CN117493583B - Method and system for generating flow operation sequence by combining event log and knowledge graph - Google Patents
Method and system for generating flow operation sequence by combining event log and knowledge graph Download PDFInfo
- Publication number
- CN117493583B CN117493583B CN202410006733.XA CN202410006733A CN117493583B CN 117493583 B CN117493583 B CN 117493583B CN 202410006733 A CN202410006733 A CN 202410006733A CN 117493583 B CN117493583 B CN 117493583B
- Authority
- CN
- China
- Prior art keywords
- event
- representing
- sequence
- candidate
- log
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 78
- 230000008569 process Effects 0.000 claims description 45
- 239000011159 matrix material Substances 0.000 claims description 20
- 230000006870 function Effects 0.000 claims description 10
- 238000011112 process operation Methods 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 5
- 238000011161 development Methods 0.000 claims description 5
- 238000005516 engineering process Methods 0.000 abstract description 6
- 230000009286 beneficial effect Effects 0.000 abstract description 3
- 238000004364 calculation method Methods 0.000 description 5
- 238000005065 mining Methods 0.000 description 3
- 238000004801 process automation Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000013105 post hoc analysis Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Animal Behavior & Ethology (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a flow operation sequence generation method and system combining an event log and a knowledge graph, which relate to the technical field of knowledge graphs, utilize historical event log information and related flow knowledge graphs related to a target generation sequence flow, flexibly and comprehensively represent events by combining a transform technology, learn context information existing in the event log by using LSTM, learn evolution mode information existing in the knowledge graph by using SGNN technology, provide next event prediction by combining the context information and the event evolution mode, and iteratively run on the basis, so as to generate a reliable flow operation sequence. The beneficial effects of the invention are as follows: by combining time sequence data and a knowledge graph method, the next event prediction is realized more flexibly and accurately, and the method has better universality and effectiveness.
Description
Technical Field
The invention relates to the technical field of knowledge graphs, in particular to a method and a system for generating a flow operation sequence by combining an event log and a knowledge graph.
Background
With the trend of digital transformation and the convenience of electronic information storage, event logs of business systems are growing on a huge scale. The massive event logs can enable organizations to discover, monitor and improve business through process mining methods, while organizations are actively investing in providing solutions through predictive analysis and automated techniques to gain insight into and improve their performance because of the rapid changes in markets, which organizations want to adapt more quickly and accurately to changes in markets, rather than improving through post-hoc analysis.
Robot process automation (Robotic Process Automation, RPA) is a novel technical concept. The system allows a software robot to simulate and execute a given business process based on interaction action of a certain rule, and can simulate operation behaviors of a person between different systems on a computer to replace the person to automatically execute the business process based on the rule and with high repeatability before the computer. Rather, it is not a real, macroscopic robot, but rather a process automation service. Generating executable code is the last step of RPA, for which it is necessary to know the sequence of flow operations that the target business flow may occur to introduce flow mining techniques. Wherein predictive business process monitoring (PREDICTIVE BUSINESS PROCESS MONITORING, PBPM) is an important area of process mining, focusing on estimating future features of an ongoing business process. Among other things, predicting the next event (Predict Next Event) is an important issue in PBPM. By predicting the next event technique, a prediction of the future operation of the operational flow may be obtained, providing a valid reference for the RPA to generate code at runtime.
In the prior art, knowledge graphs lack participation on actual time sequence data, so that the knowledge graphs cannot be flexibly adapted to event prediction in an actual scene, the relation reflected by log data is often linear, but complex information such as branches and the like may exist in event development.
Disclosure of Invention
In order to solve the problems, the invention provides a flow operation sequence generation method combining an event log and a knowledge graph, which combines the event log as a time sequence data carrier and the knowledge graph verified by expert knowledge, so that the advantages of time sequence data prediction are reserved, and meanwhile, an event development rule and an evolution mode are fused to improve the accuracy of operation sequence prediction.
A flow operation sequence generation method combining event logs and knowledge maps comprises the following steps:
S1: acquiring a knowledge graph Q and an event log which are associated with a target flow P of which an operation sequence needs to be generated;
S2: selecting events from the event log according to the running state of the target process P to obtain a predicted event sequence eQueue;
s3: according to the predicted event sequence eQueue, vector representations of each event and vector representations of candidate event sets of the next event are obtained, and all event representations are obtained;
S4: converting the predicted event sequence obtained in the step S3 into event representation fused with context time sequence information through a standard LSTM;
s5: converting all event representations and knowledge patterns Q obtained in the step S3 into event representations fused with global evolution information through SGNN;
S6: obtaining the event with highest occurrence probability according to the event representation H obtained in the step S4 and the event representation E obtained in the step S5 ；
S7: handle eventAdded to the predicted event sequence eQueue and performed on event/>Making a judgment if event/>If the event is an event representing the end of the flow or a signal of artificial termination is received, the predicted event sequence eQueue is output as a predicted flow operation sequence; otherwise, go back to step S3 to continue with the above steps.
