CN113034158A

CN113034158A - Machine responsibility judgment method and device, electronic equipment and readable storage medium

Info

Publication number: CN113034158A
Application number: CN202110298092.6A
Authority: CN
Inventors: 黎铨祺
Original assignee: Beijing Didi Infinity Technology and Development Co Ltd
Current assignee: Beijing Didi Infinity Technology and Development Co Ltd
Priority date: 2021-03-19
Filing date: 2021-03-19
Publication date: 2021-06-25

Abstract

In the embodiment of the application, at least one intermediate responsibility judgment result can be determined through at least one intermediate responsibility judgment network and at least one second type of data to be processed in the responsibility judgment data set, and then a target responsibility judgment result is determined at least based on each first type of data to be processed, each intermediate responsibility judgment result and a main responsibility judgment network. In the process, the intermediate responsibility judgment result is determined based on the information interaction attributes of the parties to be judged, so that the target responsibility judgment result is more accurate, and meanwhile, the intermediate responsibility judgment result can represent the responsibility judgment reason corresponding to the responsibility judgment data set, so that the intermediate responsibility judgment result can explain the target responsibility judgment result, and the interpretability of the target responsibility judgment result is improved.

Description

Machine responsibility judgment method and device, electronic equipment and readable storage medium

Technical Field

The present disclosure relates to the field of computer technologies, and in particular, to a method and an apparatus for determining responsibility of a machine, an electronic device, and a readable storage medium.

Background

At present, with the development of internet technology, more and more online service platforms are available, wherein a user can perform order placing operation on the online service platform through a terminal device such as a smart phone, and after receiving an order of the user, the online service platform can provide corresponding services. For example, taking the network car booking service platform as an example, the user may perform order placing operation on the network car booking service platform through terminal equipment such as a smart phone, and after receiving the order of the user, the network car booking service platform may provide network car booking service.

However, as the number of orders of the online service platform increases, a large number of problem orders (such as complaints from users to the online service platform or service providers) inevitably occur, and if a problem order occurs, the online service platform needs to determine the responsible party of the problem order, i.e. to make a responsibility for the problem order.

In the related art, the reason for the reason can be determined by machine learning, but this method is not interpretable, and when any party related to the problem order does not approve the reason for the reason, more time and labor are required to explain the reason for the reason, thereby reducing the efficiency of the reason for the reason.

Disclosure of Invention

In view of this, embodiments of the present disclosure provide a machine responsibility determination method, an apparatus, an electronic device, and a readable storage medium, which can make a target responsibility determination result more accurate and improve interpretability of the target responsibility determination result.

In a first aspect, a machine accountability method is provided, where the method is applied to an electronic device, and the method includes:

acquiring a liability judgment data set, wherein the liability judgment data set comprises a plurality of predetermined categories of data to be processed, the data to be processed comprises at least one first category of data to be processed and at least one second category of data to be processed, the first category of data to be processed is used for representing objective attributes, and the second category of data to be processed is used for representing information interaction attributes of each party to be judged;

determining at least one intermediate responsibility judgment result based on at least one intermediate responsibility judgment network and the second type of data to be processed, wherein the intermediate responsibility judgment result is used for representing a responsibility judgment reason corresponding to the responsibility judgment data set; and

and determining a target responsibility judgment result at least based on each first type of data to be processed, each intermediate responsibility judgment result and the master responsibility judgment network.

In a second aspect, a machine accountability apparatus is provided, which is applied to an electronic device, and includes:

the system comprises a responsibility judgment data set module, a responsibility judgment data set module and a responsibility judgment module, wherein the responsibility judgment data set comprises a plurality of predetermined categories of data to be processed, the data to be processed comprises at least one first category of data to be processed and at least one second category of data to be processed, the first category of data to be processed is used for representing objective attributes, and the second category of data to be processed is used for representing information interaction attributes of each party to be judged;

the intermediate responsibility judgment result module is used for determining at least one intermediate responsibility judgment result based on at least one intermediate responsibility judgment network and the second type of data to be processed, wherein the intermediate responsibility judgment result is used for representing a responsibility judgment reason corresponding to the responsibility judgment data set; and

and the target responsibility judgment result module is used for determining the target responsibility judgment result at least based on each first type of data to be processed, each intermediate responsibility judgment result and the master responsibility judgment network.

In a third aspect, an embodiment of the present application provides an electronic device, including a memory and a processor, where the memory is used to store one or more computer program instructions, where the one or more computer program instructions are executed by the processor to implement the method according to the first aspect.

In a fourth aspect, embodiments of the present application provide a computer-readable storage medium on which computer program instructions are stored, which when executed by a processor implement the method according to the first aspect.

In a fifth aspect, embodiments of the present application provide a computer program product comprising computer programs/instructions which, when executed by a processor, implement the method according to the first aspect.

In the embodiment of the application, at least one intermediate responsibility determining result may be determined by at least one intermediate responsibility determining network and at least one second type of data to be processed in the responsibility determining data set, and then the target responsibility determining result is determined based on at least each first type of data to be processed, each intermediate responsibility determining result, and the master responsibility determining network. In the process, the intermediate responsibility judgment result is determined based on the information interaction attributes of the parties to be judged, so that the target responsibility judgment result is more accurate, and meanwhile, the intermediate responsibility judgment result can represent the responsibility judgment reason corresponding to the responsibility judgment data set, so that the intermediate responsibility judgment result can explain the target responsibility judgment result, and the interpretability of the target responsibility judgment result is improved.

