Flow node data prediction method, device and equipment and readable storage medium

Abstract

The invention provides a flow node data prediction method and device, computer equipment and a computer readable storage medium, and belongs to the field of artificial intelligence processing. The data prediction method for the process nodes comprises the steps of obtaining process node case data of each time span in a historical time period; performing coding processing on the process node case data of each time span to obtain coding information of each time span; performing multi-granularity extraction processing according to the coding information of each time span through a trained granularity feature extraction module to obtain a multi-granularity coding feature tree of the historical time period; and performing prediction according to the multi-granularity coding feature tree to obtain process node case data of a target time span. The prediction precision of the process node case data can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device, computer equipment and computer-readable storage medium for predicting data at a process node. Background Art

[0002] The trend data of business process nodes can help allocate business processing resources, reduce the waste of business processing resources and improve business processing efficiency. Therefore, the accurate prediction of the trend data of business process nodes has important guiding significance for the allocation of business processing resources. For example, the prediction of trend data (such as the total number of cases entering the reporting node on a certain day) of key claim process nodes (such as reporting nodes, case filing nodes, review nodes, payment nodes, etc.) in the insurance claim process can help allocate processing resources for the insurance claim process. On the one hand, it can avoid the problem of resource waste caused by excessive allocation of processing resources to a certain claim process node, and on the other hand, it can avoid the problem of timeout processing of the claim process node caused by too little allocation of processing resources to a certain claim process node.

[0003] In the related art, the historical data of the process nodes are used to directly fit and predict the trend data of the business process nodes. However, the inventors of the present application found in the actual research and development process that there are certain differences between the trend data obtained by directly fitting and predicting the historical data of the process nodes and the actual data, and the prediction accuracy of the trend data of the business process nodes is relatively low. Summary of the invention

[0004] The present application provides a data prediction method, device, computer equipment and computer-readable storage medium for a process node, which belongs to the field of artificial intelligence technology and can improve the prediction accuracy of case data at process nodes.

[0005] In a first aspect, the present application provides a data prediction method for a process node, the method comprising:

[0006] Obtain process node case data for each time span within the historical time period;

[0007] Encoding the process node case data of each time span to obtain encoding information of each time span;

[0008] By using the trained granularity feature extraction module, multi-granularity extraction processing is performed according to the coding information of each time span to obtain a multi-granularity coding feature tree of the historical time period;

[0009] Prediction is performed based on the multi-granularity coding feature tree to obtain process node case data of the target time span.

[0010] In a second aspect, the present application provides a data prediction device for a process node, the data prediction device for a process node comprising:

[0011] An acquisition unit is used to acquire process node case data for each time span in a historical time period;

[0012] An encoding unit, used for encoding the process node case data of each time span to obtain encoding information of each time span;

[0013] An extraction unit, configured to perform multi-granularity extraction processing according to the coding information of each time span through a trained granularity feature extraction module to obtain a multi-granularity coding feature tree of the historical time period;

[0014] The prediction unit is used to make predictions based on the multi-granularity coding feature tree to obtain process node case data of a target time span.

[0015] In a third aspect, the present application also provides a computer device, comprising a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement the data prediction method of the process node when executing the computer program.

[0016] In a fourth aspect, the present application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is loaded by a processor to execute the data prediction method for the process node.

[0017] The present application first encodes the process node case data of each time span in the historical time period to obtain the encoding information of each time span; then uses the trained granular feature extraction module to perform multi-granularity extraction processing according to the encoding information of each time span to obtain a multi-granularity encoding feature tree for the historical time period, which is used to predict the process node case data of the target time span. On the first hand, since the multi-granularity extraction processing is performed according to the encoding information of each time span to obtain the multi-granularity encoding feature tree of the historical time period, a multi-level encoding information sequence representation from fine-grained time to coarse-grained time can be achieved, thereby capturing and encoding information at different time granularities, improving the information capture capability of the process node case data of the historical time period, and further improving the prediction accuracy of the process node case data. Secondly, multi-granularity extraction processing can extract features from fine-grained time to coarse-grained time dimensions, and can effectively integrate global and local features. It has stronger expression capabilities for process node case data with nonlinearity and long-term temporal dependencies, and avoids directly using historical data of process nodes for fitting and prediction, thereby reducing the problem of low prediction accuracy of process node case data caused by direct fitting and prediction based on historical data of process nodes, and to a certain extent improves the prediction accuracy of trend data of business process nodes. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 This is a flow chart of a data prediction method for a process node provided in an embodiment of the present application;

[0020] Figure 2 This is a flowchart of an embodiment of step 103 in the embodiment of the present application;

[0021] Figure 3 It is a schematic diagram of an explanation framework for extracting a multi-granularity coding feature tree in an embodiment of the present application;

[0022] Figure 4 This is a schematic diagram of a principle framework of a case data prediction model in an embodiment of the present application;

[0023] Figure 5 A schematic diagram of the structure of an embodiment of a data prediction device for a process node provided in an embodiment of the present application;

[0024] Figure 6 It is a schematic block diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0026] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0027] In the description of the embodiments of the present application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0028] In order to enable any technician in the field to implement and use the application, the following description is given. In the following description, details are listed for the purpose of explanation. It should be understood that those of ordinary skill in the art can recognize that the application can also be implemented without using these specific details. In other examples, the known process will not be elaborated in detail to avoid unnecessary details that make the description of the embodiment of the application obscure. Therefore, the application is not intended to be limited to the embodiments shown, but is consistent with the widest range of principles and features disclosed in accordance with the embodiments of the application.