Further, the step S1 specifically includes:
S1.1: sending a request for acquiring the related knowledge graph to a knowledge graph base according to the flow description of the target flow P, and obtaining a knowledge graph Q associated with the target flow P;
S1.2: and selecting event log records log 1,…,logm of m executed process instances from the historical event log according to the process description of the target process P, wherein log 1 represents the event log record of the 1 st process instance, and log m represents the event log record of the m th process instance in the event log.
Further, the step S2 specifically includes:
S2.1: when the target process P is in an operation state, namely the target process P is an executing process, taking out an event log, taking out events in the event log according to time sequence and adding the events into a set predicted event sequence eQueue;
S2.2: when the target flow P is in an unoperated state, the event log is fetched according to the description of the target flow P, and part of the events in the event log are fetched and added to the predicted event sequence eQueue.
Further, the step S3 specifically includes:
s3.1: learning respective embeddings of event content e c of event e and subscript e t of event e in predicted event sequence eQueue by using a transducer encoder to obtain event content embeddings And subscript embedding/>；
S3.2: according to the formulaObtaining a vector representation of each event, obtaining a vector representation of the predicted event sequence as/>N is a positive integer;
S3.3: obtaining event logs of m finished process instances from the event logs log, and obtaining vector representations of m candidate events of the predicted event sequence through steps S3.1 and S3.2:
Where n represents the number of events, Last 1 event representing flow instance 1,/>Last 1 event representing flow instance 2,/>The last 1 event representing the mth flow instance;
and further obtaining vector representation of the candidate event set corresponding to the next event Wherein the subscript c represents the next event, i.e., c=n+1;
S3.4: the vector representation of the predicted event sequence and the vector representation of the candidate event set for the next event are concatenated as all event representations E 0:
Wherein, Representing the next event of flow instance 1,/>Representing the next event of flow instance 2,/>Representing the next event of the mth flow instance.
Further, step S4 specifically includes:
s4.1: the predicted event sequence obtained in the step S3 is processed Input to an LSTM layer by the formula/>Get the ith event/>Hidden representation/>，/>Hidden representations representing the last event, i=1, 2,..n, and finally obtaining a hidden representation set/>, corresponding to the predicted event sequence；
S4.2: vector representation of candidate event sets based on next eventInputting each candidate event into the LSTM layer to obtain the hidden representation/>, of the current candidate event，，/>Representing the next event of the jth flow instance,/>A hidden representation representing an nth event;
S4.3: handle Hidden representation and/>, of individual eventsThe hidden representations of the candidate events are calculated as event representations H fused with context timing information by the following formula:
Where n represents the number of events, Hidden representation representing event 1,/>Hidden representation representing event 2,/>Hidden representation representing nth event,/>Representing the next event of flow instance 1,/>Representing the next event of flow instance 2,/>Representing the next event of the mth flow instance.
Further, step S5 specifically includes:
S5.1: extracting an adjacency matrix A of the subgraph from the knowledge graph Q obtained in the step S1 by using all event representations E 0 obtained in the step S3:
Wherein, Representing the ith event,/>Representing the j-th event,/>Representation/>Post-emergence/>Probability of occurrence;
S5.2: all event representations E 0 and an adjacency matrix A are input into SGNN, and the event representation E fused with global evolution information is obtained by learning the development rule and the evolution mode among the events through SGNN.