Drawings

The foregoing and other objects, features and advantages of the embodiments of the present application will be apparent from the following description of the embodiments of the present application with reference to the accompanying drawings in which:

FIG. 1 is a flowchart illustrating a method for determining responsibility of a machine according to an embodiment of the present disclosure;

FIG. 2 is a flowchart illustrating another method for determining machine responsibilities according to embodiments of the present disclosure;

FIG. 3 is a flowchart illustrating a process of determining and outputting an intermediate discriminant result according to an embodiment of the present disclosure;

FIG. 4 is a flowchart illustrating a method for determining an intermediate liability result according to an embodiment of the present disclosure;

FIG. 5 is a flowchart illustrating a method for determining a target liability result according to an embodiment of the present disclosure;

FIG. 6 is a flowchart illustrating another method for determining a target liability result according to an embodiment of the present disclosure;

FIG. 7 is a schematic diagram illustrating a machine responsibility determination device according to an embodiment of the present disclosure;

fig. 8 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Detailed Description

The present application is described below based on examples, but the present application is not limited to only these examples. In the following detailed description of the present application, certain specific details are set forth in detail. It will be apparent to one skilled in the art that the present application may be practiced without these specific details. Well-known methods, procedures, components and circuits have not been described in detail so as not to obscure the present application.

Further, those of ordinary skill in the art will appreciate that the drawings provided herein are for illustrative purposes and are not necessarily drawn to scale.

Unless the context clearly requires otherwise, throughout the description, the words "comprise", "comprising", and the like are to be construed in an inclusive sense as opposed to an exclusive or exhaustive sense; that is, what is meant is "including, but not limited to".

In the description of the present application, it is to be understood that the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance. In addition, in the description of the present application, "a plurality" means two or more unless otherwise specified.

In order to solve the above problem, embodiments of the present application provide a machine accountability method, which can determine at least one intermediate accountability result in determining a target accountability result, thereby increasing interpretability of the target accountability result.

In the embodiment of the present application, if a problem order occurs, the online service platform needs to perform machine responsibility determination for each responsible party to be determined corresponding to the problem order, where the number of responsible parties to be determined is a natural number greater than or equal to 2. For example, taking a network car appointment scenario as an example, each responsible party may be a network car appointment driver (service provider) and a passenger (service provider). As another example, taking a take-out scenario as an example, each responsible party may be a take-out rider (one of the service providers), a merchant (one of the service providers), and a customer (a service-provided provider).

The machine responsibility determination method can be applied to an electronic device, the electronic device can be a terminal or a server, the terminal can be a smartphone, a tablet Computer, a Personal Computer (PC), or the like, and the server can be a single server, a server cluster configured in a distributed manner, or a cloud server.

Specifically, the machine responsibility determination method may be as shown in fig. 1, where fig. 1 is a flowchart of a machine responsibility determination method according to an embodiment of the present disclosure.

Wherein, in the process of making accountability, the accountability data set 11 can be acquired. The liability data set 11 may include at least two types of data to be processed, and the first type of data to be processed may be used to represent objective attributes, which may be used to represent the fact that there is objectivity, such as the time, place, weather condition, and the like of a dispute event or a problem order. The second type of data to be processed may be used to characterize the information interaction attributes of the parties to be blamed, such as the service provider or evidence that the service is uploaded by the provider (chat records or live video recordings, etc.).

It should be noted that the second type of data to be processed is data normally acquired through a legal approach. For example, in a network car booking scenario, the second type of data to be processed may be audio data recorded by a network car booking dedicated recording device installed inside the network car booking, in practical applications, in order to ensure driver and passenger safety, the network car booking service platform may configure a dedicated recording device for each network car booking, before a passenger uses the network car booking service, the passenger may be informed in advance that the recording device is installed in a passenger car and is opened in the whole course, and after the passenger receives the recording device, the recording device is opened in the whole course in the service process.

After acquiring the accountability data set 11, the accountability data set 11 may be input into a pre-trained accountability model 12.

The first type of data to be processed may be input into the master responsibilities network 122, and the second type of data to be processed may be input into the intermediate responsibilities network 121. When the intermediate responsibility network 121 receives the second type of data to be processed, the intermediate responsibility result 131 may be output, and the intermediate responsibility result 131 is input into the master responsibility network 122.

After the primary accountability network 122 receives the first type of data to be processed and the intermediate accountability result 131, the target accountability result 132 may be determined based on at least the first type of data to be processed and the intermediate accountability result 131.

A machine responsibility determination method provided in the embodiment of the present application will be described in detail below with reference to a specific embodiment, as shown in fig. 2, the specific steps are as follows:

in step 21, an accountability dataset is acquired.

The responsibility judging data set comprises a plurality of predetermined categories of data to be processed, the data to be processed comprises at least one first category of data to be processed and at least one second category of data to be processed, the first category of data to be processed is used for representing objective attributes, and the second category of data to be processed is used for representing information interaction attributes of each party to be judged and responsible.

In an alternative embodiment, the second type of data to be processed may include voice interaction information of each party to be blamed and/or text interaction information of each party to be blamed.

Taking a network appointment scenario as an example, if the machine accountability method of the embodiment of the present application is applied to order accountability cancellation (i.e. responsibility determination after the passenger/driver actively cancels an order), the first type of data to be processed may include order basic features (e.g. time of order occurrence, appointed boarding place, destination, etc.), passenger features (e.g. passenger's issuing position, distance of passenger's issuing position from appointed boarding place, number of times of initiative cancellation of the passenger in a past period, etc.), and driver features (e.g. driver's taking position, distance of driver's taking position from appointed boarding place, number of times of initiative cancellation of the driver in a past period, etc.).

