Business process activity determination method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a business process activity determination method and device, electronic equipment and a storage medium, and the method comprises the steps: determining a first attribute prefix sequence corresponding to a target business process based on the attribute of at least one first historical event in a first event prefix sequence corresponding to the target business process; wherein the length of the first event prefix sequence is the same as that of the first attribute prefix sequence, and the first attribute prefix sequence comprises at least one of a first active prefix sequence, a first resource prefix sequence and a first time prefix sequence; and determining a prediction result of the target business process based on the first attribute prefix sequence by using a trained activity prediction model, the prediction result of the target business process comprising a next activity of the target business process and explanation information of the next activity of the target business process, the interpretation information of the next activity of the target business process comprises at least one of attribute influence, event influence and mutual influence.

Description

Technical Field

[0001] The present application relates to, but is not limited to, the field of information technology, and in particular to a method and device for determining a business process activity, an electronic device, and a storage medium. Background Art

[0002] In related technologies, "black box models" are basically used to predict the next activity of a business process. Due to the lack of in-depth understanding of the business, the accuracy of the prediction results is not high. At the same time, due to the lack of explainable information that generates the prediction results, the credibility and usage rate of the prediction results are reduced. Summary of the Invention

[0003] In view of this, embodiments of the present application provide a method and apparatus for determining a business process activity, an electronic device, a storage medium, and a program product.

[0004] The technical solution of the embodiment of the present application is implemented as follows:

[0005] This embodiment of the present application provides a method for determining a business process activity, including:

[0006] Determine a first attribute prefix sequence corresponding to the target business process based on an attribute of at least one first historical event in a first event prefix sequence corresponding to the target business process; wherein the length of the first event prefix sequence is the same as the length of the first attribute prefix sequence, and the first attribute prefix sequence includes at least one of the following: a first activity prefix sequence, a first resource prefix sequence, and a first time prefix sequence;

[0007] Utilizing the trained activity prediction model, based on the first attribute prefix sequence, the prediction result of the target business process is determined. The prediction result of the target business process includes the next activity of the target business process and the explanatory information of the next activity of the target business process. The explanatory information of the next activity of the target business process includes at least one of the following: attribute influence, event influence, and mutual influence. The attribute influence represents the degree of influence of the attributes of each first historical event in the first attribute prefix sequence on the next activity of the target business process. The event influence represents the degree of influence of each first historical event in the first event prefix sequence on the next activity of the target business process. The mutual influence represents the degree of mutual influence between each first historical event in the first event prefix sequence and the degree of mutual influence between the attributes of each first historical event in the first attribute prefix sequence.

[0008] This embodiment of the present application provides a device for determining a business process activity, including:

[0009] A first determining module is configured to determine a first attribute prefix sequence corresponding to the target business process based on an attribute of at least one first historical event in a first event prefix sequence corresponding to the target business process; wherein the length of the first event prefix sequence is the same as the length of the first attribute prefix sequence, and the first attribute prefix sequence includes at least one of the following: a first activity prefix sequence, a first resource prefix sequence, and a first time prefix sequence;

[0010] The second determination module is used to use the trained activity prediction model to determine the prediction result of the target business process based on the first attribute prefix sequence. The prediction result of the target business process includes the next activity of the target business process and the explanatory information of the next activity of the target business process. The explanatory information of the next activity of the target business process includes at least one of the following: attribute influence, event influence, and mutual influence. The attribute influence represents the degree of influence of the attributes of each first historical event in the first attribute prefix sequence on the next activity of the target business process. The event influence represents the degree of influence of each first historical event in the first event prefix sequence on the next activity of the target business process. The mutual influence represents the degree of mutual influence between each first historical event in the first event prefix sequence and the degree of mutual influence between the attributes of each first historical event in the first attribute prefix sequence.

[0011] An embodiment of the present application provides an electronic device, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and the above method is implemented when the processor executes the computer program.

[0012] An embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the above method when executed by a processor.

[0013] An embodiment of the present application provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program. When the computer program is read and executed by a computer, the above method is implemented.

[0014] In an embodiment of the present application, first, the attribute prefix sequence is dynamically determined through the attributes of each historical event of the target business process, thereby improving the accuracy of the attribute prefix sequence; secondly, the trained activity prediction model is used to predict the next activity based on the attribute prefix sequence. Since the attribute prefix sequence is determined according to the attributes of each historical event in the business process, the prediction result is more in line with business needs, which improves the accuracy of the prediction result while also improving the prediction efficiency; finally, the activity prediction model also generates interpretable information including attribute influence, event influence, mutual influence, etc. when generating the prediction result, which is used to describe the important influencing factors (for example, certain events, certain attributes, etc.) that produce the prediction result and the degree of internal influence between process-related events and attributes, which not only improves the credibility of the prediction result, but also increases the usage rate of the prediction result by the user.

[0015] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and do not limit the technical solutions of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings herein are incorporated into and constitute a part of the specification. These drawings illustrate embodiments consistent with the present application and, together with the specification, are used to illustrate the technical solutions of the present application.

[0017] Figure 1 Schematic diagram of the implementation process of a method for determining a business process activity provided in an embodiment of the present application Figure 1 ;

[0018] Figure 2 Schematic diagram of the implementation process of a method for determining a business process activity provided in an embodiment of the present application Figure 2 ;

[0019] Figure 3 Schematic diagram of the implementation process of a method for determining a business process activity provided in an embodiment of the present application Figure 3 ;

[0020] Figure 4A Schematic diagram 4 of a method for determining a business process activity according to an embodiment of the present application;

[0021] Figure 4B A schematic diagram of a loan application process provided in an embodiment of the present application;

[0022] Figure 4C A schematic diagram of an activity prediction model provided in an embodiment of the present application;

[0023] Figure 4D The implementation process frame of a method for determining a business process activity provided in an embodiment of the present application Figure 1 ;

[0024] Figure 4E A schematic diagram of an explanation information provided in an embodiment of the present application;

[0025] Figure 5 A schematic diagram of the structure of a device for determining a business process activity provided in an embodiment of the present application;

[0026] Figure 6 A hardware entity diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solutions and advantages of this application clearer, the technical solutions of this application are further elaborated in detail below with reference to the accompanying drawings and embodiments. The described embodiments should not be regarded as limiting this application. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0028] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0029] The terms "first / second / third" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It is understandable that "first / second / third" can be interchanged with a specific order or sequence where permitted so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing this application only and are not intended to limit this application.

[0031] With the increasing adoption of enterprise information systems, more and more companies are emphasizing the use of data-driven decision-making. Process mining is a series of technologies that use event logs to analyze business processes, extract valuable information, and ultimately empower them. Predictive process analysis, a form of process mining, uses process execution logs to predict the next activity in a process and identify development trends.

[0032] In related technologies, most mainstream business process activity prediction models are "black box" models. Due to a lack of in-depth understanding of the business, the accuracy of their predictions is low. Furthermore, because these prediction models only provide predictions without explaining the reasons for their results, users (e.g., businesses and institutions) are skeptical of their reliability, resulting in a low rate of trust and adoption of these predictions in actual decision-making.

[0033] An embodiment of the present application provides a method for determining business process activities. First, an attribute prefix sequence is dynamically determined through the attributes of each historical event of the target business process, thereby improving the accuracy of the attribute prefix sequence; second, a trained activity prediction model is used to predict the next activity based on the attribute prefix sequence. Since the attribute prefix sequence is determined according to the attributes of each historical event in the business process, the prediction result is more in line with business needs, which improves the accuracy of the prediction result while also improving the prediction efficiency; finally, the activity prediction model also generates interpretable information including attribute influence, event influence, mutual influence, etc. when generating the prediction result, which is used to describe the important influencing factors (for example, certain events, certain attributes, etc.) that produce the prediction result and the degree of internal influence between process-related events and attributes, which not only improves the credibility of the prediction result, but also increases the utilization rate of the prediction result by the user.

[0034] The method provided in the embodiment of the present application can be performed by an electronic device. Among them, the electronic device can be various types of terminals such as a laptop computer, a tablet computer, a desktop computer, a set-top box, a mobile device (for example, a mobile phone, a portable music player, a personal digital assistant, a dedicated messaging device, a portable gaming device), and can also be implemented as a server. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0035] Below, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the drawings in the embodiments of the present application.

[0036] Figure 1 Schematic diagram of the implementation process of a method for determining a business process activity provided in an embodiment of the present application Figure 1 ,like Figure 1 As shown, the method includes step S11 and step S12, wherein:

[0037] Step S11 : Determine a first attribute prefix sequence corresponding to the target business process based on an attribute of at least one first historical event in a first event prefix sequence corresponding to the target business process.

[0038] Here, the business (including the target business and other businesses mentioned later) can be a business in any appropriate industry, for example, a recruitment business, a production business, a loan business, etc. Different businesses may include different processes.

[0039] A historical event (including the first historical event and other historical events mentioned below) refers to an event that has occurred in the process. A process can contain at least one event. For example, for a recruitment business, the event may include resume screening, recruitment plan development, etc. For another example, for a loan business, the event may include application submission, loan registration, etc.

[0040] The attributes of a historical event may include, but are not limited to, at least one of the activities performed by the historical event, the resources occupied by the historical event, and the time corresponding to the historical event. An activity may be any appropriate activity in the event. A resource may be any appropriate resource used in the event, such as an activity executor, resources used by an activity, etc. The time corresponding to the event may be any appropriate time. For example, the completion time of the event, the completion time of an activity in the event, the difference between the completion time of an activity in the event and the completion time of a target activity, etc. The target activity may be the activity of any event before the event. For example, the target activity may be the activity of the first event. For another example, the target activity may be the activity of the event before the event.

[0041] The length of the event prefix sequence (including the first event prefix sequence and other event prefix sequences described below) may be any appropriate length, for example, 3, 5, etc. In some implementations, the number of event prefix sequences may be at least one.

[0042] The event prefix sequence may be determined in any appropriate manner.

[0043] In some implementations, the event prefix sequence can be determined based on multiple historical events of the business process. For example, a portion of multiple first historical events can be randomly selected as the first event prefix sequence. For another example, a portion of the first historical events can be customized as the first event prefix sequence. For another example, all first historical events can be selected as the first event prefix sequence.