Further, the step S6 specifically includes:
S6.1: dividing the event representation H fused with the context time sequence information into a first context event H 1 and a first candidate event H 2, and calculating the weight of each context event for each candidate event to obtain a first weight matrix alpha:
Wherein, Representing a transpose of the first candidate event H 2;
S6.2: obtaining a first representative vector of a sequence of predicted events ：
Wherein,Representing a transpose of the first weight matrix;
S6.3: taking the cosine similarity of the first representative vector of the predicted event sequence and the first candidate event as the first score of the first candidate event ：
Wherein,Representing cosine similarity;
S6.4: dividing the event representation E fused with the global evolution information into a second context event E 1 and a second candidate event E 2, and calculating the weight of each context event for each candidate event to obtain a second weight matrix beta:
Wherein, Representing a transpose of the second candidate event E 2;
s6.5: obtaining a second representative vector of the predicted event sequence ：
Wherein,Representing a transpose of the second weight matrix β;
s6.6: taking the cosine similarity of the cosine of the candidate event and the second representative vector of the predicted event sequence as the second score of the candidate event ：
Wherein,Representing cosine similarity;
s6.7: for candidate events Candidate event/>First score/>And a second score/>Weighted summation as probability of occurrence/>, of the candidate event：
Wherein,Representing candidate event/>First score/>Coefficients of (2);
s6.8: calculating the final occurrence probability of each candidate event, and selecting an event e p with the highest occurrence probability:
where softmax () represents the activation function, relu () represents the linear rectification function, T represents the transpose, Representing j-th candidate event/>The probability of final occurrence, argmax j p (), represents the function that finds the probability maximum,/>Representing j-th candidate event/>Probability of occurrence of/>Represents the kth candidate event/>Is a probability of occurrence of (a).
The storage device stores instructions and data for implementing a flow operation sequence generation method combining an event log and a knowledge graph.
A system for generating a sequence of flow operations in combination with an event log and a knowledge graph, comprising: a processor and the storage device; the processor loads and executes the instructions and the data in the storage device to realize a flow operation sequence generation method combining the event log and the knowledge graph.
The technical scheme provided by the invention has the beneficial effects that: the invention uses the historical event log information and the related process knowledge graph related to the target generation sequence process, combines the Transformer technology to flexibly and comprehensively represent the event, uses the context information existing in the LSTM learning event log, learns the evolution mode information existing in the knowledge graph through SGNN technology, provides the next event prediction by combining the context information and the event evolution mode, and generates a reliable process operation sequence on the basis of the next event prediction. By combining time sequence data and a knowledge graph method, the next event prediction is realized more flexibly and accurately, and the method has better universality and effectiveness.
Drawings
FIG. 1 is a flowchart of a method for generating a sequence of flow operations in combination with an event log and a knowledge graph, in an embodiment of the invention;
FIG. 2 is a schematic diagram of the operation of a hardware device in an embodiment of the invention.
Detailed Description
For a clearer understanding of technical features, objects and effects of the present invention, a detailed description of embodiments of the present invention will be made with reference to the accompanying drawings.
The invention provides a method and a system for generating a flow operation sequence by combining an event log and a knowledge graph.
Referring to fig. 1, a method for generating a flow operation sequence by combining an event log and a knowledge graph specifically includes:
S1: acquiring a knowledge graph Q and an event log which are associated with a target flow P of which an operation sequence needs to be generated; the step S1 specifically comprises the following steps:
S1.1: sending a request for acquiring the related knowledge graph to a knowledge graph base according to the flow description of the target flow P, and obtaining a knowledge graph Q associated with the target flow P;
S1.2: selecting event log records log 1,…,logm of m executed process instances from the historical event log according to the process description of the target process P to obtain event log, wherein log 1 represents the event log record of the 1 st process instance, and log m represents the event log record of the m th process instance;
S2: initializing a set predicted event sequence eQueue: selecting certain event-in predicted event sequences eQueue from the event log according to the running state of the target process P; when the events are selected, the events need to be satisfied, namely the events in the actually completed flow track event logs of the flow described by the target flow P, for example, the target flow is 'payment', and the events need to be selected from the track event logs log1, log2 and the like of which the 'payment' is completed in the past; the step S2 specifically comprises the following steps:
S2.1: if the target process P is an executing process, taking out an event log of a current executing process track, taking out events in the event log according to time sequence, and adding the events into a predicted event sequence eQueue, otherwise, performing step S2.2;
S2.2: the target flow P is not in an operation state, an event log of a history flow track is taken out according to the description of the target flow P, the first few events in the event log are taken out and added into a predicted event sequence eQueue; the first few events can be flexibly selected according to the scene and the actual use condition, and one event can be taken, and multiple events can be taken. Generally, the first event is "beginning", the first events of some processes can be selected directly to avoid multiple calculations if the first events are relatively fixed, but if the event change in the beginning stage is relatively large, the subsequent events of the first event are selected directly and obtained through prediction.
