CN115134466A

CN115134466A - Intention recognition method and device and electronic equipment

Info

Publication number: CN115134466A
Application number: CN202210635380.0A
Authority: CN
Inventors: 王雪强; 赵国庆; 杨锋; 蒋宁
Original assignee: Mashang Xiaofei Finance Co Ltd
Current assignee: Mashang Xiaofei Finance Co Ltd
Priority date: 2022-06-07
Filing date: 2022-06-07
Publication date: 2022-09-30

Abstract

The disclosure provides an intention identification method, an intention identification device and electronic equipment, wherein the method comprises the following steps: acquiring a call recording file corresponding to a target outbound task; the target outbound task is completed by the intelligent customer service, and the online intention identification result of the target outbound task is identified and obtained in the process of the target outbound task; respectively extracting customer service voice content corresponding to a first sound channel and user voice content corresponding to a second sound channel from the call recording file; determining an off-line intention recognition result of a target outbound task according to the first user voice information; the first user voice information is user voice information which is contained in the user voice content and is coincident with the customer service voice content in time; and determining an actual intention recognition result according to the offline intention recognition result and the online intention recognition result. By adopting the embodiment of the disclosure, the identification result can be more accurate.

Description

Intention recognition method and device and electronic equipment

Technical Field

The present disclosure relates to the field of information processing technologies, and in particular, to an intention identification method and apparatus, and an electronic device.

Background

The customer service system is widely applied to various fields, and can provide various services for users through customer service. In order to save labor cost, intelligent customer service is introduced in some business scenes. For example, intelligent services are provided to the user by the intelligent robot according to preconfigured standard dialogs. The intelligent customer service can be widely applied to various service scenes such as information pushing, after-sale service, user consultation, problem recording and the like.

In order to provide better service for users in subsequent business links, the user intentions of the users communicating with the intelligent customer service are required to be identified so as to fully understand the user requirements and provide targeted service for the users in subsequent processes. At present, in order to identify the user intention, the following method is generally adopted: in the process of the communication between the intelligent customer service and the user, the speaking content of the user is obtained in real time, and the intention of the user is identified according to the obtained speaking content of the user. However, due to the influence of factors such as network transmission abnormality, the user speaking content acquired in real time may have abnormal situations such as content loss, and the recognition result of the user intention may be inaccurate.

Disclosure of Invention

The present disclosure provides a method, an apparatus and an electronic device for intelligent customer service based intention recognition that overcome the above problems or at least partially solve the above problems.

In a first aspect, the present disclosure provides an intention identification method based on intelligent customer service, including:

acquiring a call recording file corresponding to a target outbound task; the target outbound task is completed by the intelligent customer service, and an online intention identification result of the target outbound task is identified and obtained in the process of the target outbound task;

respectively extracting customer service voice content corresponding to a first sound channel and user voice content corresponding to a second sound channel from the call recording file;

determining an off-line intention recognition result of the target outbound task according to the first user voice information; the first user voice information is user voice information which is contained in the user voice content and is time-coincident with the customer service voice content;

and determining an actual intention recognition result of the target outbound task according to the offline intention recognition result and the online intention recognition result.

In a second aspect, the present disclosure provides an intention recognition apparatus based on intelligent customer service, including:

the recording acquisition module is suitable for acquiring a call recording file corresponding to the target outbound task; the target outbound task is completed by the intelligent customer service, and an online intention identification result of the target outbound task is identified and obtained in the process of the target outbound task;

the extraction module is suitable for respectively extracting customer service voice content corresponding to a first sound channel and user voice content corresponding to a second sound channel from the call recording file;

the offline intention recognition module is suitable for determining an offline intention recognition result of the target outbound task according to the first user voice information; the first user voice information is user voice information which is contained in the user voice content and is time-coincident with the customer service voice content;

and the actual intention recognition module is suitable for determining the actual intention recognition result of the target outbound task according to the offline intention recognition result and the online intention recognition result.

In a third aspect, the present disclosure provides an electronic device comprising: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores one or more computer programs executable by the at least one processor, the one or more computer programs being executable by the at least one processor to enable the at least one processor to perform the above-mentioned intent recognition method.

In a fourth aspect, the present disclosure provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the intent recognition method described above.

According to the embodiment provided by the disclosure, the call recording file corresponding to the target outbound task can be obtained; and respectively extracting customer service voice content corresponding to the first sound channel and user voice content corresponding to the second sound channel from the call recording file, thereby determining an offline intention recognition result of the outbound task according to first user voice information which is contained in the user voice content and is time-coincident with the customer service voice content, and determining an actual intention recognition result of the target outbound task according to the offline intention recognition result and the online intention recognition result. Therefore, the online intention recognition result can be verified through the offline intention recognition result. The offline intention recognition result is realized according to the call recording file, and all call information is contained in the call recording file, so that the first user voice information which cannot be acquired in the online intention recognition process can be acquired, and the accuracy of the offline intention recognition result is higher than that of the online intention recognition result. Correspondingly, the actual intention recognition result jointly determined according to the off-line intention recognition result and the on-line intention recognition result is more accurate and is not influenced by objective factors such as network speed and the like.

It should be understood that the statements in this section do not necessarily identify key or critical features of the embodiments of the present disclosure, nor do they limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description.

Drawings

The accompanying drawings are included to provide a further understanding of the disclosure and are incorporated in and constitute a part of this specification, illustrate embodiments of the disclosure and together with the description serve to explain the principles of the disclosure and not to limit the disclosure. The above and other features and advantages will become more apparent to those skilled in the art by describing in detail exemplary embodiments thereof with reference to the attached drawings, in which:

fig. 1 is a flowchart of an intention identifying method provided by an embodiment of the disclosure;

FIG. 2 is a flow chart of an intent recognition method according to yet another embodiment of the present disclosure;

FIG. 3 illustrates an interface diagram of an offline intent recognition configuration interface;

FIG. 4 illustrates an interface diagram of a call log presentation interface;

FIG. 5 is a schematic diagram showing a point in time of customer service voice content and user voice content;

FIG. 6 is a block diagram of an intent recognition apparatus provided in an embodiment of the present disclosure;

fig. 7 is a block diagram of an electronic device provided in an embodiment of the present disclosure.

Detailed Description

To facilitate a better understanding of the technical aspects of the present disclosure, exemplary embodiments of the present disclosure are described below in conjunction with the accompanying drawings, wherein various details of the embodiments of the present disclosure are included to facilitate an understanding, and they should be considered as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

Embodiments of the present disclosure and features of embodiments may be combined with each other without conflict.

As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. The terms "connected" or "coupled" and the like are not restricted to physical or mechanical connections, but may include electrical connections, whether direct or indirect.

Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and the present disclosure, and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

The intention recognition method according to the embodiments of the present disclosure may be performed by an electronic device such as a terminal device or a server, the terminal device may be a vehicle-mounted device, a User Equipment (UE), a mobile device, a User terminal, a cellular phone, a cordless phone, a Personal Digital Assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, or the like, and the method may be implemented by a processor calling a computer-readable program instruction stored in a memory. Alternatively, the method may be performed by a server.

