal behavior label of the abnormal call record according to own experience.

In another embodiment of the present application, the similar call vector refers to a vector having a similarity with the abnormal call vector greater than a similarity threshold, and further, if there is a plurality of candidate call vectors having a similarity with the abnormal call vector greater than the similarity threshold, the similarity between the plurality of candidate call vectors and the abnormal call vector is the largest, or the vector distance between the plurality of candidate call vectors and the abnormal call vector is the smallest.

Because the similar call vector is similar to the abnormal call vector, the service labels and the behavior labels of the similar call vector are similar to each other, and therefore, the similar service labels and the similar behavior labels corresponding to the similar call vector can be used as the abnormal service labels and the abnormal behavior labels corresponding to the abnormal call records.

Because each candidate call record includes a call record to be analyzed that does not belong to an abnormal call record, and/or a history call record for which a history analysis result is known; therefore, when the similar call vector is a call record to be analyzed which does not belong to the abnormal call record, the similar service label and the similar behavior label corresponding to the similar call vector are the predicted service label and the predicted behavior label corresponding to the call record to be analyzed which are determined by the target analysis model. When the similar call vector is the history call record of the known history analysis result, the similar service label and the similar behavior label corresponding to the similar call vector are the history service label and the history behavior label recorded in the history analysis result.

Further, the reason why the abnormal call vector is not similar to each candidate call vector may be: the business label training sample and the behavior label training sample for training the target analysis model may not contain the abnormal business label and the abnormal behavior label of the abnormal call record. Or the accuracy of the predicted business label and the predicted behavior label of each call record to be analyzed, which are output by the target analysis model, is lost.

According to the semantic analysis method, if the prediction business label and/or the prediction behavior label of each call record to be analyzed have the error label with the prediction error, the label correction is carried out on the error label by determining the abnormal business label and the abnormal behavior label corresponding to the abnormal call record, so that the accuracy of the semantic analysis result of the call record to be analyzed is ensured, and the condition that the accuracy of the semantic analysis result is lower due to the error label is prevented.

With the increasing of enterprise scale and the increasing of enterprise related business, the number of customer calls received every day is increased, so that the number of label types corresponding to the customer calls is increased, the probability of making semantic analysis errors on the customer calls is increased, and the semantic analysis result of each call record to be analyzed is accurately analyzed, so that the computer device in this embodiment may determine the semantic analysis result of each call record to be analyzed by a manner shown in fig. 4 based on the determination result, and the predicted business label and the predicted behavior label of each call record to be analyzed, and further may include the following steps:

step 401, if there is no error label of prediction error in the predicted service label and/or predicted behavior label of each call record to be analyzed, using the predicted service label and predicted behavior label of each call record to be analyzed as the target service label and target behavior label of each call record to be analyzed.

The target service labels and the target behavior labels of the call records to be analyzed are as follows: business labels corresponding to the call records to be analyzed correctly and behavior labels corresponding to the call records to be analyzed correctly.

Further, if the predicted business label and/or the predicted behavior label of each call record to be analyzed have no error label of prediction error, the predicted business label and the predicted behavior label of each call record to be analyzed are labels of positive prediction, so that modification of the predicted business label and the predicted behavior label of each call record to be analyzed is not needed, that is, the predicted business label and the predicted behavior label of each call record to be analyzed are used as the target business label and the target behavior label of each call record to be analyzed.

Step 402, determining a semantic analysis result of each call record to be analyzed based on the target business label and the target behavior label of each call record to be analyzed.

It should be noted that, since the semantic analysis result of each call record to be analyzed refers to the service tag and the behavior tag corresponding to each call record to be analyzed, and the target service tag and the target behavior tag of each call record to be analyzed refer to the service tag corresponding to each call record to be analyzed correctly and the behavior tag corresponding to each call record to be analyzed correctly, the target service tag and the target behavior tag of each call record to be analyzed can be used as the semantic analysis result of each call record to be analyzed.

According to the semantic analysis method, if the predicted business label and/or the predicted behavior label of each call record to be analyzed do not have the error label with the error prediction, the target business label and the target behavior label of each call record to be analyzed are determined, the business label correctly corresponding to each call record to be analyzed and the behavior label correctly corresponding to each call record to be analyzed are determined, the accuracy of the semantic analysis result of the call record to be analyzed is ensured, and the condition that the semantic analysis result is lower in accuracy due to the error label is prevented.