The second type of data to be processed may include driver and passenger voice information and driver and passenger text information, where the second type of data to be processed is data normally acquired through a legal approach. In the present scenario, the driving voice information may be voice communication information between the driver and the passenger, and after the passenger and the driver authorize, the driving voice information may be used as the second type of data to be processed. The driver and passenger text information can be text communication information between the driver and the passenger, and after the passenger and the driver authorize, the driver and passenger text information can be used as second type data to be processed.

Of course, the accountability data set may also include other types of data to be processed, for example, data for characterizing the action trajectory of each party to be accountable over a period of time, and the like, and these data are also data legally obtained after being authorized.

In an optional implementation manner, after the liability assessment data set is acquired, various types of data in the liability assessment data set may be preprocessed, for example, data cleaning processing, voice recognition processing, word segmentation processing, and the like, and invalid or meaningless data in the liability assessment data set may be filtered out through the preprocessing, so that only valid data is retained, and the liability assessment efficiency is improved.

At step 22, at least one intermediate accountability result is determined based on the at least one intermediate accountability network and the second type of data to be processed.

The intermediate disclaimer result is used to characterize the disclaimer reason corresponding to the disclaimer data set, that is, the intermediate disclaimer result may be intermediate data used to explain the target disclaimer result, and in practical applications, the intermediate disclaimer result may be an explanatory text corresponding to the target disclaimer result or may be other types of explanatory data.

In an alternative implementation manner, the embodiment of the present application may output the intermediate discipline determination result, and then perform operations such as displaying the intermediate discipline determination result, so as to explain the target discipline determination result.

In an alternative embodiment, the information interaction attribute may be interaction information between the parties to be assessed, that is, the second type of data to be processed includes voice interaction information of the parties to be assessed and/or text interaction information of the parties to be assessed.

Step 22 may then be performed as: and inputting the second type of data to be processed into at least one intermediate responsibility-judging network, so that each intermediate responsibility-judging network outputs an intermediate responsibility-judging result according to the voice interaction information of each party to be judged and responsible and/or the text interaction information of each party to be judged and responsible.

In the process of making accountability, the intermediate accountability result can be used as an accountability reason of the accountability result of the corresponding target. In practical applications, each intermediate responsibility judgment network may correspond to a predetermined responsibility judgment task, and each predetermined responsibility judgment task may correspond to a responsibility judgment reason. That is, the intermediate responsibility judgment network corresponds to a predetermined responsibility judgment task, wherein the predetermined responsibility judgment task is used for judging a responsibility judgment reason corresponding to the second type of data to be processed.

Specifically, step 22 may be performed as: and respectively inputting the second type of data to be processed into each intermediate responsibility judgment network, so that each intermediate responsibility judgment network outputs an intermediate responsibility judgment result for representing responsibility judgment reasons based on the predetermined responsibility judgment service.

Specifically, as shown in fig. 3, fig. 3 is a flowchart for determining and outputting an intermediate discriminant result according to an embodiment of the present disclosure.

After determining the second type of to-be-processed data 31 in the responsivity data set, the second type of to-be-processed data 31 may be input into the intermediate responsivity network 321, the intermediate responsivity network 322, and the intermediate responsivity network 323, respectively.

The intermediate responsibility determining network 321, the intermediate responsibility determining network 322 and the intermediate responsibility determining network 323 respectively correspond to different tasks, specifically, the intermediate responsibility determining network 321 corresponds to the predetermined responsibility determining service a, the intermediate responsibility determining network 322 corresponds to the predetermined responsibility determining service B, and the intermediate responsibility determining network 323 corresponds to the predetermined responsibility determining service C.

For example, in the network appointment vehicle scenario, for the predetermined accountability tasks A, B and C, the predetermined accountability task a may be a task of determining whether the driver has an action of inducing the passenger to cancel the order (for example, the driver performs a false description, the driver's driving willingness is not clear), and the intermediate accountability network 321 may determine an intermediate accountability result corresponding to the predetermined accountability task a according to an information interaction attribute (voice communication information, instant communication text, and the like) between the driver and the passenger, where the intermediate accountability result corresponding to the predetermined accountability task a may be "yes" or "no".

The predetermined accountability task B may be to determine whether the driver has an action of adding price and negotiating price (for example, the driver requests the passenger to add price for a reason such as queuing), and the intermediate accountability network 322 may determine an intermediate accountability result corresponding to the predetermined accountability task B according to an information interaction attribute (voice communication information, instant messaging text, etc.) between the driver and the passenger, where the intermediate accountability result corresponding to the predetermined accountability task B may be "yes" or "no".

The task C may be to determine whether the driver meets the exemption condition (for example, the passenger cancels the order due to not using the vehicle, the passenger cancels the order due to sending a wrong order, etc.), and the intermediary accountability network 323 may determine an intermediary accountability result corresponding to the task C according to the information interaction attribute (voice communication information, instant communication text, etc.) between the driver and the passenger, where the intermediary accountability result corresponding to the task C may be "yes" or "no".

In the above case, the intermediate responsibility determination result is "yes" or "no", that is, in this case, the intermediate responsibility determination network may be a network constructed based on a binary classification model, where each intermediate responsibility determination network (i.e., each predetermined responsibility determination task) corresponds to two possible intermediate responsibility determination results, for example, the intermediate responsibility determination network may be a network constructed based on a softmax logistic regression model.

Of course, in another case, the intermediate disclaimer result may be a specific result other than "yes" and "no", such as "false description by the driver", "the driver asks the passenger to add price for queuing reason", or "the passenger cancels the order for taking a wrong order", etc. That is, in this case, the intermediate accountability network may be a network constructed based on other classification models, wherein each intermediate accountability network (i.e., each predetermined accountability task) may correspond to two or more possible results.

