A task relationship determination method, apparatus and device

Abstract

A task relationship determination method, device and equipment are used to determine the association relationship between tasks that can be automatically determined. In the present application, the relationship discovery device can first determine the sample attributes in each task and the corresponding attribute values. Then, according to the sample attributes and the corresponding attribute values of each task, the task attributes of each task and the corresponding attribute values are determined. After determining each task attribute and the corresponding attribute value, the hybrid attributes of each task are determined based on the sample attributes and / or the task attributes of each task. Then, the associated tasks, i.e. the association relationship between tasks, are determined based on the task hybrid attributes of each task. In the whole process, it does not depend on relevant personnel and can efficiently determine the associated tasks.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of communication, and in particular to a task relationship determination method, device and equipment. BACKGROUND

[0002] Machine learning is the core of artificial intelligence, which mainly simulates human learning behavior by means of a computer. In essence, machine learning is to construct a target function based on the error of the prediction result of the training sample as small as possible. In the process of machine learning, a sufficient training sample is needed to fit a target function to ensure that the target function can completely show the statistical distribution of the data.

[0003] However, the machine learning is currently only used to construct a target function of a single scene (which can also be understood as a single task). However, in actual application, the target function constructed by machine learning needs to predict the data of multiple scenes, that is, the target function needs to conform to the statistical distribution of the data of multiple scenes, which will lead to the deterioration of the accuracy of the target function.

[0004] Therefore, multi-task transfer learning is proposed. Through multi-task transfer learning, the target function learned from one task can be transferred to another task. A key point of multi-task transfer learning is to determine the association relationship between the multiple tasks, based on which the transfer relationship between the tasks is described to realize the transfer of the target function.

[0005] Currently, the association relationship between multiple tasks usually depends on the suggestion of an expert in the field to which the task belongs. The expert determines the association relationship between multiple tasks based on his own knowledge, which usually has poor credibility and cannot guarantee the accuracy of multi-task transfer learning. SUMMARY

[0006] The present application provides a task relationship determination method, device and equipment for automatically determining the association relationship between tasks.

[0007] In a first aspect, the present application provides a task relationship determination method, which can be executed by a relationship discovery device. The relationship discovery device can be a hardware device or a software program running on a hardware device. In the method, the relationship discovery device can first determine the sample attributes in each task and the corresponding attribute values. Then, according to the sample attributes and the corresponding attribute values of each task, the task attributes and the corresponding attribute values of each task are determined. After determining each task attribute and the corresponding attribute value, the following three ways can be used to determine the task mixed attribute of each task, wherein the task mixed attribute is used to describe the overall characteristics of the task:

[0008] In a first mode, the relationship discovery apparatus determines the task mixed attribute and the corresponding attribute value of each task according to the task attribute and the corresponding attribute value of each task.

[0009] In a second mode, the relationship discovery apparatus determines the task mixed attribute and the corresponding attribute value of each task according to the sample attribute and the corresponding attribute value of each task, and the task mixed attribute is used to describe the overall characteristics of the task.

[0010] In a third mode, the relationship discovery apparatus determines the task mixed attribute and the corresponding attribute value of each task according to the task attribute and the corresponding attribute value of each task, and the sample attribute and the corresponding attribute value of each task.

[0011] After determining the task mixed attribute of each task, the relationship discovery apparatus can determine the association relationship between tasks according to the task mixed attribute and the corresponding attribute value of each task.

[0012] Through the above method, the relationship discovery apparatus can determine the task attribute of the task based on the sample attribute of the task, then determine the mixed attribute of the task based on the sample attribute and / or the task attribute of the task, and then determine the associated tasks based on the task mixed attribute of each task, that is, the association relationship between tasks. In the whole process, it does not depend on personnel experience and can efficiently determine the associated tasks.

[0013] In a possible implementation, when determining the association relationship between tasks according to the task mixed attribute and the corresponding attribute value of each task, the relationship discovery apparatus can calculate the similarity or difference between tasks according to the task mixed attribute and the corresponding attribute value of each task; then, determine the association relationship between tasks based on the similarity or difference between tasks.

[0014] Through the above method, the relationship discovery apparatus can more conveniently determine the association relationship between tasks through the similarity or difference between tasks, and the efficiency is higher.

[0015] In a possible implementation, when determining the task mixed attribute and the corresponding attribute value of each task according to the task attribute and the corresponding attribute value of each task, and the sample attribute and the corresponding attribute value of each task, the relationship discovery apparatus can determine the target task attribute and the corresponding attribute value of each task from the task attribute and the corresponding attribute value of each task; determine the target sample attribute and the corresponding attribute value of each task from the sample attribute and the corresponding attribute value of each task; then, determine the task mixed attribute and the corresponding attribute value of each task according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value of each task.

[0016] Through the above method, the target task attribute and the target sample attribute can be selected, and then the task mixed attribute can be determined based on the target task attribute and the target sample attribute.

[0017] In a possible implementation, the relationship discovery apparatus can determine the target task attribute and the corresponding attribute value of each task in many ways from the task attribute and the corresponding attribute value of each task, and the following lists some of them.

[0018] Firstly, for any one of the tasks, the relationship discovery apparatus can select part of the task attributes as the target task attribute of the task from the task attribute of the task, wherein one task attribute in the part of the task attributes corresponds to one target task attribute, and the attribute value corresponding to the target task attribute of the task is the attribute value corresponding to the task attribute.

[0019] Secondly, for any one of the tasks, the relationship discovery apparatus can analyze or convert the task attribute of the task to generate the target task attribute of the task, and the attribute value corresponding to the target task attribute is determined by analyzing the attribute value corresponding to the task attribute of the task.

[0020] Through the above method, the relationship discovery apparatus can flexibly determine the target task attribute and the corresponding attribute value of each task in many different ways.

[0021] In a possible implementation, the relationship discovery apparatus can determine the target sample attribute and the corresponding attribute value of each task in many ways from the sample attribute and the corresponding attribute value of each task, and the following lists some of them.

[0022] Firstly, for any one of the tasks, the relationship discovery apparatus can select part of the sample attributes as the target sample attribute of the task from the sample attribute of the task, wherein one sample attribute in the part of the sample attributes corresponds to one target sample attribute, and the attribute value corresponding to the target sample attribute is determined by analyzing the attribute value corresponding to the sample attribute.

[0023] Secondly, for any one of the tasks, the relationship discovery apparatus can analyze or convert the sample attribute of the task to generate the target sample attribute of the task, wherein one sample attribute in the part of the sample attributes corresponds to one target sample attribute, and the attribute value corresponding to the target sample attribute is determined by analyzing the attribute value corresponding to the sample attribute.

[0024] Through the above method, the relationship discovery apparatus can flexibly determine the target sample attribute and the corresponding attribute value of each task in many different ways.

[0025] In a possible implementation, there are many ways for the relationship discovery apparatus to determine the task mixed attribute and the corresponding attribute value of each task according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value. Some of them are listed as follows.

[0026] Firstly, for any one of the tasks, the relationship discovery apparatus takes the target task attribute or the target sample attribute of the task as the task mixed attribute of the task. One target task attribute or one target sample attribute of the task corresponds to one task mixed attribute of the task. The attribute value corresponding to one task mixed attribute of the task is determined after analyzing the attribute value corresponding to one target task attribute or one target sample attribute of the task.