Because the number of the customer calls received every day is large, when the number of the customer calls is counted by a manual method, more time and labor are wasted for workers, and in order to ensure efficient and convenient implementation of the semantic recognition result of the customer calls, a target analysis model can be used to determine a prediction service label and a prediction behavior label corresponding to each call record to be analyzed, wherein the computer equipment of the embodiment can train the target analysis model in a manner shown in fig. 5, and specifically comprises the following steps:

step 501, performing professional phrase training on the initial analysis model based on the professional phrase sample of the service field to which the call record to be analyzed belongs, so as to obtain an intermediate analysis model.

The professional phrase sample may include call content of the client and call content of the agent, or only include call content of the client, and the call content of the client in the professional phrase sample includes professional phrases in the service domain to which the call record to be analyzed belongs.

For example, if the call record to be analyzed is a call record between a client and a bank agent, the service domain to which the call record to be analyzed belongs is the bank domain, and therefore, the professional phrase sample is a professional phrase sample in the bank domain.

The initial analysis model can be a voice model in the general field, and because the voice model in the general field has a certain training basis, the training of the voice model in the general field from a zero base is not needed, and the improvement of the training efficiency of the target analysis model can be realized.

It should be noted that, the professional phrase sample in the service field to which the call records belong is analyzed to perform professional phrase training on the initial analysis model, so that the intermediate analysis model has the capability of identifying the professional phrases in the service field to which the call records to be analyzed belong, and the prediction accuracy of the follow-up target analysis model for predicting the service labels and the prediction behavior labels of the call records to be analyzed is improved.

In one embodiment of the application, when the professional phrase sample in the service field of the call record to be analyzed is needed to perform the professional phrase training on the initial analysis model, the text of the professional phrase sample is corrected, and the wrongly written words in the professional phrase sample are removed; vectorizing the text corrected professional phrase sample to obtain a professional phrase vector; and inputting the professional phrase vector into the initial analysis model for model training to obtain an intermediate analysis model.

Step 502, training a business analysis layer of the intermediate analysis model based on the business label training sample in the conversation training sample, and training a behavior analysis layer of the intermediate analysis model based on the behavior label training sample in the conversation training sample to obtain a trained target analysis model.

The service tag training sample may include call content of the client and call content of the agent, or only include call content of the client, and further include a service tag corresponding to the call content of the client.

The behavior tag training sample may include call content of the client and call content of the agent, or only include call content of the client, and further include a behavior tag corresponding to the call content of the client.

As an implementation manner, when training is performed on the service analysis layer of the intermediate analysis model based on the service tag training sample in the call training sample, text error correction can be performed on the service tag training sample, and wrongly written words in the service tag training sample are removed; carrying out vectorization processing on the service tag training samples subjected to text error correction to obtain service tag training vectors; and inputting the service tag training vector into the intermediate analysis model to perform model training, and completing training of a service analysis layer of the intermediate analysis model.

As another implementation manner, when training the behavior analysis layer of the intermediate analysis model based on the behavior tag training sample in the call training sample, text error correction can be performed on the behavior tag training sample to remove wrongly written words in the behavior tag training sample; vectorizing the behavior label training sample subjected to text error correction to obtain a behavior label training vector; and inputting the behavior label training vector into the intermediate analysis model to perform model training, and completing training of a behavior analysis layer of the intermediate analysis model.

It should be noted that, when the service analysis layer of the intermediate analysis model and the behavior analysis layer of the intermediate analysis model are both trained, the trained target analysis model can be obtained.

According to the semantic analysis method, the intermediate analysis model is obtained by performing professional phrase training on the initial analysis model, so that the intermediate analysis model has the capability of identifying professional phrases in the service field of the call record to be analyzed, and the prediction accuracy of the prediction service label and the prediction behavior label of the call record to be analyzed by the subsequent target analysis model is improved; and training the intermediate analysis model through the service label training sample and the behavior label training sample, so as to ensure the accuracy of the predicted service label and the predicted behavior label corresponding to the call record to be analyzed according to the target analysis model.