Fig. 1 is a flowchart of an intent recognition method based on intelligent customer service according to an embodiment of the present disclosure. Referring to fig. 1, the method includes:

step S110: acquiring a call recording file corresponding to a target outbound task; the target outbound task is completed by the intelligent customer service, and the online intention identification result of the target outbound task is identified and obtained in the process of the target outbound task.

Wherein, the target outbound task is as follows: the intelligent customer service calls the appointed user and carries out the task of conversation, and also calls the outbound task to be identified. And the outbound task obtains an online intention identification result in a call process in an online intention identification mode. However, since the online intention recognition result is determined in real time, the content is easily lost due to network abnormality or the like, and the accuracy of the result is affected. In addition, in the process of determining the online intention recognition result, the following method is generally adopted: and alternately acquiring customer service voice content corresponding to the first sound channel and user voice content corresponding to the second sound channel, and performing online intention recognition according to the alternately acquired customer service voice content and user voice content. By alternate acquisition is meant: and after the customer service voice content of the first sound channel is completely broadcasted, acquiring the user voice content of the second sound channel. Therefore, only the voice content of one channel is acquired at any one time, and the voice content of two channels is not acquired at the same time.

Therefore, in order to improve the identification accuracy, the call recording file corresponding to the target outbound task is acquired. The call recording file is a complete recording of both parties of the call and simultaneously contains the voice contents of two sound channels corresponding to each time point.

Step S120: and respectively extracting customer service voice content corresponding to the first sound channel and user voice content corresponding to the second sound channel from the call recording file.

Wherein the first channel corresponds to a customer service channel and the second channel corresponds to a user channel. Correspondingly, the contents of different sound channels in the call recording file are separated in a sound channel separation mode, so that customer service voice contents corresponding to the first sound channel and user voice contents corresponding to the second sound channel are obtained. Therefore, the customer service voice content refers to: content transmitted over a first channel; the user voice content means: content transmitted over the second channel.

Step S130: determining an offline intention recognition result of the outbound task according to the voice information of the first user; the first user voice information is user voice information which is contained in the user voice content and is coincident with the customer service voice content in time.

In the real-time call process, when a call user and an intelligent customer service speak simultaneously, due to the fact that signals are transmitted in the first sound channel and the second sound channel simultaneously, the situations such as abnormal transmission and the like are prone to occurring. Moreover, because an identification mode of alternately acquiring the voice content of each channel is usually adopted in the online intention identification process, and the voice content of two channels cannot be acquired simultaneously, the first user voice information which is included in the user voice content and is time-coincident with the customer service voice content, namely the first user voice information, is focused in the offline identification process. As can be seen, the first user voice information refers to: and under the condition that the customer service voice content is not completely broadcasted, the voice information sent by the user is sent by the user, and the sending time is coincident with the sending time of the customer service voice. By analyzing the first user voice information, an offline intention recognition result of the outbound task can be determined.

Because the off-line intention identification result is determined according to the call recording file, and the call recording file contains the complete contents of both parties of the call, the off-line intention identification result is not influenced by factors such as network abnormity and the like, and the identification result is more accurate. Therefore, the online intention recognition result can be verified through the offline intention recognition result, and the more accurate actual intention recognition result is obtained.

Step S140: and determining an actual intention recognition result of the target outbound task according to the offline intention recognition result and the online intention recognition result.

Because the off-line intention recognition result can be recognized according to the first user voice information which cannot be obtained in the on-line recognition process, the recognition accuracy is higher. Correspondingly, the actual intention recognition result of the target outbound task can be determined according to the off-line intention recognition result and the on-line intention recognition result. The actual intention recognition result is a final recognition result obtained by integrating the offline intention recognition result and the online intention recognition result and is used for reflecting the real conversation intention of the called user corresponding to the target outbound task.

When the actual intention recognition result of the target outbound task is determined, the method can be flexibly realized in various ways, for example, the offline intention recognition result is compared with the online intention recognition result, and if the offline intention recognition result is consistent with the online intention recognition result, the online intention recognition result is directly used as the actual intention recognition result; if the two are not consistent, the offline intention recognition result can be used as the actual intention recognition result. In short, the present invention is not limited to a specific determination method as long as a more accurate result can be obtained.

Therefore, in the embodiment, the call recording file corresponding to the target outbound task can be obtained; and respectively extracting customer service voice content corresponding to the first sound channel and user voice content corresponding to the second sound channel from the call recording file, thereby determining an offline intention recognition result of the outbound task according to first user voice information which is contained in the user voice content and is time-coincident with the customer service voice content, and determining an actual intention recognition result of the target outbound task according to the offline intention recognition result and the online intention recognition result. The method can verify the online intention recognition result through the offline intention recognition result. Since the off-line intention recognition result is performed according to the call recording file and all call information is contained in the call recording file, the first user voice information which cannot be acquired in the on-line intention recognition process can be acquired, and therefore the accuracy of the off-line intention recognition result is higher than that of the on-line intention recognition result. Correspondingly, the actual intention recognition result jointly determined according to the off-line intention recognition result and the on-line intention recognition result is more accurate and is not influenced by objective factors such as network speed and the like.

Fig. 2 is a flowchart of an intent recognition method based on intelligent customer service according to another embodiment of the present disclosure. In the embodiment of the disclosure, in order to save cost, the intelligent customer service is an intelligent robot which does not support an intelligent interrupt function. Intelligent interruption means: and the intention of the client is identified in the speaking process of the user by utilizing a real-time speaker voice identification algorithm and a noise reduction interruption technology. In the intelligent interruption scene, the robot can acquire the voice content sent by the user before the end of the broadcasting of the customer service voice content, so that the robot can perform targeted processing according to the voice content sent by the user before the end of the broadcasting of the customer service voice content under the condition that the broadcasting of the customer service voice content is not finished. Therefore, the robot supporting the intelligent interrupt function has the following characteristics: the voice contents of the two channels can be simultaneously obtained. Namely: in the broadcasting process of the customer service voice content corresponding to the first sound channel, the customer service voice content of the first sound channel is obtained, and meanwhile, the user voice content transmitted through the second sound channel is detected and obtained.

By not supporting smart interruptions, we mean: in the process of executing intelligent voice broadcast by the robot, the robot cannot receive voice fed back by the user, and only after the intelligent voice broadcast of the robot is finished, the robot can receive the voice fed back by the user. In other words, in the process of performing voice broadcast as an intelligent customer service by the robot, the user cannot interrupt the intelligent voice broadcast process. Therefore, the robot which does not support the intelligent interrupt function has the following characteristics: the speech content of both channels cannot be acquired simultaneously. At any point in time, only the voice content of one channel is acquired, and the voice content of the other channel cannot be acquired. Namely: in the broadcasting process of the customer service voice content corresponding to the first channel, the user voice content transmitted through the second channel cannot be acquired. That is, the user voice content transmitted through the second channel cannot be acquired until the customer service voice content broadcasting of the first channel is finished. And only under the condition that the broadcasting of the customer service voice content corresponding to the first channel is detected to be finished, the user voice content transmitted through the second channel can be acquired. For the robot, the intention identification method in the embodiment is provided, and specifically includes the following steps:

step S200: and carrying out online intention identification on the outbound task to obtain an online intention identification result.