[0029] Embodiments of the present application provide a method, apparatus, computer device, and computer-readable storage medium for predicting data at a process node.

[0030] The executor of the data prediction method for the process node of the embodiment of the present application may be the data prediction device for the process node provided in the embodiment of the present application, or the computer device provided in the embodiment of the present application, wherein the data prediction device for the process node may be implemented in hardware or software.

[0031] Some embodiments of the present application are described in detail below in conjunction with the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0032] See also Figure 1 , Figure 1 1 is a flow chart of a data prediction method for a process node provided in an embodiment of the present application. The data prediction method for a process node includes steps 101 to 104, wherein:

[0033] 101. Obtain process node case data for each time span within the historical time period.

[0034] Among them, the process node is a process node in the business process (such as the insurance claim process). The process node can be determined according to the actual business process scenario. For example, it can be a process node in the insurance claim process (such as the reporting node, the case filing node, the review node, the compensation node, etc.).

[0035] Among them, the process node case data can be the total number of cases, case processing time, etc. of the process nodes (such as the reporting node, filing node, review node, and payment node in the insurance claim process). The process node case data is determined according to the actual business scenario requirements. For example, in the scenario where the "data prediction task of the process node is to predict the total number of cases at the reporting node in the insurance claim process on a certain day", the process node case data is the total number of cases at the reporting node; for another example, in the scenario where the "data prediction task of the process node is to predict the case processing time required for the filing node in the insurance claim process on a certain day", the process node case data is the case processing time of the reporting node.

[0036] The historical time period may be the past n hours, n days, n weeks, n months, etc. There is no restriction on the historical time period.

[0037] The time span is the period for counting process node case data, for example, it can be 1 hour, 1 day, etc.

[0038] For example, taking "the process node is the filing node in the insurance claims process, the time span is 1 day, the historical time period is the past 7 days, and the process node case data is the total number of cases" as an example, the total number of cases entering the filing node every day in the past 7 days (such as January 1, January 2, January 3, January 4, January 5, January 6, January 7) can be counted, thereby obtaining the process node case data (i.e. the total number of cases in the filing node) for each time span (every day) in the historical time period (the past 7 days).

[0039] 102. Encode the process node case data of each time span to obtain the encoding information of each time span.

[0040] There are many ways to perform encoding processing in step 102, illustratively including:

[0041] (1) Encoding is performed using the process node case data to obtain the encoding information of each time span. At this time, step 102 may specifically include the following steps 1021A to 1022A:

[0042] 1021A. Encode the process node case data of each time span to obtain the case data code of each time span.

[0043] Exemplarily, the process node case data of each time span can be encoded using a One-Hot vector model, a Word2Vec model, or an n-gram model to obtain the case data encoding Xi for each time span i.

[0044] For example, taking "the process node is the case filing node in the insurance claims process, the time span is 1 day, the historical time period is the past 7 days, and the process node case data is the total number of cases" as an example, assuming that the total number of cases at the case filing node in the past 7 days (such as January 1, January 2, January 3, January 4, January 5, January 6, and January 7) are 10, 20, 30, 40, 50, 60, and 70 respectively, then the One-Hot vector model can be used to encode the total number of cases at the case filing node 10, 20, 30, 40, 50, 60, and 70 respectively, and obtain case data code X1, case data code X2, case data code X3, case data code X4, case data code X5, case data code X6, and case data code X7.

[0045] For another example, taking "the process node is the case filing node in the insurance claims process, the time span is 1 day, the historical time period is the past 7 days, and the process node case data is the case processing time" as an example, assuming that the case processing time of the case filing node in the past 7 days (such as January 1, January 2, January 3, January 4, January 5, January 6, January 7) are: 1, 2, 3, 4, 5, 6, 7 respectively, then the One-Hot vector model can be used to encode the case processing time 1, 2, 3, 4, 5, 6, 7 of the case filing node respectively, and obtain case data code X1, case data code X2, case data code X3, case data code X4, case data code X5, case data code X6, case data code X7.

[0046] 1022A. Encode the case data of each time span as the encoding information of each time span.

[0047] (2) Encoding is performed using the process node case data and sequence information to obtain the encoding information of each time span. At this time, step 102 may specifically include the following steps 1021B to 1023B:

[0048] 1021B. Encode the process node case data of each time span to obtain the case data code of each time span.

[0049] Step 1021B is implemented similarly to step 1021A. For details, please refer to the relevant instructions above and will not be repeated here.

[0050] 1022B. Encode the sequence information of each time span to obtain the sequence code of each time span.

[0051] In some embodiments, first, a sequence information is assigned to each time span using a preset sorting rule; then, the sequence information of each time span can be encoded using a One-Hot vector model, a Word2Vec model, or an n-gram model to obtain a sequence code Zi for each time span i.

[0052] For example, take "the process node is the filing node in the insurance claim process, the time span is 1 day, the historical time period is the past 7 days, and the process node case data is the total number of cases" as an example. Assuming that the total number of cases at the filing node in the past 7 days (such as January 1, January 2, January 3, January 4, January 5, January 6, and January 7) are 10, 20, 30, 40, 50, 60, and 70 respectively, first, use the preset sorting rules to assign a sequence information to each time span, for example, January 1, January 2 The sequence information of January 1, January 3, January 4, January 5, January 6, and January 7 are the sorting codes p0, p1, p2, p3, p4, p5, and p6, respectively; then, the One-Hot vector model can be used to encode the sequence information assigned to each time span (i.e., sorting codes p0, p1, p2, p3, p4, p5, and p6) to obtain sequence code Z1, sequence code Z2, sequence code Z3, sequence code Z4, sequence code Z5, sequence code Z6, and sequence code Z7.

[0053] In some embodiments, in order to improve the spatial characteristic capture capability, the sequential information of each time span may be encoded using a rotational position encoding method. In this case, step 1022B may specifically include: assigning a value to each time span according to a preset sorting rule to obtain the sequential information of each time span; encoding the sequential information of each time span through a rotational position encoding module to obtain a rotational position encoding of each time span as the sequential encoding of each time span. For example, the ROPE (Rotary Positional Encoding) method may be used to first calculate the position vector for the sequential information of each time span, and then perform a rotation operation to obtain a rotational position encoding of each time span as the sequential encoding of each time span. This enables the case data prediction model to better understand and utilize the time sequence and relative distance of elements in the information sequence (such as the target encoding information sequence, the sub-information sequence), and improve the ability to capture periodicity and trends in the information sequence with time sequence information (such as the target encoding information sequence, the sub-information sequence).

[0054] 1023B. Concatenate the case data code of each time span and the sequence code of each time span to obtain the coding information of each time span.

[0055] For example, the case data codes for each time span in the past 7 days (such as January 1, January 2, January 3, January 4, January 5, January 6, January 7) are X1, X2, X3, X4, X5, X6, X7, and the sequential codes are Z1, Z2, Z3, Z4, Z5, Z6, Z7. Then, the coding information of each time span (such as January 1, January 2, January 3, January 4, January 5, January 6, January 7) can be obtained in sequence: [X1, Z1], [X2, Z2], …, [X7, Z7], and so on.

[0056] (3) Encoding is performed using the process node case data and process node case attributes to obtain the encoding information of each time span. At this time, step 102 may specifically include the following steps 1021C to 1024C:

[0057] 1021C. Encode the process node case data of each time span to obtain the case data code of each time span.

[0058] Step 1021C is implemented similarly to step 1021A. For details, please refer to the relevant instructions above and will not be repeated here.

[0059] 1022C. Encode the case attributes of the process nodes of each time span to obtain the case attribute code of each time span.

[0060] Among them, case attributes can be customer attributes, accident types, case complexity, customer cooperation level, etc. By introducing case attributes as auxiliary information for process node case data prediction, nonlinear and uncertainty problems can be reduced, and the ability to predict nonlinear and uncertain case data can be improved. Reduce the complexity and heterogeneity of business processes (such as claims processes), such as reducing the problem that different types of insurance (such as auto insurance, health insurance, life insurance, etc.) and different claims cases have different processing processes and time spans, and improve the prediction accuracy of case data prediction models for different cases.

[0061] In addition, in order to address the problems of incomplete data and uneven data quality in actual business processes, such as incomplete data records in the claims process (for example, missing reporting time, filing time, review time, payment time, etc. at key time nodes, and data entry errors, inconsistent formats, etc.), in order to reduce the impact on the prediction accuracy and difficulty of case data, a missing data processing strategy (such as supplementing according to default values) can be designed to enable the model to effectively deal with incomplete data in actual business and improve the adaptability of the model's case data prediction.

[0062] Exemplarily, the case attributes of the process nodes of each time span can be encoded using a One-Hot vector model, a Word2Vec model, or an n-gram model to obtain the case data encoding Yi of each time span i.

[0063] For example, taking "the process node is the case filing node in the insurance claims process, the time span is 1 day, the historical time period is the past 7 days, and the process node case attribute is the total number of cases" as an example, assuming that the process node case attributes of the case filing node in the past 7 days (such as January 1, January 2, January 3, January 4, January 5, January 6, January 7) are: attribute 1, attribute 2, attribute 3, attribute 4, attribute 5, attribute 6, attribute 7, then the One-Hot vector model can be used to encode attribute 1, attribute 2, attribute 3, attribute 4, attribute 5, attribute 6, and attribute 7 respectively to obtain case attribute code Y1, case attribute code Y2, case attribute code Y3, case attribute code Y4, case attribute code Y5, case attribute code Y6, and case attribute code Y7.

[0064] 1023C. Concatenate the case data code of each time span and the case attribute code of each time span to obtain the coding information of each time span.

[0065] For example, the case data codes for each time span in the past 7 days (such as January 1, January 2, January 3, January 4, January 5, January 6, and January 7) are X1, X2, X3, X4, X5, X6, and X7, and the case attribute codes are Y1, Y2, Y3, Y4, Y5, Y6, and Y7, respectively. Then, the coding information for each time span (such as January 1, January 2, January 3, January 4, January 5, January 6, and January 7) can be obtained in sequence: [X1, Y1], [X2, Y2], …, [X7, Y7], and so on.

[0066] (4) Encoding is performed using process node case data, process node case attributes, and sequence information to obtain encoding information for each time span. At this time, step 102 may specifically include the following steps 1021D to 1024D:

[0067] 1021D. Encode the process node case data of each time span to obtain the case data code of each time span.

[0068] Step 1021D is implemented similarly to step 1021A. For details, please refer to the relevant instructions in the previous text and will not be repeated here.

[0069] 1022D. Encode the case attributes of the process nodes of each time span to obtain the case attribute code of each time span.

[0070] Step 1022D is similar to step 1022C in implementation. For details, please refer to the relevant instructions in the previous text and will not be repeated here.

[0071] 1023D. Encode the sequence information of each time span to obtain the sequence code of each time span.

[0072] Step 1023D is implemented similarly to step 1022B. For details, please refer to the relevant instructions in the previous text and will not be repeated here.

[0073] 1024D. Concatenate the case data code of each time span, the case attribute code of each time span and the sequence code of each time span to obtain the coding information of each time span.

[0074] For example, the case data codes for each time span in the past 7 days (such as January 1, January 2, January 3, January 4, January 5, January 6, January 7) are X1, X2, X3, X4, X5, X6, X7, the case attribute codes are Y1, Y2, Y3, Y4, Y5, Y6, Y7, and the sequence codes are Z1, Z2, Z3, Z4, Z5, Z6, Z7. Then, the coding information of each time span (such as January 1, January 2, January 3, January 4, January 5, January 6, January 7) can be obtained in sequence: [X1, Y1, Z1], [X2, Y2, Z2], …, [X7, Y7, Z7], and so on.

[0075] Therefore, on the first hand, by using the process node case attributes of each time span to encode and obtain the coding information of each time span, the process node case attributes (such as customer attributes, accident type, etc.) can be used as auxiliary information to reduce the interference of the process node case attributes on the process node case data, thereby improving the prediction accuracy of the process node case data. On the second hand, by using the sequential encoding of each time span to encode and obtain the coding information of each time span, the temporal characteristics of the process node case attributes can be captured to improve the prediction accuracy of the process node case data. On the third hand, by encoding the sequential information of each time span through the rotational position encoding method, the process node case data and / or process node case attribute embedding of each time span can be dynamically adjusted through the complex rotation matrix, which effectively solves the limitations of traditional position encoding in long sequences, maintains the relative time sequence relationship of the process node case data and / or process node case attributes of each time span, improves the ability to capture the sequence time sequence characteristics, and thereby improves the prediction accuracy of the process node case data.

[0076] 103. Perform multi-granularity extraction processing according to the coding information of each time span through the trained granularity feature extraction module to obtain a multi-granularity coding feature tree of the historical time period.

[0077] There are many ways to implement step 103, illustratively including:

[0078] (1) For example, Figure 2 As shown, step 103 may specifically include the following steps 1031 to 1034:

[0079] 1031. Concatenate the coded information of each time span in chronological order to obtain a target coded information sequence of the historical time period.

[0080] 1032. Using the trained granularity feature extraction module, divide the target coded information sequence to obtain the sth granularity sub-information sequence of each historical time period.

[0081] 1033. Divide the sth granularity sub-information sequence through the granularity feature extraction module to obtain s+1th granularity sub-information sequences of the historical time period.

[0082] 1034. Perform fusion processing on each s-th granularity sub-information sequence of the historical time period and each s+1-th granularity sub-information sequence of the historical time period to obtain a multi-granularity coding feature tree of the historical time period.

[0083] like Figure 3 As shown, taking "the process node is the case filing node in the insurance claim process, the time span is 1 day, the historical time period is the past 7 days, and the process node case attribute is the total number of cases" as an example, assuming that the coding information of each time span in the past 7 days (such as January 1, January 2, January 3..., January 6, January 8) is: R1, R2, R3, R4, R5, R6, R7, R8, first, in step 1032, the coding information (i.e., R1, R2, R3, R4, R5, R6, R7, R8) of each time span (such as January 1, January 2, January 3, January 4, January 5, January 6, January 7, January 8) is sequentially spliced ​​together to obtain a spliced ​​coding information sequence (such as {R1, R2, R3, R4, R5, R6, R7, R8}) as the target coding information sequence for the historical time period.

[0084] Then, in steps 1032 to 1033, the target coded information sequence (such as {R1, R2, R3, R4, R5, R6, R7, R8}) is first divided by the trained granularity feature extraction module to obtain the s=1th granularity sub-information sequence of each historical time period (reference Figure 3 , for example, two s=1 granularity sub-information sequences A1={R1, R2, R3, R4}, A2={R5, R6, R7, R8}) can be obtained.

[0085] Next, in steps 1032 to 1033, the s=1th granularity sub-information sequence is divided by the trained granularity feature extraction module to obtain the s+1=2th granularity sub-information sequences of the historical time period (reference Figure 3For example, the s=1th granularity sub-information sequence A1={R1, R2, R3, R4} can be divided into two s+1=2th granularity sub-information sequences B1={R1, R2}, B2={R1, R2}, and so on. The s=1th granularity sub-information sequence A2={R5, R6, R7, R8} can be divided into two s+1=2th granularity sub-information sequences B3, B4).

[0086] By analogy, the trained granularity feature extraction module is used to divide the sth granularity sub-information sequence to obtain each s+1th granularity sub-information sequence of the historical time period, until the division stop condition is met, and each s=1th granularity sub-information sequence, each s+1=2th granularity sub-information sequence, ..., each S-1th granularity sub-information sequence, each Sth granularity sub-information sequence of the historical time period is obtained. Among them, 1≤s≤S-1.

[0087] Finally, through step 1034, each s=1 granularity sub-information sequence, each s+1=2 granularity sub-information sequence, ..., each S-1 granularity sub-information sequence, and each S granularity sub-information sequence of the historical time period are fused to obtain a multi-granularity coding feature tree for the historical time period. For example, the target coding information sequence of the historical time period is taken as the root node, and a sub-information sequence of the historical time period (such as R, A1, A2, B1, B2, B3, B4) is taken as a tree node. According to the hierarchical relationship of the divided granularity, each tree node (tree node of the s granularity) is respectively connected with its previous granularity node (i.e., the tree node of the s-1 granularity) and its next granularity node (i.e., the tree node of the s+1 granularity); wherein, the previous granularity node of the tree node of the s=1 granularity is the root node (i.e., the target coding information sequence). The structure of the multi-granularity coding feature tree finally obtained by fusion is as follows: Figure 3 At this time, the obtained multi-granularity coding feature tree contains the coding information of each node (such as each s=1 granularity sub-information sequence, each s+1=2 granularity sub-information sequence, ..., each S-1 granularity sub-information sequence, each S granularity sub-information sequence), and the upper and lower granularity hierarchical relationship between each node.

[0088] (2) In order to achieve cross-scale information fusion, information transmission is ensured while reducing computational complexity. Step 103 may specifically include: taking the target coding information sequence as the root node and the sub-information sequence as the tree node, establishing a preliminary coding feature tree for the historical time period; performing connection processing on each node in the preliminary coding feature tree through a fully connected layer to obtain a fully connected post-coding feature tree for the historical time period; performing convolution processing on each node in the fully connected post-coding feature tree according to a preset convolution kernel size and a preset convolution step size to obtain a multi-granularity coding feature tree for the historical time period. Therefore, since each node in the preliminary coding feature tree is connected through a fully connected layer, the number of data processing dimensions of each node can be reduced, thereby reducing the amount of calculation, and thus the amount of data processing of the convolution operation can be reduced, thereby improving the data processing efficiency when cross-scale information fusion is achieved by using the convolution operation.

[0089] 104. Predictions are made based on the multi-granularity coding feature tree to obtain process node case data of the target time span.

[0090] See also Figure 4 , for example, in this embodiment, Figure 4 The case data prediction model shown is used to calculate the process node case data of the target time span, such as Figure 4 As shown, the process node case data, process node case attributes, and sequence information are encoded to obtain the encoding information of each time span; then they are input into the trained granular feature extraction module for multi-granular extraction processing to obtain a multi-granular encoding feature tree for the historical time period; the multi-granular encoding feature tree is then input into the trained attention feature extraction module, and the attention encoding features of the historical time period are extracted through the trained attention feature extraction module; finally, through the prediction module, the process node case data of the target time span (for example, in the next 1 day) is obtained (such as the total number of cases at the filing node in the insurance claims process).

[0091] There are many ways to implement step 104, illustratively including:

[0092] (1) In some embodiments, a multi-granularity coding feature tree is directly used for prediction to obtain process node case data of a target time span.

[0093] (2) In some embodiments, in order to improve the prediction accuracy of process node case data, such as Figure 4 As shown, an attention feature extraction module is introduced to extract the attention information of the node to learn the weight distribution between nodes of different granularities and automatically focus on the granularity information that is most valuable for predicting case data at the process node. At this time, step 104 may specifically include the following steps 1041 to 1044:

[0094] 1041. Obtain an information sequence of adjacent nodes of the same granularity of each node in the multi-granularity coding feature tree, an information sequence of nodes of a previous granularity of each node, and an information sequence of nodes of a next granularity of each node.

[0095] 1042. Through the trained attention feature extraction module, weight extraction processing is performed based on the information sequence of adjacent nodes of the same granularity, the information sequence of the node of the previous granularity, and the information sequence of the node of the next granularity to obtain the attention coefficient of each node.

[0096] Exemplarily, the attention feature extraction module can calculate the attention coefficient of each node in the multi-granularity coding feature tree according to the following formula 1:

[0097]

[0098] In formula 1, represents the lth node at granularity s; Indicates the set of nodes that the lth node on the granularity s can follow, including: ① On the same granularity s, the node The adjacent nodes that can be paid attention to (at this time, the model parameters learned by the attention feature extraction module can set how much information each node pays attention to the adjacent nodes at the same granularity), ② at the next granularity s+1, the node The child nodes that can be paid attention to (at this time, the model parameters learned by the attention feature extraction module can set how much information each node pays attention to the adjacent nodes at a finer granularity), ③ At the previous granularity s-1, the node The direct parent node that can be paid attention to (at this time, the model parameters learned by the attention feature extraction module can set how much information each node pays attention to the adjacent nodes at a coarser granularity).

[0099] 1043. By using the trained attention feature extraction module, information extraction processing is performed according to the attention coefficient of each node and the information sequence of each node to obtain the attention coding features of the historical time period.

[0100] Therefore, on the one hand, through the trained granular feature extraction module, multi-granular extraction processing is performed according to the encoded information of each time span to obtain a multi-granular encoding feature tree of the historical time period, which can realize the multi-level encoding information sequence representation from fine-grained time to coarse-grained time, thereby capturing and encoding information at different time granularities. On the second hand, through the trained attention feature extraction module, the weight distribution between nodes of different granularities can be learned, so that on the one hand, it can learn to pay attention to the information sequence between nodes of different granularities to capture global context time series information, and on the other hand, it can learn to pay attention to the information sequence between nodes of the same granularity to capture local time series information, improve the ability to capture long-term and short-term dependent information, and then automatically focus on the most valuable granular information for predicting process node case data, thereby improving the prediction accuracy of process node case data.

[0101] 1044. Predictions are made based on the attention coding features to obtain process node case data of the target time span.

[0102] For example, Figure 4 As shown, first, the attention feature extraction module extracts attention features according to the multi-granularity coding feature tree to obtain the attention coding features of the historical time period; then, the attention coding features of the historical time period are input into the prediction module, and the prediction module predicts according to the attention coding features of the historical time period to obtain the process node case data of the target time span. Among them, the prediction module can be a Transformer model or a fully connected layer.

[0103] Furthermore, regularization and loss function optimization strategies can be introduced to ensure the generalization ability and prediction accuracy of the model.

[0104] From the above content, it can be seen that by first encoding the process node case data of each time span in the historical time period, the encoding information of each time span is obtained; then, through the trained granular feature extraction module, multi-granularity extraction processing is performed according to the encoding information of each time span to obtain a multi-granularity encoding feature tree of the historical time period, which is used to predict the process node case data of the target time span. On the one hand, since the multi-granularity encoding feature tree of the historical time period is obtained by multi-granularity extraction processing according to the encoding information of each time span, a multi-level encoding information sequence representation from fine-grained time to coarse-grained time can be realized, thereby capturing and encoding information at different time granularities, improving the information capture capability of the process node case data of the historical time period, and thus improving the prediction accuracy of the process node case data. Avoid the problem of resource waste caused by excessive allocation of processing resources to a certain claim process node, and avoid the problem of overtime processing of the claim process node caused by insufficient allocation of processing resources to a certain claim process node.

[0105] Secondly, multi-granularity extraction processing can extract features from fine-grained time to coarse-grained time dimensions, and can effectively integrate global and local features. It has stronger expression capabilities for process node case data with nonlinearity and long-term temporal dependencies, and avoids directly using historical data of process nodes for fitting and prediction, thereby reducing the problem of low prediction accuracy of process node case data caused by direct fitting and prediction based on historical data of process nodes, and to a certain extent improves the prediction accuracy of trend data of business process nodes.

[0106] Thirdly, by accurately predicting each process node of a business process (such as the insurance claims process), business managers (such as insurance companies) can allocate resources more effectively and reduce unnecessary delays, thereby shortening processing cycles and improving customer satisfaction.

[0107] Fourthly, the case data model based on the rotational position encoding module and the attention feature extraction module can quickly and accurately predict case data at business process nodes, thereby significantly improving the efficiency of business processes (such as claims processes) and directly improving customer satisfaction.

[0108] Fifthly, through the case data model based on modules and attention feature extraction modules, we can deeply explore the multi-level time dependencies in business process nodes (claims process) while ensuring efficient calculation, and improve the prediction accuracy of case data at process nodes.

