Business process prediction method, device and equipment and readable storage medium

Abstract

The present application relates to the field of machine learning, and provides a business process prediction method, device, equipment and readable storage medium, the method comprising: constructing each business process in an event log as an event trajectory; transforming the event trajectory to obtain a high-dimensional sparse feature vector, and converting the high-dimensional sparse feature vector into a low-dimensional dense feature vector through a preset embedding layer; inputting the low-dimensional dense feature vector into a business process prediction model to obtain a business process prediction result, wherein the business process prediction model is constructed based on a heterogeneous neural network. The present application realizes the acquisition of feature representations of tasks through different ways at the same time, reduces the loss of information, and improves the adaptive dynamic weighting of the weighting algorithm of the optimization part of the multi-task learning model, thereby improving the training efficiency and the parallel prediction accuracy of multiple tasks.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine learning, and in particular to a business process prediction method and device, equipment and a readable storage medium. BACKGROUND

[0002] The common business process prediction methods at present include: a single task learning model is used to predict a business process, and a homogeneous multi-task learning model is used to predict a business process.

[0003] The business process prediction method based on the single task learning model can only learn one task at a time, and when facing a complex problem, the task is usually divided into multiple independent subtasks, and the final output is obtained by integrating the results, without considering the relevance between the subtasks. Because the correlation between tasks is not considered, the common information between different tasks is not mined, so that the learning ability is reduced in the model training process.

[0004] The business process prediction method based on the homogeneous multi-task learning model assumes that the training data of each task has the same feature representation, but there are a large number of problems that do not meet such restrictions, for example, there are text data and image data related to the task. In the case of insufficient sample size, we want to use shared information to obtain a classifier with high reliability. For such problems, the homogeneous multi-task learning method cannot solve. At the same time, due to the complexity of the correlation between subtasks in reality, the commonly used multi-task learning model often has negative transfer, resulting in the phenomenon of "seesaw" of the model effect, that is, the performance of one task usually improves by damaging the performance of other tasks. SUMMARY

[0005] The present application provides a business process prediction method, device, equipment and readable storage medium to solve the technical problem of low prediction accuracy existing in the prior art business process prediction method.

[0006] The present application provides a business process prediction method, comprising:

[0007] Each business process in the event log is constructed as an event trajectory;

[0008] The event trajectory is transformed to obtain a high-dimensional sparse feature vector, and the high-dimensional sparse feature vector is converted into a low-dimensional dense feature vector through a preset embedding layer;

[0009] The low-dimensional dense feature vector is input into a business process prediction model to obtain a business process prediction result, and the business process prediction model is constructed based on a heterogeneous neural network.

[0010] According to the business process prediction method provided by the application, the event log is extracted into multi-column data containing track identification, activity identification and timestamp identification.

[0011] The event log is extracted into multi-column data containing track identification, activity identification and timestamp identification.

[0012] The multi-column data is constructed into event tracks based on preset task requirements.

[0013] According to the business process prediction method provided by the application, the event log is extracted into multi-column data containing track identification, activity identification and timestamp identification.

[0014] Each event track is converted into a high-dimensional sparse feature vector based on a one-hot encoding mode.

[0015] According to the business process prediction method provided by the application, the event log is extracted into multi-column data containing track identification, activity identification and timestamp identification.

[0016] The high-dimensional sparse feature vector is input into a preset embedding layer, and the high-dimensional sparse feature vector is converted into a low-dimensional dense feature vector through linear mapping.

[0017] According to the business process prediction method provided by the application, the event log is extracted into multi-column data containing track identification, activity identification and timestamp identification.

[0018] The expert network in the preset homogeneous multi-task learning model is replaced by a heterogeneous neural network, and a business process prediction model is constructed.

[0019] According to the business process prediction method provided by the application, the event log is extracted into multi-column data containing track identification, activity identification and timestamp identification.

[0020] According to the business process prediction method provided by the application, the event log is extracted into multi-column data containing track identification, activity identification and timestamp identification.

[0021] During the construction of the business process prediction model, adaptive dynamic weighting is used as a weighted part of model optimization.

[0022] The application also provides a business process prediction device, comprising:

[0023] An event track construction module is configured to construct each business process in an event log into an event track.

[0024] A feature vector conversion module is configured to convert the event track into a high-dimensional sparse feature vector, and convert the high-dimensional sparse feature vector into a low-dimensional dense feature vector through a preset embedding layer.

[0025] The business process prediction method comprises the following steps: inputting the low-dimensional dense feature vector into a business process prediction model to obtain a business process prediction result, wherein the business process prediction model is constructed based on a heterogeneous neural network.