[0027] Secondly, the relationship discovery apparatus selects part of the target task attributes of the tasks and part of the target sample attributes of the tasks. According to the part of the target task attributes and the part of the target sample attributes, the relationship discovery apparatus determines the task mixed attribute of each task. The attribute value corresponding to the task mixed attribute of the task is determined after analyzing the attribute value corresponding to the part of the target task attributes or the part of the target sample attributes.

[0028] Through the above method, the relationship discovery apparatus can flexibly determine the task mixed attribute and the corresponding attribute value of each task through various different ways.

[0029] In a possible implementation, the task mixed attribute and the corresponding attribute value of each task determined by the relationship discovery apparatus can be one or more levels of task mixed attribute and the corresponding attribute value. When the relationship discovery apparatus calculates the similarity or difference between each task according to the task mixed attribute and the corresponding attribute value of each task, the relationship discovery apparatus can calculate the similarity or difference between the tasks at the same level according to the task mixed attribute and the corresponding attribute value of each task at the same level.

[0030] Through the above method, the similarity or difference between the tasks at different levels can show the similarity or difference between the tasks from multiple different granularities.

[0031] In a possible implementation, when the relationship discovery apparatus determines the association relationship between the tasks based on the similarity or difference between the tasks, the relationship discovery apparatus can determine the association relationship between the tasks based on the similarity or difference between the tasks at one or more levels.

[0032] Through the above method, the relationship discovery apparatus can more accurately determine the association relationship between the tasks by using the similarity or difference between the tasks at different levels.

[0033] In a possible implementation, when determining the association relationship between the tasks in each level based on the similarity between the tasks in one or more levels, the relationship discovery apparatus can construct a similarity matrix in each level based on the similarity between the tasks in each level, the similarity matrix in a level is constructed based on the similarity between the tasks in the level, then determine a matching mapping relationship, the matching mapping relationship is used to indicate the correspondence between the similarity matrix in different levels and a clustering method, then determine the clustering method corresponding to the similarity matrix in one or more levels based on the matching mapping relationship, and determine the candidate associated tasks in one or more levels by using the clustering method corresponding to the similarity matrix based on the similarity matrix in one or more levels, and then determine the association relationship between the tasks based on the candidate associated tasks in one or more levels.

[0034] By the above method, the relationship discovery apparatus can determine the candidate associated tasks in each level based on the matching mapping relationship, and then determine the association relationship between the tasks based on the candidate associated tasks in each level.

[0035] In a second aspect, the embodiments of the present application further provide a relationship discovery apparatus, which has functions of implementing the behaviors in the method examples of the first aspect. The functions can be implemented by hardware, or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above functions. In a possible design, the structure of the apparatus can include a sample attribute extraction module, a task attribute extraction module, a task mixed attribute determination module, and an adaptation module, and optionally, a task similarity measurement module. These modules can perform the corresponding functions in the method examples of the first aspect, and details are referred to the detailed description in the method examples, which will not be repeated here. The above modules are only an example of function division, and cannot limit the protection scope of the present application. According to the actual situation, other logical division manners can also be used to obtain different functional modules, which are all within the protection scope of the present application.

[0036] In a third aspect, the embodiments of the present application further provide a computing device, which includes a processor and a memory, and can further include a communication interface and a display screen. The processor executes program instructions in the memory to perform the method provided by the first aspect or any possible implementation of the first aspect. The memory is coupled with the processor, and stores program instructions and data necessary in the method execution process. The communication interface is used to communicate with other devices, for example, to receive samples of tasks, and to send associated tasks in the tasks. The display screen is used to display information to the user under the trigger of the processor, for example, to display information such as associated tasks.

[0037] In a fourth aspect, the present application provides a computing device cluster, which comprises at least one computing device. Each computing device comprises a memory and a processor. The processor of the at least one computing device is configured to access the code in the memory to execute the method provided in the first aspect or any possible implementation manner of the first aspect.

[0038] In a fifth aspect, the present application provides a non-transitory readable storage medium, which is executed by a computing device, and the computing device can execute the method provided in the first aspect or any possible implementation manner of the first aspect. The storage medium stores a program. The storage medium comprises, but is not limited to, a volatile memory such as a random access memory, a non-volatile memory such as a flash memory, a hard disk drive (HDD), and a solid state drive (SSD).

[0039] In a sixth aspect, the present application provides a computing device program product, which comprises computer instructions, and when executed by a computing device, the computing device executes the method provided in the first aspect or any possible implementation manner of the first aspect. The computer program product can be a software package, and when the method provided in the first aspect or any possible implementation manner of the first aspect is needed, the computer program product can be downloaded and executed on the computing device.

[0040] In a seventh aspect, the present application further provides a computer chip, which is connected with a memory, and the chip is configured to read and execute a software program stored in the memory to execute the method provided in the first aspect or any possible implementation manner of the first aspect.

[0041] The beneficial effects achieved by any one of the second aspect to the seventh aspect and any possible solution of any one of the second aspect to the seventh aspect can be referred to the beneficial effect description of the corresponding solution in the first aspect, and will not be described here. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 A structural schematic diagram of a relationship discovery device provided by the present application;

[0043] Figure 2 A schematic diagram of a task relationship determination method provided by the present application;

[0044] Figure 3 A schematic diagram of a sample set provided by the present application;

[0045] Figure 4 A schematic diagram of various attributes of a task and corresponding attribute value distribution provided by the present application;

[0046] Figure 5 An example of various attributes of another task provided by the present application and a corresponding attribute value distribution diagram;

[0047] Figure 6 An example of a computer cluster structure provided by the present application;

[0048] Figure 7 An example of a system structure provided by the present application. DETAILED DESCRIPTION

[0049] Before describing the task relationship determination method and device provided by the embodiments of the present application, the concepts involved in the embodiments of the present application are described:

[0050] (1) task, sample

[0051] In the embodiments of the present application, the data processing process in a scene can be referred to as a task, different tasks correspond to different scenes, and the rules (i.e., objective functions) followed by the data in different tasks are different. The data processing process in a scene can be performed by a model, for example, a defect detection model can be used to detect defects on a printed circuit board (PCB), and a face detection model can be used to detect faces in an image.

[0052] Taking PCB defect detection as an example, the defect detection of a PCB produced by a production line can be regarded as a task, and the defect detection of PCBs produced by different production lines can be regarded as different tasks. This is because the production environment of different production lines can be different, and the types and distribution positions of defects in the generated PCBs can also be different.

[0053] Taking face detection as an example, face detection in the Asian region can be regarded as a task, and face detection in the European region can be regarded as another task. Since the growth environment of personnel in the Asian region is different from that of personnel in the European region, the face shape and facial features of personnel in the Asian region are different from those of personnel in the European region, so face detection in the Asian region and face detection in the European region can be regarded as different tasks.

[0054] In the embodiments of the present application, some data required to be processed by each task can be obtained in advance, and such data can be referred to as a sample.

[0055] Taking PCB defect detection as an example, the image or structured data of a PCB produced by a production line can be regarded as a sample, and a sample set of images or structured data of multiple PCBs can be selected.

[0056] (2) sample attribute, task attribute, task mixed attribute

[0057] Sample attribute is used to describe the characteristics of one or more samples in a task. In the embodiments of the present application, one sample can have multiple sample attributes. For an image type sample, the sample attribute can be the name, pixels, grayscale value, shape of the object (e.g., PCB) shown in the image, etc. For a structured data type sample, the sample attribute can be a field in the table header, and the value under the table header can be the attribute value of the sample attribute.