[0109] In addition, in order to better implement the data prediction method for the process node in the embodiment of the present application, based on the data prediction method for the process node, the embodiment of the present application also provides a data prediction device for the process node, such as Figure 5 FIG. 1 is a schematic diagram of a structure of a data prediction device for a process node according to an embodiment of the present application. The data prediction device 500 for a process node includes:

[0110] The acquisition unit 501 is used to acquire the process node case data of each time span in the historical time period;

[0111] The encoding unit 502 is used to encode the process node case data of each time span to obtain the encoding information of each time span;

[0112] The extraction unit 503 is used to perform multi-granularity extraction processing according to the coding information of each time span through the trained granularity feature extraction module to obtain a multi-granularity coding feature tree of the historical time period;

[0113] The prediction unit 504 is used to perform prediction according to the multi-granularity coding feature tree to obtain process node case data of a target time span.

[0114] In some embodiments, the extraction unit 503 is specifically used to:

[0115] According to the chronological order, the coding information of each time span is spliced ​​to obtain the target coding information sequence of the historical time period;

[0116] By using the trained granularity feature extraction module, the target coded information sequence is divided to obtain the s-th granularity sub-information sequence of each historical time period;

[0117] By means of the granularity feature extraction module, the s-th granularity sub-information sequence is divided to obtain the s+1-th granularity sub-information sequence of each historical time period;

[0118] Each s-th granularity sub-information sequence of the historical time period and each s+1-th granularity sub-information sequence of the historical time period are fused to obtain a multi-granularity coding feature tree of the historical time period.

[0119] In some embodiments, the extraction unit 503 is specifically used to: establish a preliminary coding feature tree for the historical time period with the target coding information sequence as a root node and the sub-information sequence as a tree node;

[0120] Performing connection processing on each node in the preliminary coding feature tree through a fully connected layer to obtain a fully connected post-coding feature tree for the historical time period;

[0121] According to a preset convolution kernel size and a preset convolution step size, convolution processing is performed on each node in the fully connected coded feature tree to obtain a multi-granularity coded feature tree for the historical time period.

[0122] In some embodiments, the prediction unit 504 is specifically used to:

[0123] Obtaining an information sequence of adjacent nodes of the same granularity of each node in the multi-granularity coding feature tree, an information sequence of nodes of the previous granularity of each node, and an information sequence of nodes of the next granularity of each node;

[0124] By using the trained attention feature extraction module, weight extraction processing is performed based on the information sequence of the adjacent nodes of the same granularity, the information sequence of the node of the previous granularity, and the information sequence of the node of the next granularity to obtain the attention coefficient of each node;

[0125] By using the trained attention feature extraction module, information extraction processing is performed according to the attention coefficient of each node and the information sequence of each node to obtain the attention coding feature of the historical time period;

[0126] Predictions are made based on the attention encoding features to obtain process node case data for the target time span.

[0127] In some embodiments, the encoding unit 502 is specifically used to:

[0128] Encoding the process node case data of each time span to obtain the case data code of each time span;

[0129] Encoding the case attributes of the process nodes of each time span to obtain the case attribute code of each time span;

[0130] Encoding the sequence information of each time span to obtain the sequence code of each time span;

[0131] The case data code of each time span, the case attribute code of each time span and the sequence code of each time span are concatenated to obtain the coding information of each time span.

[0132] In some embodiments, the encoding unit 502 is specifically used to:

[0133] According to a preset sorting rule, assigning a value to each time span to obtain sequence information of each time span;

[0134] The sequence information of each time span is coded by a rotational position coding module to obtain the rotational position coding of each time span as the sequence coding of each time span.

[0135] In some embodiments, the prediction unit 504 is specifically used to:

[0136] The process node case data of the target time span is obtained by predicting based on the attention coding features through the trained prediction module, wherein the prediction module is a Transformer model or a fully connected layer.

[0137] In specific implementation, the above units can be implemented as independent entities, or can be arbitrarily combined to be implemented as the same or several entities. The specific implementation of the above units can refer to the data prediction method embodiment of the previous process node, which will not be repeated here.

[0138] See also Figure 6 , Figure 6 A schematic block diagram of the structure of a computer device provided in an embodiment of the present application. The computer device may be a server.

[0139] like Figure 6As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0140] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any data prediction method of a process node.

[0141] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0142] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute the data prediction method of any process node.

[0143] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0144] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0145] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0146] Acquire the process node case data of each time span within the historical time period; encode the process node case data of each time span to obtain the encoding information of each time span; perform multi-granularity extraction processing according to the encoding information of each time span through the trained granularity feature extraction module to obtain a multi-granularity encoding feature tree of the historical time period; perform prediction according to the multi-granularity encoding feature tree to obtain the process node case data of the target time span.

[0147] A person of ordinary skill in the art can understand that all or part of the steps in the data prediction method of the above-mentioned process node can be completed by instructions, or by controlling related hardware through instructions. The instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0148] To this end, an embodiment of the present application provides a computer-readable storage medium, which stores multiple computer programs. The computer programs can be loaded by a processor to execute the data prediction method of any process node provided in the embodiment of the present application.