According to the embodiment of the application, after the intermediate result of discriminant judgment is output as the reason of discriminant judgment, the reason of discriminant judgment can be used for specifically explaining the target result of discriminant judgment without manually explaining, so that the interpretability of the process of machine discriminant judgment is enhanced.

In this embodiment of the present application, in order to enable the intermediate responsibility determining network to better process the second type of data to be processed, feature extraction and data concatenation may be performed on the second type of data to be processed, specifically, the process may be performed as: and respectively inputting the second merged vectors into each intermediate responsibility-judging network so as to determine an intermediate responsibility-judging result output by each intermediate responsibility-judging network.

Specifically, as shown in fig. 4, fig. 4 is a flowchart for determining an intermediate discriminant result according to an embodiment of the present disclosure.

In fig. 4, after the second type feature data 411 and the second type feature data 412 are obtained, feature extraction may be performed on the second type feature data 411 and the second type feature data 412, specifically, an embedding (embedding) operation may be performed on the second type feature data 411 and the second type feature data 412 to determine a second feature vector 421 and a second feature vector 422.

The embedding is a feature extraction means commonly used in deep learning, specifically, the feature extraction is to map high-dimensional raw data (images, characters, and the like) to a low-dimensional Manifold (manifeld), so that the high-dimensional raw data becomes separable after being mapped to the low-dimensional Manifold, and this mapping process may be called as embedding.

After determining the second eigenvector 421 and the second eigenvector 422, data splicing may be performed on the second eigenvector 421 and the second eigenvector 422, specifically, data splicing may be performed on the second eigenvector 421 and the second eigenvector 422 through the concat function 43, and the second merged vector 44 is determined.

Wherein, the concat function can be used for connecting two or more arrays to realize data splicing.

After the second merged vector 44 is determined, the second merged vector 44 may be input to each of the intermediate responsibilities networks (intermediate responsibilities network 451-intermediate responsibilities network 453), so that each of the intermediate responsibilities networks outputs an intermediate responsivity result, i.e., intermediate responsibilities result 461-intermediate responsibilities result 463.

According to the embodiment of the application, before the intermediate responsibility judgment result is determined through the intermediate responsibility judgment network, the feature extraction and the data splicing can be performed on each second type of data to be processed, so that the intermediate responsibility judgment network can process the second type of data to be processed more efficiently, and the responsibility judgment efficiency of the machine is improved.

In another alternative embodiment, the data to be processed of each second type may be pre-processed first, and then each second feature vector may be determined. Specifically, the process may be performed as: and performing data preprocessing on each second type of data to be processed, determining preprocessed data corresponding to each second type of data to be processed, and performing embedding processing on each preprocessed data to determine a second feature vector corresponding to each preprocessed data.

Wherein the data preprocessing comprises at least one of data cleaning processing, voice recognition processing and word segmentation processing.

Through data preprocessing, the second type of data to be processed can be sorted, so that the processing efficiency of the intermediate responsibility judgment network is higher.

In step 23, a target responsibility judgment result is determined based on at least each first type of data to be processed, each intermediate responsibility judgment result and the master responsibility judgment network.

In an alternative embodiment, the master accountability network may determine the target accountability result based on each first type of data to be processed and each intermediate accountability result, and specifically, the process may be performed as: the method comprises the steps of extracting features of each first type of data to be processed, determining a first feature vector corresponding to each first type of data to be processed, splicing each first feature vector and each intermediate responsibility judgment result, determining a first combined vector, and inputting the first combined vector into a master responsibility judgment network to determine a target responsibility judgment result output by the master responsibility judgment network.

As shown in fig. 5, fig. 5 is a flowchart for determining a target liability determination result according to an embodiment of the present application. The responsibility determination data set comprises first type data to be processed 511, first type data to be processed 512, first type data to be processed 513, second type data to be processed 531 and second type data to be processed 532.

After the disclaimer data set is obtained, feature extraction may be performed on each first type of data to be processed and each second type of data to be processed, so as to determine a first feature vector 521, a first feature vector 522, a first feature vector 523, a second feature vector 541, and a second feature vector 542.

Before feature extraction is performed on each first type of data to be processed and each second type of data to be processed, preprocessing operation can be performed on each first type of data to be processed and each second type of data to be processed, so that the processing efficiency of each network is higher.

After determining each second eigenvector, data splicing may be performed on each second eigenvector through the concat function 55 to determine a second merged vector, and then based on that the second merged vector is input into the intermediate responsivity determining network 561, the intermediate responsivity determining network 562 and the intermediate responsivity determining network 563, each intermediate responsivity determining network outputs an intermediate responsivity determining result, that is, the intermediate responsivity determining result 571, the intermediate responsivity determining result 572 and the intermediate responsivity determining result 573.

After each intermediate discipline result is determined, each intermediate discipline result (the intermediate discipline result 571, the intermediate discipline result 572, and the intermediate discipline result 573) and each first feature vector (the first feature vector 521, the first feature vector 522, and the first feature vector 523) may be input into the concat function 58 for data splicing, a first merged vector is determined, and then the first merged vector is input into the master discipline network 59, so that the master discipline network 59 outputs the target discipline result 510.

The master responsibility network can be a network constructed based on a classification model, for example, the master responsibility network can be a network constructed based on a softmax logistic regression model.

In the embodiment of the application, at least one intermediate responsibility determining result may be determined by at least one intermediate responsibility determining network and at least one second type of data to be processed in the responsibility determining data set, and then the target responsibility determining result is determined based on at least each first type of data to be processed, each intermediate responsibility determining result, and the master responsibility determining network. In the process, the intermediate responsibility judgment result is determined based on the information interaction attributes of the parties to be judged, so that the target responsibility judgment result is more accurate, and meanwhile, the intermediate responsibility judgment result can explain the target responsibility judgment result, so that the interpretability of the target responsibility judgment result is improved.