[0058] Task attribute is used to describe the characteristics of the task. Different task attributes describe the characteristics of the task from different perspectives. In the embodiments of the present application, the definition of the task attribute is not limited. For example, the task attribute of a task can be determined according to the metadata describing the sample in advance, or can be defined based on the sample attribute of the task.

[0059] Taking the task of PCB defect detection as an example, the number of components in a PCB produced by a production line can be a task attribute. The size of a PCB produced by a production line can also be a task attribute. The model of a PCB produced by a production line or the shape of a component can also be a task attribute.

[0060] For any task, the task attribute of the task can be one or multiple. In order to more comprehensively describe the characteristics of the task, multiple task attributes can be set for a task.

[0061] Task mixed attribute can also be used to describe the characteristics of the task. The task mixed attribute can be determined by multiple different task attributes and / or sample attributes of a task. The task mixed attribute can describe the overall characteristics of the task. The task mixed attribute can be a task attribute, a sample attribute, or a combination of a task attribute and a sample attribute.

[0062] In the embodiments of the present application, the task mixed attribute can be divided into different levels. Each level corresponds to a group of task mixed attributes, and the group of task mixed attributes includes one or more task mixed attributes. The larger the level, the more accurate the overall characteristics of the task described by the task mixed attribute. The level can also be understood as granularity. The task mixed attribute with larger level can describe the fine-grained characteristics of the task, and the task mixed attribute with smaller level can describe the coarse-grained characteristics of the task.

[0063] (3) Similarity, difference degree

[0064] Similarity is used to measure the degree of similarity between two tasks. The similarity between tasks is that the data distribution in the task is close, or the rule followed by the data in the task is the same or the difference is small.

[0065] The difference degree is used to measure the difference between two tasks, and the difference between tasks is that the data distribution in the task is different, or the rule followed by the data in the task is different or the gap is large.

[0066] It should be noted that similarity and difference degree are two opposite concepts, and the difference degree is determined when the similarity is determined. Similarly, the similarity is determined when the difference degree is determined. One of the similarity and the difference degree is selected to represent the distribution of data in two tasks.

[0067] (4) Association relationship, associated task, candidate associated task

[0068] In the embodiments of the present application, there is an association relationship between the tasks that can use the same model for data processing. The tasks with an association relationship can also be referred to as associated tasks, and the similarity of the associated tasks is within a threshold range. The candidate associated task refers to the associated task determined according to the similarity matrix at different levels.

[0069] Figure 1 As shown in the figure, a relationship discovery device 100 provided by the embodiments of the present application includes a sample attribute extraction module 101, a task attribute extraction module 102, a task mixed attribute determination module 103, a task similarity measurement module 104, and an adaptation module 105.

[0070] For example, the sample attribute extraction module 101 is used to extract the sample attribute of each task in each different task and the corresponding attribute value. The task attribute extraction module 102 is used to determine the task attribute and the corresponding attribute value of each task based on the characteristics of each task itself and the sample attribute and the corresponding attribute value in the task. The task mixed attribute determination module 103 can determine the task mixed attribute of any task according to the task attribute and the corresponding attribute value of the task and / or the sample attribute and the corresponding attribute value of the task. The task similarity measurement module 104 determines the similarity or difference degree between tasks according to the task mixed attribute of each task. The adaptation module 105 determines the association relationship between each task based on the similarity or difference degree between each task.

[0071] The embodiments of the present application will be described below in conjunction with Figure 1 and Figure 2 A task relationship determination method provided by the embodiments of the present application will be described. Referring to Figure 2 , the method includes the following steps:

[0072] Step 201: The sample attribute extraction module 101 first determines the sample attribute of each task in each task and the corresponding attribute value. The sample attribute extraction module 101 can transmit the sample attribute of each task in each different task and the corresponding attribute value to the task attribute extraction module 102.

[0073] In step 201, the sample attribute extraction module 101 first acquires the sample set of each task, extracts the sample attribute of the sample set of each task and the corresponding attribute value. The sample attribute determined by the sample attribute extraction module 101 is related to the data type of the sample.

[0074] For example, the sample is an image, and the sample attribute extraction module 101 can be a module with image analysis capability, which can analyze image type samples and can take the features of the image as sample attributes, such as image gray scale and image texture. For example, the sample attribute extraction module 101 can include a neural network model and can also include other image analysis models.

[0075] For another example, the sample set is structured data, and in the structured data, the data is in units of rows, and each row of data represents the information of a sample. Each column is a type of feature data of the sample. If the structured data includes a table header, that is, a field is set on each column of the structured data, the field can indicate a type of feature of the sample. For this type of structured data, the field can be used as a sample attribute, and the value in the column where the field is located can be used as the attribute value corresponding to the sample attribute.

[0076] As shown in FIG. 1, the sample set of a task is shown, which is in the form of structured data, and the structured data includes a table header. Each row is a sample in the sample set, and each column is a feature of the sample. Figure 3

[0077] Step 202: For any task in each task, the task attribute extraction module 102 determines the task attribute of the task and the corresponding attribute value based on the sample attribute of the task and the corresponding attribute value, and can also determine the task attribute of the task and the corresponding attribute value according to the metadata for describing the sample, wherein the metadata for describing the sample is a type of data for describing the sample.

[0078] The task attribute extraction module 102 sends the task attribute of each task and the sample attribute and the corresponding attribute value to the task mixed attribute determination module 103.

[0079] ​The task attribute extraction module 102 can determine sample attributes and corresponding attribute values according to the sample set of the task, and then determine the task attributes and corresponding attribute values according to the sample attributes and corresponding attribute values. The sample attributes can be divided into common attributes and difference attributes. The common attributes refer to the attributes that all samples in the sample set have and the corresponding attribute values are the same. Taking the task of PCB defect detection as an example, the size of the PCB generated on a production line is the same, and the size of the PCB can be used as a task attribute, and the actual size value of the PCB is the corresponding attribute value.

[0080] The difference attributes refer to the attributes that all samples in the sample set have, but the corresponding attribute values are different. Taking the task of PCB defect detection as an example, although the size of the PCB generated on a production line is the same, the number of components on the PCB is different, and the number of components on the PCB can be used as a sample attribute, and the component quantity value can be used as the corresponding attribute value.

[0081] The task attribute extraction module 102 can use the common attributes as the task attributes, and the attribute values corresponding to the task attributes are the attribute values corresponding to the sample attributes. The task attribute extraction module 102 can also use the difference attributes as the task attributes, and determine the attribute values of the task attributes according to the attribute values corresponding to the difference attributes.

[0082] Taking the task of PCB defect detection as an example, the task attribute extraction module 102 can use the size of the PCB as the task attribute, and the actual size of the PCB is the corresponding attribute value. The task attribute extraction module 102 can also use the number of components on the PCB as the task attribute, and the average of the component quantities of each PCB in the sample set can be used as the corresponding attribute value.

[0083] In addition to the common attributes and difference attributes of each sample in the sample set, the task attribute extraction module 102 can also directly define the task attributes according to the task characteristics. For example, the data type of the sample is an image, and the task attribute extraction module 102 can directly use the data type of the sample as the task attribute, and the image is the attribute value corresponding to the task attribute.

[0084] Step 203: For any task in the plurality of tasks, the task mixed attribute determination module 103 determines the task mixed attribute and the corresponding attribute value of the task according to the task attribute and the corresponding attribute value and / or the sample attribute and the corresponding attribute value of the task. The task mixed attribute determination module 103 sends the task mixed attribute and the corresponding attribute value of each task in each task to the task similarity measurement module 104.

[0085] For any task, the task attributes or the sample attributes can reflect the characteristics of the task. The task attributes can be regarded as the common characteristics of the samples, and the sample attributes reflect the characteristics of the task from the sample granularity. Different task attributes and sample attributes of a task describe the characteristics of the task from different angles.

[0086] In order to more clearly understand the relationship between the task attributes, the sample attributes and the task mixed attributes of any task, please refer to Figure 4 and Figure 5 .

[0087] Figure 4 is a representation of the various attributes of a task and the distribution of the corresponding attribute values (normalized attribute values), wherein u represents a statistical characteristic, which can be a characteristic value such as the mean, the median or the mode of the attribute values corresponding to the sample attributes which are the same as the task attributes; and σ represents the deviation of the sample attribute values from the statistical characteristic u, which can be the variance of the sample attribute values from the statistical characteristic u. The subscripts of u and σ correspond to different attributes. Figure 4 A curve in the figure corresponds to a task attribute of the task, and the ordinate can be regarded as the number distribution of the attribute values of the samples under the task attribute (here, the attribute values of the samples refer to the attribute values of the sample attributes which are the same as the task attributes). The ordinate of each point on the curve can be used as the attribute value corresponding to the task attribute, or other attribute values derived from the curve. Figure 4 It can be seen that for a task, multiple task attributes and the corresponding attribute value distributions of each sample under the task attribute are relatively scattered, and cannot represent the characteristics of the task as a whole. The task mixed attribute can be regarded as a curve determined by the comprehensive curve and the ordinates of each point on each curve, which can cover the attribute values of each sample under the task attribute.

[0088] It should be noted that theoretically, the comprehensive curve and the ordinates of each point on each curve can determine multiple different curves, and each of the multiple different curves can cover the attribute values of each sample under the multiple task attributes of the task. However, the fitting degrees of the different curves to the attribute values of each sample under the multiple task attributes of the task are different. The greater the fitting degree, the more accurate the task characteristics represented by the curve. In essence, the curve with a large fitting degree can be regarded as a fine-grained task mixed attribute, and similarly, the curve with a small fitting degree can be regarded as a coarse-grained task mixed attribute.

[0089] Figure 5 is another representation of the various attributes of another task and the distribution of the corresponding attribute values, Figure 5one of the tasks attributes of the task, the area surrounded by the solid line includes the distance distribution between the attribute value of each sample and the attribute value of the task attribute. Similarly, from Figure 5 It can be seen that for a task, multiple task attributes and corresponding attribute value distributions are relatively scattered and cannot represent the characteristics of the task as a whole. The task mixed attribute can be regarded as an envelope line determined by the comprehensive curve and the area surrounded by each curve (indicated by the dashed line in Figure 5 It needs to be explained here that theoretically, the comprehensive curve and the area surrounded by each curve can determine multiple different envelope lines, and each of the multiple different envelope lines can cover the attribute values of each sample under the task attribute, but the degrees of fit of different envelope lines to the attribute values of each sample under the task attribute are different. The greater the degree of fit, the more accurate the task characteristics represented by the envelope line. In essence, the envelope line with a large degree of fit can be regarded as a fine-grained task mixed attribute, and similarly, the envelope line with a small degree of fit can be regarded as a coarse-grained task mixed attribute.

[0090] When determining the task mixed attribute of any task, the task mixed attribute determination module 103 can determine the task mixed attribute and the corresponding attribute value of the task based on only the task attribute and the corresponding attribute value of the task, or based on only the sample attribute and the corresponding attribute value of the task, or based on the task attribute and the corresponding attribute value of the task and the sample attribute and the corresponding attribute value of the task. The following describes the three methods:

[0091] First, the task mixed attribute determination module 103 determines the task mixed attribute and the corresponding attribute value of the task based on the task attribute and the corresponding attribute value of the task. The task mixed attribute determination module 103 can first determine a target task attribute according to the task attribute of the task, and then determine the task mixed attribute of the task based on the target task attribute.

[0092] The task mixed attribute determination module 103 has many ways to determine the target task attribute, which is not limited in the embodiments of the present application, and some of them are listed as follows:

[0093] 1. When determining the target task attribute of the task, the task mixed attribute determination module 103 can regard each task attribute of the task as a target task attribute. The attribute value of the target task attribute is the attribute value corresponding to the task attribute.

[0094] 2. The task mixed attribute determination module 103 can also select part of the task attributes of the task as target task attributes, and the attribute value of the target task attribute is the attribute value corresponding to the task attribute.

[0095] 3、The task mixed attribute determination module 103 can also analyze the task attributes of the task, determine a new task attribute, and take the new task attribute as the target task attribute. The attribute value corresponding to the target task attribute is determined according to the attribute value corresponding to the task attribute of the task.

[0096] One or more task attributes of the task can comprehensively reflect the new task attribute of the task. For example, in the motor fault analysis task, the real-time voltage, current and power of the motor can reflect the real-time fault state of the motor. The task mixed attribute determination module 103 can take the fault state as a new task attribute, and the attribute value corresponding to the fault state can be determined according to the real-time voltage value, current value and power value of the motor.

[0097] 4、The task mixed attribute determination module 103 converts the task attributes of the task to generate a new task attribute, and takes the new task attribute as the target task attribute.

[0098] One task attribute of the task can be characterized by the new task attribute of the task, that is, the task attribute can be converted into a new task attribute. For example, in the motor fault analysis task, the frequency of the motor is a determined task attribute, and the frequency of the motor can be converted into the speed of the motor. The performance of the motor can be more intuitively reflected by using the speed of the motor.

[0099] After determining the target task attribute of the task, the task mixed attribute determination module 103 can determine the task mixed attribute of the task based on the target task attribute of the task.

[0100] There are many ways for the task mixed attribute determination module 103 to determine the task mixed attribute based on the target task attribute, some of which are listed below:

[0101] 1)、The task mixed attribute determination module 103 can directly take the target task attribute of the task as the task mixed attribute of the task.

[0102] 2)、If there are multiple target task attributes, the task mixed attribute determination module 103 can reduce the dimension of the multiple target task attributes of the task to determine the task mixed attribute of the task.

[0103] The embodiments of the present application do not limit the way of reducing the dimension of the multiple target task attributes, for example, principal component analysis (PCA), auto-encoder, etc. can be used.

[0104] 3) If there are multiple target task attributes, the task mixed attribute determination module 103 can select part of the target task attributes of the task as the task mixed attribute of the task.

[0105] There are many ways for the task mixed attribute determination module 103 to select part of the target task attributes, for example, the task mixed attribute determination module 103 can select the target task attribute that can best reflect the characteristics of the task as the task mixed attribute of the task. The task mixed attribute determination module 103 can set an evaluation index for each target task attribute of the task, which can indicate the reliability of the task attribute in describing the characteristics of the task, and the task mixed attribute determination module 103 can select part of the target task attributes as the task mixed attribute of the task according to the evaluation index of the target task attribute of the task. The embodiments of the present application do not limit the calculation method of the evaluation index, for example, the task mixed attribute determination module 103 can analyze the sample set of the task, and determine the evaluation index according to the distribution of the attribute value of the target task attribute in the sample set.