In one embodiment of the present application, as shown in fig. 6, fig. 6 is a flowchart of another semantic analysis method provided in the embodiment of the present application, and when it is required to determine the semantic analysis result of each call record to be analyzed, the method specifically includes the following steps:

step 601, obtaining at least one section of call record to be analyzed.

Step 602, determining a predicted service label and a predicted behavior label corresponding to each call record to be analyzed through a target analysis model.

Step 603, judging whether error labels with wrong prediction exist in the prediction business labels and/or the prediction behavior labels of the call records to be analyzed; if yes, go to step 604; if not, step 611 is performed.

Step 604, determining an abnormal call record corresponding to the error label from the call records to be analyzed.

Step 605, determining a candidate call vector corresponding to each candidate call record and an abnormal call vector corresponding to the abnormal call record.

Step 606, the similarity between the abnormal call vector and each candidate call vector is obtained by performing similarity operation on the abnormal call vector and each candidate call vector.

In step 607, it is determined whether the similarity between the abnormal call vector and each candidate call vector is less than the similarity threshold. If yes, go to step 608; if not, go to step 609.

Step 608, obtaining an abnormal service tag and an abnormal behavior tag corresponding to the abnormal call record.

Step 609, using the similar service label and the similar behavior label corresponding to the similar call vector as the abnormal service label and the abnormal behavior label corresponding to the abnormal call record.

Step 610, determining a semantic analysis result of each call record to be analyzed based on the predicted service label and the predicted behavior label of each call record to be analyzed and the abnormal service label and the abnormal behavior label corresponding to the abnormal call record.

In step 611, the predicted service label and the predicted behavior label of each call record to be analyzed are used as the target service label and the target behavior label of each call record to be analyzed.

Step 612, determining the semantic analysis result of each call record to be analyzed based on the target business label and the target behavior label of each call record to be analyzed.

It should be understood that, although the steps in the flowcharts related to the above embodiments are sequentially shown as indicated by arrows, these steps are not necessarily sequentially performed in the order indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in the flowcharts described in the above embodiments may include a plurality of steps or a plurality of stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of the steps or stages is not necessarily performed sequentially, but may be performed alternately or alternately with at least some of the other steps or stages.

Based on the same inventive concept, the embodiment of the application also provides a semantic analysis device for realizing the semantic analysis method. The implementation of the solution provided by the device is similar to the implementation described in the above method, so the specific limitation in one or more embodiments of the semantic analysis device provided below may refer to the limitation of the semantic analysis method described above, and will not be repeated here.

In one embodiment, as shown in fig. 7, there is provided a first semantic analysis apparatus comprising: an acquisition module 10, a first determination module 20, a judgment module 30, and a second determination module 40, wherein:

the obtaining module 10 is configured to obtain at least one section of call record to be analyzed.

The first determining module 20 is configured to determine, according to the target analysis model, a predicted service tag and a predicted behavior tag corresponding to each call record to be analyzed.

The judging module 30 is configured to judge whether an error label with a prediction error exists in the predicted service label and/or the predicted behavior label of each call record to be analyzed.

The second determining module 40 is configured to determine a semantic analysis result of each call record to be analyzed based on the determination result and the predicted traffic label and the predicted behavior label of each call record to be analyzed.

According to the semantic analysis device, after the call record to be analyzed of the client is obtained, the prediction service label and the prediction behavior label corresponding to the call record to be analyzed are determined through the target analysis model, so that the prediction service label and the prediction behavior label corresponding to the call record to be analyzed can be obtained without participation of staff in the process, labor and time consumed in statistics of incoming calls of the client are reduced, resource consumption in statistics of the incoming calls of the client is greatly reduced, semantic analysis accuracy of the call record to be analyzed is improved by judging whether the error label exists or not, and real incoming call intention of the client in the call record to be analyzed is guaranteed to be accurately and effectively reflected by a semantic analysis result, so that service requirements of the user are accurately obtained.

In one embodiment, as shown in fig. 8, there is provided a second semantic analysis apparatus in which the second determination module 40 includes: a first determination unit 41 and a second determination unit 42, wherein:

the first determining unit 41 is configured to determine, if a prediction service label and/or a prediction behavior label of each call record to be analyzed has a wrong label with a prediction error, an abnormal call record corresponding to the wrong label from each call record to be analyzed, and perform semantic verification on the abnormal call record based on at least one segment of candidate call record, so as to obtain a verification result; each candidate call record includes call records to be analyzed that do not belong to an abnormal call record and/or historical call records for which historical analysis results are known.