S3: setting an event presentation layer: obtaining a vector representation of each event e in the predicted event sequence eQueue and a vector representation of a candidate event set for a next event; the step S3 specifically comprises the following steps:
s3.1: learning respective embeddings of event content e c of event e and subscript e t of event e in predicted event sequence eQueue by using a transducer encoder to obtain event content embeddings And subscript embedding/>；
S3.2: according to the formulaObtaining a vector representation of each event e, obtaining a vector representation/>, of the predicted event sequence eQueue，/>Vector representation representing event 1,/>Vector representation representing event 2,/>A vector representation representing an nth event, i.e., a vector representation of the last 1 event, n representing the number of events, being a positive integer;
S3.3: obtaining event logs of m finished process instances from the event logs log, and obtaining vector representations of m candidate event sets of the predicted event sequence through steps S3.1 and S3.2:
where n represents the number of events and m represents the number of flow instances; because the event length for completion of different procedures is not determined, denoted by n, Last 1 event representing flow instance 1,/>Last 1 event representing flow instance 2,/>The last 1 event representing the mth flow instance; for example, the 1 st flow example has 3 events, then the last event is e 31, the 4 th flow example has 5 events, then the last event is e 54. It can be seen that the number of candidate events is equal to the number of flow instances, and is denoted by m.
And then get the next eventVector representation of corresponding candidate event setsWherein c=n+1,/>Representing the next event after the 1 st flow instance,Representing the next event after the 2 nd flow instance,/>Representing a next event after the mth flow instance; n is the number of the last event in the currently existing predicted event sequence eQueue, i.e. the first n events to occur are already known, and n+1 events of the event logs of m flow instances need to be selected as candidate event sets to be referred to for knowing what n+1 events are.
S3.4: the vector representation of the predicted event sequence and the vector representation of the candidate event set for the next event are concatenated as all event representations E 0:
Where subscript c denotes the sequence number of the next event, e cm is the reference completed mth flow instance (a flow instance is a sequence of a series of historically occurring events) for the event at number c (i.e., n + 1), Representing the next event of flow instance 1,/>Representing the next event of flow instance 2,/>The next event (one event after the nth event) representing the mth flow instance.
And step S3, word embedding is carried out on various elements representing the event through a transducer, and then vector representations obtained by word embedding are spliced to obtain the representation of the whole event, so that flexible and comprehensive application of various information in log data is realized, meanwhile, candidate events of the next event are extracted from a historical event log to carry out word embedding, the historical data is fully utilized, and the combination of prediction of the next event and an actual scene is ensured.
S4: a set time sequence representation module: the output of the input event presentation layer (i.e. the predicted event sequence output in step S3) is used to obtain an event presentation H fused with the context timing information using a standard LSTM; the LSTM is used for learning time sequence information in the context event, so that the real continuous relation between the events influences the event prediction result, and the prediction accuracy is improved by combining the use of the candidate event set. The specific steps of step S4 are as follows:
s4.1: inputting the predicted event sequence obtained in the step S3 ；
S4.2: handle sequenceInput to an LSTM layer, get each event/>Hidden representation/>I.e./>，/>Hidden representations representing the last event, i=1, 2,..n, and finally obtaining a hidden representation set/>, corresponding to the predicted event sequence；
S4.3: vector representation of candidate event sets based on next eventEach candidate event/>, thereinInputting the hidden representation of the current candidate event into an LSTM layer to obtain the hidden representation/>，；
S4.4: handleHidden representation and/>, of individual context eventsHidden representations of the candidate events the event representation H fused with the context timing information is obtained by the following calculation formula:
Wherein n represents the number of times, Hidden representation representing event 1,/>Hidden representation representing event 2,/>Hidden representation representing nth event,/>Representing the next event of flow instance 1,/>Representing the next event of flow instance 2,/>Representing the next event of the mth flow instance.
S5: setting an event evolution module: inputting the output of the event representation layer (namely all event representations output in the step S3) and the knowledge graph Q, and obtaining an event representation E fused with global evolution information by using SGNN; firstly, event representation and a knowledge graph are subjected to sub-graph matching to obtain a knowledge graph mode sub-graph related to a current event sequence, then SGNN is used for event prediction, global evolution information contained in the knowledge graph is well fused, and prediction accuracy is further improved. The specific steps of step S5 are as follows:
S5.1: extracting an adjacency matrix A of the subgraph from the knowledge graph Q obtained in the step S1 by using all event representations E 0 obtained in the step S3, wherein the extraction formula is as follows:
wherein the formalization of the map Q is expressed as Representing node set,/>Representing an edge set, wherein the node set comprises all events occurring in an event chain of the training set, and the edge set comprises all continuous edges among nodes,/>Namely, the nodes in the node set represent an event, event/>And event/>Weights of edges/>Representation/>Post-emergence/>Probability of occurrence, calculation formula: /(I)，/>Representing event/>And/>And the number of occurrences in the flow instance of the training set.