[0106] Secondly, the task mixed attribute determination module 103 determines the task mixed attribute and the corresponding attribute value of the task based on the sample attribute and the corresponding attribute value of the task. The task mixed attribute determination module 103 can first determine the target sample attribute of the task according to the sample attribute of the task, and then determine the task mixed attribute of the task based on the target sample attribute of the task.

[0107] The task mixed attribute determination module 103 determines the target sample attribute in a similar way to determining the target task attribute, and details can be referred to the foregoing content, which will not be repeated here.

[0108] Since the same sample attribute can have different attribute values, the task mixed attribute determination module 103 can determine the attribute value corresponding to the target sample attribute after determining the target sample attribute. The embodiments of the present application do not limit the way to determine the attribute value corresponding to the target sample attribute, for example, the attribute value corresponding to the target sample attribute can be determined by taking the mean, variance or median of different attribute values.

[0109] After the task mixed attribute determination module 103 determines the target sample attribute, the way to determine the task mixed attribute and the corresponding attribute value based on the target sample attribute and the corresponding attribute value is similar to the way to determine the task mixed attribute and the corresponding attribute value based on the target task attribute and the corresponding attribute value, and details can be referred to the foregoing content, which will not be repeated here.

[0110] Thirdly, the task mixed attribute determination module 103 determines the task mixed attribute and the corresponding attribute value of the task based on the task attribute and the corresponding attribute value, and the sample attribute and the corresponding attribute value of the task.

[0111] The task mixed attribute determination module 103 may first determine the target task attribute of the task according to the task attribute of the task, determine the target sample attribute of the task according to the sample attribute of the task, and then determine the task mixed attribute of the task based on the target task attribute and the target sample attribute of the task.

[0112] The manner in which the task mixed attribute determination module 103 determines the target task attributes and the target sample attributes can be found in the above content and will not be described in detail here.

[0113] The way in which the task mixed attribute determination module 103 determines the task mixed attribute and the corresponding attribute value of the task based on the target task attribute and the target sample attribute and the corresponding attribute value of the task is similar to the way in which the task mixed attribute determination module 103 determines the task mixed attribute and the corresponding attribute value based on the target task attribute and the corresponding attribute value of the task. For details, please refer to the aforementioned content and will not be repeated here.

[0114] The following further explains the third method of determining task promiscuity attributes based on specific scenarios:

[0115] The task miscellaneous attribute can be used to describe the overall characteristics of the task. In the embodiment of the present application, the task miscellaneous attribute and the corresponding attribute value can be determined by a miscellaneous function based on one or more task attributes and corresponding attribute values ​​and one or more sample attributes and corresponding attribute values ​​of the task.

[0116] The confusion function can be a predefined function used to calculate the task confusion attribute. The task confusion attribute and its calculation process can be expressed as follows:

[0117]

[0118] Among them, F τ is the mixing function. τ is the level factor, and τ∈N is an integer. Represent task attributes and sample attributes respectively. They represent the task attribute extraction function and sample attribute extraction function respectively. They represent the original task attributes and task attribute values, and the original sample attributes and sample attribute values ​​respectively.

[0119] The process of determining task mixed attributes is divided into two stages: the first stage is to extract the target task attributes and the corresponding attribute values ​​and the target sample attributes and the corresponding attribute values; the second stage is to mix the target task attributes and the corresponding attribute values ​​and the target sample attributes and the corresponding attributes to construct the task mixed attributes and the corresponding attribute values.

[0120] The first stage is to extract target task attributes and corresponding attribute values and to extract target sample attributes and corresponding attribute values.

[0121] For task attribute extraction function The input is task attributes and corresponding attribute values, and the output is target task attributes and corresponding attribute values.

[0122] The method characterized includes limiting to:

[0123] a) Output the task attribute without change. For example, in PCB defect detection, the production line name of the PCB board can be directly used as the target task attribute without change, the target task attribute is the production line, and the target task attribute value is the production line name.

[0124] b) Select part of the task attribute as the target task attribute. For example, input the current, voltage, and power of the motor, select the power as the target task attribute, the target task attribute is the power, and the target task attribute value is the motor power value.

[0125] c) Determine a new task attribute according to the task attribute, and the new task attribute is the target task attribute. For example, input the current, voltage, and power of the motor, use the fault detection algorithm to detect whether there is a fault according to the current, voltage, and power, output the current, voltage, power, and fault state of the motor, and add a new task attribute of the motor fault state, and the new task attribute value is "yes or no".

[0126] d) Convert the task attribute to a new task attribute, and the new task attribute is the target task attribute. For example, convert the motor frequency to the motor speed, and the motor speed is the new task attribute.

[0127] Sample attribute extraction function The input can be a sample, The sample can be analyzed to determine the sample attribute and the corresponding attribute value, and then the target sample attribute and the corresponding attribute value are output. The input can also be a sample attribute and a sample attribute corresponding attribute value, and the output is a target sample attribute and a corresponding attribute value, and the output target sample attribute and the corresponding attribute value. It can be regarded as a dimension reduction process.

[0128] The method characterized includes limiting to: mean, variance, mode, median, etc.

[0129] For example, in the PCB defect inspection scene, 100 pieces of PCB production process monitoring data are input, including the number of components of each PCB, and the target sample attribute is output: the number of components of the PCB, and the target sample attribute value is the average of the number of components of the 100 pieces of PCB.

[0130] In the second stage, the target task attribute and the corresponding attribute value and the target sample attribute and the corresponding attribute value are mixed to determine the mixed task attribute and the corresponding attribute value.

[0131] For the τ-level mixed function F τ (·), it can be a linear function or a nonlinear function.

[0132] F τ The method characterized includes limiting to:

[0133] a) directly splicing the target task attribute and the target sample attribute as the task mixed attribute, for example, in the PCB defect inspection scene, the target task attribute "production line" and the corresponding attribute value "production line name" and the target sample attribute "PCB component quantity" and the corresponding attribute value "average of 100 data sample corresponding attribute values" are input, and the output is the mixed attribute "input production line, PCB component quantity" and the corresponding attribute value "production line name, average of 100 data sample corresponding attribute values".

[0134] b) reducing the target task attribute and the target sample attribute together to determine the task mixed attribute, such as using principal component analysis, autoencoder, etc. for dimension reduction.

[0135] For example, in the motor fault analysis task, the target task attribute and the target sample attribute determined according to the monitoring parameters of the motor at different times are motor current, voltage and power. Taking the motor current, voltage and power as input, it is found by principal component analysis that the voltage is basically unchanged, and the current and power are positively correlated. The principal component analysis dimension reduction output task mixed attribute is motor current and power.

[0136] c) processing each target task attribute and target sample attribute and combining the output as the task mixed attribute.

[0137] Taking the task of PCB defect detection as an example, the attribute values corresponding to each target task attribute are output as the task attribute without any change, and the attribute values corresponding to the target sample attribute are the average {u1, u2, …, u n}, and the corresponding variance (the variance can be understood as the similarity of the sample) is {σ1, σ2, …, σ n}, and the τ-level mixed function F τ (·) takes the principal component analysis function.