In another alternative embodiment, the master responsivity network may determine the target responsivity result based on each first type of data to be processed, each intermediate responsivity result, and the second consolidated vector, and specifically, the process may be performed as follows: the method comprises the steps of extracting features of each first type of data to be processed, determining a first feature vector corresponding to each first type of data to be processed, performing data splicing on each first feature vector, each intermediate responsibility judgment result and a second merged vector, determining a first merged vector, and inputting the first merged vector into a main responsibility judgment network so as to determine a target responsibility judgment result output by the main responsibility judgment network.

As shown in fig. 6, fig. 6 is a flowchart of another method for determining a target liability result according to the embodiment of the present application. The responsibility determination data set includes a first type of data to be processed 611, a first type of data to be processed 612, a first type of data to be processed 613, a second type of data to be processed 631, and a second type of data to be processed 632.

After the disclaimer data set is obtained, feature extraction may be performed on each first type of data to be processed and each second type of data to be processed first, so as to determine a first feature vector 621, a first feature vector 622, a first feature vector 623, a second feature vector 641, and a second feature vector 642.

After determining each second eigenvector, data splicing may be performed on each second eigenvector through the concat function 65 to determine a second merged vector, and then based on that the second merged vector is input into the intermediate responsivity network 661, the intermediate responsivity network 662 and the intermediate responsivity network 663, each intermediate responsivity network outputs an intermediate responsivity result, that is, the intermediate responsivity result 671, the intermediate responsivity result 672 and the intermediate responsivity result 673, respectively.

After each intermediate discipline result is determined, each intermediate discipline result (the intermediate discipline result 671, the intermediate discipline result 672 and the intermediate discipline result 673), each first feature vector (the first feature vector 621, the first feature vector 622 and the first feature vector 623) and the second combined vector output by the concat function 65 may be input into the concat function 68 for data splicing, a first combined vector is determined, and then the first combined vector is input into the master discipline network 69, so that the master discipline network 69 outputs the target discipline result 610.

Before the main accountability network and the intermediate accountability network are online, the main accountability network and the intermediate accountability network can be trained.

In an alternative embodiment, the master and intermediate accountability networks may be trained separately based on a training set.

For the master responsibility network, the master responsibility network can be trained based on the following steps: the method comprises the steps of obtaining a training set, determining at least one historical intermediate result according to at least one intermediate responsibility network and second-class historical data aiming at each historical data set, determining a historical result according to at least each first-class historical data, each historical intermediate result and a main responsibility network, and training the main responsibility network according to the historical result and marks corresponding to the historical result.

The training set can comprise a plurality of historical data sets and a plurality of labels corresponding to each historical data set, each historical data set comprises a plurality of preset categories of historical data, each historical data comprises at least one first category of historical data and at least one second category of historical data, the first category of historical data is used for representing objective attributes, the second category of historical data is used for representing information interaction attributes of each party to be responsible, and the labels are used for representing real responsibility judgment results of the historical data sets.

In practical application, accurate real responsibility judgment results (for example, judgment results considered by professional persons) can be used as labels in a training set, so that the main responsibility judgment network can learn based on the accurate real responsibility judgment results, and the training effect of the main responsibility judgment network is improved.

For the intermediate accountability network, the intermediate accountability network can be trained based on the following steps: obtaining a training set, determining at least one historical intermediate result based on at least one intermediate responsibility network and the second type historical data aiming at each historical data set, and training the intermediate responsibility network based on the historical intermediate result and the corresponding label of the historical intermediate result.

The training set comprises a plurality of historical data sets and a plurality of labels corresponding to each historical data set, the historical data sets at least comprise second-class historical data, the second-class historical data are used for representing information interaction attributes of each party to be subjected to liability judgment, and the labels are used for representing real intermediate results of the historical data sets.

In the embodiment of the application, because the main responsibility judging network and the intermediate responsibility judging network are trained independently, when the intermediate responsibility judging network is trained, the intermediate responsibility judging network can be trained only by acquiring the historical intermediate result, so that the efficiency of training the intermediate responsibility judging network is improved.

In another alternative embodiment, the master and intermediate discriminant networks may be jointly trained based on a training set.

Specifically, the master responsivity network and the intermediate responsivity network may perform joint training based on the following steps: the method comprises the steps of obtaining a training set, determining at least one historical intermediate result according to each historical data set and based on at least one intermediate responsibility network and second-class historical data, determining historical results according to at least each first-class historical data, each historical intermediate result and a main responsibility network, and adjusting network parameters of the main responsibility network and the intermediate responsibility network according to a preset joint loss function, labels corresponding to the historical results and the historical results, labels corresponding to the historical intermediate results and labels corresponding to the historical intermediate results.

The training set comprises a plurality of historical data sets and a plurality of labels corresponding to each historical data set, the historical data sets comprise a plurality of preset categories of historical data, the historical data comprises at least one first category of historical data and at least one second category of historical data, the first category of historical data is used for representing objective attributes, the second category of historical data is used for representing information interaction attributes of each party to be subjected to responsibility judgment, and the labels are used for representing real responsibility judgment results and real intermediate results of the historical data sets.

In one case, the joint loss function may be a summation of multiple losses, e.g., the loss L may be determined from the historical results and corresponding annotations for the historical results₁Determining the loss L according to the intermediate results of each history and the labels corresponding to the intermediate results of each history₂Then, based on the joint loss function L ═ L₁+L₂A final loss L is determined, and then network parameters of the master and intermediate responsibilities networks may be adjusted based on the loss L.

In another case, the joint loss function may further include a weight corresponding to each loss, i.e., L ═ w₁L₁+w₂L₂Wherein w is₁And w₂The weight parameters are characterized.

In practical application, the second type of to-be-processed data is used for representing the information interaction attribute of each party to be responsible for judgment, so that the second type of to-be-processed data can more clearly represent the actual situation of a dispute event, and therefore, in the process of joint training, the L loss can be reduced₂The weight of (2) is increased so that each network can process the information interaction attribute more accurately.

Based on the same technical concept, an embodiment of the present application further provides a machine responsibility determination device, as shown in fig. 7, the device includes: an accountability data set module 71, an intermediate accountability result module 72 and a target accountability result module 73.

The accountability data set module 71 is configured to obtain an accountability data set, where the accountability data set includes multiple predetermined categories of to-be-processed data, the to-be-processed data includes at least one first category of to-be-processed data and at least one second category of to-be-processed data, the first category of to-be-processed data is used to represent objective attributes, and the second category of to-be-processed data is used to represent information interaction attributes of each party to be accountable.

And an intermediate responsibility determination result module 72, configured to determine at least one intermediate responsibility determination result based on the at least one intermediate responsibility determination network and the second type of data to be processed, where the intermediate responsibility determination result is used to characterize a responsibility determination reason corresponding to the responsibility determination data set.

And the target responsibility judgment result module 73 is configured to determine a target responsibility judgment result at least based on each first type of data to be processed, each intermediate responsibility judgment result, and the master responsibility judgment network.

Optionally, the second type of data to be processed includes voice interaction information of each party to be assessed and/or text interaction information of each party to be assessed and assessed;

the intermediate disclaimer result module 72 is specifically configured to:

and inputting the second type of data to be processed into at least one intermediate responsibility-judging network, so that each intermediate responsibility-judging network outputs an intermediate responsibility-judging result according to the voice interaction information of each party to be judged and responsible and/or the text interaction information of each party to be judged and responsible.

Optionally, each intermediate responsibility judgment network corresponds to a predetermined responsibility judgment task;

the intermediate disclaimer result module 72 is specifically configured to:

and inputting the second type of data to be processed into each intermediate responsibility judgment network respectively so as to obtain an intermediate responsibility judgment result corresponding to each predetermined responsibility judgment service respectively.

Optionally, the intermediate responsibility determination network is constructed based on a two-classification model.

Optionally, the intermediate disclaimer result module 72 is specifically configured to:

and performing feature extraction on each second type of data to be processed, and determining a second feature vector corresponding to each second type of data to be processed.

And performing data splicing on the second feature vectors to determine a second merging vector.

And respectively inputting the second merged vector into each intermediate responsibility judgment network so as to determine an intermediate responsibility judgment result output by each intermediate responsibility judgment network.

Optionally, the target liability assessment result module 73 is specifically configured to:

and performing feature extraction on each first type of data to be processed, and determining a first feature vector corresponding to each first type of data to be processed.

And performing data splicing on each first feature vector, each intermediate discriminant result and the second merged vector to determine a first merged vector.

And inputting the first merged vector into a master responsibility network so as to determine a target responsibility judgment result output by the master responsibility network.

And performing data splicing on each first feature vector and each intermediate discriminant result to determine a first merged vector.

Optionally, the master accountability network and the intermediate accountability network are trained separately based on a training set.

Optionally, the master responsibility network is trained based on the following modules:

the system comprises a first obtaining module, a training set and a second obtaining module, wherein the training set comprises a plurality of historical data sets and a plurality of labels corresponding to each historical data set, the historical data sets comprise a plurality of preset categories of historical data, the historical data comprises at least one first category of historical data and at least one second category of historical data, the first category of historical data is used for representing objective attributes, the second category of historical data is used for representing information interaction attributes of parties to be responsible, and the labels are used for representing real responsibility judgment results of the historical data sets.

And the first historical intermediate result module is used for determining at least one historical intermediate result based on the at least one intermediate accountability network and the second type of historical data aiming at each historical data set.

And the first history result module is used for determining a history result at least based on each first type of history data, each history intermediate result and the main responsibility network.

And the first training module is used for training the main accountability network based on the historical result and the label corresponding to the historical result.

Optionally, the intermediate responsibility determination network is trained based on the following modules:

the second obtaining module is configured to obtain a training set, where the training set includes a plurality of historical data sets and a plurality of labels corresponding to each historical data set, the historical data sets at least include a second type of historical data, the second type of historical data is used to represent information interaction attributes of each party to be responsible, and the labels are used to represent real intermediate results of the historical data sets.

And the second historical intermediate result module is used for determining at least one historical intermediate result according to each historical data set and based on the at least one intermediate accountability network and the second type of historical data.

And the second training module is used for training the intermediate responsibility network based on the historical intermediate result and the label corresponding to the historical intermediate result.

Optionally, the master responsivity network and the intermediate responsivity network are jointly trained based on a training set.

Optionally, the master responsibility network and the intermediate responsibility network perform joint training based on the following modules:

the third obtaining module is configured to obtain a training set, where the training set includes a plurality of historical data sets and a plurality of labels corresponding to each historical data set, the historical data set includes a plurality of predetermined categories of historical data, the historical data includes at least one first category of historical data and at least one second category of historical data, the first category of historical data is used to represent objective attributes, the second category of historical data is used to represent information interaction attributes of each party to be responsible for judging, and the labels are used to represent real responsibility judgment results and real intermediate results of the historical data sets.

And the third history intermediate result module is used for determining at least one history intermediate result according to each history data set and based on the at least one intermediate accountability network and the second type history data.

And the second history result module is used for determining a history result at least based on each first-class history data, each history intermediate result and the main responsibility network.

And the third training module is used for adjusting network parameters of the main responsibility judgment network and the intermediate responsibility judgment network based on a preset joint loss function, the historical results, labels corresponding to the historical results, the historical intermediate results and labels corresponding to the historical intermediate results.

Fig. 8 is a schematic diagram of an electronic device according to an embodiment of the present application. As shown in fig. 8, the electronic device shown in fig. 8 is a general address query device, which includes a general computer hardware structure, which includes at least a processor 81 and a memory 82. The processor 81 and the memory 82 are connected by a bus 83. The memory 82 is adapted to store instructions or programs executable by the processor 81. Processor 81 may be a stand-alone microprocessor or a collection of one or more microprocessors. Thus, the processor 81 implements the processing of data and the control of other devices by executing instructions stored in the memory 82 to perform the method flows of the embodiments of the present application as described above. The bus 83 connects the above components together, and also connects the above components to a display controller 84 and a display device and an input/output (I/O) device 85. Input/output (I/O) devices 85 may be a mouse, keyboard, modem, network interface, touch input device, motion sensing input device, printer, and other devices known in the art. Typically, the input/output devices 85 are coupled to the system through an input/output (I/O) controller 86.

As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, apparatus (device) or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may employ a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations of methods, apparatus (devices) and computer program products according to embodiments of the application. It will be understood that each flow in the flow diagrams can be implemented by computer program instructions.

These computer program instructions may be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows.

These computer program instructions may also be provided to a processor of a general purpose computer, special purpose computer, embedded processor, 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 specified in the flowchart flow or flows.

Another embodiment of the present application is directed to a non-transitory storage medium storing a computer-readable program for causing a computer to perform some or all of the above-described method embodiments.

That is, as can be understood by those skilled in the art, all or part of the steps in the method for implementing the embodiments described above may be accomplished by specifying the relevant hardware through a program, where the program is stored in a storage medium and includes several instructions to enable a device (which may be a single chip, a chip, or the like) or a processor (processor) to execute all or part of the steps of the method described in the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.

Another embodiment of the present application relates to a computer program product comprising computer programs/instructions which, when executed by a processor, may implement some or all of the above-described method embodiments.

That is, as can be understood by those skilled in the art, the embodiments of the present application may specify related hardware (including the processor itself) by the processor executing the computer program product (computer program/instruction), so as to implement all or part of the steps in the method of the above embodiments.

The above description is only a preferred embodiment of the present application and is not intended to limit the present application, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

The embodiment of the application discloses a TS1 and a machine responsibility judgment method, wherein the method comprises the following steps:

The TS2, where the method in TS1, includes voice interaction information of each party to be blamed and/or text interaction information of each party to be blamed;

the determining at least one intermediate responsibility judgment result based on the at least one intermediate responsibility judgment network and the second type of data to be processed comprises:

TS3, the method of TS1 or TS2, wherein each of the intermediate accountability networks corresponds to a predetermined accountability task;

TS4, the method of TS1, wherein the intermediate liability network is constructed based on a two-classification model.

The TS5, the method as set forth in TS1, wherein the determining at least one intermediate accountability result based on the at least one intermediate accountability network and the second type of data to be processed includes:

extracting the features of the second type of data to be processed, and determining a second feature vector corresponding to the second type of data to be processed;

performing data splicing on each second feature vector to determine a second merged vector; and inputting the second merged vector into each intermediate responsibility judgment network respectively so as to determine an intermediate responsibility judgment result output by each intermediate responsibility judgment network.

The TS6, the method as in TS5, wherein the determining the target responsibility result based on at least each first type of data to be processed, each intermediate responsibility result, and the master responsibility network includes:

extracting the characteristics of each first type of data to be processed, and determining a first characteristic vector corresponding to each first type of data to be processed;

performing data splicing on each first feature vector, each intermediate discriminant result and the second merged vector to determine a first merged vector; and

The TS7, the method as in TS1, wherein the determining the target responsibility result based on at least each first type of data to be processed, each intermediate responsibility result, and the master responsibility network includes:

performing data splicing on each first feature vector and each intermediate discriminant result to determine a first merged vector; and

TS8, the method of TS1, wherein the master and intermediate accountability networks are trained separately based on a training set.

TS9, the method of TS8, wherein the master liable network is trained based on the steps of:

acquiring a training set, wherein the training set comprises a plurality of historical data sets and a plurality of labels corresponding to each historical data set, the historical data sets comprise a plurality of preset categories of historical data, the historical data comprises at least one first category of historical data and at least one second category of historical data, the first category of historical data is used for representing objective attributes, the second category of historical data is used for representing information interaction attributes of each party to be accountable, and the labels are used for representing real accountability results of the historical data sets;

for each historical data set, determining at least one historical intermediate result based on at least one intermediate accountability network and the second type historical data;

determining a historical result at least based on each first-class historical data, each historical intermediate result and the main responsibility network; and

and training the main responsibility network based on the historical result and the label corresponding to the historical result.

TS10, the method of TS8, wherein the intermediate disclaimer network is trained based on the steps of:

acquiring a training set, wherein the training set comprises a plurality of historical data sets and a plurality of labels corresponding to each historical data set, the historical data sets at least comprise second-class historical data, the second-class historical data is used for representing information interaction attributes of each party to be assessed and blamed, and the labels are used for representing real intermediate results of the historical data sets;

for each historical data set, determining at least one historical intermediate result based on at least one intermediate accountability network and the second type historical data; and

and training the intermediate liability assessment network based on the historical intermediate result and the label corresponding to the historical intermediate result.

The TS11, the method as recited in TS1, wherein the master and intermediate accountability networks are jointly trained based on a training set.

The TS12, the method as in TS11, wherein the master and intermediate disclaimers networks are jointly trained based on:

acquiring a training set, wherein the training set comprises a plurality of historical data sets and a plurality of labels corresponding to each historical data set, the historical data sets comprise a plurality of preset categories of historical data, the historical data comprises at least one first category of historical data and at least one second category of historical data, the first category of historical data is used for representing objective attributes, the second category of historical data is used for representing information interaction attributes of each party to be responsible, and the labels are used for representing real responsibility judgment results and real intermediate results of the historical data sets;

and adjusting network parameters of the main responsibility judgment network and the intermediate responsibility judgment network based on a preset joint loss function, the historical result, the label corresponding to the historical result, each historical intermediate result and the label corresponding to each historical intermediate result.

TS13, a machine accountability apparatus, wherein the apparatus comprises:

The TS14, the apparatus according to TS13, wherein the second type of data to be processed includes voice interaction information of each party to be blamed and/or text interaction information of each party to be blamed;

the intermediate responsibility determination result module is specifically configured to:

TS15, the apparatus of TS13 or TS14, wherein each of the intermediate responsibilities networks corresponds to a predetermined responsivity task;

TS16, the apparatus of TS13, wherein the intermediate liability network is constructed based on a two-classification model.

The TS17, the apparatus of TS13, wherein the intermediate disclaimer result module is specifically configured to:

The TS18, the apparatus of TS17, wherein the target liability determination result module is specifically configured to:

The TS19, the apparatus of TS13, wherein the target liability determination result module is specifically configured to:

The TS20, the apparatus as recited in TS13, wherein the master and intermediate accountability networks are trained separately based on a training set.

TS21, the apparatus of TS20, wherein the master liability network is trained based on the following modules:

the system comprises a first acquisition module, a second acquisition module and a third acquisition module, wherein the training set comprises a plurality of historical data sets and a plurality of labels corresponding to each historical data set, the historical data sets comprise a plurality of preset categories of historical data, the historical data comprises at least one first category of historical data and at least one second category of historical data, the first category of historical data is used for representing objective attributes, the second category of historical data is used for representing information interaction attributes of parties to be responsible, and the labels are used for representing real responsibility judgment results of the historical data sets;

the first historical intermediate result module is used for determining at least one historical intermediate result according to each historical data set and based on at least one intermediate accountability network and the second type of historical data;

the first history result module is used for determining a history result at least based on each first type of history data, each history intermediate result and the main responsibility network; and

TS22, the apparatus of TS20, wherein the intermediate disclaimer network is trained based on the following modules:

the second obtaining module is used for obtaining a training set, wherein the training set comprises a plurality of historical data sets and a plurality of labels corresponding to each historical data set, the historical data sets at least comprise second-class historical data, the second-class historical data are used for representing information interaction attributes of each party to be responsible, and the labels are used for representing real intermediate results of the historical data sets;

the second historical intermediate result module is used for determining at least one historical intermediate result according to each historical data set and based on at least one intermediate responsibility judgment network and the second type of historical data; and

The apparatus of TS23, such as TS13, wherein the master and intermediate accountability networks are jointly trained based on a training set.

The TS24, the apparatus as in TS23, wherein the master and intermediate responsibilities networks are jointly trained based on:

a third obtaining module, configured to obtain a training set, where the training set includes multiple historical data sets and multiple labels corresponding to each historical data set, the historical data set includes multiple predetermined categories of historical data, the historical data includes at least one first category of historical data and at least one second category of historical data, the first category of historical data is used to represent objective attributes, the second category of historical data is used to represent information interaction attributes of each party to be responsible for judging, and the labels are used to represent real responsibility judgment results and real intermediate results of the historical data sets;

the third history intermediate result module is used for determining at least one history intermediate result according to each history data set based on at least one intermediate responsibility judgment network and the second type history data;

the second historical result module is used for determining a historical result at least based on each first-class historical data, each historical intermediate result and the main responsibility judgment network; and

TS25, an electronic device comprising a memory and a processor, wherein the memory is configured to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement a method as set forth in any one of TS1-TS 12.

TS26, a computer readable storage medium, wherein the computer readable storage medium has stored therein a computer program which, when executed by a processor, implements the method of any one of TS1-TS 12.

TS27, a computer program product comprising computer programs/instructions, wherein the computer programs/instructions, when executed by a processor, implement the method of any one of TS1-TS 12.

Claims

1. A method for machine accountability, the method comprising:

2. The method according to claim 1, wherein the second type of data to be processed includes voice interaction information of each party to be blamed and/or text interaction information of each party to be blamed;

3. The method according to claim 1 or 2, wherein each of the intermediate accountability networks corresponds to a predetermined accountability task;

4. The method of claim 1, wherein the intermediate accountability network is constructed based on a binary classification model.

5. The method of claim 1, wherein the master and intermediate accountability networks are trained separately based on a training set.

6. The method of claim 1, wherein the master and intermediate accountability networks are jointly trained based on a training set.

7. A machine accountability apparatus, the apparatus comprising:

8. An electronic device comprising a memory and a processor, wherein the memory is configured to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method of any of claims 1-6.

9. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the method of any one of claims 1 to 6.

10. A computer program product comprising computer programs/instructions, characterized in that the computer programs/instructions, when executed by a processor, implement the method of any of claims 1-6.