[0149] The computer-readable storage medium may include: a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0150] In the above-mentioned data prediction device and computer-readable storage medium embodiments of the process node, the description of each embodiment has its own emphasis. For the part not described in detail in a certain embodiment, reference can be made to the relevant description of other embodiments. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process and beneficial effects of the data prediction device, computer-readable storage medium and corresponding units of the process node described above can refer to the description of the data prediction method of the process node in the above embodiment, and will not be repeated here.

[0151] The above is a detailed introduction to a data prediction method, device, computer equipment and computer-readable storage medium for a process node provided in an embodiment of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A data prediction method for a process node, characterized in that: The method comprises: Obtain process node case data for each time span within the historical time period; Encoding the process node case data of each time span to obtain encoding information of each time span; By using the trained granularity feature extraction module, multi-granularity extraction processing is performed according to the coding information of each time span to obtain a multi-granularity coding feature tree of the historical time period; Prediction is performed based on the multi-granularity coding feature tree to obtain process node case data of the target time span.

2. The data prediction method for a process node according to claim 1, characterized in that: The trained granular feature extraction module performs multi-granularity extraction processing according to the coding information of each time span to obtain a multi-granularity coding feature tree of the historical time period, including: According to the chronological order, the coding information of each time span is spliced ​​to obtain the target coding information sequence of the historical time period; By using the trained granularity feature extraction module, the target coded information sequence is divided to obtain the s-th granularity sub-information sequence of each historical time period; By means of the granularity feature extraction module, the s-th granularity sub-information sequence is divided to obtain the s+1-th granularity sub-information sequence of each historical time period; Each s-th granularity sub-information sequence of the historical time period and each s+1-th granularity sub-information sequence of the historical time period are fused to obtain a multi-granularity coding feature tree of the historical time period.

3. The data prediction method for a process node according to claim 2, characterized in that: The fusing process of each s-th granularity sub-information sequence of the historical time period and each s+1-th granularity sub-information sequence of the historical time period to obtain a multi-granularity coding feature tree of the historical time period includes: Taking the target coding information sequence as a root node and the sub-information sequence as a tree node, a preliminary coding feature tree for the historical time period is established; Performing connection processing on each node in the preliminary coding feature tree through a fully connected layer to obtain a fully connected post-coding feature tree for the historical time period; According to a preset convolution kernel size and a preset convolution step size, convolution processing is performed on each node in the fully connected coded feature tree to obtain a multi-granularity coded feature tree for the historical time period.

4. The data prediction method for a process node according to claim 1, characterized in that: The step of performing prediction according to the multi-granularity coding feature tree to obtain process node case data of the target time span includes: Obtaining an information sequence of adjacent nodes of the same granularity of each node in the multi-granularity coding feature tree, an information sequence of nodes of the previous granularity of each node, and an information sequence of nodes of the next granularity of each node; By using the trained attention feature extraction module, weight extraction processing is performed based on the information sequence of the adjacent nodes of the same granularity, the information sequence of the node of the previous granularity, and the information sequence of the node of the next granularity to obtain the attention coefficient of each node; By using the trained attention feature extraction module, information extraction processing is performed according to the attention coefficient of each node and the information sequence of each node to obtain the attention coding feature of the historical time period; Predictions are made based on the attention encoding features to obtain process node case data for the target time span.

5. The data prediction method for a process node according to claim 1, characterized in that: The encoding process is performed on the process node case data of each time span to obtain the encoding information of each time span, including: Encoding the process node case data of each time span to obtain the case data code of each time span; Encoding the case attributes of the process nodes of each time span to obtain the case attribute code of each time span; Encoding the sequence information of each time span to obtain the sequence code of each time span; The case data code of each time span, the case attribute code of each time span and the sequence code of each time span are concatenated to obtain the coding information of each time span.

6. The data prediction method for a process node according to claim 5, characterized in that: The encoding of the sequence information of each time span to obtain the sequence code of each time span includes: According to a preset sorting rule, assigning a value to each time span to obtain sequence information of each time span; The sequence information of each time span is coded by a rotational position coding module to obtain the rotational position coding of each time span as the sequence coding of each time span.

7. The data prediction method for a process node according to claim 5, characterized in that: The prediction based on the attention encoding feature to obtain the process node case data of the target time span includes: The process node case data of the target time span is obtained by predicting based on the attention coding features through the trained prediction module, wherein the prediction module is a Transformer model or a fully connected layer.

8. A data prediction device for a process node, characterized in that: The data prediction device of the process node comprises: An acquisition unit is used to acquire process node case data for each time span in a historical time period; An encoding unit, used for encoding the process node case data of each time span to obtain encoding information of each time span; An extraction unit, configured to perform multi-granularity extraction processing according to the coding information of each time span through a trained granularity feature extraction module to obtain a multi-granularity coding feature tree of the historical time period; The prediction unit is used to make predictions based on the multi-granularity coding feature tree to obtain process node case data of a target time span.

9. A computer device, characterized in that: The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the data prediction method for the process node according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program is loaded by a processor to execute the data prediction method for a process node as described in any one of claims 1 to 7.

Images