[0138] If the level factor τ is 1, the task mixed attribute is essentially the attribute with the most important information content among the task attributes and the sample attributes, which can be derived from the task attributes or the sample attributes. Shape identifier The physical meaning is that the information distribution of a task is reflected on the target task attribute or the target sample attribute, and the most important attribute is used as the task mixed attribute of the task to describe the overall characteristics of the task.

[0139] Step 204: The task similarity measurement module 104 determines the similarity or difference between each task according to the task mixed attribute of each task. The task similarity measurement module 104 sends the similarity or difference between each task to the adaptation module 105.

[0140] After determining the task mixed attribute of each task and the corresponding attribute value, the characteristics of the task can be marked by the task mixed attribute, and the similarity between tasks can be represented by the closeness between the attribute values corresponding to the task mixed attributes of each task. The difference between tasks can also be represented by the closeness between the attribute values corresponding to the task mixed attributes of each task.

[0141] The following are several similarity measurement methods:

[0142] Method one, using a measurement model to determine the similarity.

[0143] The task similarity measurement module 104 can include a measurement model, which can be obtained by training a preset model using supervised learning or unsupervised learning and a training sample set. The training sample set includes the task mixed attribute and the corresponding attribute value of the known task. The input of the measurement model is the attribute value corresponding to the task mixed attribute of different tasks, and the output is the similarity between the different tasks.

[0144] The embodiments of the present application do not limit the specific type of the metric model, for example, the metric model can be a model of a Mahalanobis metric learning problem based on linear transformation, including but not limited to: information theoretic metric learning (ITML), local linear discriminant analysis (local LDA), principal components analysis (PCA), multidimensional scaling (MDS), locality preserving projection (LPP); the metric model can also be a learning model based on non-linear metric, wherein the non-linear metric includes but is not limited to: isometric mapping (ISOMAP), laplacian eigenmaps (LE), metric learning based on neural network, etc.

[0145] Method two, method of distance calculation:

[0146] The task similarity measurement module 104 calculates a distance matrix according to the attribute values of the mixed attributes of different tasks, wherein the elements in the distance matrix are distance values between two tasks determined based on the attribute values of the mixed attributes of the two tasks, and the distance values can represent the similarity between the two tasks. The embodiments of the present application do not limit the type of distance value, which can be Euclidean distance, Manhattan distance, Chebyshev distance, Minkowski distance, bulldozer distance, or divergence, etc.

[0147] Method three, method of similarity function:

[0148] The task similarity measurement module 104 can determine the similarity of different tasks by means of a similarity function. The embodiments of the present application do not limit the type of similarity function, which can be cosine similarity, Pearson correlation coefficient, or log-likelihood similarity, etc.

[0149] It should be noted that the above three similarity measurement methods are only examples, and the embodiments of the present application do not limit the task similarity measurement module 104 to use other methods to measure the similarity between different tasks. Any method that can measure the similarity between different tasks is applicable to the embodiments of the present application.

[0150] The task similarity measurement module 104 can determine the similarity between tasks through the same level of task mixed attributes of each task, and since one task can have multiple levels of task mixed attributes, the task similarity measurement module 104 can determine the similarity at different levels between tasks through the different levels of task mixed attributes of each task.

[0151] Step 205: The adaptation module 105 determines the association relationship between tasks based on the similarity or difference between tasks.

[0152] In step 205, the adaptation module 105 can determine the associated tasks based on the similarity or difference of each task at one level, or can determine the associated tasks based on the similarity or difference of each task at multiple levels.

[0153] When the adaptation module 105 can determine the associated tasks based on the similarity or difference of each task at one level, the adaptation module 105 can determine a plurality of task sets in which the similarity is in a preset range based on the similarity or difference between tasks at the level, each task set includes at least one task, and the tasks included in each task set are different. The tasks in one task set have an association relationship, and the tasks in the task set can use the same model for data processing. The similarity between tasks in different task sets is low, and the same model cannot be used for data processing.

[0154] Generally, the adaptation module 105 can quickly determine the association relationship between tasks based on the similarity or difference between tasks at one level. In order to accurately determine the association relationship between tasks, the adaptation module 105 can determine the association relationship between tasks based on the similarity or difference between tasks at multiple levels.

[0155] The embodiments of the present application provide two ways of determining the association relationship between tasks based on the similarity or difference between tasks at multiple levels, which will be introduced as follows:

[0156] Method one, the adaptation module 105 first reduces the dimension of the similarity or difference between tasks at multiple levels, and then determines the association relationship between tasks.

[0157] For example, the adaptation module 105 first reduces the dimension of the similarity between tasks at multiple levels, and then determines the associated tasks. The dimension of the similarity or difference between tasks at multiple levels is reduced, and then the association relationship between tasks is determined.

[0158] The similarity between tasks at one level can form a similarity matrix, and an element in the similarity matrix indicates the similarity between two tasks.

[0159] The similarity matrix can be a symmetric matrix or an asymmetric matrix, which is related to the similarity measurement method selected when calculating the similarity. For example, the similarity is determined by using the KL divergence method. The similarity of A to B is different from the similarity of B to A, so the similarity matrix can be asymmetric.

[0160] The similarities between the tasks at each level form a plurality of similarity matrices, each of which corresponds to the similarities between the tasks at a level.

[0161] The adaptation module 105 can reduce and combine the plurality of similarity matrices into a target similarity matrix. The embodiments of the present application do not limit the manner in which the adaptation module 105 reduces and combines the plurality of similarity matrices. For example, the adaptation module 105 can use an averaging method to average the elements at the same position in the plurality of similarity matrices, and then combine them into a target similarity matrix. For another example, the adaptation module 105 can also use a weight allocation method to configure weights for each similarity matrix, multiply the elements at the same position in each similarity matrix in the plurality of similarity matrices by the corresponding weights, sum the values, and then average them to combine them into a target similarity matrix.

[0162] After determining the target similarity matrix, the adaptation module 105 can perform clustering based on the target similarity matrix to determine the association relationship between the tasks, and divide the tasks having an association relationship into a task set.

[0163] Method two: The adaptation module 105 first determines the candidate associated tasks at each level based on the similarities or differences at each level, and then determines the association relationship between the tasks based on the candidate associated tasks at each level, that is, determines the associated tasks in each task.

[0164] Before determining the candidate associated tasks at each level based on the similarities or differences between the tasks at each level, the adaptation module 105 can pre-determine a matching mapping table of the similarity matrix and the clustering method, which indicates the correspondence between the similarity matrix at different levels and the clustering method.

[0165] The embodiments of the present application do not limit the manner in which the adaptation module 105 pre-determines the matching mapping table. Some of them are listed as follows:

[0166] Method one: The matching mapping table can be pre-configured in the adaptation module 105. When the adaptation module 105 needs to determine the associated tasks, the matching mapping table can be obtained.

[0167] The second mode is that the adaptation module 105 can detect the similarity matrix and the clustering method input by the user according to the trigger of the user, establish a corresponding relationship between the similarity matrix and the clustering method, and generate a matching mapping table.

[0168] The third mode is that the adaptation module 105 can determine the level of each similarity matrix based on a similarity matrix set, determine the level of each clustering method based on a clustering method set, establish a corresponding relationship between the similarity matrix and the clustering method of the same level, and generate a matching mapping table. The similarity matrix set includes multiple similarity matrices, which can be classified into multiple levels, and each level can have one or more similarity matrices. The clustering method set includes multiple clustering methods, which can be classified into multiple levels, and each level can have one or more clustering methods.