Wherein, the outbound task means: the task of the specified user is called by the intelligent customer service. In the process of executing the outbound task, the intelligent customer service communicates with the user, and in the process of communicating, the outbound task is subjected to online intention identification. Wherein the online intention recognition result is determined by: after the fact that the broadcasting of the customer service voice information of the intelligent customer service through the first sound channel is finished is detected, second user voice information transmitted by a user through a second sound channel is obtained; and determining an online intention recognition result according to the acquired second user voice information. Different from the first user voice information mentioned above, the second user voice information is obtained after the completion of the customer service voice information broadcast is detected, so that the user voice information which is time-coincident with the customer service voice content is not included, and only the user voice information which is sent out after the completion of the customer service voice content broadcast is included. Therefore, the second user voice information and the first user voice information have different acquisition opportunities, acquisition modes and information contents. In addition, the online intention recognition result of the outbound task is recognized and obtained in the process of the outbound task.

The customer service voice information of the intelligent customer service is usually standard voice information configured in advance, contains a preset number of sentences and corresponds to a preset voice playing time length. Correspondingly, under the condition that the customer service voice information is not completely broadcasted, the execution main body for executing the online intention recognition operation cannot acquire the user voice information transmitted by the user through the second channel. After the customer service voice information is completely broadcasted, the execution main body for executing the online intention recognition operation can acquire the user voice information transmitted by the user through the second channel.

The executing agent for executing the online intention recognition operation may be a robot serving as an intelligent customer service, or may be an Internet Call Center (ICC). Alternatively, the execution subject for executing the online intention recognition operation may also be a network device installed with an online intention recognition model. In summary, the present invention is not limited to the execution subject of the online intent recognition operation. However, in the present application, since the intelligent customer service does not support the intelligent interruption function, the execution main body for executing the online intention recognition operation can acquire the user voice information only after the customer service voice information is completely broadcasted. In specific implementation, after the completion of the broadcasting of the customer service voice information is detected, a corresponding user voice acquisition event is triggered, so that an execution main body executing online intention recognition operation acquires the user voice information.

In addition, if the customer service voice information includes a plurality of customer service voice paragraphs, the corresponding user voice obtaining event may be triggered after the completion of the broadcasting of the current customer service voice paragraph is detected each time, so that the executing body executing the online intention identifying operation obtains the user voice information. For example, if the smart customer service is a robot that supports a question and a answer, then a plurality of customer service speech passages correspond to different questions or answers, respectively. Correspondingly, when the intelligent customer service finishes broadcasting the first customer service voice section, triggering a first user voice obtaining event to obtain user voice content sent by a user aiming at the first customer service voice section; and after the intelligent customer service finishes broadcasting the second customer service voice section, triggering a second user voice obtaining event to obtain the user voice content … … sent by the user aiming at the second customer service voice section, and so on.

Therefore, the online intention recognition result is obtained according to the user voice content obtained after the customer service voice information is broadcasted. Therefore, the user voice content generated before the customer service voice information is completely broadcasted cannot be acquired. For example, some users may have responded in advance when the customer service voice information is not completely broadcast due to time emergency, and the content of the part cannot be acquired, so that the online intention recognition result is inaccurate.

In this embodiment, the number of outbound tasks is usually multiple, and after online intention recognition is performed on each outbound task, an online intention recognition result corresponding to each outbound task is obtained. In order to facilitate the query of the online intention recognition result of each outbound task, a corresponding online intention recognition tag can be set for each outbound task.

Step S210: and acquiring task parameters contained in the received offline identification configuration instruction, and setting offline identification configuration information according to the task parameters.

The offline recognition configuration instruction can be triggered through an offline recognition configuration entry provided in an offline recognition configuration interface. The number of the offline identification configuration entries can be multiple. Accordingly, the number of task parameters included in the offline identification configuration instruction may be multiple. Also, the number of offline identification configuration instructions may be one or more.

In a first implementation, the offline recognition configuration entry includes a first configuration entry named "offline recognition switch" for configuring whether to turn on the offline intent recognition operation: if the switch state of the off-line identification switch is the opening state, executing off-line intention identification operation; if the off-line recognition switch is in the off state, the off-line intention recognition operation is not executed. Correspondingly, the task parameters contained in the offline identification configuration instruction include: and the switch parameter is used for indicating whether the offline intention identification operation needs to be executed.

In a second implementation, the offline recognition configuration entry includes a second configuration entry named "timed start time" for configuring the timed start time of the offline intention recognition operation, and the offline intention recognition operation is automatically started each time the timed start time is reached. For example, the timed start time is 19 per day: 00 or weekly weekdays, etc. In summary, the timing start time may be set to a traffic low peak period to reduce the impact on traffic. Correspondingly, the task parameters contained in the offline identification configuration instruction include: a timed initiation parameter for indicating when an offline intent recognition operation needs to be timed initiated.

In a third implementation, the offline recognition configuration entries include a third configuration entry named "type of intent to be recognized" for configuring which types of outbound tasks belonging to online intent recognition results need to be performed offline intent recognition operations. Correspondingly, the task parameters contained in the offline identification configuration instruction include: a first parameter indicating an online intent type of an online intent recognition result of the outbound task. As mentioned above, in the present embodiment, each outbound task has performed the online intent recognition operation in real time during the call, and has obtained the online intent recognition result. Accordingly, in the offline recognition configuration interface, it may be further defined by a third configuration entry named "type of intent to be recognized" for which types of online intent recognition results offline intent recognition operations need to be performed.

For example, in the case that the task type of the outbound task is the information push type, each outbound task may correspond to the following four types of online intention recognition results: intent of interest, busy intent, to follow intent, deny intent. If the online intention identification result is an interest intention, the fact that the user corresponding to the outbound task is interested in the pushed information content is indicated, the user is classified into the type of interest in the user classification table, and then the same type of information can be pushed for the user again. If the online intention identification result is a rejection intention, the fact that the user corresponding to the outbound task is not interested in the pushed information content is indicated, the user is classified into rejection types in the user classification table, and the same type of information is not pushed for the user in the following process. If the online intention identification result is a busy intention, the user is in a busy state when answering the call, so that the user is not yet aware of whether the pushed information is interesting or not; if the online intention recognition result is the intention to follow, the intention of the user for the pushed information is not recognized through the online intention recognition operation. It follows that in this example, the "busy intent" and the "to-follow intent" may be set to the "type of intent to be recognized". Accordingly, when the offline intention recognition operation is performed, it is only necessary to perform the outbound task whose online intention recognition result is "busy intention" and "to-follow intention", and it is not necessary to perform the offline intention recognition for the outbound task whose online intention recognition result is "intention of interest" and "intention of rejection". Therefore, the first parameter used for indicating the online intention type of the online intention recognition result of the outbound task is contained in the offline recognition configuration instruction, so that the problem of resource waste caused by secondary recognition of the accurate online intention recognition result can be avoided, and limited system resources are fully used for secondary recognition of the outbound task with the inaccurate online intention recognition result.