The second determining unit 42 is configured to determine a semantic analysis result of each call record to be analyzed based on the verification result, the predicted traffic label and the predicted behavior label of each call record to be analyzed.

In one embodiment, as shown in fig. 9, there is provided a third semantic analysis apparatus in which the first determination unit 41 includes: a first determination subunit 411, an operation subunit 412, and a second determination subunit 413, wherein:

the first determining subunit 411 is configured to determine a candidate call vector corresponding to each candidate call record, and an abnormal call vector corresponding to the abnormal call record.

The operation subunit 412 is configured to obtain the similarity between the abnormal call vector and each candidate call vector by performing similarity operation on the abnormal call vector and each candidate call vector.

The second determining subunit 413 is configured to determine the verification result based on the similarity between the abnormal call vector and each candidate call vector, and a preset similarity threshold.

In one embodiment, as shown in fig. 10, there is provided a fourth semantic analysis apparatus in which the second determination unit 42 includes: a third determination subunit 421 and a fourth determination subunit 422, wherein:

The third determining subunit 421 is configured to determine, according to the verification result, an abnormal service tag and an abnormal behavior tag corresponding to the abnormal call record.

The fourth determining subunit 422 is configured to determine a semantic analysis result of each call record to be analyzed based on the predicted traffic label and the predicted behavior label of each call record to be analyzed, and the abnormal traffic label and the abnormal behavior label corresponding to the abnormal call record.

The third determining subunit is specifically configured to: if the verification result is that the similarity of each candidate call vector and the abnormal call vector is smaller than the similarity threshold value, acquiring an abnormal service label and an abnormal behavior label corresponding to the abnormal call record.

The third determination subunit is further specifically configured to: if the verification result is that the similarity between the candidate call vectors and the abnormal call vector is greater than the similarity threshold, the similar service label and the similar behavior label corresponding to the similar call vector are used as the abnormal service label and the abnormal behavior label corresponding to the abnormal call record.

In one embodiment, as shown in fig. 11, there is provided a fifth semantic analysis apparatus in which the second determination module 40 includes: a third determination unit 43 and a fourth determination unit 44, wherein:

The third determining unit 43 is configured to, if there is no error label with a prediction error in the predicted traffic label and/or the predicted behavior label of each call record to be analyzed, take the predicted traffic label and the predicted behavior label of each call record to be analyzed as the target traffic label and the target behavior label of each call record to be analyzed.

The fourth determining unit 44 is configured to determine a semantic analysis result of each call record to be analyzed based on the target service tag and the target behavior tag of each call record to be analyzed.

In one embodiment, as shown in fig. 12, there is provided a sixth semantic analysis apparatus, further comprising: a first training module 50 and a second training module 60, wherein:

the first training module 50 is configured to perform professional phrase training on the initial analysis model based on the professional phrase sample of the service domain to which the call record to be analyzed belongs, so as to obtain an intermediate analysis model.

The second training module 60 is configured to train the business analysis layer of the intermediate analysis model based on the business label training sample in the conversation training sample, and train the behavior analysis layer of the intermediate analysis model based on the behavior label training sample in the conversation training sample, so as to obtain a trained target analysis model.

The respective modules in the above-described semantic analysis apparatus may be implemented in whole or in part by software, hardware, and combinations thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.

In one embodiment, a computer device is provided, which may be a terminal, and the internal structure thereof may be as shown in fig. 13. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The communication interface of the computer device is used for carrying out wired or wireless communication with an external terminal, and the wireless mode can be realized through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. The computer program is executed by a processor to implement a semantic analysis method. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, can also be keys, a track ball or a touch pad arranged on the shell of the computer equipment, and can also be an external keyboard, a touch pad or a mouse and the like.

It will be appreciated by those skilled in the art that the structure shown in FIG. 13 is merely a block diagram of some of the structures associated with the present inventive arrangements and is not limiting of the computer device to which the present inventive arrangements may be applied, and that a particular computer device may include more or fewer components than shown, or may combine some of the components, or have a different arrangement of components.