[0169] It should be noted that the adaptation module 105 can analyze the similarity (or similarity matrix) between the tasks in the same level by the clustering method, cluster the tasks having a correlation relationship from the tasks, that is, determine the candidate associated tasks in the level. The type of clustering method is not limited in the embodiment of the present application, and any clustering method that can cluster the tasks having a correlation relationship is applicable to the embodiment of the present application.

[0170] The adaptation module 105 can generate a similarity matrix between the tasks in each level according to the similarity or difference between the tasks in each level, wherein the similarity between the tasks in one level can constitute a similarity matrix in the level, and one element in the similarity matrix indicates the similarity between two tasks.

[0171] The adaptation module 105 determines the candidate associated tasks in each level based on the similarity matrix in each level based on the matching mapping table, and then determines the associated tasks based on the candidate associated tasks in each level.

[0172] The adaptation module 105 determines the candidate associated tasks in each level based on the similarity matrix in each level based on the matching mapping table, and then determines the associated tasks based on the candidate associated tasks in each level.

[0173] The first mode is that the adaptation module 105 determines a target level, determines the corresponding relationship between the similarity matrix and the clustering method in the target level from the matching mapping table, determines the candidate associated tasks by the clustering method based on the similarity matrix in the target level, and takes the candidate associated tasks as the associated tasks.

[0174] The embodiments of the present application do not limit the manner in which the adaptation module 105 determines the target level. For example, the target level can be determined by the adaptation module 105 according to the operation of the user under the triggering of the user, such as the user selecting a level as the target level. For another example, the adaptation module 105 can randomly select a level as the target level, or determine a level as the target level by a simulation method or a simulation method.

[0175] In the second manner, the adaptation module 105 starts from the lowest level and gradually ascends, determines the clustering method corresponding to the similarity matrix of each level according to the matching mapping table, and determines the candidate associated tasks of each level by clustering according to the similarity matrix of each level through the corresponding clustering method in sequence. After the clustering based on the similarity matrix of each level to determine the candidate associated tasks of each level, the clustering result of the similarity matrix of the previous level (i.e., the candidate associated tasks of the previous level) is analyzed. If the clustering result of the similarity matrix of the current level is consistent with the clustering result of the similarity matrix of the previous level, the clustering can be stopped, and the clustering result of the similarity matrix of the current level (i.e., the candidate associated tasks of the current level) is taken as the associated tasks in each task, thereby the association relationship between each task can be determined. Otherwise, the candidate associated tasks of the subsequent level are determined by clustering according to the similarity matrix of the subsequent level through the corresponding clustering method, until the clustering result of the similarity matrix of the previous level is consistent.

[0176] Based on the same technical concept as the method embodiments, the embodiments of the present application also provide a computer cluster for executing the method shown in the above method embodiments. The related features can be referred to the above method embodiments, and will not be described here again, such as Figure 6 As shown in the above method embodiments, the computer cluster provided by the embodiments of the present application includes at least one computing device 600, and each computing device 600 is connected through a communication network to establish a communication path.

[0177] Each computing device 600 includes a bus 601, a processor 602, a communication interface 603, and a memory 604. Optionally, the computing device 600 can also include a display screen 605. The processor 602, the memory 604, and the communication interface 603 communicate through the bus 601.

[0178] The processor 602 can be constituted by one or more general-purpose processors, such as a central processing unit (CPU), or a combination of a CPU and a hardware chip. The hardware chip can be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The PLD can be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0179] The memory 604 can include a volatile memory, such as a random access memory (RAM). The memory 604 can also include a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD), or a solid-state drive (SSD). The memory 604 can also include a combination of the above-mentioned types.

[0180] The memory 604 stores executable code, and the processor 602 can read the executable code in the memory 604 to implement functions, and can also communicate with other computing devices through the communication interface 603. In an embodiment of the present application, the processor 602 can implement the functions of one or more modules of the relationship discovery device 100 (such as one or more of the sample attribute extraction module 101, the task attribute extraction module 102, the task mixed attribute determination module 103, the task similarity measurement module 104, and the adaptation module 105). In this case, the memory 604 stores one or more modules of the relationship discovery device 100 (such as one or more of the sample attribute extraction module 101, the task attribute extraction module 102, the task mixed attribute determination module 103, the task similarity measurement module 104, and the adaptation module 105).

[0181] In an embodiment of the present application, the processors 602 in the plurality of computing devices 600 can work in coordination to execute the task relationship determination method provided in an embodiment of the present application.

[0182] As Figure 7As shown, a system architecture provided by an embodiment of the present application includes a client 200 and a cloud device 300 deployed with a relationship discovery apparatus 100. The client 200 and the cloud device 300 are connected via a network. The cloud device 300 is located in a cloud environment and can be a server or a virtual machine deployed in a cloud data center. Figure 7 In the figure, the relationship discovery apparatus 100 is deployed on one cloud device 300 as an example. As a possible implementation, the relationship discovery apparatus 100 can be deployed on multiple cloud devices 300 in a distributed manner.

[0183] like Figure 7 As shown, the client 200 includes a bus 201, a processor 202, a communication interface 203, a memory 204 and a display screen 205. The processor 202, the memory 204 and the communication interface 203 communicate with each other via the bus 201. The types of the processor 202 and the memory 204 can be found in the relevant descriptions of the processor 602 and the memory 604, which will not be repeated here. The memory 204 stores executable code, and the processor 202 can read the executable code in the memory 204 to implement functions. The processor 202 can also communicate with the cloud device through the communication interface 203. For example, the processor 202 can prompt the user to input a sample set of multiple tasks through the display screen 205, and feed back the sample set of multiple tasks to the cloud device 300 through the communication interface 203.

[0184] like Figure 7 As shown, the cloud device 300 includes a bus 301, a processor 302, a communication interface 303 and a memory 304. The processor 302, the memory 304 and the communication interface 303 communicate with each other through the bus 301. Among them, the types of the processor 302 and the memory 304 can refer to the relevant descriptions of the processor 602 and the memory 604, which will not be repeated here. The memory 304 stores executable code, and the processor 302 can read the executable code in the memory 304 to implement functions, and can also communicate with the client 200 through the communication interface 303. In an embodiment of the present application, the processor 302 can implement the functions of the relationship discovery device 100. In this case, the memory 304 stores one or more modules of the information sample attribute extraction module 101, the task attribute extraction module 102, the task mixed attribute determination module 103, the task similarity measurement module 104, and the adaptation module 105 of the relationship discovery device 100.

[0185] After receiving a sample set of multiple tasks from the client 200 through the communication interface 303 , the processor 302 may call a module stored in the memory 304 to implement the task relationship determination method provided in the embodiment of the present application.

[0186] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. For example, the division of the above-described device embodiment is only a logical function division, and there can be another division manner for actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different units, or the among different units, can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0187] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0188] The above is only a specific implementation of the present application. Those skilled in the art can think of changes or substitutions based on the specific implementation provided by the present application, which should be covered within the protection scope of the present application.

Claims

1. A task relationship determination method characterized by comprising: The method comprises: determining sample attributes and corresponding attribute values in each task; determining task attributes and corresponding attribute values of each task according to the sample attributes and corresponding attribute values of each task; determining task mixed attributes and corresponding attribute values of each task according to the task attributes and corresponding attribute values of each task, and the sample attributes and corresponding attribute values of each task, the task mixed attributes being used to describe overall characteristics of the task; calculating similarity or difference degrees between the tasks according to the task mixed attributes and corresponding attribute values of each task; determining a correlation between the tasks based on the similarity or difference degrees between the tasks.