In a fourth implementation manner, the offline recognition configuration entries include a fourth configuration entry named "recording range to be recognized", and are used for configuring which call records of the outbound task need to be subjected to offline intention recognition operation. Correspondingly, the task parameters contained in the offline identification configuration instruction include: a second parameter for indicating a call time for the outbound task. For example, it is set to perform an offline intent recognition operation for an outbound task of ten to twelve am every day. Because ten to twelve am in each day is the peak period of the business, the online intention identification operation is more easily influenced by factors such as network abnormality and the like, so that the accuracy of online intention identification is low; also, outbound tasks performed ten to twelve am are generally more important, and therefore performing offline intent recognition operations for that portion of the outbound task helps to improve accuracy. Therefore, information such as time period of the target outbound task can be flexibly configured through the fourth configuration entry of the recording range to be identified, so that the task amount of the target outbound task is reduced, and the offline intention identification efficiency is improved.

In addition, for the fourth configuration entry named "recording range to be identified", the task parameters included in the offline identification configuration instruction may further include: a third parameter indicating a task type of the outbound task. For example, assume that the task types of the outbound task include: information push type, call return type, after-sale consultation type. Accordingly, the outbound task of "information push type" may be determined as the type of task that needs to perform offline intent recognition. In short, the target outbound task can be configured by other factors such as task type besides the time period, which is not limited in this disclosure.

In a fifth implementation, the offline recognition configuration entries include a fourth configuration entry named "update mode" for configuring which call records of outbound tasks are to be subjected to offline intent recognition operations. Correspondingly, the task parameters contained in the offline identification configuration instruction include: a second parameter indicative of a call time for the outbound task.

For ease of understanding, fig. 3 is a schematic diagram of an offline recognition configuration interface, which is shown in fig. 3 and includes: the off-line identification switch, the timing starting time, the intention type to be identified and the recording range to be identified can store each task parameter as off-line identification configuration information due to the fact that the types and the number of configurable task parameters are large. For example, parameter values of each task parameter are written into a configuration file, so that each item of offline identification configuration information can be conveniently acquired in an offline intention identification process in a manner of loading the configuration file.

Step S220: and determining a target outbound task according to the offline identification configuration information.

The steps S210 and S220 are optional steps, and the information such as the number and the type of the target outbound task can be flexibly set through the steps S210 and S220, and the execution time of the offline intention recognition operation can also be set. If step S210 and step S220 are omitted, the offline intention recognition operation is performed for all completed outbound tasks by default.

In addition, the skilled person may also set the execution condition of step S220. For example, the execution conditions of step S220 are: the off-line recognition switch has an open state. Specifically, whether the offline intention identifying operation needs to be executed is judged according to the switch parameters contained in the offline identifying configuration command. And if the switch parameter contained in the offline identification configuration instruction is the opening state value, executing offline intention identification operation.

The execution time of step S220 may also be flexibly configured. For example, the execution time or execution cycle of the offline intention identifying operation may be preset, and when the execution time or execution cycle of the offline intention identifying operation predetermined in advance arrives, step S220 is executed, otherwise, step S220 is not executed. The execution timing of step S220 is a task start time set for an execution subject of the offline intention recognition operation, where the execution subject of the offline intention recognition operation may be various types of network devices such as an offline intention recognition robot and an offline intention recognition server. In summary, the execution time or the execution cycle of the offline intention recognition operation is used to set the start time of the offline intention recognition task.

In addition, when the offline intention recognition task is started, it may be further determined when to start executing the offline intention recognition operation by the timing start parameter included in the offline recognition configuration instruction mentioned above.

The execution time or execution cycle of the offline intention identifying operation is used for setting the starting time of the offline intention identifying task, and the timing starting parameter contained in the offline identifying configuration instruction is used for setting when the offline intention identifying operation is started after the offline intention identifying task is started. In practical cases, only the execution time or the execution cycle of the offline intention recognition operation may be set, or only the timing start parameter included in the offline recognition configuration instruction may be set. The present disclosure is not limited to the specific details.

For example, in this step, it may be further determined whether the current time is within an executable time period, where the time period is divided into three segments: points 0-6 are the time at which the system update is released, which allows the offline intent recognition task to execute immediately; points 6 to 19 are normal operation time of the business in the daytime, and off-line intention identification execution is strictly forbidden (the execution in the daytime can cause the contention of server resources, so that the normal business is influenced); point 19-0 belongs to a time period during which the offline intention recognition task can be executed, and during this time period, whether the timing start time in the offline recognition setting is reached or not can be checked by polling every preset time (for example, 5 seconds), and if so, the next step is executed. The 5 seconds are set to avoid unnecessary stress on the system due to multiple executions in a short time.

Specifically, the target outbound task is screened when the offline intention identifying task is started and the offline intention identifying operation is started, and specifically, the outbound task which is included in the outbound history record and is matched with the task parameter in the offline identifying configuration instruction may be screened as the target outbound task. In the embodiment of the present disclosure, the target outbound task may be screened through several screening methods as follows:

in a first screening method, a target outbound task is screened according to a first parameter contained in an offline identification configuration instruction. The first parameter is used for indicating the online intention type of the online intention recognition result of the outbound task. For example, according to the online intention type contained in the first parameter, the outbound task with the online intention identification result matched with the online intention type is screened for offline intention identification.

In a second screening method, the target outbound task is screened according to a second parameter included in the offline identification configuration instruction. Wherein the second parameter is used for indicating the calling time of the outbound task. For example, according to the parameter values contained in the second parameters, outbound tasks performed from ten to twelve am every day are screened for offline intention identification.

In a third screening mode, the target outbound task is screened according to a third parameter contained in the offline identification configuration instruction. Wherein the third parameter is used for indicating the task type of the outbound task. For example, according to the parameter value contained in the third parameter, the outbound task of the information push type is screened to perform offline intention identification.

The screening methods can be used in combination or individually, and when the screening methods are used in combination, the execution sequence among the screening methods is not limited.

In addition, in order to avoid repeatedly executing the offline intention identification operation on the same outbound task, an execution mark can be set for the outbound task which has executed the offline intention identification operation, correspondingly, whether each outbound task has executed the offline intention identification on the same day is judged through the execution mark, if the outbound task has executed the offline intention identification, the offline intention identification is not carried out, and the problems of data repetition and resource waste caused by repeated identification are avoided.