[0044] In some embodiments, the event prefix sequence can be determined based on an event sequence (including a first event sequence and other event sequences described below), and the event sequence is obtained by sorting each historical event according to the time of occurrence of multiple historical events. For example, the last few first historical events in the first event sequence are used as the first event prefix sequence. For another example, the multiple first historical events in the first event sequence are divided according to a length of 3 to obtain at least one first event prefix sequence. For example, if the first event sequence includes ABCDEF, then the first event prefix sequence may include ABC and DEF, or may include ABC, BCD, CDE, and DEF.

[0045] During implementation, those skilled in the art can independently set the method for determining the event prefix sequence according to actual needs, and the embodiments of the present application do not limit this.

[0046] The length of the attribute prefix sequence (including the first attribute prefix sequence and other attribute prefix sequences described below) can be any suitable length, for example, 3, 5, etc. The length of the attribute prefix sequence is the same as the length of the event prefix sequence. For example, the length of the first event prefix sequence and the length of the first attribute prefix sequence are both 5.

[0047] The attribute prefix sequence may include but is not limited to at least one of an activity prefix sequence (including the first activity prefix sequence and other activity prefix sequences thereafter), a resource prefix sequence (including the first resource prefix sequence and other resource prefix sequences thereafter), a time prefix sequence (including the first time prefix sequence and other time prefix sequences thereafter), etc. The activity prefix sequence is determined based on the activities performed by each historical event. The resource prefix sequence is determined based on the resources occupied by each historical event. The time prefix sequence is determined based on the time corresponding to each historical event. In some embodiments, the time corresponding to each historical event can be normalized to each time within a target range as an element in the time prefix sequence. The target range can be any suitable range, for example, [0, 1], [-1, 1], etc. The time corresponding to the i-th historical event can be the difference Δt between the completion time of the activity of the i-th historical event and the completion time of the activity of the first historical event. i .

[0048] For example, the first attribute prefix sequence includes at least one of the following: a first activity prefix sequence, a first resource prefix sequence, and a first time prefix sequence.

[0049] In some implementations, different services may include the same or different attribute prefix sequences. For example, for service X1, the attribute prefix sequence includes an activity prefix sequence and a resource prefix sequence; for service X2, the attribute prefix sequence includes an activity prefix sequence, a resource prefix sequence, and a time prefix sequence.

[0050] In some implementations, the number of attribute prefix sequences may be at least one, for example, the number of attribute prefix sequences matches the number of event prefix sequences.

[0051] The attribute prefix sequence may be determined in any suitable manner. In some implementations, the attribute prefix sequence may be determined according to the attributes of each historical event in the event prefix sequence.

[0052] For example, the attributes of each historical event in the event prefix sequence are respectively used as an element in the attribute prefix sequence.

[0053] For example, the first event prefix sequence is (e1, e2, e3), then:

[0054] Activity a1 executed by the first historical event e1, activity a2 executed by the first historical event e2, and activity a3 executed by the first historical event e3 can be respectively used as an element of the first activity prefix sequence. In this case, the first activity prefix sequence can be (a1, a2, a3);

[0055] The resource r1 occupied by the first historical event e1, the resource r2 occupied by the first historical event e2, and the resource r3 occupied by the first historical event e3 may be respectively used as an element of the first activity prefix sequence. In this case, the first resource prefix sequence may be (r1, r2, r3);

[0056] The time t1 corresponding to the first historical event e1, the time t2 corresponding to the first historical event e2, and the time t3 corresponding to the first historical event e3 can be respectively used as an element of the first activity prefix sequence. In this case, the first activity prefix sequence can be (t1, t2, t3).

[0057] For another example, the attributes of each historical event in the event prefix sequence after duplication is removed are used as an element in the attribute prefix sequence.

[0058] For example, if the first event prefix sequence is (e1, e2, e3), then activity a1 executed by the first historical event e1, activity a2 executed by the first historical event e2, and activity a3 executed by the first historical event e3 can be deduplicated, and the resulting activity a1 and activity a3 can be used as an element of the first activity prefix sequence respectively. At this time, the first activity prefix sequence can be (a1, a3).

[0059] Step S12: Utilize the trained activity prediction model to determine the prediction result of the target business process based on the first attribute prefix sequence. The prediction result of the target business process includes the next activity of the target business process and the explanatory information of the next activity of the target business process. The explanatory information of the next activity of the target business process includes at least one of the following: attribute influence, event influence, and mutual influence. The attribute influence represents the degree of influence of the attributes of each first historical event in the first attribute prefix sequence on the next activity of the target business process. The event influence represents the degree of influence of each first historical event in the first event prefix sequence on the next activity of the target business process. The mutual influence represents the degree of mutual influence between each first historical event in the first event prefix sequence and the degree of mutual influence between the attributes of each first historical event in the first attribute prefix sequence.

[0060] Here, the activity prediction model may be any suitable neural network model capable of achieving this function. In some embodiments, the activity prediction model may include an encoding part, an attribute part, a connection part, an event part, a prediction part, and the like.

[0061] The encoding part is used to encode the input information to obtain the encoding feature corresponding to the input information. The encoding part can be any suitable part that can achieve this function. For example, the first attribute prefix sequence is encoded by the encoding part to obtain the attribute encoding feature. The number of encoding parts can be at least one. For example, the number of encoding parts is two, and the first encoding part is used to encode the first activity prefix sequence to obtain the encoded activity encoding feature. The second encoding part is used to encode the first resource prefix sequence to obtain the encoded resource encoding feature In some implementations, the encoding portion may encode the first attribute prefix sequence and the first activity prefix sequence, but may not encode the first time prefix sequence.

[0062] In some implementations, the activity, resource, and time attributes of each historical event in a process instance can be encoded. These activity and resource attributes can be input as text. In some implementations, each attribute can be indexed. For example, the activity set A formed by the activities of each historical event can be represented by numbers from 1 to |A|, where |A| represents the length of activity set A. For another example, the resource set R formed by the resources of each historical event can be represented by numbers from 1 to |R|, where |R| represents the length of resource set R.

[0063] The attribute part is mainly used to process the attributes. The attribute part may include but is not limited to the attribute transformer part, the attribute attention weight part, the multiplication part, etc.

[0064] The attribute transformer part is used to determine the attribute intermediate state feature of the input information and the attribute multi-head attention score corresponding to the input information. The number of attribute transformer parts can be at least one. For example, the number of attribute transformer parts is three, and the first attribute transformer part is used to determine the activity intermediate state feature g of the first activity prefix sequence. a The multi-head attention score of the activity corresponding to the first activity prefix sequence The second attribute Transformer part is used to determine the resource intermediate state feature g of the first resource prefix sequence r The resource multi-head attention score corresponding to the first resource prefix sequence The third attribute Transformer part is used to determine the time intermediate state feature g of the first time prefix sequence t The temporal multi-head attention score corresponding to the first temporal prefix sequence

[0065] The attribute attention weight portion is used to determine the attribute attention weight of the input information. The attribute attention weight portion can be any suitable portion. For example, any suitable activation function, such as a tanh function, a softmax function, etc. The number of attribute attention weight portions can be at least one. In some embodiments, the number of attribute attention weight portions is adapted to the number of attribute transformer portions. For example, if the number of attribute transformer portions is 3, then the number of attribute attention weight portions can also be 3.

[0066] The multiplication part mainly performs product operations on the input information. The multiplication part can be any suitable part that can achieve this function, for example, a multiplication layer.

[0067] The event part is mainly used to operate on events. The event part may include but is not limited to the event transformer part and the event attention weight part.

[0068] The event Transformer part is used to determine the event intermediate state feature g of the input information. e The event multi-head attention score corresponding to the input information For example, the target feature corresponding to the first attribute prefix sequence By inputting it into the Transformer part of the event, the intermediate state feature g of the event can be obtained. e The event multi-head attention score corresponding to the input information In some implementations, the event Transformer portion and the attribute Transformer portion may be the same or different.

[0069] The event attention weight portion is used to determine the event attention weight of the input information. The event attention weight portion can be any suitable portion. For example, any suitable activation function, such as a tanh function, a softmax function, etc. In some embodiments, the event attention weight portion and the attribute attention weight portion can be the same or different.

[0070] The connection part is used to connect multiple input information to obtain the target features corresponding to the input information. The connection part can be any suitable part that can achieve this function.

[0071] For example, the connection part can be used to connect multiple attribute features to obtain the target feature corresponding to the attribute prefix sequence. For example, the connection part can connect the activity feature Resource characteristics and time characteristics To obtain the target feature corresponding to the attribute prefix sequence Among them, the characteristics of this activity Based on activity coding features and activity attention weight β a Determined, the resource characteristics Based on resource encoding characteristics and resource attention weight β r Determined, the time feature is based on the time prefix sequence and the time attention weight β t Sure.

[0072] The prediction part is used to predict the input information to obtain the prediction result. The prediction part can be any suitable part that can achieve this function. For example, the prediction part can include but is not limited to the product part, activation function, etc.

[0073] In some embodiments, the activity prediction model may include but is not limited to an encoding part, a Transformer part, an attention weight part, a connection part, a prediction part, and the like. The Transformer part is used to determine the intermediate state features of the input information and the multi-head attention score corresponding to the input information. The number of the Transformer parts may be at least one. In some embodiments, the Transformer part may include an attribute Transformer part, an event Transformer part, and the like. The attention weight part is used to determine the attention weight of the input information, and the attention weight part may be any suitable part. For example, any suitable activation function, such as a tanh function, a softmax function, and the like. The number of the attention weight parts may be at least one. In some embodiments, the attention weight part may include an attribute attention weight part, an event attention weight part, and the like.

[0074] In some embodiments, the first attribute prefix sequence may be input into the activity prediction model to obtain the prediction result. In some embodiments, when the activity prediction model does not include the encoding portion, the first attribute prefix sequence may be encoded and then input into the activity prediction model to obtain the prediction result.

[0075] The next activity of the target business process refers to the predicted activity that will occur in the target business process.

[0076] The explanatory information is mainly used to explain the reason for the next activity, that is, to explain the factors that affect the next activity. In some embodiments, the explanatory information may include attribute influence, event influence, and mutual influence. During implementation, the explanatory information can be used to determine which attributes in the attribute prefix sequence and which events in the event prefix sequence have a greater impact on the prediction results, as well as which attributes in the attribute prefix sequence and which events in the event prefix sequence have a greater mutual influence. In some embodiments, the explanatory information may include attribute influence and event influence.