In one embodiment, a computer device is provided comprising a memory and a processor, the memory having stored therein a computer program, the processor when executing the computer program performing the steps of:

acquiring at least one section of call record to be analyzed;

In one embodiment, the processor when executing the computer program further performs the steps of:

In one embodiment, a computer readable storage medium is provided having a computer program stored thereon, which when executed by a processor, performs the steps of:

Acquiring at least one section of call record to be analyzed;

In one embodiment, the computer program when executed by the processor further performs the steps of:

In one embodiment, a computer program product is provided comprising a computer program which, when executed by a processor, performs the steps of:

acquiring at least one section of call record to be analyzed;

The user information (including but not limited to user equipment information, user personal information, etc.) and the data (including but not limited to data for analysis, stored data, presented data, etc.) related to the present application are information and data authorized by the user or sufficiently authorized by each party.

Those skilled in the art will appreciate that implementing all or part of the above-described methods in accordance with the embodiments may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed may comprise the steps of the embodiments of the methods described above. Any reference to memory, database, or other medium used in embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high density embedded nonvolatile Memory, resistive random access Memory (ReRAM), magnetic random access Memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric Memory (Ferroelectric Random Access Memory, FRAM), phase change Memory (Phase Change Memory, PCM), graphene Memory, and the like. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, and the like. By way of illustration, and not limitation, RAM can be in the form of a variety of forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), and the like. The databases referred to in the embodiments provided herein may include at least one of a relational database and a non-relational database. The non-relational database may include, but is not limited to, a blockchain-based distributed database, and the like. The processor referred to in the embodiments provided in the present application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, or the like, but is not limited thereto.

The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.

The foregoing examples illustrate only a few embodiments of the application, which are described in detail and are not to be construed as limiting the scope of the application. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the application, which are all within the scope of the application. Accordingly, the scope of the application should be assessed as that of the appended claims.

Claims

1. A method of semantic analysis, the method comprising:

acquiring at least one section of call record to be analyzed;

2. The method of claim 1, wherein determining the semantic analysis result of each call record to be analyzed based on the determination result and the predicted traffic label and the predicted behavior label of each call record to be analyzed comprises:

if the prediction business label and/or the prediction behavior label of each call record to be analyzed have a wrong prediction label, determining an abnormal call record corresponding to the wrong label from each call record to be analyzed, and carrying out semantic verification on the abnormal call record based on at least one section of candidate call record to obtain a verification result; each candidate call record comprises a call record to be analyzed which does not belong to an abnormal call record, and/or a history call record with known history analysis results;

3. The method according to claim 2, wherein the performing semantic verification on the abnormal call record based on the at least one segment of candidate call record to obtain a verification result includes:

4. A method according to claim 2 or 3, wherein determining the semantic analysis result of each call record to be analyzed based on the verification result, the predicted traffic label and the predicted behavior label of each call record to be analyzed comprises:

5. The method of claim 4, wherein the determining, according to the verification result, an abnormal service tag and an abnormal behavior tag corresponding to an abnormal call record includes:

And if the verification result is that the similarity of each candidate call vector and the abnormal call vector is smaller than the similarity threshold value, acquiring an abnormal service label and an abnormal behavior label corresponding to the abnormal call record.

6. The method of claim 4, wherein determining an abnormal service tag and an abnormal behavior tag corresponding to an abnormal call record according to the verification result, further comprises:

and if the verification result is that the similar call vectors with the similarity larger than the similarity threshold exist in the candidate call vectors, using the similar service labels and the similar behavior labels corresponding to the similar call vectors as the abnormal service labels and the abnormal behavior labels corresponding to the abnormal call records.

7. The method according to claim 1, wherein the method further comprises:

8. The method of claim 1, wherein the training process of the target analytical model comprises:

based on the professional phrase sample of the service field of the call record to be analyzed, performing professional phrase training on the initial analysis model to obtain an intermediate analysis model;

and training the business analysis layer of the intermediate analysis model based on the business label training sample in the conversation training sample, and training the behavior analysis layer of the intermediate analysis model based on the behavior label training sample in the conversation training sample to obtain the trained target analysis model.

9. A semantic analysis apparatus, the apparatus comprising:

10. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any one of claims 1 to 8 when the computer program is executed.

11. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 8.

12. A computer program product comprising a computer program, characterized in that the computer program, when executed by a processor, implements the steps of the method of any one of claims 1 to 8.