Step S230: and acquiring each target outbound task from the outbound history record, and sequentially storing the acquired target outbound tasks into a task queue.

Because the number of the target outbound tasks is large, each target outbound task is stored through the task queue, and the outbound tasks can be processed in order. In the embodiment, in order to improve the task processing efficiency, a plurality of task execution instances are created and executed in parallel.

The multiple task execution instances need to sequentially store each target outbound task acquired from the outbound history record into the task queue in a queue initialization mode. The queue initialization operation in this step is usually performed by one task execution instance, and therefore, the initialization operation is performed by combining a distributed lock and a blocking queue. Correspondingly, the task queue comprises: and the first blocking queue is provided with task waiting time and is used for storing each target outbound task. The blocking queue has the following characteristics: when the queue is empty, the thread that gets the element may wait for the queue to become non-empty and may set the latency. Correspondingly, in the step, one of the multiple task execution examples is controlled through the distributed lock, each target outbound task is obtained from the outbound history record, and the obtained multiple target outbound tasks are sequentially stored in the first blocking queue included in the task queue according to the sequence of the obtaining time of each target outbound task. Through the distributed lock, the task execution instances in the task execution instances can be ensured not to repeatedly execute the same task, and the operation of storing the target outbound task to the first blocking queue is ensured to be executed by one or only one task execution instance. And by blocking the queue, the task waiting time can be set, so that other task execution instances are prevented from exiting in the process of waiting for the initialization to be completed.

The inventor finds that initialization fails due to abnormal reasons such as downtime and the like in the process of implementing the invention, so that if other task execution examples are enabled to wait endlessly, system resources are wasted. The conventional task queue can not set the task waiting time, so that other task execution instances can only wait endlessly or quit early before the initialization is finished. In order to solve the problem, the blocking queue is arranged, and the task waiting time length is reasonably set according to the estimated time length of the initialization, so that other task execution instances can not exit before the initialization is finished, and can not wait endlessly when the initialization is abnormal.

Step S240: and a plurality of task execution instances acquire a call recording file corresponding to the target outbound task according to a first blocking queue contained in the task queue in a parallel mode.

Specifically, a plurality of task execution instances acquire a call recording file corresponding to the target outbound task according to the first blocking queue in a parallel mode. In one implementation, a plurality of target outbound tasks form a task sequence, and correspondingly, when the acquired target outbound tasks are sequentially stored in the task queue, the plurality of target outbound tasks are stored in the task queue by taking the task sequence as a whole. Wherein, one task sequence comprises a plurality of target outbound tasks. For example, a sequence of tasks corresponds to a plurality of targeted outbound tasks corresponding to a set of outbound lists. The number of target outbound tasks included in one task sequence may be several tens or even more. Correspondingly, in order to avoid abnormal situations such as downtime during the execution of the task execution instance, in this embodiment, the processing is performed by taking the task sequence as a unit, and the task queue further includes: and the second execution queue is used for storing each outbound task in the current task execution instance being processed by means of the second execution queue. Correspondingly, when the call recording file corresponding to the target outbound task is obtained according to the first blocking queue, the method is realized in the following mode: according to the sequence of each task sequence stored in the first blocking queue, one task sequence is sequentially obtained from the first blocking queue to serve as a current task sequence, and each outbound task contained in the obtained current task sequence is stored in a second execution queue; according to the second execution queue, sequentially acquiring call recording files corresponding to all outbound tasks contained in the current task sequence; deleting the current task sequence from the second execution queue and taking the next task sequence in the first blocking queue as a new current task sequence every time each outbound task in the current task sequence stored in the second execution queue is processed; and, in case of system abnormality, the recovery can be performed according to the current task sequence in the second execution queue.

For example, first, an ith task sequence is acquired from a first blocking queue, and each outbound task included in the acquired ith task sequence is stored in a second execution queue. And then, according to the second execution queue, sequentially acquiring the call recording files corresponding to all outbound tasks contained in the ith task sequence. After each outbound task in the ith task sequence stored in the second execution queue is processed, deleting the ith task sequence from the second execution queue, acquiring an (i + 1) th task sequence from the first blocking queue, and storing each outbound task contained in the acquired (i + 1) th task sequence into the second execution queue; wherein i is a natural number. And, when the execution is performed for the first time, the value of i is 1, and the value of i gradually increases with the increase of the execution times. As can be seen, the second execution queue is used to store the sequence of tasks that are currently being processed. The advantage of introducing a second execution queue is that: if abnormal conditions such as downtime occur in the process of processing the current task sequence, the processing is carried out according to the task sequence stored in the second execution queue after the system is recovered, so that the loss of the task sequence being processed due to factors such as system restart or fault is prevented. In addition, when the system is down, part of the target outbound tasks in the task sequence are processed, so that the task identifiers of the processed outbound tasks can be further recorded, and correspondingly, when the system is restored and then the system is processed again according to the task sequence stored in the second execution queue, the next unprocessed target outbound task is determined according to the recorded task identifiers of the processed target outbound tasks, so that the tasks are prevented from being processed repeatedly.

In addition, when a plurality of task execution instances acquire a call recording file corresponding to the target outbound task according to the first blocking queue in a parallel mode, the method is specifically realized by the following modes: after each task execution instance obtains the target outbound task from the first blocking queue, obtaining a token from a token bucket; and if the token is successfully acquired, sending a recording acquisition request to the call center server so that the call center server can return a call recording file corresponding to the target outbound task acquired from the first blocking queue. When the call record file is stored in the call center server, considering that the resource of the call center server is limited, and therefore, the call record file cannot be acquired without limitation, a query rate per second (QPS) needs to be controlled by a token bucket method. The current limiting function can be realized through the token bucket, and the task execution instance which takes the token from the token bucket can send a request for acquiring the call recording file to the call center server only by controlling the number of tokens generated in each second in the token bucket, so that the current limiting purpose is realized.

Step S250: and executing voice-to-text (ASR) processing aiming at the acquired call recording file.

The speech-to-text processing of the call recording file can be realized by calling the ASR service. Because the number of the call recording files is large, the ASR processing result of each call recording file can be stored in the thread pool, so that the subsequent processing can be performed in an asynchronous processing mode.

Step S250 is an optional step, and in other embodiments of the present invention, the subsequent processing may also be performed directly on the speech without performing a text conversion operation, and details of the present invention are not limited.

For example, in one implementation, the network device performing the offline intention recognition operation sends the recording storage address of the call recording file of each outbound task in the call center server to the ASR service, and the ASR service performs the speech-to-text processing. And after the ASR service obtains the ASR processing result of each call recording file, the ASR processing result of each call recording file is called back to the network equipment executing the offline intention recognition operation. Namely: and the network equipment executing the off-line intention recognition operation performs callback through an HTTP interface mode to obtain an ASR processing result of each call recording file processed by ASR service. Because the ASR processing results of the call recording files are numerous (more than 30W records need to be recognized in one day), in order to avoid blocking when receiving ASR recognition results sent by the ASR service, the received ASR recognition results can be put into a thread pool for asynchronous processing.