[0026] The application further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor implements the business process prediction method according to any one of the preceding embodiments when executing the program.

[0027] The application further provides a non-transitory computer-readable storage medium, which stores a computer program, wherein the computer program is executed by a processor to implement the business process prediction method according to any one of the preceding embodiments.

[0028] The application provides a business process prediction method, device, equipment and readable storage medium, which replaces the expert network in the multi-task learning model with three basic component models of convolutional neural network, time recurrent neural network and attention neural network, simultaneously obtains the feature representation of the task in different ways, and improves the weighting algorithm of the optimization part of the multi-task learning model to adaptive dynamic weighting. The application converts the execution of each business process in the event log into a set of event trajectories, reconstructs the input data conforming to the prediction model, converts the reconstructed event trajectory into feature encoding, and converts it into a low-dimensional dense and interrelated vector through an embedding layer. The expert network in the multi-task learning model is replaced with three basic component models of convolutional neural network, time recurrent neural network and attention neural network, the feature representation of the task is simultaneously obtained in different ways to reduce the loss of information, the weighting algorithm of the optimization part of the multi-task learning model is improved to adaptive dynamic weighting, the training efficiency is improved, the computing resources are saved, and the parallel prediction accuracy of multiple tasks is improved. BRIEF DESCRIPTION OF DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0030] Figure 1 is one of the flowcharts of the business process prediction method provided by the application;

[0031] Figure 2 is the second flowchart of the business process prediction method provided by the application;

[0032] Figure 3 is the structural schematic diagram of the business process prediction device provided by the application;

[0033] Figure 4 is a structural schematic diagram of an electronic device provided by the present application. DETAILED DESCRIPTION

[0034] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below with reference to the drawings in the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all the other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

[0035] Please refer to Figure 1 The present application provides a business process prediction method, comprising:

[0036] Step 100, constructing each business process in the event log as an event track;

[0037] The above step 100 further comprises:

[0038] Step 110, extracting the event log as multi-column data containing track identification, activity identification and timestamp identification;

[0039] Step 120, constructing the multi-column data as an event track based on the preset task requirement.

[0040] Specifically, as Figure 2 shown, the reconstruction process of the event log is as follows:

[0041] Firstly, three-column data containing only track identification, activity identification and timestamp identification are extracted from the event log L, and the representation form can be L={CaseID, active, time}, wherein CaseID represents a process instance, active represents each activity executed in the process instance, and time represents the execution time corresponding to each activity executed in the process instance.

[0042] Then, an event log is constructed according to the task requirements, conforming to the input format of the prediction model. For example, in the task flow result prediction, an event trajectory is {(active1, time1), (active2, time2), (active3, time3)}. The input data corresponding to this event trajectory can be {[(active1, time1)], [(active1, time1), (active2, time2)], [(active1, time1), (active2, time2), (active3, time3)]}, and its corresponding label value is {active3, active3, comp}, where comp is the termination symbol, indicating that this activity sequence has been completed.

[0043] Step 200: The event trajectory is transformed to obtain a high-dimensional sparse feature vector, and the high-dimensional sparse feature vector is transformed into a low-dimensional dense feature vector through a preset embedding layer.

[0044] Step 200 above also includes:

[0045] Step 210: Based on one-hot encoding, the event trajectories are converted into high-dimensional sparse feature vectors.

[0046] Step 220: Input the high-dimensional sparse feature vector into a preset embedding layer, and transform the high-dimensional sparse feature vector into a low-dimensional dense feature vector through linear mapping.

[0047] Specifically, the process of encoding event trajectories into feature vectors is as follows:

[0048] The symbolic representation of the event log is L={σ1,σ2,σ3,…,σ n}, where σ i σ represents the trajectory of the i-th event. i = <e1,e2,e3,…,e n >(n=|σ i |), transform each event trajectory into a feature vector x = [x1, x2, x3, ..., x...]. p ], where p represents the number of samples, x i It represents a set of features.

[0049] The business process prediction task contains event activities and time features, and each feature can be processed into a one-dimensional vector or a two-dimensional matrix according to the needs of the prediction model. Among them, the activity feature adopts One-Hot encoding, that is, it is marked as 1 at the position with activity and 0 at other positions; the time interval is taken as the time feature, and the time feature is normalized by the range of all values in the event log. Since One-Hot encoding has high-dimensional sparse features, and there is no inherent relationship in the conversion process, the feature vector is input into the embedding layer, and the high-dimensional and sparse vector is converted into a low-dimensional and dense feature vector e = [e1, e2, e3, …, e n ] through linear mapping n ∈R d , where d represents the encoding dimension of the feature vector.