[0077] In some implementations, since the attributes of each historical event may have varying degrees of influence on the upcoming activity, during implementation, the attribute influence can be used to explain which attributes in the attribute prefix sequence have a greater impact on the prediction result.

[0078] In some implementations, since each historical event may have varying degrees of impact on an upcoming activity, during implementation, the event influence may be used to explain which events in the event prefix sequence have a greater impact on the prediction result.

[0079] In an embodiment of the present application, first, the attribute prefix sequence is dynamically determined through the attributes of each historical event of the target business process, thereby improving the accuracy of the attribute prefix sequence; secondly, the trained activity prediction model is used to predict the next activity based on the attribute prefix sequence. Since the attribute prefix sequence is determined according to the attributes of each historical event in the business process, the prediction result is more in line with business needs, which improves the accuracy of the prediction result while also improving the prediction efficiency; finally, the activity prediction model also generates interpretable information including attribute influence, event influence, mutual influence, etc. when generating the prediction result, which is used to describe the important influencing factors (for example, certain events, certain attributes, etc.) that produce the prediction result and the degree of internal influence between process-related events and attributes, which not only improves the credibility of the prediction result, but also increases the usage rate of the prediction result by the user.

[0080] In some embodiments, step S11 includes steps S111 to S113, wherein:

[0081] Step S111 : determining a plurality of first historical events of the target business process based on the event log of the target business process.

[0082] Here, the event log includes at least the log of the target business process, and the event log can record logs of different business processes. Multiple events will occur during the execution of each process.

[0083] The method for determining multiple historical events can be any suitable method. In some embodiments, the historical events recorded in the log can be classified, grouped, or otherwise processed according to the time of occurrence based on the process instance identifier to obtain multiple historical events for each process. In some embodiments, clustering can be performed based on the correlation between the historical events to obtain a cluster for each process, and the cluster is used as the multiple historical events for the process. The cluster of the process includes at least two historical events.

[0084] Step S112: Determine a first event prefix sequence based on a first event sequence formed by multiple first historical events of the target business process.

[0085] Here, the first event sequence is obtained by sorting the first historical events in order of occurrence from earliest to latest, or from latest to earliest. The length of the first event sequence can be any suitable length, such as 6 or 10, and is determined based on the number of first historical events included in the target business process.

[0086] The first event prefix sequence includes at least one first historical event. In some embodiments, the number of the first event prefix sequence may be at least one. The length of the first event prefix sequence is not greater than the length of the first event sequence.

[0087] The first event prefix sequence may be determined in any suitable manner. In some embodiments, the first event sequence is used as the first event prefix sequence. In some embodiments, some first historical events in the first event sequence are used as the first event prefix sequence.

[0088] For example, the last several first historical events in the first event sequence are used as the first event prefix sequence.

[0089] For another example, multiple first historical events in the first event sequence are divided according to a length of 3 to obtain at least one first event prefix sequence. For example, if the first event sequence includes ABCDEF, then the first event prefix sequence may include ABC and DEF, or may include ABC, BCD, CDE, and DEF.

[0090] Step S113: Determine a first attribute prefix sequence based on the attribute of at least one first historical event in the first event prefix sequence.

[0091] Here, the length of the first attribute prefix sequence is the same as the length of the first event prefix sequence. The first attribute prefix sequence may include but is not limited to at least one of a first activity prefix sequence, a first resource prefix sequence, a first time prefix sequence, etc. In some embodiments, different services may include the same or different first attribute prefix sequences. For example, for an event prefix sequence (e1, e2, ..., e L ), its active prefix sequence can be expressed as (a1,a2,...,a L ), the resource prefix sequence can be expressed as (r1,r2,...,r L ), the time prefix sequence can be expressed as (Δt1, Δt2, ..., Δt L ).

[0092] The first attribute prefix sequence may be determined in any suitable manner. For implementation, reference may be made to the specific implementation in the aforementioned step S11.

[0093] In an embodiment of the present application, the first event prefix sequence is determined based on multiple historical events recorded in the event log, which improves the accuracy of the first event prefix sequence, thereby improving the accuracy of the first attribute prefix sequence determined based on the first event prefix sequence, and further improving the prediction efficiency and accuracy of the prediction results.

[0094] In some embodiments, the attributes of the first historical event include at least one of the following: an activity performed by the first historical event, resources occupied by the first historical event, and a time corresponding to the first historical event. Step S113 includes steps S1131 to S1133, wherein:

[0095] Step S1131 : When the first attribute prefix sequence includes a first activity prefix sequence, determine a first activity prefix sequence based on an activity performed by at least one first historical event in the first event prefix sequence.

[0096] Here, the activities of each first historical event in the target business process instance may be extracted first to obtain the activities executed by each first historical event.

[0097] The first activity prefix sequence may be determined in any suitable manner. In some embodiments, the activities performed by each first historical event in the first event prefix sequence may be used as an element in the first activity prefix sequence. In some embodiments, at least one activity obtained by deduplicating the activities performed by each first historical event in the first event prefix sequence may be used as an element in the first activity prefix sequence.

[0098] Step S1132: When the first attribute prefix sequence includes a first resource prefix sequence, determine a first resource prefix sequence based on resources occupied by at least one first historical event in the first event prefix sequence.

[0099] Here, the resources of each first historical event in the target business process instance may be extracted first to obtain the resources occupied by each first historical event.

[0100] The first resource prefix sequence may be determined in any suitable manner. In some embodiments, the resources occupied by each first historical event in the first event prefix sequence may be used as an element in the first resource prefix sequence. In some embodiments, at least one resource obtained by deduplicating the resources occupied by each first historical event in the first event prefix sequence may be used as an element in the first resource prefix sequence.

[0101] Step S1133: When the first attribute prefix sequence includes a first time prefix sequence, determine a first time prefix sequence based on a time corresponding to at least one first historical event in the first event prefix sequence.

[0102] Here, the time of each first historical event in the target business process instance can be extracted first to obtain the time corresponding to each first historical event. The time corresponding to the first historical event is determined based on the completion time of the activity executed by the first historical event and the completion time of the activity executed by the target first historical event. The target first historical event includes one of the following: the previous first historical event of the first historical event, the first first historical event. The time corresponding to the first historical event may include but is not limited to the first difference, the weighting of the first difference, etc. The first difference means that the time corresponding to the first historical event may be the difference between the completion time of the activity executed by the first historical event and the completion time of the activity executed by the target first historical event.

[0103] The first time prefix sequence can be determined in any suitable manner. In some embodiments, the time corresponding to each first historical event in the first event prefix sequence can be used as an element in the first time prefix sequence. In some embodiments, at least one time obtained by normalizing the time corresponding to each first historical event in the first event prefix sequence to within a target range can be used as an element in the first time prefix sequence.

[0104] In the implementation manner of the present application, each attribute prefix sequence is determined separately according to dimensions such as activity, resource, and time of each historical event in the event prefix sequence, thereby improving the accuracy of each attribute prefix sequence.

[0105] In some embodiments, step S12 includes steps S121 to S123, wherein:

[0106] Step S121: Encode the first attribute prefix sequence to obtain attribute encoding features.

[0107] Here, the attribute coding feature may include but is not limited to a vector, a matrix, etc. The attribute coding feature may include but is not limited to an activity coding feature, a resource coding feature, a time coding feature, etc. The activity coding feature is a feature obtained by encoding an activity prefix sequence. The resource coding feature is a feature obtained by encoding a resource prefix sequence. The time coding feature is a feature obtained by encoding a time prefix sequence. In some embodiments, the attribute coding feature can be obtained by encoding the first attribute prefix sequence through the encoding part. In some embodiments, the encoding part includes a first encoding part and a second encoding part, the first encoding part is used to encode the activity prefix sequence, and the second encoding part is used to encode the resource prefix sequence.

[0108] In some embodiments, the activity coding feature can be determined by the following formula (1-1): Right now:

[0109]

[0110] Among them, (a1,...,a l ) represents the active prefix sequence, l represents the length of the active prefix sequence, and |A| represents the length of the active set A.

[0111] In some implementations, the resource coding feature can be determined by the following formula (1-2): Right now:

[0112]

[0113] Among them, (r1,...,r l ) represents the resource prefix sequence, l represents the length of the resource prefix sequence, and |R| represents the length of the resource set R.

[0114] Step S122: Determine the target feature corresponding to the first attribute prefix sequence based on the attribute coding feature.

[0115] Here, the activity prediction model includes an attribute part and a connection part. During implementation, the attribute encoding feature is input into the attribute part to obtain the attribute features; the attribute features are input into the connection part to obtain the target feature. Attribute features may include, but are not limited to, activity features, resource features, and time features. The attribute part may include, but is not limited to, an attribute transformer part, an attribute attention weight part, and a multiplication part.

[0116] The attribute transformer is used to determine attribute intermediate state features and attribute multi-head attention scores. Attribute intermediate state features may include, but are not limited to, activity intermediate state features, resource intermediate state features, and time intermediate state features. Attribute multi-head attention scores may include, but are not limited to, activity multi-head attention scores, resource multi-head attention scores, and time multi-head attention scores.

[0117] The attribute Transformer part may include but is not limited to at least one of an activity Transformer part, a resource Transformer part, a time Transformer part, etc. The activity Transformer part is used to encode the characteristics according to the activity. Determine the intermediate state characteristics of the activity g a and active multi-head attention scores The resource transformer part is used to encode features according to the resource Determine the intermediate state characteristics of resources g r and resource multi-head attention scores The temporal Transformer part is used to calculate the time prefix sequence v t Determine the time intermediate state characteristics g t and temporal multi-head attention scores

[0118] In some embodiments, the activity intermediate state feature g can be determined by the following formula (1-3): a and active multi-head attention scores Right now:

[0119]

[0120] in, Represents activity encoding features.

[0121] In some embodiments, the active intermediate state feature g a Includes the activity intermediate state features that affect each other between the i-th activity and the j-th activity in the activity prefix sequence i and j are not greater than l (the length of the event prefix sequence).

[0122] In some embodiments, the resource intermediate state characteristic g can be determined by the following formula (1-4): r and resource multi-head attention scores Right now:

[0123]

[0124] in, Indicates resource encoding characteristics.

[0125] In some embodiments, the resource intermediate state feature g r Including resource intermediate state characteristics of the mutual influence between the i-th resource and the j-th resource in the resource prefix sequence i and j are not greater than l (the length of the resource prefix sequence).

[0126] In some embodiments, the temporal intermediate state feature g can be determined by the following formula (1-5): t and temporal multi-head attention scores Right now:

[0127]

[0128] Among them, v t Represents a time prefix sequence.

[0129] In some embodiments, the temporal intermediate state feature g t Including the time intermediate state characteristics of the mutual influence between the i-th time and the j-th time in the time prefix sequence i and j are not greater than l (the length of the time prefix sequence).

[0130] The attribute attention weight part is used to determine the attribute attention weight. Attribute attention weights may include but are not limited to activity attention weight, resource attention weight, time attention weight, etc.

[0131] The attribute attention weight part may include but is not limited to the activity attention weight part, the resource attention weight part, the time attention weight part, etc. The activity attention weight part is used to calculate the activity intermediate state feature g. a Determine the attention weight β of the activity a , the resource attention weight part is used to calculate the resource intermediate state feature g r Determine the resource attention weight β r , the time attention weight part is used to calculate the time intermediate state feature g t Determine the time attention weight β t .

[0132] In some embodiments, the activity attention weight of the mutual influence between the i-th activity and the j-th activity in the activity prefix sequence can be determined by the following formula (1-6): Right now:

[0133]

[0134] in, represents the weight matrix of the activity, represents the bias vector of the activity, The intermediate state feature of an activity that represents the mutual influence between the i-th activity and the j-th activity in the activity prefix sequence.

[0135] In some embodiments, the resource attention weight of the mutual influence between the i-th resource and the j-th resource in the resource prefix sequence can be determined by the following formula (1-7): Right now:

[0136]

[0137] in, represents the weight matrix of resources, represents the bias vector of the resource, The resource intermediate state characteristics that represent the mutual influence between the i-th resource and the j-th resource in the resource prefix sequence.

[0138] In some embodiments, the time attention weight of the interaction between the i-th time and the j-th time in the time prefix sequence can be determined by the following formula (1-8): Right now:

[0139]

[0140] in, represents the weight matrix of time, represents the time bias vector, It represents the temporal intermediate state characteristics of the mutual influence between the i-th time and the j-th time in the time prefix sequence.

[0141] The multiplication part is used to determine each attribute feature according to the attribute attention weight. The multiplication part may include but is not limited to at least one of the activity multiplication part, resource multiplication part, time multiplication part, etc. The activity multiplication part is used to determine the attribute features according to the activity encoding feature. and activity attention weight β a To determine the characteristics of the activity The resource multiplication part is used to encode characteristics according to the resource and resource attention weight β r To determine resource characteristics The time multiplication part is used to calculate the time prefix sequence v t and temporal attention weight β t To determine the time characteristics

[0142] In some embodiments, the activity characteristics can be determined by the following formula (1-9): Right now:

[0143]

[0144] in, represents the activity encoding feature, β a represents the activity attention weight.

[0145] In some embodiments, the resource characteristics can be determined by the following formula (1-10): Right now:

[0146]

[0147] in, represents resource encoding features, β r represents the resource attention weight.

[0148] In some embodiments, the time characteristic can be determined by the following formula (1-11): Right now:

[0149]

[0150] Among them, v t represents the time prefix sequence, βt represents the temporal attention weight.

[0151] The connection part is used to connect multiple attribute features to obtain the target feature. For example, after merging the activity feature, resource feature, and time feature, the feature of all attribute features after the attribute attention weight is applied is obtained.

[0152] In some embodiments, the target feature can be determined by the following formula (1-12): Right now:

[0153]

[0154] in, Indicates the characteristics of the activity, Represents resource characteristics, Indicates the characteristics of the activity,

[0155] Step S123: Determine a prediction result of the target business process based on the target feature corresponding to the first attribute prefix sequence.

[0156] Here, the activity prediction model also includes a prediction part, and the prediction result can be obtained by inputting the target feature into the prediction part.

[0157] In some embodiments, the activity prediction model also includes an event part. During implementation, the target feature is first input into the event part to obtain the event attention weight, and the event attention weight and target feature are input into the prediction part to obtain the prediction result.

[0158] The event part may include but is not limited to an event Transformer part, an event attention weight part, and the like.

[0159] The event Transformer part is used to determine the intermediate state feature g of the event. e and event multi-head attention scores In some embodiments, the event transformer and the attribute transformer may be the same or different. The process for determining the event intermediate state features and the event multi-head attention score is similar to the process for determining the attribute intermediate state features and the attribute multi-head attention score. When implementing this process, please refer to the specific implementation of the aforementioned step S122.

[0160] In some embodiments, the event intermediate state feature g can be determined by the following formula (1-13): e and event multi-head attention scores Right now:

[0161]

[0162] in, Represents the target feature.

[0163] In some embodiments, the event intermediate state feature g e Includes the event intermediate state features of the mutual influence between the i-th event and the j-th event in the event prefix sequence

[0164] The event attention weight portion is used to determine the event attention weight. In some embodiments, the event attention weight portion and the attribute attention weight portion may be the same or different. The process for determining the event attention weight is similar to the process for determining the attribute attention weight. When implementing, please refer to the specific implementation of the aforementioned step S122.

[0165] In some embodiments, the event attention weight includes the event attention weight α of the i-th event in the event prefix sequence i .

[0166] In some embodiments, the event attention weight α of the i-th event can be determined by the following formula (1-14): i ,Right now:

[0167]

[0168] Among them, W α Represents the weight matrix of the event, b α represents the bias vector of the event, The event intermediate state feature represents the mutual influence between the i-th event and the j-th event in the event prefix sequence.

[0169] In the implementation manner of the present application, on the one hand, the attribute prefix sequence is first encoded to obtain the attribute coding feature, so as to facilitate the subsequent data processing and improve the efficiency of data processing; on the other hand, the prediction result is determined by the target feature corresponding to the attribute prefix sequence. Since the target feature is determined according to the attributes of each historical event in the business process, the prediction result is more in line with business needs, which improves the accuracy of the prediction result and the prediction efficiency.

[0170] In some embodiments, step S122 includes steps S1221 to S1223, wherein:

[0171] Step S1221: Determine the attribute intermediate state feature based on the attribute coding feature.

[0172] Here, the attribute coding feature may include but is not limited to activity coding feature, resource coding feature, time coding feature, etc.

[0173] Attribute intermediate state characteristics may include but are not limited to activity intermediate state characteristics, resource intermediate state characteristics, time intermediate state characteristics, etc.

[0174] The attribute intermediate state feature can be determined in any suitable manner. In some embodiments, the activity prediction model may include an attribute transformer component, through which the attribute intermediate state feature can be obtained. In implementation, each attribute intermediate state feature can be obtained using the aforementioned formulas (1-3), (1-4), and (1-5).

[0175] Step S1222: Determine the attribute attention weight based on the attribute intermediate state characteristics.

[0176] Here, the attribute attention weight may include but is not limited to activity attention weight, resource attention weight, time attention weight, etc. The attribute attention weight may be determined in any suitable manner. In some embodiments, the activity prediction model may include an attribute attention weight portion, through which the attribute attention weight portion may be used to obtain the attribute attention weight. In implementation, each attribute attention weight may be obtained by the aforementioned formulas (1-6), (1-7), and (1-8).

[0177] Step S1223: Determine the target feature corresponding to the first attribute prefix sequence based on the attribute encoding feature and the attribute attention weight.

[0178] Here, the target feature can be determined in any suitable manner. In some embodiments, the activity prediction model can include a multiplication part and a concatenation part, wherein the attribute feature can be obtained by the multiplication part, and the target feature can be obtained by the concatenation part. In implementation, each feature can be obtained by the aforementioned formulas (1-9), (1-10), and (1-11), and the target feature can be obtained by the aforementioned formula (1-12).

[0179] In the embodiment of the present application, the target feature is determined based on the attribute coding feature and the attribute attention weight, which improves the accuracy of the target feature and thus improves the accuracy of the prediction result determined based on the target feature.

[0180] In some embodiments, step S123 includes steps S1231 to S1233, wherein:

[0181] Step S1231: Determine the event intermediate state feature based on the target feature corresponding to the first attribute prefix sequence.

[0182] Here, the event intermediate state feature can be determined in any suitable manner. In some embodiments, the activity prediction model may include an event transformer component, through which the event intermediate state feature can be obtained. In implementation, the event intermediate state feature can be obtained using the aforementioned formula (1-13).

[0183] Step S1232: Determine the event attention weight based on the event intermediate state characteristics.

[0184] Here, the event attention weight can be determined in any suitable manner. In some embodiments, the activity prediction model may include an event attention weight portion, through which the event attention weight portion can be used to obtain the event attention weight. In implementation, the event attention weight can be obtained using the aforementioned formula (1-14).

[0185] Step S1233: Determine the prediction result of the target business process based on the target features and event attention weights corresponding to the first attribute prefix sequence.

[0186] Here, the activity prediction model may also include a prediction part. The prediction part may include but is not limited to a multiplication layer, an activation function, etc. The multiplication layer can be used to obtain the context, and the activation function can be used to obtain the prediction result.

[0187] In some implementations, the context c can be obtained by the following formula (1-15), namely:

[0188]

[0189] Among them, α i represents the event attention weight of the i-th event, Represents the target feature.

[0190] In the implementation manner of the present application, on the one hand, the event attention weight is determined based on the target features corresponding to the attribute prefix sequence, thereby improving the accuracy of the event attention weight; on the other hand, the prediction result is comprehensively determined by the target features and the event attention weight, thereby improving the accuracy of the prediction result and the prediction efficiency.

[0191] In some embodiments, the determination method further includes steps S131 to S133, wherein:

[0192] Step S131: Determine the attribute influence based on the attribute encoding feature and the attribute attention weight.

[0193] Here, the attribute influence characterizes the degree of influence of the attributes of each first historical event in the first attribute prefix sequence on the next activity of the target business process. The attribute influence may include but is not limited to activity influence, resource influence, time influence, etc. The activity influence can be represented by activity characteristics, and when implemented, the activity influence can be determined by the aforementioned formula (1-9). The resource influence can be represented by resource characteristics, and when implemented, the resource influence can be determined by the aforementioned formula (1-10). The time influence can be represented by time characteristics, and when implemented, the time influence can be determined by the aforementioned formula (1-11).

[0194] Step S132: Determine the event influence based on the event attention weight.

[0195] Here, event influence represents the degree of influence of each first historical event in the first event prefix sequence on the next activity of the target business process. This event influence can be represented by the event attention weight. During implementation, the event influence can be determined using the aforementioned formula (1-13).

[0196] Step S133: Determine the mutual influence based on the attribute multi-head attention score corresponding to the first attribute prefix sequence, the event multi-head attention score corresponding to the first event prefix sequence, and the event attention weight.

[0197] Here, the attribute multi-head attention score may include but is not limited to the activity multi-head attention score, the resource multi-head attention score, the time multi-head attention score, etc.

[0198] The mutual influence represents the degree of mutual influence between the first historical events in the first event prefix sequence and the degree of mutual influence between the attributes of the first historical events in the first attribute prefix sequence. This mutual influence may include, but is not limited to, a first mutual influence between activities, a second mutual influence between resources, a third mutual influence between times, and a fourth mutual influence between events.

[0199] The first mutual influence can be determined by the activity multi-head attention score and the event attention weight. The determination method of the first mutual influence can be any suitable method. For example, the first mutual influence can be the product between the activity multi-head attention score and the event attention weight, the weight of the product, etc.

[0200] In some embodiments, the first mutual influence m can be determined by the following formula (1-16): a ,Right now:

[0201]

[0202] in, represents the activity multi-head attention score, and α represents the event attention weight.

[0203] The second mutual influence can be determined by the resource multi-head attention score and the event attention weight. The second mutual influence can be determined in any suitable manner. For example, the second mutual influence can be the product between the resource multi-head attention score and the event attention weight, the weight of the product, etc.

[0204] In some embodiments, the second interaction influence m can be determined by the following formula (1-17): r ,Right now:

[0205]

[0206] in, represents the resource multi-head attention score, and α represents the event attention weight.

[0207] The third mutual influence can be determined by the time multi-head attention score and the event attention weight. The third mutual influence can be determined in any suitable manner. For example, the third mutual influence can be the product between the time multi-head attention score and the event attention weight, the weight of the product, etc.

[0208] In some embodiments, the third mutual influence m can be determined by the following formula (1-18): t ,Right now:

[0209]

[0210] in, represents the temporal multi-head attention score, and α represents the event attention weight.

[0211] The fourth mutual influence can be determined by the event multi-head attention score and the event attention weight. The fourth mutual influence can be determined in any suitable manner. For example, the fourth mutual influence can be the product of the event multi-head attention score and the event attention weight, the weight of the product, etc.

[0212] In some embodiments, the fourth interaction influence m can be determined by the following formula (1-19): e ,Right now:

[0213]

[0214] in, Event multi-head attention score, α represents the event attention weight.

[0215] In the implementation mode of the present application, each influence is determined based on the attribute coding feature, attribute attention weight, event attention weight, attribute multi-head attention score and event multi-head attention score, thereby improving the accuracy of each influence, thereby improving the credibility of the prediction result while also improving the utilization rate of the prediction result.

[0216] Figure 2 Schematic diagram of the implementation process of a method for determining a business process activity provided in an embodiment of the present application Figure 2 ,like Figure 2 As shown, the method includes steps S21 to S24, wherein:

[0217] Step S21 : Determine a first attribute prefix sequence corresponding to the target business process based on an attribute of at least one first historical event in a first event prefix sequence corresponding to the target business process.

[0218] Step S22: Utilize the trained activity prediction model to determine the prediction result of the target business process based on the first attribute prefix sequence. The prediction result of the target business process includes the next activity of the target business process and the explanatory information of the next activity of the target business process. The explanatory information of the next activity of the target business process includes at least one of the following: attribute influence, event influence, and mutual influence. The attribute influence represents the degree of influence of the attributes of each first historical event in the first attribute prefix sequence on the next activity of the target business process. The event influence represents the degree of influence of each first historical event in the first event prefix sequence on the next activity of the target business process. The mutual influence represents the degree of mutual influence between each first historical event in the first event prefix sequence and the degree of mutual influence between the attributes of each first historical event in the first attribute prefix sequence.

[0219] Here, the above steps S21 and S22 correspond to the above steps S11 and S12 respectively. When implementing, please refer to the specific implementation of the above steps S11 and S12.

[0220] Step S23: Determine the credibility corresponding to the explanation information of the next activity of the target business process and the fidelity corresponding to the explanation information of the next activity of the target business process.

[0221] Here, credibility is to evaluate the trustworthiness of the explanation information by perturbing the most important attributes in the explanation information and observing the changes in the prediction of the activity prediction model.

[0222] In some implementations, each attribute in the attribute prefix sequence can be sequentially set to a target value to obtain multiple perturbation attribute prefix sequences. The credibility is then determined based on the prediction results corresponding to each perturbation prefix sequence. The target value can be any suitable value, such as 0. In practice, removing unimportant attributes will have a minimal impact on the prediction results; removing important attributes will have a significant impact on the prediction results.

[0223] In some implementations, the credibility can be determined based on a prediction result corresponding to at least one disturbance attribute prefix sequence and a prediction result of the target business process. The disturbance attribute prefix sequence is a sequence obtained by setting an attribute in the first attribute prefix sequence to a target value.

[0224] For example, the difference between the prediction result corresponding to a disturbance attribute prefix sequence and the prediction result of the target business process is used as the credibility.

[0225] For another example, the average value of the difference between the prediction results corresponding to each disturbance attribute prefix sequence and the prediction result of the target business process is used as the credibility.

[0226] In some embodiments, the credibility S can be determined according to the following formula (2-1): faithfulness ,Right now:

[0227]

[0228] Among them, y(x (i) ) represents the prediction result of the i-th sample, represents the prediction result corresponding to the perturbation prefix sequence corresponding to the i-th sample, and N represents the total number of samples to be evaluated.

[0229] Fidelity reflects the consistency between the predicted trend and the importance sign after the perturbation of the attribute. If the predicted trend and the importance sign are consistent after the perturbation of these attributes, the interpretation information is considered to be true. This fidelity can be determined in any suitable manner. For example, the mean of the target number can be used as the fidelity, where the target number refers to the number of predicted results whose trend and importance sign are consistent. Another example is the weighted value of the mean.

[0230] In some embodiments, the target quantity C may be determined according to the following formula (2-2):

[0231]

[0232] Among them, y(x (i) ) represents the prediction result of the i-th sample, represents the prediction result corresponding to the perturbation prefix sequence corresponding to the i-th sample, N represents the total number of samples to be evaluated, and sign(·) is the sign function.

[0233] In some embodiments, the fidelity S can be determined according to the following formula (2-3): truthfulness ,Right now:

[0234]

[0235] Where N is the total number of samples to be evaluated and C is the target number.

[0236] Step S24 : Determine an evaluation result of the next activity of the target business process based on the credibility corresponding to the interpretation information of the next activity of the target business process and the fidelity corresponding to the interpretation information of the next activity of the target business process.

[0237] Here, the evaluation result may include but is not limited to a first evaluation result, a second evaluation result, etc. The first evaluation result indicates that the interpretation information is credible, and the second evaluation result indicates that the interpretation information is uncredible. During implementation, the evaluation result can be used as a reference for whether to adopt the prediction result.

[0238] The evaluation result may be determined in any suitable manner. In some embodiments, if the credibility exceeds the credibility threshold and the fidelity exceeds the fidelity threshold, the first evaluation result is used as the evaluation result; conversely, if the credibility does not exceed the credibility threshold and / or the fidelity does not exceed the fidelity threshold, the second evaluation result is used as the evaluation result. The credibility threshold may be any suitable value. In some embodiments, different attributes may have the same or different credibility thresholds. For example, the credibility threshold of the activity may be 0.2, the credibility threshold of the resource may be 0.1, and the credibility threshold of the time may be 0.02. The fidelity threshold may be any suitable value. For example, the fidelity threshold may be 0.4.

[0239] In the embodiment of the present application, the interpretation information is evaluated based on two indicators: the credibility of the interpretation information and the fidelity of the interpretation information. This not only achieves a quantitative evaluation of the interpretation information, but also improves the utilization rate of the prediction results by the user.

[0240] In some embodiments, step S23 includes steps S231 to S233, wherein:

[0241] Step S231 : Determine at least one disturbance attribute prefix sequence based on the first attribute prefix sequence and the explanation information of the next activity of the target business process.

[0242] Here, the length of the disturbance attribute prefix sequence is the same as the length of the first attribute prefix sequence. In implementation, important attributes can be set as target values according to the interpretation information to obtain at least one disturbance attribute prefix sequence.

[0243] For example, the explanation information believes that activity a1 in the activity prefix sequence has a greater impact on the prediction result. Then, activity a3 in the activity prefix sequence can be set to 0 to obtain the disturbance attribute prefix sequence.

[0244] For another example, the interpretation information believes that activity a1 in the activity prefix sequence and resource r2 in the resource prefix sequence have a greater impact on the prediction result. Then, activity a1 in the activity prefix sequence can be set to 0 to obtain a disturbance attribute prefix sequence, resource r2 in the resource prefix sequence can be set to 0 to obtain a disturbance attribute prefix sequence, and / or activity a1 in the activity prefix sequence can be set to 0 and resource r2 in the resource prefix sequence can be set to 0 to obtain a disturbance attribute prefix sequence.

[0245] Step S232: For each disturbance prefix sequence, using the trained activity prediction model, based on the disturbance attribute prefix sequence, determine a prediction result corresponding to the disturbance attribute prefix sequence.

[0246] Here, the prediction result corresponding to the disturbance attribute prefix sequence may include, but is not limited to, the next activity, explanation information for the next activity, etc. The process for determining the prediction result corresponding to the disturbance attribute prefix sequence can be found in the specific implementation of step S12 above. During implementation, the disturbance attribute prefix sequence is input into the activity prediction model to obtain the prediction result corresponding to the disturbance attribute prefix sequence.

[0247] Step S233: Determine the credibility and fidelity of the explanation information of the next activity of the target business process based on the prediction result of the target business process and the prediction result corresponding to each disturbance attribute prefix sequence.

[0248] Here, the credibility is to evaluate the trustworthiness of the explanatory information by perturbing the most important attributes in the explanatory information and observing the predicted changes of the activity prediction model. The method for determining the credibility can be any appropriate method. In some embodiments, the difference between the prediction result corresponding to a certain perturbation attribute prefix sequence and the prediction result of the target business process is used as the credibility. In some embodiments, the average value of the difference between the prediction result corresponding to each perturbation attribute prefix sequence and the prediction result of the target business process is used as the credibility. In some embodiments, the credibility can be determined according to the above formula (2-1).

[0249] Fidelity reflects the consistency between the trend of change in the predicted result and the sign of the importance after the perturbation attribute. The fidelity can be determined in any suitable manner. In some embodiments, the mean of the target number, a weighted value of the mean, etc. is used as the fidelity. In some embodiments, the fidelity can be determined according to the above formula (2-3).

[0250] In the implementation manner of the present application, on the one hand, the disturbance attribute prefix sequence is dynamically determined based on the interpretation information, thereby improving the accuracy of the disturbance attribute prefix sequence; on the other hand, the credibility and fidelity are comprehensively determined based on the prediction results of the target business process and the prediction results corresponding to each disturbance attribute prefix sequence, thereby improving the accuracy of the credibility and fidelity.

[0251] Figure 3 Schematic diagram of the implementation process of a method for determining a business process activity provided in an embodiment of the present application Figure 3 ,like Figure 3 As shown, the method includes steps S31 to S35, wherein:

[0252] Step S31: Determine a training sample set based on multiple second historical events of multiple business processes; wherein the training sample set includes a training sample corresponding to at least one business process, and the training sample corresponding to the business process includes a second attribute prefix sequence corresponding to the business process and a label value corresponding to the business process.

[0253] Here, each business process may include multiple second historical events. During implementation, the process of determining multiple second historical events of the business process may refer to the specific implementation of the aforementioned step S111.

[0254] The second attribute prefix sequence may include, but is not limited to, at least one of a second activity prefix sequence, a second resource prefix sequence, a second time prefix sequence, etc. In some implementations, different services may include the same or different second attribute prefix sequences.

[0255] The second attribute prefix sequence may be determined in any appropriate manner.

[0256] In some embodiments, the second attribute prefix sequence can be determined based on the second event prefix sequence. The second event prefix sequence includes at least one second historical event. The length of the second attribute prefix sequence is the same as the length of the second event prefix sequence. In some embodiments, the number of second event prefix sequences can be at least one. The method for determining the second event prefix sequence can be any suitable method. During implementation, the process for determining the second event prefix sequence and the second attribute prefix sequence can refer to the specific implementation of the aforementioned step S11.

[0257] Step S32: using the activity prediction model to be trained, based on the second attribute prefix sequence corresponding to each business process, determine the prediction result of each business process.

[0258] Here, the prediction result of the business process may include but is not limited to the next activity of the business process, explanation information of the next activity of the business process, etc. During implementation, the process of determining the prediction result of the business process may refer to the specific implementation of the aforementioned step S12.

[0259] Step S33: Based on the prediction result of each business process and the label value corresponding to each business process, the model parameters of the activity prediction model to be trained are updated at least once to obtain a trained activity prediction model.

[0260] Here, the updating method of the model parameters may include but is not limited to at least one of the gradient descent method, the momentum update method, the Newton momentum method, etc. During implementation, those skilled in the art can independently determine the updating method according to actual needs, and the embodiments of this application are not limited thereto.

[0261] In some embodiments, whether the model parameters of the activity prediction model need to be updated can be determined based on preset conditions. If the model parameters of the activity prediction model need to be updated, an appropriate parameter update algorithm is used to update the model parameters of the activity prediction model. After the model parameters are updated, a new prediction result is re-determined based on the next training sample. Based on the new prediction result, it is then determined whether the model parameters of the activity prediction model need to be further updated. If the model parameters of the activity prediction model do not need to be updated, the final updated activity prediction model is determined as the trained activity prediction model. The preset conditions may include, but are not limited to, at least one of the following: the number of iterations, the loss value being no greater than a threshold, and the loss value converging.

[0262] For example, a target loss value can be determined based on the prediction results of each business process and the label value corresponding to each business process. If the target loss value does not meet a first preset condition, the model parameters of the activity prediction model are updated. If the target loss value meets the first preset condition, updating the model parameters of the activity prediction model is stopped, and the final updated activity prediction model is determined as the first trained model. The preset condition may include, but is not limited to, a threshold value and a value less than the previous target loss value.

[0263] The target loss value may include, but is not limited to, at least one of a mean square error loss value, a cross entropy loss value, a contrast loss value, etc. In some embodiments, the target loss value may be calculated using the Pearson Linear Correlation Coefficient (PLCC), the Spearman Rank-order Correlation Coefficient (SRCC), the Kendall Rank-order Correlation Coefficient (KRCC), etc.

[0264] Step S34: Determine the first attribute prefix sequence corresponding to the target business process based on the attributes of at least one first historical event in the first event prefix sequence corresponding to the target business process; wherein the length of the first event prefix sequence is the same as the length of the first attribute prefix sequence, and the first attribute prefix sequence includes at least one of the following: a first activity prefix sequence, a first resource prefix sequence, and a first time prefix sequence.

[0265] Step S35: Utilize the trained activity prediction model to determine the prediction result of the target business process based on the first attribute prefix sequence. The prediction result of the target business process includes the next activity of the target business process and the explanatory information of the next activity of the target business process. The explanatory information of the next activity of the target business process includes at least one of the following: attribute influence, event influence, and mutual influence. The attribute influence represents the degree of influence of the attributes of each first historical event in the first attribute prefix sequence on the next activity of the target business process. The event influence represents the degree of influence of each first historical event in the first event prefix sequence on the next activity of the target business process. The mutual influence represents the degree of mutual influence between each first historical event in the first event prefix sequence and the degree of mutual influence between the attributes of each first historical event in the first attribute prefix sequence.

[0266] Here, the above steps S34 and S35 correspond to the above steps S11 and S12 respectively. When implementing, please refer to the specific implementation of the above steps S11 and S12.

[0267] In an embodiment of the present application, on the one hand, the overall network structure of the activity prediction model is trained through a training sample set, thereby enhancing the robustness of the activity prediction model and improving the performance of the activity prediction model; on the other hand, the model parameters of the activity prediction model are updated at least once, which can improve the consistency and accuracy of the prediction of the trained activity prediction model for the same sample, thereby enabling the trained activity prediction model to more accurately predict business process activities.

[0268] The following uses the loan application process to illustrate the application of the method provided in the embodiment of the present application.

[0269] Figure 4A Schematic diagram 4 of the implementation flow of a method for determining a business process activity provided in an embodiment of the present application, such as Figure 4A As shown, the method includes steps S401 to S405, wherein:

[0270] Step S401: Based on the historical execution log of the loan application process (corresponding to the aforementioned event log), traverse each process instance to generate a training sample set;

[0271] Here, using the identifiers of each loan application process instance, historical events in the historical execution log are aggregated to obtain individual process instances. Each process instance contains multiple events. Each event is then converted into a vector, and the activity and resource attributes of the event are processed using index encoding. During implementation, for each process instance, a 5-length event prefix sequence and the corresponding next activity are constructed as label values to facilitate training the activity prediction model. Based on the event prefix sequence, activity prefix sequences, resource prefix sequences, and time prefix sequences are constructed as training samples.

[0272] Figure 4B A schematic diagram of a loan application process provided in this application embodiment, such as Figure 4B As described, the loan application process includes application submission 411 (A_SUBMITTED), partial application submission 412 (A_PARTLYSUBMITTED), application pre-acceptance 413 (A_PREACCEPTED), application acceptance 414 (A_ACCEPTED), acceptance rejection 415, loan activation 416 (A_ACTIVATED), loan approval 417 (A_APPROVED), application completion 418 (A_FINALIZED) and loan registration (A_REGISTERED) 419.

[0273] Step S402: training the activity prediction model using the training sample set to obtain a trained activity prediction model;

[0274] Here, since time is a discrete value and cannot be indexed and coded, only the activity vector (corresponding to the aforementioned activity coding feature) obtained after encoding the activity prefix sequence, the resource vector (corresponding to the aforementioned resource coding feature) obtained after encoding the resource prefix sequence, and the time vector (the vector after the time prefix sequence is converted) are input into the activity prediction model to obtain the prediction result. Based on the prediction result and the label value, the model parameters of the activity prediction model are updated at least once to obtain the trained activity prediction model.

[0275] Figure 4C A schematic diagram of an activity prediction model provided in an embodiment of the present application is shown in FIG. Figure 4C The activity prediction model includes an encoding part, an attribute part 12, a connection part 13, an event part 14, a multiplication part 15 and a prediction part 16. The encoding part includes an activity encoding part 111 and a resource encoding part 112. The attribute part 12 includes an activity transformer part 121, a resource transformer part 122, a time transformer part 123, an activity attention weight part 124, a resource attention weight part 125, a time attention weight part 126, an activity multiplication part 127, a resource multiplication part 128 and a time multiplication part 129. The event part 14 includes an event transformer part 141 and an event attention weight part 142, wherein:

[0276] The activity coding part 111 is used to encode the activity prefix sequence to obtain an activity coding feature;

[0277] The resource coding part 112 is used to encode the resource prefix sequence to obtain a resource coding feature;

[0278] The activity Transformer part 121 is used to encode features according to the activity Determine the intermediate state characteristics of the activity g a and active multi-head attention scores

[0279] The resource transformer part 122 is used to encode features according to the resource Determine the intermediate state characteristics of resources g r and resource multi-head attention scores

[0280] The time transformer part 123 is used to determine the time intermediate state feature g according to the time vector t and temporal multi-head attention scores

[0281] The activity attention weight part 124 is used to calculate the activity attention weight according to the activity intermediate state feature g. a Determine activity attention weight β a ;

[0282] The resource attention weight part 125 is used to calculate the resource intermediate state feature g. r Determine resource attention weight β r ;

[0283] The time attention weight part 126 is used to calculate the time intermediate state feature g t Determine the temporal attention weight β t ;

[0284] The activity multiplication section 127 is used to encode the characteristics of the activity and activity attention weight β a Determine the characteristics of the activity

[0285] The resource multiplication part 128 is used to encode the resource characteristics according to the resource and resource attention weight β r Determine the characteristics of the resource

[0286] The time multiplication part 129 is used to calculate the time vector v t and temporal attention weight β t Determine the time characteristics

[0287] The connection part 13 is used to Resource characteristics and time characteristics Determine the target characteristics

[0288] The event transformer portion 141 is used to Determine the intermediate state characteristics g of the event e and event multi-head attention scores

[0289] The event attention weight part 142 is used to calculate the event attention weight according to the target feature. Determine the event attention weight α i ;

[0290] The multiplication part 15 is used to calculate the target feature and event attention weight α i Determine context c;

[0291] The prediction part 16 is used to determine the prediction result according to the context c.

[0292] Figure 4D The implementation process frame of a method for determining a business process activity provided in an embodiment of the present application Figure 1 ,like Figure 4D As shown, the method includes: performing information extraction processing on the event log 441 to obtain an attribute prefix sequence 442, inputting the attribute prefix sequence 442 into an interpretable activity prediction model 443, and then obtaining the next activity 444 and the explanation information 445 of the next activity.

[0293] Step S403: Determine the attribute prefix sequence based on the operation log of the loan application process (corresponding to the aforementioned event log);

[0294] Step S404: Input the attribute coding features obtained by encoding the attribute prefix sequence into the activity prediction model to obtain the next activity of the loan application process and explanation information of the next activity;

[0295] Here, the explanation information includes attribute influence, event influence, and mutual influence. Figure 4E A schematic diagram of an explanation information provided in an embodiment of the present application, such as Figure 4E As shown in Figure 2, each event and its attributes will affect the prediction results, and events and attributes will also affect each other.

[0296] for Figure 4B In the loan application process shown in FIG. 1 , if the attributes of each event in the event prefix sequence are as shown in Table 1 below, then the next activity is predicted to be A_REGISTERED (loan registration).

[0297] Table 1 Events and attributes in the event prefix sequence

[0298]

[0299] In some embodiments, different events have different effects on the predicted outcome.

[0300] In some implementations, the attributes of an event may include the activities performed, the resources used, the time difference from the start time of the process, etc. The value of each attribute has different degrees of impact on the upcoming activities. The impact size includes positive impact and negative impact. The larger the absolute value, the greater the impact on the next activity.

[0301] In some implementations, the activities performed by an event may have an impact on the predicted results. Activity influence represents the degree of influence of the activities performed by an event on the upcoming activities.

[0302] In some implementations, the resources (eg, performers) occupied by an event may have an impact on the prediction result. Resource influence indicates the degree of influence of the resources occupied by an event on the upcoming activity.

[0303] In some embodiments, the time corresponding to the event (e.g., the time interval between the occurrence of the first event) may have an impact on the prediction result. The time influence indicates the degree of influence of the time corresponding to the event on the upcoming activity.

[0304] In some implementations, various events and various attributes may affect each other.

[0305] Step S405: Evaluate the credibility and fidelity of the interpretation information of the next activity to obtain an evaluation result.

[0306] Here, credibility is an assessment of the trustworthiness of an explanation by perturbing its most important attributes and observing the changes in the predictions of the activity prediction model. Fidelity reflects the consistency between the changing trend of the prediction results after the perturbation and the sign of the importance.

[0307] The credibility is determined according to the above formula (2-1) and the fidelity is determined according to formula (2-3). The results are shown in Table 2 below, where:

[0308] Table 2 Credibility and fidelity of interpretation information

[0309] Activity Role time Credibility 0.431 0.129 0.020 Fidelity 0.732 0.428 0.564

[0310] In terms of credibility, since the activity credibility of 0.431 exceeds 0.2 (activity credibility threshold), the role credibility of 0.129 exceeds 0.1 (resource credibility threshold), and the time credibility of 0.020 exceeds 0.02 (time credibility threshold), the credibility of the explanation information is relatively credible;

[0311] In terms of fidelity, since the fidelity of each attribute exceeds 0.4 (fidelity threshold), the fidelity of the explanation information is also credible.

[0312] Combining the two types of evaluation indicators, it can be considered that the explanatory information obtained is more credible.

[0313] In an embodiment of the present application, first, the attribute prefix sequence is dynamically determined by the attributes of each historical event of the target business process, thereby improving the accuracy of the attribute prefix sequence; secondly, the next activity is predicted based on the attribute prefix sequence using the trained activity prediction model. Since the attribute prefix sequence is determined according to the attributes of each historical event in the business process, the prediction result is more in line with business needs, which improves the accuracy of the prediction result while also improving the prediction efficiency; thirdly, the activity prediction model also generates interpretable information including attribute influence, event influence, mutual influence, etc. when generating the prediction result, which is used to describe the important influencing factors (for example, certain events, certain attributes, etc.) that produce the prediction result and the degree of internal influence between process-related events and attributes, thereby not only improving the credibility of the prediction result, but also improving the usage rate of the prediction result by users; finally, the explanation information is comprehensively evaluated through credibility and fidelity, which not only achieves quantitative evaluation of the explanation information, but also improves the usage rate of the prediction result by users.

[0314] Based on the above embodiments, the present application also provides a device for determining a business process activity. Figure 5 A schematic diagram of the structure of a device for determining a business process activity provided in an embodiment of the present application is shown in FIG. Figure 5 As shown, the business process activity determination device 50 includes a first determination module 51 and a second determination module 52, wherein:

[0315] A first determining module 51 is configured to determine a first attribute prefix sequence corresponding to the target business process based on an attribute of at least one first historical event in a first event prefix sequence corresponding to the target business process; wherein the length of the first event prefix sequence is the same as the length of the first attribute prefix sequence, and the first attribute prefix sequence includes at least one of the following: a first activity prefix sequence, a first resource prefix sequence, and a first time prefix sequence;

[0316] The second determination module 52 is used to use the trained activity prediction model to determine the prediction result of the target business process based on the first attribute prefix sequence. The prediction result of the target business process includes the explanatory information of the next activity of the target business process. The explanatory information of the next activity of the target business process includes at least one of the following: attribute influence, event influence, and mutual influence. The attribute influence represents the degree of influence of the attributes of each first historical event in the first attribute prefix sequence on the next activity of the target business process. The event influence represents the degree of influence of each first historical event in the first event prefix sequence on the next activity of the target business process. The mutual influence represents the degree of mutual influence between each first historical event in the first event prefix sequence and the degree of mutual influence between the attributes of each first historical event in the first attribute prefix sequence.

[0317] In some embodiments, the first determination module 51 is also used to: determine multiple first historical events of the target business process based on the event log of the target business process; determine a first event prefix sequence based on a first event sequence formed by multiple first historical events of the target business process; wherein the length of the first event prefix sequence is not greater than the length of the first event sequence; determine a first attribute prefix sequence based on the attributes of at least one first historical event in the first event prefix sequence.

[0318] In some embodiments, the attributes of the first historical event include at least one of the following: the activities performed by the first historical event, the resources occupied by the first historical event, and the time corresponding to the first historical event; the first determination module 51 is further used to: when the first attribute prefix sequence includes the first activity prefix sequence, determine the first activity prefix sequence based on the activities performed by at least one first historical event in the first event prefix sequence; when the first attribute prefix sequence includes the first resource prefix sequence, determine the first resource prefix sequence based on the resources occupied by at least one first historical event in the first event prefix sequence; when the first attribute prefix sequence includes the first time prefix sequence, determine the first time prefix sequence based on the time corresponding to at least one first historical event in the first event prefix sequence; wherein the time corresponding to the first historical event is determined based on the completion time of the activity performed by the first historical event and the completion time of the activity performed by the target first historical event, and the target first historical event includes one of the following: the previous first historical event of the first historical event, the first first historical event.

[0319] In some embodiments, the second determination module 52 is further used to: encode the first attribute prefix sequence to obtain attribute coding features; determine the target features corresponding to the first attribute prefix sequence based on the attribute coding features; and determine the prediction results of the target business process based on the target features corresponding to the first attribute prefix sequence.

[0320] In some embodiments, the second determination module 52 is further used to: determine the attribute intermediate state feature based on the attribute encoding feature; determine the attribute attention weight based on the attribute intermediate state feature; and determine the target feature corresponding to the first attribute prefix sequence based on the attribute encoding feature and the attribute attention weight.

[0321] In some embodiments, the second determination module 52 is also used to: determine the intermediate state characteristics of the event based on the target characteristics corresponding to the first attribute prefix sequence; determine the event attention weight based on the intermediate state characteristics of the event; and determine the prediction result of the target business process based on the target characteristics and event attention weight corresponding to the first attribute prefix sequence.

[0322] In some embodiments, the second determination module 52 is also used to: determine the attribute influence based on the attribute coding features and the attribute attention weight; determine the event influence based on the event attention weight; determine the mutual influence based on the attribute multi-head attention score corresponding to the first attribute prefix sequence, the event multi-head attention score corresponding to the first event prefix sequence, and the event attention weight.

[0323] In some embodiments, the device also includes a third determination module, which is used to: determine the credibility corresponding to the explanatory information of the next activity of the target business process and the fidelity corresponding to the explanatory information of the next activity of the target business process; and determine the evaluation result of the next activity of the target business process based on the credibility corresponding to the explanatory information of the next activity of the target business process and the fidelity corresponding to the explanatory information of the next activity of the target business process.

[0324] In some embodiments, the third determination module is further used to: determine at least one disturbance attribute prefix sequence based on the first attribute prefix sequence and the explanatory information of the next activity of the target business process; wherein the length of the disturbance attribute prefix sequence is the same as the length of the first attribute prefix sequence; for each disturbance prefix sequence, using the trained activity prediction model, based on the disturbance attribute prefix sequence, determine the prediction result corresponding to the disturbance attribute prefix sequence; based on the prediction result of the target business process and the prediction result corresponding to each disturbance attribute prefix sequence, determine the credibility corresponding to the explanatory information of the next activity of the target business process and the fidelity corresponding to the explanatory information of the next activity of the target business process.

[0325] In some embodiments, the device also includes a training module, which is used to: determine a training sample set based on multiple second historical events of multiple business processes; wherein the training sample set includes at least one training sample corresponding to a business process, and the training sample corresponding to the business process includes a second attribute prefix sequence corresponding to the business process and a label value corresponding to the business process; using the activity prediction model to be trained, based on the second attribute prefix sequence corresponding to each business process, determine the prediction result of each business process; based on the prediction result of each business process and the label value corresponding to each business process, update the model parameters of the activity prediction model to be trained at least once to obtain a trained activity prediction model.