S5.2: inputting all event representations E 0 and an adjacency matrix A into SGNN, and learning a development rule and an evolution mode among events through SGNN to obtain an event representation E fused with global evolution information;
S5.2.1: and carrying out message transmission on the child graph nodes through an adjacency matrix A, wherein the formula is as follows:
Wherein A represents an adjacency matrix, Node state (activation) representing current time step t, obtained by transmitting information from adjacency matrix A,/>An event representation representing a last time step t-1, b being a bias term (bias);
S5.2.2: updating the state of each node by the information of the last time step and other nodes: the specific formula is as follows:
Wherein, Is an update gate that controls how much of the current time step information should be retained, a value between 0 and 1, for weighting the event representation of the last time step; /(I)The gate is reset, the information of the last time step is decided how much influence the node state of the current time step is, and the bit is multiplied,/>Representing a sigmoid activation function for mapping the input value between 0 and 1; /(I)、/>、/>Are weight matrixes for learning how to integrate the information of the current time step; /(I)、/>、Are weight matrices for learning how to integrate the information of the last time step; /(I)The event representation representing the current time step t integrates the information of the last time step by updating the gate and resetting the gate;
s5.2.3: and iterating the steps for k times to obtain the event representation E which is output by the event evolution module and is fused with global evolution information.
S6: and (3) setting a prediction module: inputting H and E, and outputting an event E p with highest probability of occurrence of the next event of the target flow;
S6.1: the event representation H fused with the context time sequence information is divided into a context event H 1 and a candidate event H 2, the weight of each event for each candidate event is calculated, the weight matrix alpha of the context event and the candidate event is obtained, and the calculation formula is as follows:
Wherein, Representing a transpose of candidate event H 2;
S6.2: obtaining a first representative vector of a sequence of predicted events ：
Wherein,Representing a transpose of the weight matrix α;
s6.3: taking the cosine similarity of the cosine of the candidate event and the first representative vector of the predicted event sequence as the first score of the candidate event ：
Wherein,Representing cosine similarity;
S6.4: the event representation E fused with the global evolution information is divided into a context event E 1 and a candidate event E 2, the weight of each event for each candidate event is calculated, the weight matrix beta of the context event and the candidate event is obtained, and the calculation formula is as follows:
s6.5: obtaining a second representative vector of the predicted event sequence ：
S6.6: taking the cosine similarity of the cosine of the candidate event and the second representative vector of the predicted event sequence as the second score of the candidate event：
S6.7: for candidate eventsCandidate event/>First score/>And a second score/>Weighted summation as probability of occurrence/>, of the candidate event：
Wherein,Representing event/>First score/>Coefficients of (2);
s6.8: calculating the final occurrence probability of each candidate event, and selecting an event e p with the highest occurrence probability:
Wherein softmax () represents the activation function, relu () represents the linear rectification function, T represents the transpose, argmax j p () represents the event that is required to maximize p Function of/>Representing candidate event/>Probability of occurrence of/>Representing candidate event/>K represents the k candidate event;
S7: adding the event e p into the predicted event sequence eQueue, judging whether the event e p is a process end event or the predicted process is manually interrupted (i.e. manually terminated), and outputting eQueue as a predicted process operation sequence if e p is an event representing the process end or a signal of manually termination is received; otherwise, the current eQueue is input to step S3 to continue with the above steps.
Referring to fig. 2, fig. 2 is a schematic working diagram of a hardware device according to an embodiment of the present invention, where the hardware device specifically includes: a sequence of flow operations combining event logs and knowledge maps is generated 201, processor 202, and storage 203.
A sequence of flow operations generation system 201 that combines event logs and knowledge maps: the process operation sequence generating system 201 combining the event log and the knowledge graph realizes the process operation sequence generating method combining the event log and the knowledge graph.
Processor 202: the processor 202 loads and executes the instructions and data in the storage device 203 to implement the method for generating a sequence of flow operations combining event logs and knowledge maps.
Storage device 203: the storage device 203 stores instructions and data; the storage device 203 is configured to implement the method for generating a flow operation sequence by combining an event log and a knowledge graph.