2. The method of claim 1, wherein, The determination of the task mixed attributes and corresponding attribute values of each task according to the task attributes and corresponding attribute values of each task, and the sample attributes and corresponding attribute values of each task comprises: determining target task attributes and corresponding attribute values of each task from the task attributes and corresponding attribute values of each task; determining target sample attributes and corresponding attribute values of each task from the sample attributes and corresponding attribute values of each task; determining the task mixed attributes and corresponding attribute values of each task according to the target task attributes and corresponding attribute values of each task, and the target sample attributes and corresponding attribute values of each task.

3. The method of claim 2, wherein, The determination of the target task attributes and corresponding attribute values of each task from the task attributes and corresponding attribute values of each task comprises: for any one of the tasks, selecting part of the task attributes as the target task attributes of the task, wherein one task attribute corresponds to one target task attribute, and the attribute value corresponding to the target task attribute of the task is the attribute value corresponding to the task attribute; or for any one of the tasks, performing analysis or conversion on the task attributes to generate the target task attributes of the task, and the attribute value corresponding to the target task attribute is determined after analysis of the attribute value corresponding to the task attribute.

4. The method of claim 2, wherein, The determination of the target sample attributes and corresponding attribute values of each task from the sample attributes and corresponding attribute values of each task comprises: for any one of the tasks, selecting part of the sample attributes as the target sample attributes of the task, wherein one sample attribute corresponds to one target sample attribute, and the attribute value corresponding to the target sample attribute is determined after analysis of the attribute value corresponding to the sample attribute; or for any one of the tasks, performing analysis or conversion on the sample attributes to generate the target sample attributes of the task, wherein one sample attribute corresponds to one target sample attribute, and the attribute value corresponding to the target sample attribute is determined after analysis of the attribute value corresponding to the sample attribute.

5. The method according to any one of claims 2 to 4, characterized in that, The task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value, and the task mixed attribute and the corresponding attribute value of each task are determined according to the target task attribute and the corresponding attribute value of each task, and the target sample attribute and the corresponding attribute value ​ ​ 6. The method according to any one of claims 1 to 4, characterized in that ​ ​ ​ 7. The method of claim 6, wherein, ​ ​ 8. The method of claim 7, wherein, ​ ​ ​ ​ ​ 9. A relationship discovery apparatus characterized by comprising: ​ ​ ​ a task mixed attribute determination module, configured to determine a task mixed attribute and a corresponding attribute value of each of the tasks according to the task attribute and the corresponding attribute value of each of the tasks and the sample attribute and the corresponding attribute value of each of the tasks, the task mixed attribute being used to describe an overall feature of the task; a task similarity measurement module, configured to calculate a similarity or difference between the tasks according to the task mixed attribute and the corresponding attribute value of each of the tasks; an adaptation module, configured to determine a correlation between the tasks based on the similarity or difference between the tasks.

10. The apparatus of claim 9, wherein, The task mixed attribute determination module is specifically configured to: determine a target task attribute and a corresponding attribute value of each of the tasks from the task attribute and the corresponding attribute value of each of the tasks; determine a target sample attribute and a corresponding attribute value of each of the tasks from the sample attribute and the corresponding attribute value of each of the tasks; determine a task mixed attribute and a corresponding attribute value of each of the tasks according to the target task attribute and the corresponding attribute value of each of the tasks and the target sample attribute and the corresponding attribute value.

11. The apparatus of claim 10, wherein, The task attribute extraction module is specifically configured to: for any one of the tasks, select part of the task attributes as the target task attribute of the task from the task attributes of the task, wherein one task attribute in the part of the task attributes corresponds to one target task attribute, and the corresponding attribute value of the target task attribute of the task is the corresponding attribute value of the task attribute; or for any one of the tasks, analyze or convert the task attributes of the task to generate the target task attribute of the task, and the corresponding attribute value of the target task attribute is determined after analyzing the corresponding attribute value of the task attribute of the task.

12. The apparatus of claim 10, wherein, The sample attribute extraction module is specifically configured to: for any one of the tasks, select part of the sample attributes as the target sample attribute of the task from the sample attributes of the task, wherein one sample attribute in the part of the sample attributes corresponds to one target sample attribute, and the corresponding attribute value of the target sample attribute is determined after analyzing the corresponding attribute value of the sample attribute; or for any one of the tasks, analyze or convert the sample attributes of the task to generate the target sample attribute of the task, wherein one sample attribute in the part of the sample attributes corresponds to one target sample attribute, and the corresponding attribute value of the target sample attribute is determined after analyzing the corresponding attribute value of the sample attribute.

13. The apparatus of any one of claims 10-12, wherein The task mixed attribute is specifically configured to: for any one of the tasks, take the target task attribute or the target sample attribute of the task as the task mixed attribute of the task, one target task attribute or one target sample attribute of the task corresponds to one task mixed attribute of the task, and the corresponding attribute value of one task mixed attribute of the task is determined after analyzing the corresponding attribute value of one target task attribute or one target sample attribute of the task; or Part of the target task attributes and part of the target sample attributes are selected from the target task attributes and the target sample attributes of the tasks, and a task mixed attribute of each task is determined according to the part of the target task attributes and the part of the target sample attributes. The attribute value corresponding to the task mixed attribute of the task is determined after analyzing the attribute values corresponding to the part of the target task attributes or the part of the target sample attributes.

14. The apparatus of any one of claims 9 to 12, wherein The task mixed attribute of each task and the corresponding attribute value include one or more levels of task mixed attributes and corresponding attribute values; and the task similarity measurement module is specifically configured to: According to the task mixed attribute and the corresponding attribute value of each task at the same level, the similarity or difference between the tasks at the same level is calculated.

15. The apparatus of claim 14, wherein, When determining the association relationship between the tasks based on the similarity or difference between the tasks, the adaptation module is specifically configured to: Based on the similarity or difference between each task at one or more levels, the association relationship between the tasks is determined.

16. The apparatus of claim 15, wherein, When determining the association relationship between the tasks based on the similarity between each task at one or more levels, the adaptation module is specifically configured to: A matching mapping relationship is determined, the matching mapping relationship is used to indicate the corresponding relationship between the similarity matrix at different levels and the clustering method, and the similarity matrix at one level is constructed by the similarity between the tasks at the level. Based on the matching mapping relationship, the clustering method corresponding to the similarity matrix is determined according to the similarity matrix at one or more levels. According to the similarity matrix at one or more levels, the candidate associated tasks at one or more levels are determined by the clustering method corresponding to the similarity matrix. The association relationship between the tasks is determined according to the candidate associated tasks at one or more levels.

17. A computing device, comprising: The computing device includes a processor and a memory; The memory is used to store computer program instructions; The processor is used to call the computer program instructions in the memory to execute the method in any one of claims 1 to 8.

18. A cluster of computing devices, characterized in that, The computing device cluster includes a plurality of computing devices, each computing device includes a processor and a memory; the memory in at least one of the computing devices is used to store computer program instructions; The processor in at least one of the computing devices is used to call the computer program instructions stored in the memory to execute the method in any one of claims 1 to 8.

Images