[0326] The description of the above device embodiment is similar to the description of the above method embodiment and has similar beneficial effects as the method embodiment. In some embodiments, the functions or modules included in the device provided in the embodiments of the present application can be used to perform the methods described in the above method embodiments. For technical details not disclosed in the device embodiments of the present application, please refer to the description of the method embodiments of the present application for understanding.

[0327] It should be noted that, in the embodiment of the present application, if the above method is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for making a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the methods of each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM), a magnetic disk or an optical disk. In this way, the embodiment of the present application is not limited to any specific hardware, software or firmware, or any combination of hardware, software and firmware.

[0328] An embodiment of the present application provides a computer device including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and the above method is implemented when the processor executes the computer program.

[0329] The embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the above method when executed by a processor. The computer-readable storage medium can be transient or non-transient.

[0330] The present application provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program. When the computer program is read and executed by a computer, the above method is implemented. The computer program product can be implemented in hardware, software, or a combination thereof. In some embodiments, the computer program product is embodied as a computer storage medium. In other embodiments, the computer program product is embodied as a software product, such as a software development kit (SDK).

[0331] It should be noted that Figure 6 This is a hardware entity diagram of an electronic device provided in an embodiment of the present application, such as Figure 6 As shown, the hardware entity of the electronic device 600 includes: a processor 601, a communication interface 602 and a memory 603, wherein:

[0332] The processor 601 generally controls the overall operations of the electronic device 600 .

[0333] The communication interface 602 enables the electronic device to communicate with other terminals or servers through a network.

[0334] The memory 603 is configured to store instructions and applications executable by the processor 601, and can also cache data to be processed or processed by the processor 601 and various modules in the electronic device 600 (for example, image data, audio data, voice communication data, and video communication data). This can be implemented using flash memory (FLASH) or random access memory (RAM). Data can be transmitted between the processor 601, the communication interface 602, and the memory 603 via a bus 604.

[0335] It should be noted that the description of the above storage medium and device embodiments is similar to the description of the above method embodiments and has similar beneficial effects as the method embodiments. For technical details not disclosed in the storage medium and device embodiments of this application, please refer to the description of the method embodiments of this application for understanding.

[0336] It should be understood that "one embodiment" or "an embodiment" mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned steps / processes does not mean the order of execution, and the execution order of each step / process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. The above-mentioned serial numbers of the embodiments of the present application are for description only and do not represent the advantages and disadvantages of the embodiments.

[0337] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0338] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0339] The units described above as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, the functional units in the various embodiments of the present application may all be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0340] Those skilled in the art will understand that all or part of the steps of implementing the above-mentioned method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, and other media that can store program codes.

[0341] Alternatively, if the above-mentioned integrated unit of the present application is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can essentially or in other words be embodied in the form of a software product that contributes to the relevant technology. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods of each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0342] The above is only an implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any changes or replacements that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed in this application should be covered by the scope of protection of the present application.

Claims

1. A method for determining a business process activity, characterized in that: include: Determine, based on an attribute of at least one first historical event in a first event prefix sequence corresponding to a target business process, a first attribute prefix sequence corresponding to the target business process; wherein the length of the first event prefix sequence is the same as the length of the first attribute prefix sequence, and the first attribute prefix sequence includes at least one of the following: a first activity prefix sequence, a first resource prefix sequence, and a first time prefix sequence; Utilizing the trained activity prediction model, based on the first attribute prefix sequence, the prediction result of the target business process is determined. The prediction result of the target business process includes the next activity of the target business process and the explanatory information of the next activity of the target business process. The explanatory information of the next activity of the target business process includes at least one of the following: attribute influence, event influence, and mutual influence. The attribute influence represents the degree of influence of the attributes of each first historical event in the first attribute prefix sequence on the next activity of the target business process. The event influence represents the degree of influence of each first historical event in the first event prefix sequence on the next activity of the target business process. The mutual influence represents the degree of mutual influence between each first historical event in the first event prefix sequence and the degree of mutual influence between the attributes of each first historical event in the first attribute prefix sequence.

2. The determination method according to claim 1, characterized in that The determining, based on the attribute of at least one first historical event in the first event prefix sequence corresponding to the target business process, a first attribute prefix sequence corresponding to the target business process includes: Determining a plurality of first historical events of the target business process based on the event log of the target business process; Determining the first event prefix sequence based on a first event sequence formed by a plurality of first historical events of the target business process; wherein the length of the first event prefix sequence is not greater than the length of the first event sequence; The first attribute prefix sequence is determined based on an attribute of at least one first historical event in the first event prefix sequence.

3. The determination method according to claim 2, characterized in that: The attributes of the first historical event include at least one of the following: an activity performed by the first historical event, resources occupied by the first historical event, and a time corresponding to the first historical event; The determining the first attribute prefix sequence based on the attribute of at least one first historical event in the first event prefix sequence includes: In a case where the first attribute prefix sequence includes the first activity prefix sequence, determining the first activity prefix sequence based on an activity performed by at least one first historical event in the first event prefix sequence; In a case where the first attribute prefix sequence includes the first resource prefix sequence, determining the first resource prefix sequence based on resources occupied by at least one first historical event in the first event prefix sequence; In the case where the first attribute prefix sequence includes the first time prefix sequence, the first time prefix sequence is determined based on the time corresponding to at least one first historical event in the first event prefix sequence; wherein the time corresponding to the first historical event is determined based on the completion time of the activity performed by the first historical event and the completion time of the activity performed by the target first historical event, and the target first historical event includes one of the following: the previous first historical event of the first historical event, the first first historical event.

4. The determination method according to claim 1, characterized in that The determining, based on the first attribute prefix sequence, a prediction result of the target business process includes: performing encoding processing on the first attribute prefix sequence to obtain an attribute encoding feature; Determining a target feature corresponding to the first attribute prefix sequence based on the attribute encoding feature; A prediction result of the target business process is determined based on the target feature corresponding to the first attribute prefix sequence.

5. The determination method according to claim 4, characterized in that: The determining, based on the attribute encoding feature, a target feature corresponding to the first attribute prefix sequence includes: Determining attribute intermediate state features based on the attribute encoding features; Determining attribute attention weights based on the attribute intermediate state features; Based on the attribute encoding feature and the attribute attention weight, a target feature corresponding to the first attribute prefix sequence is determined.

6. The determination method according to claim 4, characterized in that: The determining the prediction result of the target business process based on the target feature corresponding to the first attribute prefix sequence includes: Determining an event intermediate state feature based on a target feature corresponding to the first attribute prefix sequence; Determining an event attention weight based on the intermediate state characteristics of the event; Based on the target feature corresponding to the first attribute prefix sequence and the event attention weight, a prediction result of the target business process is determined.

7. The determination method according to claim 4, characterized in that: The determination method further includes: Determining the attribute influence based on the attribute encoding feature and the attribute attention weight; Determine the influence of the event based on the event attention weight; The mutual influence is determined based on the attribute multi-head attention score corresponding to the first attribute prefix sequence, the event multi-head attention score corresponding to the first event prefix sequence, and the event attention weight.

8. The determination method according to claim 1, characterized in that: The determination method further includes: Determining the credibility corresponding to the interpretation information of the next activity of the target business process and the fidelity corresponding to the interpretation information of the next activity of the target business process; An evaluation result of the next activity of the target business process is determined based on the credibility corresponding to the interpretation information of the next activity of the target business process and the fidelity corresponding to the interpretation information of the next activity of the target business process.

9. The determination method according to claim 8, characterized in that: The determining of the credibility corresponding to the interpretation information of the next activity of the target business process and the fidelity corresponding to the interpretation information of the next activity of the target business process includes: Determining at least one disturbance attribute prefix sequence based on the first attribute prefix sequence and the interpretation information of the next activity of the target business process; wherein the length of the disturbance attribute prefix sequence is the same as the length of the first attribute prefix sequence; For each disturbance prefix sequence, using the trained activity prediction model and based on the disturbance attribute prefix sequence, determining a prediction result corresponding to the disturbance attribute prefix sequence; Based on the prediction result of the target business process and the prediction result corresponding to each of the disturbance attribute prefix sequences, the credibility corresponding to the explanation information of the next activity of the target business process and the fidelity corresponding to the explanation information of the next activity of the target business process are determined.

10. The determination method according to any one of claims 1 to 9, characterized in that: The determination method further includes: Determine a training sample set based on multiple second historical events of multiple business processes; wherein the training sample set includes a training sample corresponding to at least one business process, and the training sample corresponding to the business process includes a second attribute prefix sequence corresponding to the business process and a label value corresponding to the business process; Determining a prediction result for each of the business processes based on a second attribute prefix sequence corresponding to each of the business processes using the activity prediction model to be trained; Based on the prediction result of each of the business processes and the label value corresponding to each of the business processes, the model parameters of the activity prediction model to be trained are updated at least once to obtain a trained activity prediction model.

11. A device for determining a business process activity, characterized in that: include: A first determining module is configured to determine a first attribute prefix sequence corresponding to the target business process based on an attribute of at least one first historical event in a first event prefix sequence corresponding to the target business process; wherein the length of the first event prefix sequence is the same as the length of the first attribute prefix sequence, and the first attribute prefix sequence includes at least one of the following: a first activity prefix sequence, a first resource prefix sequence, and a first time prefix sequence; The second determination module is used to use the trained activity prediction model to determine the prediction result of the target business process based on the first attribute prefix sequence. The prediction result of the target business process includes the next activity of the target business process and the explanatory information of the next activity of the target business process. The explanatory information of the next activity of the target business process includes at least one of the following: attribute influence, event influence, and mutual influence. The attribute influence represents the degree of influence of the attributes of each first historical event in the first attribute prefix sequence on the next activity of the target business process. The event influence represents the degree of influence of each first historical event in the first event prefix sequence on the next activity of the target business process. The mutual influence represents the degree of mutual influence between each first historical event in the first event prefix sequence and the degree of mutual influence between the attributes of each first historical event in the first attribute prefix sequence.

12. An electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, wherein: When the processor executes the program, the determination method according to any one of claims 1 to 10 is implemented.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the determination method according to any one of claims 1 to 10 is implemented.

14. A computer program product, characterized in that The computer program product comprises a non-transitory computer-readable storage medium storing a computer program, and when the computer program is read and executed by a computer, the method according to any one of claims 1 to 10 is implemented.

Images