The beneficial effects of the invention are as follows: the method comprises the steps of utilizing historical event log information and related process knowledge patterns related to a target generation sequence process, carrying out flexible and comprehensive event representation on an event by combining a Transformer technology, utilizing context information existing in an LSTM learning event log, learning evolution mode information existing in the knowledge patterns by SGNN technology, providing next event prediction by combining the context information and an event evolution mode, and carrying out iterative operation on the basis, so as to generate a reliable process operation sequence. According to the invention, by combining the time sequence data and the knowledge graph method, on one hand, the event time sequence information in the actual operation scene is utilized, and on the other hand, the rule and mode of the aggregated event evolution in the knowledge graph are utilized, so that more flexible and accurate prediction of the next event is realized, and the method has better universality and effectiveness.
The foregoing description of the preferred embodiments of the invention is not intended to limit the invention to the precise form disclosed, and any such modifications, equivalents, and alternatives falling within the spirit and scope of the invention are intended to be included within the scope of the invention.
Claims (7)
1. A process operation sequence generation method combining event logs and knowledge maps is characterized in that: comprising the following steps:
S1: acquiring a knowledge graph Q and an event log which are associated with a target flow P of which an operation sequence needs to be generated;
S2: selecting events from the event log according to the running state of the target process P to obtain a predicted event sequence eQueue;
s3: according to the predicted event sequence eQueue, vector representations of each event and vector representations of candidate event sets of the next event are obtained, and all event representations are obtained;
S4: converting the predicted event sequence obtained in the step S3 into event representation H fused with context time sequence information through a standard LSTM;
S5: converting all event representations and knowledge patterns Q obtained in the step S3 into event representations E fused with global evolution information through SGNN;
the step S5 specifically comprises the following steps:
S5.1: extracting an adjacency matrix A of the subgraph from the knowledge graph Q obtained in the step S1 by using all event representations E 0 obtained in the step S3:
Wherein, Representing the ith event,/>Representing the j-th event,/>Representation/>Post-emergence/>Probability of occurrence;
S5.2: inputting all event representations E 0 and an adjacency matrix A into SGNN, and learning a development rule and an evolution mode among events through SGNN to obtain an event representation E fused with global evolution information;
S6: obtaining the event with highest occurrence probability according to the event representation H obtained in the step S4 and the event representation E obtained in the step S5 ；
The step S6 specifically comprises the following steps:
S6.1: dividing the event representation H fused with the context time sequence information into a first context event H 1 and a first candidate event H 2, and calculating the weight of each context event for each candidate event to obtain a first weight matrix alpha:
Wherein, Representing a transpose of candidate event H 2;
S6.2: obtaining a first representative vector of a sequence of predicted events ：
Wherein,Representing a transpose of the weight matrix;
S6.3: taking the cosine similarity of the first representative vector of the predicted event sequence and the first candidate event as the first score of the first candidate event ：
Wherein,Representing cosine similarity;
s6.4: dividing the event representation E fused with the global evolution information into a second context event E 1 and a second candidate event E 2, and calculating the weight of each event for each candidate event to obtain a second weight matrix beta:
Wherein, Representing a transpose of the second candidate event E 2;
s6.5: obtaining a second representative vector of the predicted event sequence ：
Wherein,Representing a transpose of the second weight matrix β;
S6.6: taking the cosine similarity of the second representative vector of the predicted event sequence and the second candidate event as a second score of the second candidate event ：
Wherein,Representing cosine similarity;
s6.7: for candidate events Candidate event/>First score/>And a second score/>Weighted summation as probability of occurrence/>, of the candidate event：
Wherein,Representing candidate event/>First score/>Coefficients of (2);
s6.8: calculating the final occurrence probability of each candidate event, and selecting an event e p with the highest occurrence probability:
where softmax () represents the activation function, relu () represents the linear rectification function, T represents the transpose, Representing j-th candidate event/>The probability of final occurrence, argmax j p (), represents the function that finds the probability maximum,/>Representing j-th candidate event/>Probability of occurrence of/>Represents the kth candidate event/>Is a probability of occurrence of (2);
S7: handle event Added to the predicted event sequence eQueue and performed on event/>Making a judgment if event/>If the event is an event representing the end of the flow or a signal of artificial termination is received, the predicted event sequence eQueue is output as a predicted flow operation sequence; otherwise, go back to step S3 to continue with the above steps.