[0050] Step 300, input the low-dimensional dense feature vector into the business process prediction model to obtain a business process prediction result, wherein the business process prediction model is constructed based on a heterogeneous neural network.

[0051] Specifically, the construction process of the heterogeneous multi-task learning prediction model (i.e., the business process prediction model in this embodiment) is as follows:

[0052] The present application proposes the concept of a heterogeneous expert subnetwork, and replaces the expert network in the homogeneous multi-task learning model with CNN (Convolutional Neural Network), LSTM (Long Short-Term Memory) and Transformer model (a kind of neural network), which are three kinds of heterogeneous neural networks, to realize the acquisition of feature representation of the task through different ways at the same time and reduce the loss of effective information.

[0053] Specifically, the main reason for selecting the above three heterogeneous neural networks as the main expert network in the heterogeneous multi-task learning model is: (1) LSTM has good modeling ability for time series, and in the prediction of business processes, LSTM network can obtain the information state of the past, connect the information as the final state output, and better obtain the features related to the task itself by considering the context relationship; (2) the data contained in the business process instance is converted into spatial data, and the CNN network can better extract local features through convolution and pooling operations; (3) the Transformer network based on attention mechanism calculates the relationship between two events in the same event trajectory, and in this process, the analysis between the two events is independent of the distance, only related to the event vector itself, so there is no long-term dependence problem.

[0054] In the embodiment, the expert network in the multi-task learning model is replaced by three basic component models of convolutional neural network, time recurrent neural network and attention neural network, the feature representation of the task is obtained in different ways, and the weighting algorithm of the multi-task learning model optimization part is improved to adaptive dynamic weighting. The application converts the execution of each business process in the event log into a set of event trajectories, reconstructs the input data conforming to the prediction model, converts the reconstructed event trajectories into feature encoding, and converts them into low-dimensional dense and related vectors through an embedding layer, replaces the expert network in the multi-task learning model with three basic component models of convolutional neural network, time recurrent neural network and attention neural network, realizes the feature representation of the task in different ways at the same time, reduces the loss of information, and improves the training efficiency, saves the computing resources, and improves the parallel prediction accuracy of multiple tasks.

[0055] In one embodiment, the business process prediction method provided by the embodiment of the application can further include:

[0056] In step 400, the expert network in the preset homogeneous multi-task learning model is replaced by a heterogeneous neural network to construct a business process prediction model, and the heterogeneous neural network includes a time recurrent neural network, a convolutional neural network and an attention neural network.

[0057] The business process prediction method provided by the embodiment of the application can further include:

[0058] In step 500, when constructing the business process prediction model, adaptive dynamic weighting is used as the weighting part of the model optimization.

[0059] Specifically, the application proposes a heterogeneous multi-task learning model based on deep learning to predict business processes. The application proposes the concept of a heterogeneous expert subnetwork, replaces the expert network in the multi-task learning model with three basic component models of convolutional neural network, time recurrent neural network and attention neural network, and obtains the feature representation of the task in different ways to reduce the loss of information and improve the parallel prediction accuracy of multiple tasks.

[0060] The application improves the weighting algorithm of the multi-task learning model optimization part to adaptive dynamic weighting. Due to the introduction of multiple task labels, multiple Losses (losses) are brought in. In the past, the Losses for multiple tasks were weighted according to a specified proportion. In this case, only one proportion can be verified at a time. To get the optimal proportion, multiple rounds of experiments are needed, which is costly and it is not easy to find the optimal solution.

[0061] The adaptive dynamic weighting can automatically learn the loss weighting ratio, and the weighting ratio is dynamically adjusted in the training process, thereby improving the training efficiency, saving the computing resources, and more easily finding the optimal state, and effectively alleviating the problems in the prior art.

[0062] The adaptive dynamic weighting can automatically learn the loss weighting ratio, and the weighting ratio is dynamically adjusted in the training process, thereby improving the training efficiency, saving the computing resources, and more easily finding the optimal state, and effectively alleviating the problems in the prior art.

[0063] The business process prediction device provided by the present application is described below, and the business process prediction device described below can be correspondingly referred to the business process prediction method described above.

[0064] Please refer to Figure 3 The present application also provides a business process prediction device, comprising:

[0065] An event trajectory construction module 301 is configured to construct each business process in an event log as an event trajectory;

[0066] A feature vector conversion module 302 is configured to convert the event trajectory to obtain a high-dimensional sparse feature vector, and convert the high-dimensional sparse feature vector to a low-dimensional dense feature vector through a preset embedding layer;

[0067] A business process prediction module 303 is configured to input the low-dimensional dense feature vector into a business process prediction model to obtain a business process prediction result, and the business process prediction model is constructed based on a heterogeneous neural network.

[0068] Optionally, the event trajectory construction module comprises:

[0069] An event log extraction unit is configured to extract an event log as multi-column data containing a trajectory identifier, an activity identifier and a timestamp identifier;

[0070] An event trajectory construction unit is configured to construct the multi-column data as an event trajectory based on a preset task requirement.

[0071] Optionally, the feature vector conversion module comprises:

[0072] An event trajectory conversion unit is configured to convert each event trajectory into a high-dimensional sparse feature vector based on a one-hot encoding mode.

[0073] Optionally, the feature vector conversion module further comprises:

[0074] A feature vector conversion unit is configured to input the high-dimensional sparse feature vector into a preset embedding layer, and convert the high-dimensional sparse feature vector into a low-dimensional dense feature vector through linear mapping.

[0075] Optionally, the business process prediction device further includes:

[0076] The business process prediction model building module is used to replace the expert network in the preset homogeneous multi-task learning model with a heterogeneous neural network to build a business process prediction model.

[0077] Optionally, the heterogeneous neural network includes a time-recurrent neural network, a convolutional neural network, and an attention neural network.

[0078] Optionally, the business process prediction method further includes:

[0079] The adaptive dynamic weighting module is used to incorporate adaptive dynamic weighting as a weighting component in the model optimization when constructing the business process prediction model.

[0080] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include a processor 410, a communications interface 420, a memory 430, and a communication bus 440. The processor 410, communications interface 420, and memory 430 communicate with each other via the communication bus 440. The processor 410 can call logical instructions from the memory 430 to execute a business process prediction method.

[0081] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0082] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the business process prediction methods provided by the methods described above.

[0083] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0084] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0085] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A business process prediction method characterized by, The method comprises the following steps: constructing each business process in an event log into an event trajectory; transforming the event trajectory to obtain a high-dimensional sparse feature vector, and converting the high-dimensional sparse feature vector into a low-dimensional dense feature vector through a preset embedding layer; replacing an expert network in a preset homogeneous multi-task learning model with a heterogeneous neural network to construct a business process prediction model; during the construction of the business process prediction model, using adaptive dynamic weighting as a weighting part of model optimization; inputting the low-dimensional dense feature vector into the business process prediction model to obtain a business process prediction result, wherein the business process prediction model is constructed based on a heterogeneous neural network; and the heterogeneous neural network comprises a time recurrent neural network, a convolutional neural network and an attention neural network.

2. The business process forecasting method of claim 1, wherein, The step of constructing each business process in an event log into an event trajectory comprises the following steps: extracting an event log into multi-column data containing trajectory identification, activity identification and timestamp identification; constructing the multi-column data into an event trajectory based on a preset task requirement.

3. The business process forecasting method of claim 1, wherein, The step of transforming the event trajectory to obtain a high-dimensional sparse feature vector comprises the following step: transforming each event trajectory into a high-dimensional sparse feature vector based on a one-hot encoding mode.

4. The business process prediction method of claim 3, wherein, The step of converting the high-dimensional sparse feature vector into a low-dimensional dense feature vector through a preset embedding layer comprises the following step: inputting the high-dimensional sparse feature vector into a preset embedding layer, and converting the high-dimensional sparse feature vector into a low-dimensional dense feature vector through linear mapping.

5. A business process prediction apparatus characterized by comprising: The method comprises the following steps: an event trajectory construction module is configured to construct each business process in an event log into an event trajectory; a feature vector conversion module is configured to transform the event trajectory to obtain a high-dimensional sparse feature vector, and convert the high-dimensional sparse feature vector into a low-dimensional dense feature vector through a preset embedding layer; a business process prediction model construction module is configured to replace an expert network in a preset homogeneous multi-task learning model with a heterogeneous neural network to construct a business process prediction model; an adaptive dynamic weighting module is configured to use adaptive dynamic weighting as a weighting part of model optimization during the construction of the business process prediction model; a business process prediction module is configured to input the low-dimensional dense feature vector into the business process prediction model to obtain a business process prediction result, wherein the business process prediction model is constructed based on a heterogeneous neural network; and the heterogeneous neural network comprises a time recurrent neural network, a convolutional neural network and an attention neural network.

6. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the business process prediction method according to any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the business process prediction method according to any one of claims 1 to 4.

Images