Step S260: respectively extracting customer service voice content corresponding to a first sound channel and user voice content corresponding to a second sound channel from the call recording file; and determining an offline intention recognition result of the outbound task according to the first user voice information.

The call recording file is the complete recording content of both parties of the call, so the customer service voice content corresponding to the first sound channel and the user voice content corresponding to the second sound channel can be extracted according to the sound channel information. In addition, the customer service voice content and the user voice content each have time stamp information that can indicate the start and end times of the customer service voice content and the user voice content. Accordingly, the first user voice information can be obtained by the following method: respectively determining customer service voice content and start-stop time information of the user voice content; and determining the voice information of the first user according to the start-stop time information.

For example, the start time point of the customer service voice content is t1, and the end time point is t 2; the start time point of the user voice content is t3, and the end time point is t 4. If the starting time point t3 of the user voice content is earlier than the ending time point t2 of the customer service voice content, it indicates that the user voice content includes user voice information time-coincident with the customer service voice content, that is: the first user voice information. FIG. 5 illustrates a point-in-time diagram of customer service voice content and user voice content. As shown in fig. 5, the first user voice information refers to: user voice information located between a start time point t3 of the user voice content and an end time point t2 of the customer service voice content. When the offline intention recognition result of the outbound task is determined according to the first user voice information which is contained in the user voice content and is time-coincident with the customer service voice content, the offline intention recognition result can be determined through an offline intention recognition model obtained through pre-training. In addition, in order to improve the accuracy of the model, different offline intention recognition models can be trained in advance according to parameter values of different task parameters. Accordingly, this step can be implemented in the following way: determining an offline intent recognition model matched with parameter values of the task parameters; pre-training to obtain a plurality of offline intention recognition models respectively matched with different parameter values of the task parameters; and performing intention recognition on the voice information of the first user through the determined offline intention recognition model to obtain an offline intention recognition result of the target outbound task. For example, assume that the parameter values for the task parameters include: the system comprises a first parameter value used for indicating that the task type is an information push type, a second parameter value used for indicating that the task type is a call return type, and a third parameter value used for indicating that the task type is an after-sales consultation type; accordingly, the offline intent recognition model includes: a first intent recognition model corresponding to a first parameter value of the push type of information, a second intent recognition model corresponding to a second parameter value of the call return type, a third intent recognition model corresponding to a third parameter value of the after-sales consultation type. Because each task type has different characteristics, different intention recognition models are trained respectively aiming at different tasks, and the accuracy of results is improved.

For example, in one implementation, the intent recognition model is implemented by a three-classification robot whose knowledge is divided into three classes: complaints, reject grey lists, others. Wherein, different classification numbers and classification names can be set according to different service scenes. The three-classification robot is obtained by training in the following way: randomly extracting a batch of data generated in a conversation process with a client, for example, extracting 500 pieces of data at 3 o 'clock of today, 500 pieces of data at 10 o' clock of the last day, extracting 500 pieces of data at the current day of the last week, after extracting the data, performing intention labeling on the data, then using the labeled data as a training set or a test set, calling an algorithm model generated by Natural Language Processing (NLP) to perform intention recognition, generating a prediction report after recognition, analyzing and comparing according to the accuracy, precision, recall rate and other numerical values in the prediction report, and finally determining the robot adopting any classification mode. And after the robot is configured, conveying the robot to an algorithm side, and training by adopting a TextCNN model to generate an algorithm model. The method can be used for training through a TextCNN model, an LR model and a BERT model respectively, and selecting a better model according to a test effect. In order to improve the timeliness of offline intention recognition, for example, in some service scenarios, the offline recognition task needs to be completed within 3 hours, so in order to improve the recognition efficiency, after prediction analysis is performed on different algorithms, a TextCNN model is finally selected as an offline intention recognition model, so that the purposes of highest efficiency and shortest time consumption are achieved under the condition that intention recognition accuracy is met.

In addition, in the offline intention recognition model, corresponding classification keywords can be configured for each classification, so that the offline intention recognition type of the outbound task can be quickly determined according to the classification keywords. For example, the classification keywords corresponding to the rejection gray list include: disregard, not thought, not of interest, etc.

Step S270: and determining an actual intention recognition result of the target outbound task according to the offline intention recognition result and the online intention recognition result.

The actual intention recognition result is used for reflecting the actual call intention of the called user corresponding to the target outbound task, and the actual intention recognition result is comprehensively determined according to the offline intention recognition result and the online intention recognition result, so that the accuracy is higher. Specifically, when determining the actual intention recognition result, the offline intention recognition result may be compared with the online intention recognition result.

In one implementation, aiming at any target outbound task, comparing an offline intention recognition result of the target outbound task with an online intention recognition result; if the two are consistent, directly taking the determined online intention recognition result as an actual intention recognition result; and if the two are not consistent, determining the offline intention recognition result as an actual intention recognition result.

In yet another alternative implementation, in order to prevent some online intent recognition results that are not desired to be covered from being replaced, the permission update intent type may also be pre-configured. Namely: the result types of the online intention recognition result are previously divided into an update allowed intention type and a non-update allowed intention type. Correspondingly, if the off-line intention recognition result is inconsistent with the on-line intention recognition result, the off-line intention recognition result is determined as the actual intention recognition result of the target outbound task when the on-line intention recognition result belongs to the type of the permission updating intention. For example, in the case that the offline intention recognition result is inconsistent with the online intention recognition result, it is first determined whether the online intention recognition result belongs to an update-allowed intention type; if so, updating the online intention recognition result into an offline intention recognition result; wherein, the type of the permission updating intention can be determined according to the offline identification configuration instruction received in advance. For example, in the offline recognition configuration interface, an updatable intention configuration entry is further included, and a user can trigger an offline recognition configuration instruction containing an update-allowed intention type identifier through the updatable intention configuration entry, so that only the update-allowed intention type is updated in the step, and online intention recognition results which are not desired to be updated are prevented from being covered by the offline intention recognition results.

In addition, after the actual intention recognition result is determined, the confidence score of the actual intention recognition result can be further determined according to the comparison result between the offline intention recognition result and the online intention recognition result, and the accuracy of the actual intention recognition result is prompted through the confidence score. For example, if the comparison result between the offline intent recognition result and the online intent recognition result is consistent, the confidence score is a first score; and if the comparison result between the offline intention recognition result and the online intention recognition result is inconsistent, the confidence score is a second score, and the second score is smaller than the first score.

In the process of determining the actual intention recognition result of the target outbound task according to the offline intention recognition result and the online intention recognition result under the condition that the offline intention recognition result is inconsistent with the online intention recognition result, the following steps can be understood: and updating the online intention recognition result according to the offline intention recognition result. In addition, in the offline recognition configuration interface, the offline recognition configuration interface may further include an update operation switch, and when the update operation switch is in an on state, the online intention recognition result is updated to the offline intention recognition result through step S270, so as to obtain an actual intention recognition result; when the updating operation switch is in the closed state, the online intention recognition result is still kept for the user to view. In addition, in the execution process of step S270, in addition to performing intent updating, additional operations such as dialing decision (determining whether to dial the number again), adding a blacklist, and the like may be further performed. The intention after offline recognition can be updated to the intention generated when the user makes a call in real time in the daytime through intention updating, so that the correctness of data is ensured, and the user can see the real intention when viewing the call record. For example, in the call record presentation interface shown in fig. 4, the intention level is included, and includes two categories, i.e., "AI category" and "offline AI category". The AI classification is used for displaying an online intention recognition result; the "offline AI classification" is used to show the offline intent recognition result. In addition, the offline intention recognition result can be synchronized to the upstream information push system again, so that the information push system can decide which lists need to be subjected to information push. In addition, clients meeting negative intentions of refusing, abusing words, dirty words and the like can be added into the grey blacklist, and the occurrence of complaint events caused by repeated information push is avoided.

Step S280: updating the user classification table according to the actual intention identification result; the user classification list is used for realizing classification information pushing.

The user classification table is used for storing classification results of all users. The user classification table can be used for subsequent information push, call return visit and other operations. For example, assume that the actual intent recognition results include: the user name list of the interested intention type can be determined as a list for subsequently pushing information, so that disturbance on users who are not interested is avoided.

In summary, in the embodiments provided by the present disclosure, a call recording file corresponding to a target outbound task can be obtained; and respectively extracting customer service voice content corresponding to the first sound channel and user voice content corresponding to the second sound channel in the call recording file, so as to determine an offline intention recognition result of the outbound task according to first user voice information which is contained in the user voice content and is time-coincident with the customer service voice content. Therefore, the online intention recognition result can be verified through the offline intention recognition result, and the more accurate actual intention recognition result can be obtained. The off-line intention identification result is carried out according to the call recording file, and all call information is contained in the call recording file and is not influenced by objective factors such as network speed, so that the identification result is more accurate. In addition, the method realizes the high-efficiency processing of the outbound task of the big data magnitude through the technologies of distributed locks, blocking queues, token buckets and the like. Moreover, the accuracy of offline recognition can be improved through the offline intention recognition model. In addition, the call recording file contains the user voice content which is sent out when the customer service voice content is not completely broadcasted, so that the user intention can be more accurately identified. For example, some users hang up the phone after indicating no interest or no time when the customer service voice content is not broadcasted, and the voice content of the user cannot be acquired in the online intention recognition process, so that misjudgment is easily caused, and the problem can be better solved by means of an offline intention recognition mode.

Moreover, the intention identification method in the disclosure can also be applied to various service scenes such as a voice quality inspection scene and a user group division scene, which is not limited by the disclosure.

It is understood that the above-mentioned method embodiments of the present disclosure can be combined with each other to form a combined embodiment without departing from the logic of the principle, which is limited by the space, and the detailed description of the present disclosure is omitted. Those skilled in the art will appreciate that in the above methods of the specific embodiments, the specific order of execution of the steps should be determined by their function and possibly their inherent logic.

In addition, the present disclosure also provides an intention identification device based on intelligent customer service, an electronic device, and a computer-readable storage medium, which can be used to implement any intention identification method based on intelligent customer service provided by the present disclosure, and the corresponding technical solutions and descriptions and corresponding descriptions of the method sections are not repeated.

Fig. 6 is a block diagram of an intention identifying apparatus based on intelligent customer service according to an embodiment of the present disclosure.

Referring to fig. 6, an embodiment of the present disclosure provides an intelligent customer service-based intention recognition apparatus, including:

the recording acquisition module 61 is suitable for acquiring a call recording file corresponding to the target outbound task; the target outbound task is completed by the intelligent customer service, and an online intention identification result of the target outbound task is identified and obtained in the process of the target outbound task;

an extracting module 62, adapted to respectively extract customer service voice content corresponding to a first channel and user voice content corresponding to a second channel from the call recording file;

an offline intention recognition module 63 adapted to determine an offline intention recognition result of the target outbound task according to the first user voice information; the first user voice information is user voice information which is contained in the user voice content and is time-coincident with the customer service voice content;

and the actual intention recognition module 64 is suitable for determining an actual intention recognition result of the target outbound task according to the offline intention recognition result and the online intention recognition result.

In an optional implementation manner, the intelligent customer service is an intelligent robot which does not support an intelligent interrupt function;

wherein the online intent recognition result is determined by:

after the fact that the customer service voice information of the intelligent customer service is broadcasted through the first sound channel is detected to be finished, second user voice information transmitted through a second sound channel is obtained; and determining an online intention recognition result according to the acquired second user voice information.

In an alternative implementation, the actual intent recognition module is specifically adapted to:

if the offline intention recognition result is inconsistent with the online intention recognition result, determining the offline intention recognition result as an actual intention recognition result of the target outbound task under the condition that the online intention recognition result belongs to an intention type allowed to be updated;

wherein the permission updating intention type is determined according to the offline identification configuration instruction received in advance.

In an optional implementation manner, the recording obtaining module is further adapted to:

acquiring task parameters contained in a received offline identification configuration instruction;

the outbound task matched with the task parameter and contained in the outbound history record is screened as the target outbound task;

wherein the task parameters include at least one of: the call-out task comprises a first parameter used for indicating the online intention type of the online intention recognition result of the call-out task, a second parameter used for indicating the call time of the call-out task and a third parameter used for indicating the task type of the call-out task.

In an alternative implementation, the offline intent recognition module is specifically adapted to:

determining an offline intent recognition model that matches parameter values of the task parameters; the method comprises the following steps of training in advance to obtain a plurality of off-line intention recognition models respectively matched with different parameter values of task parameters;

and performing intention recognition on the first user voice information through the determined offline intention recognition model to obtain an offline intention recognition result of the target outbound task.

In an optional implementation manner, the target outbound task is multiple, and the recording obtaining module is further adapted to:

controlling one of a plurality of task execution examples through a distributed lock, acquiring each target outbound task from the outbound history record, and sequentially storing the acquired target outbound tasks to a first blocking queue included in a task queue according to the sequence of the acquisition time of each target outbound task, wherein the first blocking queue is provided with task waiting time;

and acquiring a call recording file corresponding to the target outbound task by a plurality of task execution instances in a parallel mode according to the first blocking queue.

In an optional implementation manner, a plurality of target outbound tasks form a task sequence, and the plurality of target outbound tasks are stored in the task queue as a whole by the task sequence;

the task queue further comprises a second execution queue; the recording acquisition module is specifically adapted to:

according to the sequence of each task sequence stored in the first blocking queue, one task sequence is sequentially obtained from the first blocking queue to serve as a current task sequence, and each outbound task contained in the obtained current task sequence is stored in the second execution queue;

sequentially acquiring call recording files corresponding to all outbound tasks contained in the current task sequence according to the second execution queue;

deleting the current task sequence from the second execution queue and taking the next task sequence in the first blocking queue as a new current task sequence every time each outbound task in the current task sequence stored in the second execution queue is processed; and when the system is abnormal, recovering according to the current task sequence in the second execution queue.

Referring to fig. 7, an embodiment of the present disclosure provides an electronic device, including: at least one processor 501; at least one memory 502, and one or more I/O interfaces 503 coupled between the processor 501 and the memory 502; the memory 502 stores one or more computer programs that can be executed by the at least one processor 501, and the one or more computer programs are executed by the at least one processor 501 to enable the at least one processor 501 to perform the above-described intention identification method.

Embodiments of the present disclosure also provide a computer-readable storage medium having stored thereon a computer program, wherein the computer program, when executed by a processor/processing core, implements the above-described intent recognition method. The computer readable storage medium may be a volatile or non-volatile computer readable storage medium.

The disclosed embodiments also provide a computer program product comprising computer readable code or a non-transitory computer readable storage medium carrying computer readable code, which when run in a processor of an electronic device, the processor in the electronic device performs the above-mentioned intent recognition method.

It will be understood by those of ordinary skill in the art that all or some of the steps of the methods, systems, functional modules/units in the devices disclosed above may be implemented as software, firmware, hardware, or suitable combinations thereof. In a hardware implementation, the division between functional modules/units mentioned in the above description does not necessarily correspond to the division of physical components; for example, one physical component may have multiple functions, or one function or step may be performed by several physical components in cooperation. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on computer readable storage media, which may include computer storage media (or non-transitory media) and communication media (or transitory media).

The term computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable program instructions, data structures, program modules or other data, as is well known to those of ordinary skill in the art. Computer storage media includes, but is not limited to, Random Access Memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), Static Random Access Memory (SRAM), flash memory or other memory technology, portable compact disc read-only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by a computer. In addition, communication media typically embodies computer readable program instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media as is well known to those skilled in the art.

The computer-readable program instructions described herein may be downloaded from a computer-readable storage medium to a respective computing/processing device, or to an external computer or external storage device via a network, such as the internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers. The network adapter card or network interface in each computing/processing device receives computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in the respective computing/processing device.

The computer program instructions for carrying out operations of the present disclosure may be assembler instructions, Instruction Set Architecture (ISA) instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider). In some embodiments, the electronic circuitry that can execute the computer-readable program instructions implements aspects of the present disclosure by utilizing the state information of the computer-readable program instructions to personalize the electronic circuitry, such as a programmable logic circuit, a Field Programmable Gate Array (FPGA), or a Programmable Logic Array (PLA).

The computer program product described herein may be embodied in hardware, software, or a combination thereof. In an alternative embodiment, the computer program product is embodied in a computer storage medium, and in another alternative embodiment, the computer program product is embodied in a Software product, such as a Software Development Kit (SDK), or the like.

Various aspects of the present disclosure are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the disclosure. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer-readable program instructions.

These computer-readable program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer-readable program instructions may also be stored in a computer-readable storage medium that can direct a computer, programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer-readable medium storing the instructions comprises an article of manufacture including instructions which implement the function/act specified in the flowchart and/or block diagram block or blocks.

The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer, other programmable apparatus or other devices implement the functions/acts specified in the flowchart and/or block diagram block or blocks.

The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

Example embodiments have been disclosed herein, and although specific terms are employed, they are used and should be interpreted in a generic and descriptive sense only and not for purposes of limitation. In some instances, features, characteristics and/or elements described in connection with a particular embodiment may be used alone or in combination with features, characteristics and/or elements described in connection with other embodiments, unless expressly stated otherwise, as would be apparent to one skilled in the art. Accordingly, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the scope of the disclosure as set forth in the appended claims.

Claims

1. An intent recognition method, comprising:

acquiring a call recording file corresponding to a target outbound task; the target outbound task is completed by an intelligent customer service, and an online intention identification result of the target outbound task is identified and obtained in the process of the target outbound task;

2. The intent recognition method of claim 1, wherein the intelligent customer service is an intelligent robot that does not support intelligent interrupt functionality;

wherein the online intent recognition result is determined by:

after detecting that the broadcasting of the customer service voice information of the intelligent customer service through the first sound channel is finished, acquiring second user voice information transmitted through the second sound channel; and determining an online intention recognition result according to the acquired second user voice information.

3. The intent recognition method according to claim 1 or 2, wherein the determining an actual intent recognition result of the target outbound task according to the offline intent recognition result and the online intent recognition result comprises:

if the offline intention recognition result is inconsistent with the online intention recognition result, determining the offline intention recognition result as the actual intention recognition result of the target outbound task under the condition that the online intention recognition result belongs to the type of the intention allowed to be updated;

wherein the permission updating intention type is determined according to a pre-received offline recognition configuration instruction.

4. The method for identifying an intention according to claim 1, wherein before the obtaining of the call record file corresponding to the target outbound task, the method further comprises:

screening the outbound task matched with the task parameter and contained in the outbound history record as the target outbound task;

wherein the task parameters include at least one of: the first parameter is used for indicating the online intention type of the online intention recognition result of the outbound task, the second parameter is used for indicating the calling time of the outbound task, and the third parameter is used for indicating the task type of the outbound task.

5. The intent recognition method according to claim 4, wherein the determining an offline intent recognition result of the target outbound task from the first user speech information comprises:

6. The intention identifying method according to any one of claims 4 to 5, wherein the number of the target outbound tasks is plural, and before the acquiring the call record file corresponding to the target outbound task, the method further comprises:

the acquiring of the call recording file corresponding to the target outbound task includes: and a plurality of task execution instances acquire a call recording file corresponding to the target outbound task according to the first blocking queue in a parallel mode.

7. The intention recognition method of claim 6, wherein a plurality of target outbound tasks constitute a task sequence, the plurality of target outbound tasks being stored in the task sequence as a whole to the task queue;

the task queue further comprises a second execution queue; the obtaining of the call recording file corresponding to the target outbound task according to the first blocking queue includes:

when each outbound task in the current task sequence stored in the second execution queue is processed, deleting the current task sequence from the second execution queue, and taking the next task sequence in the first blocking queue as a new current task sequence; and when the system is abnormal, recovering according to the current task sequence in the second execution queue.

8. An intention recognition device based on intelligent customer service is characterized by comprising:

the offline intention recognition module is suitable for determining an offline intention recognition result of the target outbound task according to the first user voice information; the first user voice information is user voice information contained in the user voice content and time-coincident with the customer service voice content;

9. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,

the memory stores one or more computer programs executable by the at least one processor, the one or more computer programs being executable by the at least one processor to enable the at least one processor to perform the intent recognition method of any of claims 1-7.

10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the intent recognition method according to any of claims 1-7.