2. The method for generating a sequence of flow operations by combining an event log and a knowledge graph as claimed in claim 1, wherein: the step S1 specifically comprises the following steps:
S1.1: sending a request for acquiring the related knowledge graph to a knowledge graph base according to the flow description of the target flow P, and obtaining a knowledge graph Q associated with the target flow P;
S1.2: and selecting event log records log 1,…,logm of m executed process instances from the historical event log according to the process description of the target process P, wherein log 1 represents the event log record of the 1 st process instance, and log m represents the event log record of the m th process instance in the event log.
3. The method for generating a sequence of flow operations combining an event log and a knowledge graph as claimed in claim 2, wherein: the step S2 specifically comprises the following steps:
S2.1: when the target process P is in an operation state, namely the target process P is an executing process, taking out an event log, taking out events in the event log according to time sequence and adding the events into a predicted event sequence eQueue;
S2.2: when the target flow P is in an unoperated state, the event log is fetched according to the description of the target flow P, and part of the events in the event log are fetched and added to the predicted event sequence eQueue.
4. The method for generating a sequence of flow operations combining an event log and a knowledge graph as claimed in claim 3, wherein: the step S3 specifically comprises the following steps:
s3.1: learning respective embeddings of event content e c of event e and subscript e t of event e in predicted event sequence eQueue by using a transducer encoder to obtain event content embeddings And subscript embedding/>；
S3.2: according to the formulaObtaining a vector representation of each event, obtaining a vector representation of the predicted event sequence as/>，/>Vector representation representing event 1,/>Vector representation representing event 2,/>A vector representation representing an nth event, n representing the number of events;
S3.3: obtaining event logs of m finished process instances from the event logs log, and obtaining vector representations of m candidate events of the predicted event sequence through steps S3.1 and S3.2:
Where n represents the number of events, Last 1 event representing flow instance 1,/>Last 1 event representing flow instance 2,/>The last 1 event representing the mth flow instance;
And further obtains a vector representation of the candidate event set corresponding to the next event Wherein the subscript c represents the next event, i.e., c=n+1;
S3.4: the vector representation of the predicted event sequence and the vector representation of the candidate event set for the next event are concatenated as all event representations E 0:
Wherein, Representing the next event after the 1 st flow instance,/>Representing the next event after the 2 nd flow instance,/>Representing the next event after the mth flow instance.
5. The method for generating a sequence of flow operations combining an event log and a knowledge graph as claimed in claim 4, wherein: the step S4 specifically comprises the following steps:
s4.1: the predicted event sequence obtained in the step S3 is processed Input to an LSTM layer by the formula/>Get the ith event/>Hidden representation/>，/>Hidden representations representing the last event, i=1, 2,..n, and finally obtaining a hidden representation set/>, corresponding to the predicted event sequence；
S4.2: vector representation of candidate event sets based on next eventInputting each candidate event into the LSTM layer to obtain the hidden representation/>, of the current candidate event，/>，Representing the next event of the jth flow instance,/>A hidden representation representing an nth event;
S4.3: handle Hidden representation and/>, of individual eventsThe hidden representations of the candidate events are calculated as event representations H fused with context timing information by the following formula:
Where n represents the number of events, Hidden representation representing event 1,/>A hidden representation of the 2 nd event is represented,Hidden representation representing nth event,/>Representing the next event of flow instance 1,/>Representing the next event of flow instance 2,/>Representing the next event of the mth flow instance.
6. A memory device, characterized by: the storage device stores instructions and data for implementing the method for generating a flow operation sequence by combining an event log and a knowledge graph according to any one of claims 1 to 5.
7. A flow operation sequence generation system combining event logs and knowledge maps is characterized in that: the system comprises: a processor and a storage device; the processor loads and executes the instructions and data in the storage device to implement the method for generating a flow operation sequence by combining the event log and the knowledge graph according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410006733.XA CN117493583B (en) | 2024-01-03 | 2024-01-03 | Method and system for generating flow operation sequence by combining event log and knowledge graph |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410006733.XA CN117493583B (en) | 2024-01-03 | 2024-01-03 | Method and system for generating flow operation sequence by combining event log and knowledge graph |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117493583A CN117493583A (en) | 2024-02-02 |
| CN117493583B true CN117493583B (en) | 2024-05-14 |
Family
ID=89676863
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410006733.XA Active CN117493583B (en) | 2024-01-03 | 2024-01-03 | Method and system for generating flow operation sequence by combining event log and knowledge graph |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117493583B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112632223A (en) * | 2020-12-29 | 2021-04-09 | 天津汇智星源信息技术有限公司 | Case and event knowledge graph construction method and related equipment |
| CN114357197A (en) * | 2022-03-08 | 2022-04-15 | 支付宝(杭州)信息技术有限公司 | Event reasoning method and device |
| CN116069947A (en) * | 2023-01-30 | 2023-05-05 | 思必驰科技股份有限公司 | Log data event map construction method, device, equipment and storage medium |
| WO2023115761A1 (en) * | 2021-12-20 | 2023-06-29 | 北京邮电大学 | Event detection method and apparatus based on temporal knowledge graph |
| CN117236677A (en) * | 2023-08-07 | 2023-12-15 | 安徽思高智能科技有限公司 | RPA process mining method and device based on event extraction |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190043500A1 (en) * | 2017-08-03 | 2019-02-07 | Nowsportz Llc | Voice based realtime event logging |
| US11294754B2 (en) * | 2017-11-28 | 2022-04-05 | Nec Corporation | System and method for contextual event sequence analysis |
| US11663498B2 (en) * | 2019-05-21 | 2023-05-30 | Sisense Ltd. | System and method for generating organizational memory using semantic knowledge graphs |
-
2024
- 2024-01-03 CN CN202410006733.XA patent/CN117493583B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112632223A (en) * | 2020-12-29 | 2021-04-09 | 天津汇智星源信息技术有限公司 | Case and event knowledge graph construction method and related equipment |
| WO2023115761A1 (en) * | 2021-12-20 | 2023-06-29 | 北京邮电大学 | Event detection method and apparatus based on temporal knowledge graph |
| CN114357197A (en) * | 2022-03-08 | 2022-04-15 | 支付宝(杭州)信息技术有限公司 | Event reasoning method and device |
| CN116069947A (en) * | 2023-01-30 | 2023-05-05 | 思必驰科技股份有限公司 | Log data event map construction method, device, equipment and storage medium |
| CN117236677A (en) * | 2023-08-07 | 2023-12-15 | 安徽思高智能科技有限公司 | RPA process mining method and device based on event extraction |
Non-Patent Citations (1)
| Title |
|---|
| 知识图谱技术在金融网络日志告警压降溯源领域的应用研究;李家炎;《中国金融电脑》;20230607;正文第73-76页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117493583A (en) | 2024-02-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11423325B2 (en) | Regression for metric dataset | |
| CN115618269B (en) | Big data analysis method and system based on industrial sensor production | |
| CN112558185A (en) | Bidirectional GRU typhoon track intelligent prediction and forecast system based on attention mechanism, computer equipment and storage medium | |
| WO2022222026A1 (en) | Medical diagnosis missing data completion method and completion apparatus, and electronic device and medium | |
| CN113239702A (en) | Intention recognition method and device and electronic equipment | |
| CN111915086A (en) | Abnormal user prediction method and equipment | |
| CN117236677A (en) | RPA process mining method and device based on event extraction | |
| CN113484882B (en) | GNSS sequence prediction method and system of multi-scale sliding window LSTM | |
| CN117493583B (en) | Method and system for generating flow operation sequence by combining event log and knowledge graph | |
| CN111753959A (en) | Neural network model optimization method, device, equipment and storage medium | |
| CN116341720A (en) | Multi-fan wind speed and direction prediction method based on dynamic graph convolution and transformation | |
| CN111161238A (en) | Image quality evaluation method and device, electronic device, and storage medium | |
| CN115660795A (en) | Data processing method, device, equipment, storage medium and program product | |
| CN115062769A (en) | Knowledge distillation-based model training method, device, equipment and storage medium | |
| CN112580885A (en) | Method, device and equipment for predicting accessory qualification rate and storage medium | |
| CN111402042A (en) | Data analysis and display method for stock market large disc state analysis | |
| CN113469244B (en) | Volkswagen app classification system | |
| CN115640336B (en) | Business big data mining method, system and cloud platform | |
| US20230146635A1 (en) | Method and Systems for Conditioning Data Sets for Efficient Computational Processing | |
| CN115630289B (en) | Target identification method and device based on evidence theory | |
| CN114154687A (en) | Trajectory prediction method and apparatus, electronic device and storage medium | |
| CN118098213A (en) | Automatic voice recognition model building method, voice recognition method and related equipment | |
| CN117787341A (en) | Social media content propagation prediction method, device, computing equipment and storage medium | |
| CN117391713A (en) | Information pushing method and device, electronic equipment and storage medium | |
| CN114881283A (en) | Training method of solar-term change forecasting model, and solar-term change forecasting method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |