Similar event determination method and device, equipment, medium and product

Abstract

The embodiment of the invention discloses a similar event determination method and device, equipment, a medium and a product. The method comprises the following steps: acquiring a plurality of recall events associated with a to-be-processed event; determining at least one target department responsibility statement corresponding to each event description statement in the to-be-processed event according to the plurality of event description statements of the to-be-processed event and a pre-generated content relation graph; determining a first confidence degree of each recall event relative to the to-be-processed event according to the event description statements of the to-be-processed event and the recall events and at least one target department responsibility statement of the to-be-processed event; and determining at least one similar event corresponding to the to-be-processed event according to the first confidence of each recall event and the recall event to which the target department responsibility statement belongs. According to the technical scheme, a plurality of similar events similar to the to-be-processed event are determined, related event processing experience can be quickly obtained through the similar events, and the event processing efficiency is improved.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the field of information processing technology, and in particular, to a method, device, equipment, medium and product for determining similar events. Background Art

[0002] Through the convenience hotline, relevant personnel can collect various life problems reported by users. When receiving calls from users, relevant personnel can record the reported problems reported by users and forward the reported problems to the corresponding execution departments so that the staff of the execution departments can properly solve the reported problems. At present, when relevant personnel determine the execution department corresponding to a reported problem, they mainly rely on human work experience.

[0003] When implementing the technical solution based on the above method, the inventor found the following problems:

[0004] When faced with a large number of reported problems, staff are unable to promptly and accurately determine the executive departments corresponding to the reported problems, and are thus unable to effectively handle them. Furthermore, if there is a staff replacement, new staff need to be retrained, which requires more manpower costs. It can be seen that this method of determining the executive departments corresponding to reported problems through manual experience not only has the problem of high labor costs, but also has the problem of time-consuming problem analysis, resulting in low efficiency in conveying reported problems, which in turn affects the efficiency of incident resolution. Summary of the invention

[0005] The present invention provides a method, device, equipment, medium and product for determining similar events, so as to automatically determine multiple similar events close to the event to be processed from a large number of events, and quickly obtain relevant event processing experience through similar events, thereby improving event processing efficiency.

[0006] In a first aspect, an embodiment of the present invention provides a method for determining a similar event, the method comprising:

[0007] Acquire multiple recall events associated with the pending event, wherein the pending event and the recall event both include multiple event description statements;

[0008] According to the multiple event description statements of the event to be processed and the pre-generated content relationship graph, at least one target department responsibility statement corresponding to each event description statement in the event to be processed is determined; wherein the content relationship graph is generated based on the historical work order event and at least part of the department responsibility statements of all execution departments associated with the historical work order event;

[0009] Determine a first confidence level of each recall event relative to the event to be processed according to the event description statements of the event to be processed and the recall event, and at least one target department responsibility statement of the event to be processed;

[0010] At least one similar event corresponding to the event to be processed is determined based on the first confidence of each recall event and the recall event to which the target responsible department belongs; wherein the similar event is an event in the recall event.

[0011] Furthermore, the method also includes:

[0012] Processing multiple event description statements of the event to be processed and historical description statements of at least part of historical work order events based on a similarity matching algorithm to obtain event similarity of the historical work order event with respect to the event to be processed;

[0013] Based on the event similarity corresponding to the historical work order event, a plurality of recall events associated with the event to be processed are determined; wherein the recall event is an event in the historical work order event.

[0014] Furthermore, the method also includes:

[0015] The content relationship diagram is generated in the following manner:

[0016] Acquire multiple historical work order events, wherein the historical work order events include multiple historical description statements and department responsibility statements of the executing department;

[0017] For the multiple historical work order events, the historical work order events are input into the pre-trained sentence embedding model to obtain the sentence embedding of each historical description sentence in the historical work order event; and based on the sentence embedding model, the department responsibility statements of the execution department corresponding to the historical work order event are processed to obtain the responsibility sentence embedding corresponding to each department responsibility statement.

[0018] The content relationship graph is determined by embedding the sentences of all historical work order events and embedding the corresponding responsibility sentences.

[0019] Furthermore, the method also includes:

[0020] By clustering all sentence embeddings, a plurality of cluster sets are obtained, wherein each cluster set includes at least one sentence embedding;

[0021] For each sentence embedding in each cluster set, determine the target responsibility sentence embedding of the sentence embedding according to the distance information between the sentence embedding and all responsibility sentence embeddings of the execution department to which the sentence embedding belongs;

[0022] The content relationship graph is determined based on all sentence embeddings in each cluster set and the corresponding target responsibility sentence embeddings.

[0023] Furthermore, the method also includes:

[0024] For each event description sentence, vectorize the event description sentence based on a pre-trained sentence embedding model to obtain a sentence embedding to be used for the event description sentence;

[0025] Determine a target cluster set to which the sentence embedding to be used belongs, and determine at least one target department responsibility sentence of the event description sentence according to the number of associations of sentence embeddings associated with each responsibility sentence embedding in the target cluster set;

[0026] The target department responsibility statements are statements in which all statements in the target cluster set are embedded in the associated department responsibility statements.

[0027] Furthermore, the method also includes:

[0028] At least one target department responsibility statement corresponding to the event to be processed is adjusted to a responsibility description statement corresponding to the execution department in the recall event, and a first confidence matrix of the event description statement relative to the department responsibility statement is determined.

[0029] Furthermore, the method also includes:

[0030] Creating a first vector matrix based on the departmental responsibility statements, and determining a second vector matrix based on the event description statements;

[0031] Determine a matching coefficient matrix of the event description statement relative to the department responsibility statement based on the first vector matrix and the second vector matrix;

[0032] By normalizing the matching coefficient matrix columns, a first unidirectional attention matrix is ​​obtained;

[0033] Determine the first confidence matrix based on the first unidirectional attention matrix.

[0034] Furthermore, the method also includes:

[0035] Determine a sentence matching matrix based on the at least one target department responsibility statement and the multiple event description statements, and determine an opinion matching matrix based on the multiple event description statements and the handling opinions in the recall event, wherein the element values ​​in the sentence matching matrix are used to characterize the similarity between the description statement and the target department responsibility statement, and the element values ​​in the opinion matching matrix are used to characterize the matching degree between the description statement and the handling opinions;

[0036] Determine a statement set according to the statement matching matrix and the opinion matching matrix; wherein the first statement in the statement set is a statement whose relevance to the target responsible department and the handling opinion is lower than a preset threshold;

[0037] Determining a second confidence level of the recalled event relative to the event to be processed based on the statement set, the recalled event, and the event description statement of the event to be processed;

[0038] According to the second confidence, the first confidence of the recalled event relative to the event to be processed is updated to determine the at least one similar event based on the updated first confidence.

[0039] Furthermore, the method also includes:

[0040] By normalizing the columns of the sentence matching matrix, a sentence unidirectional attention matrix of the event description sentence relative to the target department responsibility sentence is obtained; and by normalizing the rows of the sentence matching matrix, a responsibility unidirectional attention matrix is ​​obtained;

[0041] By normalizing the columns of the opinion matching matrix, a first opinion one-way attention matrix of the event description sentence relative to the handling opinion is obtained, and by normalizing the rows of the opinion matching matrix, a second opinion one-way attention matrix is ​​obtained;

[0042] Determining a second sentence from the event description sentence based on the sentence unidirectional attention matrix and the responsibility unidirectional attention matrix; and determining a third sentence from the event description sentence based on the first opinion unidirectional attention matrix and the second opinion unidirectional attention matrix;

[0043] The statements in the event description statements except the second statement and the third statement are used as the first statement in the statement set.

[0044] Furthermore, the method also includes:

[0045] For each recall event, the event description statement of the pending event is updated based on the handling opinion of the recall event, and at least part of the first statement in the statement set in the pending event is replaced with the event description statement in the recall event, to obtain an updated pending event;

[0046] Determine an opinion matrix regarding handling opinions between the updated pending events and the recalled events;

[0047] Based on the opinion matrix, a second confidence level of the updated event to be processed corresponding to the recalled event is determined.

[0048] Furthermore, the method also includes:

[0049] For each recall event, determining a third confidence level between the event description statement of the event to be processed and the event description statement of the recall event;

[0050] The first confidence level or the second confidence level is updated based on the third confidence level to determine the at least one similar event based on the updated first confidence level.

[0051] Furthermore, the method also includes:

[0052] A preset number of recalled events are determined in descending order according to the first confidence as similar events to the event to be processed.

[0053] Furthermore, the method also includes:

[0054] Based on the at least one similar event, a target processing result corresponding to the event to be processed is determined.

[0055] In a second aspect, an embodiment of the present invention further provides a device for determining a similar event, the device comprising:

[0056] A recall event determination module, used to obtain multiple recall events associated with the pending event, wherein the pending event and the recall event both include multiple event description statements;

[0057] A responsibility statement determination module is used to determine at least one target department responsibility statement corresponding to each event description statement in the event to be processed based on multiple event description statements of the event to be processed and a pre-generated content relationship graph; wherein the content relationship graph is generated based on historical work order events and at least part of the department responsibility statements of all execution departments associated with the historical work order events;

[0058] A confidence determination module, configured to determine a first confidence of each recall event relative to the event to be processed based on event description statements of the event to be processed and the recall event, and at least one target department responsibility statement of the event to be processed;

[0059] A similar event determination module is used to determine at least one similar event corresponding to the event to be processed based on the first confidence of each recall event and the recall event to which the target responsible department belongs; wherein the similar event is an event in the recall event.

[0060] In a third aspect, an embodiment of the present invention further provides an electronic device, the electronic device comprising:

[0061] one or more processors;

[0062] a storage device for storing one or more programs,

[0063] When one or more programs are executed by one or more processors, the one or more processors implement a method for determining similar events as described in any one of the embodiments of the present invention.

[0064] In a fourth aspect, an embodiment of the present invention further provides a storage medium comprising computer executable instructions, which, when executed by a computer processor, are used to execute a method for determining a similar event as described in any one of the embodiments of the present invention.

[0065] The technical solution of the embodiment of the present invention is to obtain multiple recall events associated with the pending event, wherein the pending event and the recall event both include multiple event description statements, and then, according to the multiple event description statements of the pending event and the pre-generated content relationship graph, determine at least one target department responsibility statement corresponding to each event description statement in the pending event, wherein the content relationship graph is generated based on historical work order events and at least part of the department responsibility statements of all execution departments associated with the historical work order events. Since the content relationship graph is determined based on a large number of historical work order events, it has captured the potential causal relationship between the event description statement and the department responsibility statement. Through the event description statement of the pending event and the content relationship graph, the target department responsibility statement involved in the pending event can be determined. Further, according to the event description statements of the pending event and the recall event, and at least one target department responsibility statement of the pending event, determine the first confidence of each recall event relative to the pending event, thereby, according to the first confidence of each recall event and the recall event to which the target responsibility department belongs, determine at least one similar event from multiple recall events. The technical solution provided in this embodiment can automatically determine multiple similar events that are close to the event to be processed from a large number of events. Through similar events, relevant event processing experience can be quickly obtained, which not only reduces labor costs but also improves event processing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required to describe the embodiments. Obviously, the drawings introduced are only drawings of a part of the embodiments to be described in the present invention, rather than all the drawings. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0067] Figure 1 A schematic flow chart of a method for determining a similar event provided by an embodiment of the present invention;

[0068] Figure 2 This is an example diagram of a content relationship diagram involved in an embodiment of the present invention;

[0069] Figure 3 A schematic flow chart of another method for determining similar events provided by an embodiment of the present invention;

[0070] Figure 4 A flowchart of another method for determining a similar event provided by an embodiment of the present invention;

[0071] Figure 5 A schematic diagram of recorded text description information of a recall event involved in an embodiment of the present invention;

[0072] Figure 6 A schematic diagram of the structure of a device for determining similar events provided by an embodiment of the present invention;

[0073] Figure 7 The present invention is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0074] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0075] Before introducing the technical solution, an exemplary description of the application scenario can be given. Embodiments of the Invention The technical solution provided by the embodiments of the present invention can be applied to any scenario where it is necessary to determine similar events that are highly similar to a certain event to be processed from a large number of historical events, so as to provide a reference basis for the event to be processed according to the processing process of the similar events.

[0076] For example, the public service platform is designed to efficiently handle user calls and ensure that the problems are properly handled. The government service hotline can collect various life problems reported by users. At present, some hotline staff mainly rely on accumulated experience to determine events similar to the events to be handled, or search for corresponding similar events from a large number of historical events in sequence, which is inefficient.

[0077] For a problem reported by a user, the entire process of handling the problem can be called an event. The event handling process covers three parts, namely, the reported problem raised by the user, the operator dispatching the reported problem to the corresponding execution department, and the execution department providing handling opinions. The following is a detailed description of these three parts. In the first step, the reported problem told by the user through the convenience hotline will be recorded as text content by the operator. Among them, the text content recorded by the operator generally includes: problem description text, event background text and event impact text; the problem description text refers to the text record corresponding to the problem or dissatisfaction encountered by the user, and the event background text provides a text record of background information related to the problem, such as the specific location, time, relevant history, etc.; the impact part text shows the impact of the problem on the user's life and emphasizes the problems that need to be solved or improved. In the second step, the operator will determine the corresponding execution department according to the nature and field of the reported problem, and transfer the text content of the reported problem to the corresponding execution department so that the execution department can coordinate and handle the reported problem. Finally, after the execution department handles and solves the reported problem, it will inform the hotline platform of the handling opinion, which will be conveyed to the reporting user through the operator.

[0078] At present, the reported problems of users are very colloquial, with redundant information and key information mixed together. Moreover, the logic of user expression is unclear, the cause and effect of the event are not organized, and there are difficulties in information extraction. In addition, some complex events need to involve multiple executive departments, and there are also overlapping responsibilities between multiple departments, which leads to an event often having multiple impacts, and different executive departments can handle it, and there is a problem of confusion in the delegation of executive departments. Further, the information associated with the responsibilities of different departments in the reported problem text will interfere with each other and affect the similarity judgment between events. Based on the interference of the above-mentioned problems, the staff cannot efficiently convey the reported problems to the corresponding executive departments. Through the embodiment of the present invention, when the reported problem text is given a pending time, similar cases similar to it can be automatically retrieved from a large number of historical cases, so as to provide the predecessors' processing process (including the dispatched executive departments) for the operator to provide reference, so as to quickly obtain relevant event processing experience and improve the processing efficiency of the event.

[0079] Figure 1 A flow chart of a method for determining similar events provided in an embodiment of the present invention. This embodiment can be applied to situations where it is necessary to determine similar events that are highly similar to an event to be processed from a large number of historical events, so as to provide a reference basis for the event to be processed based on the processing process of the similar events. The method can be executed by a similar event determination device, which can be implemented in the form of software and / or hardware. The hardware can be an electronic device, and the electronic device can be a mobile terminal, a PC or a server, etc.

[0080] like Figure 1 As shown, the method includes:

[0081] S110: Acquire multiple recall events associated with the event to be processed.

[0082] Among them, pending events refer to events that need to be further processed or resolved. These events require relevant personnel or execution departments to perform corresponding operations to complete the processing. The pending events include multiple event description statements. Event description statements refer to statements used to describe the process or situation of an event. This description usually includes the time, place, participants and / or specific content of the event, which is used to record the ins and outs of the event in detail. The role of the event description statement is to help understand the background, process and results of the event, so as to better analyze and summarize experience. In particular, since the pending events are in an unresolved state, the multiple event description statements corresponding to the pending events are description statements of the occurrence of a certain event, and do not include the processing process and results of the event.

[0083] In actual applications, multiple work order events may have been processed before the current moment, and the processed work order events can be regarded as historical work order events. Furthermore, the historical work order events also include event description statements. The recall event is a work order event that has a certain similarity with the event to be processed, which is determined from the historical work order events. Each recall event includes multiple event description statements. In particular, the event description statement corresponding to the recall event refers to the historical event description statement, which is used to characterize the reported problem corresponding to the recall event.

[0084] It should be particularly noted that since the number of recall events associated with the event to be processed may include multiple ones, and the recall events and the event to be processed have some similarities in some aspects, it does not mean that the recall events and the event to be processed are highly related similar events. The purpose of this embodiment is to determine similar events that are more relevant to the event to be processed from the numerous recall events, so as to provide a reference basis for the event to be processed according to the processing process of similar events.

[0085] Specifically, a text matching algorithm can be used to perform text matching on the event to be processed and a large number of historical work order events to obtain the similarity between the event to be processed and each historical work order event, and multiple historical work order events whose similarity meets a preset similarity threshold are screened out as recall events.

[0086] S120: Determine at least one target department responsibility statement corresponding to each event description statement in the event to be processed based on multiple event description statements of the event to be processed and a pre-generated content relationship diagram.

[0087] Among them, the content relationship diagram is a graphical tool used to show the complex relationship between the specific content contained in different events. For example, it can represent these relationships through circles (or boxes) and arrows. The department responsibility statement is a short and concise expression used to summarize and convey the core responsibilities and goals of the executive department. These statements usually reflect the work focus of a certain executive department. The target department responsibility statement is a department responsibility statement that is highly related to the event description statement of the event to be processed.

[0088] In this embodiment, the content relationship graph is generated based on historical work order events and at least part of the department responsibility statements of all execution departments associated with the historical work order events.

[0089] Among them, the historical work order event is the entire event of collecting problems, assigning execution departments, and properly solving problems in response to problems raised by users in the historical period. The historical work order event has been solved, and the entire event resolution process can be recorded in text. These texts used to record the resolution process of an event may include but are not limited to the problem description text, the identification text corresponding to the execution department that solved the problem, the description text of the specific solution to the problem, etc. It can be seen that the execution department associated with each historical work order event is determined, and then all the execution departments associated with the historical work order event can be determined. The department responsibility statements corresponding to these execution departments are text information that can be directly obtained.

[0090] In this embodiment, the pre-generated content relationship diagram has captured the potential mapping relationship between the event description statement and the department responsibility statement. Through the event description statement of the event to be processed and the content relationship diagram, the target department responsibility statement corresponding to the event description statement can be quickly and conveniently determined.

[0091] Specifically, since the pre-generated content relationship graph contains a large number of correspondences between historical event description statements and at least one corresponding departmental responsibility statement, based on this, for a certain event description statement of the event to be processed, it can be determined which historical event description statements in the content relationship graph the event description statement is similar to; based on the content relationship graph, the departmental responsibility statements corresponding to these historical event description statements can be determined, and then these departmental responsibility statements can be determined as at least one target departmental responsibility statement corresponding to the event description statement, and based on the same determination method, at least one target departmental responsibility statement corresponding to each event description statement in the event to be processed can be determined.

[0092] For example, see the content relationship diagram example diagram Figure 2 ,like Figure 2As shown, the content relationship diagram includes 7 nodes, among which the nodes filled with black represent the nodes corresponding to the historical event description statements, and the nodes filled with white represent the nodes corresponding to the department responsibility statements. It can be seen from the figure that the historical event description statement node 0 corresponds to the department responsibility statement node 3, the department responsibility statement node 2 and the department responsibility statement node 1, and the historical event description statement node 4 corresponds to the department responsibility statement node 5 and the department responsibility statement node 6. For a certain event description statement S of the event to be processed, the event description statement S is closer to the node 0 corresponding to the historical event description statement, then at least one target department responsibility statement corresponding to the event description statement S is the department responsibility statement corresponding to node 3, node 2 and node 1.

[0093] S130: Determine a first confidence level of each recall event relative to the event to be processed based on event description statements of the event to be processed and the recall event, and at least one target department responsibility statement of the event to be processed.

[0094] The first confidence level is used to characterize the similarity between the recalled event and the event to be processed.

[0095] Specifically, the processing process for each recall event is consistent, and any one of the recall events is used as an example for explanation. For a certain recall event, the execution department corresponding to the recall event can be determined based on the recorded text description information corresponding to the recall event. Based on this, multiple duty description statements corresponding to the execution department are easy to obtain. Furthermore, at least one target department duty statement of the event to be processed can be replaced with multiple duty description statements corresponding to the execution department. Further, each target department duty statement after replacement can be converted into a corresponding first eigenvector, and the first eigenvectors form a first vector matrix, and each event description statement of the event to be processed can be converted into a corresponding second eigenvector, and the second eigenvectors form a second vector matrix; and the transpose of the first vector matrix and the second vector matrix is ​​multiplied by a dot multiplication operation to obtain a matching coefficient matrix of the event description statement relative to the target department duty statement, and then the matching coefficient matrix is ​​column-normalized to obtain a first unidirectional attention matrix, wherein each element in the first row of the first unidirectional attention matrix represents the degree of matching between the first event description statement in the event to be processed and each target department duty statement. Then, the first unidirectional attention matrix and the transpose of the first unidirectional attention matrix are multiplied, and the confidence value obtained by the partial correlation CI-test on the final matrix is ​​the first confidence of the recall event relative to the event to be processed. Based on the same processing method, the first confidence of each recall event relative to the event to be processed can be obtained.

[0096] Exemplarily, the event to be processed can be represented as E(x), and the event description statement corresponding to the event to be processed is {X=[x1,x2,...,x n ]}, at least one target department responsibility statement of the event to be processed can be expressed as {D=[d1,d2,...,d m ]}, where x1 represents the first event description statement corresponding to the event to be processed, and d1 represents the first target department responsibility statement corresponding to the event to be processed. Any recall event can be represented by E(y), and the event description statement corresponding to the recall event is {Y=[y1,y2,...,y u ]}, where y1 represents the first event description statement of the reported problem in the recall event. V is the executive department corresponding to the recall event, A=[A1,A2,...,A v ] represents the responsibility description statement corresponding to the execution department V, wherein A1 represents the first responsibility description statement. At least one target department responsibility statement {D=[d1, d2, ..., d m ]} is replaced by A=[A1,A2,...,A v ]. For the ith target department responsibility statement A after replacement i , which can be converted into the first eigenvector a i ; For the jth event description statement X of the event to be processed j , which can be converted into the second eigenvector x j ; Then, calculate the matching coefficient between the two Subsequently, the matching coefficient matrix M can be column normalized to obtain the first unidirectional attention matrix α. The Pth row in the first unidirectional attention matrix can be expressed as:

[0097] α(p)=softmax(M(1,p),M(2,p),...M(n,p))

[0098] In the formula, α(p) represents x p The degree of match with the target department's responsibility statement, x p The second feature vector corresponding to the p-th event description statement of the event to be processed.

[0099] Furthermore, the covariance matrix C can be calculated z =α·α T , the final covariance matrix C z The partial correlation CI-test is performed to obtain a confidence value, which is the first confidence that the recalled event E(y) can be expressed as E(x) relative to the event to be processed.

[0100] S140: Determine at least one similar event corresponding to the event to be processed based on the first confidence level of each recall event and the recall events to which the target responsible department belongs.

[0101] Among them, similar events are events in the recall event. The purpose of determining similar events is that since similar events are similar to the pending events, when relevant personnel handle the pending events, they can use the handling methods of similar events as a reference to help relevant personnel smoothly complete the pending events.

[0102] Specifically, the recall event to which the target department responsibility statement belongs is determined. For all recall events, the target department responsibility statements corresponding to each recall event can be determined by statistical means, and then a reliability value can be assigned to it according to the number of target department responsibility statements corresponding to a certain recall event. For example, if a certain recall event corresponds to a large number of target department responsibility statements, a higher reliability is assigned to the recall event; if a certain recall event corresponds to a small number of target department responsibility statements, a lower reliability is assigned to the recall event; if a certain recall event does not have a corresponding target department responsibility statement, the reliability of the recall event is assigned to 0. Based on this, each recall event corresponds to a first confidence and a reliability. Therefore, for each recall event, the first confidence and reliability can be weighted and summed to obtain the event similarity attribute value of each recall event, and then, one or more similar events corresponding to the event to be processed are determined according to the numerical value of the similarity attribute value of each event.

[0103] For example, the five recall events are respectively recall event A, recall event B, recall event C, recall event D and recall event E. There are 3 target department responsibility statements belonging to recall event A, 2 target department responsibility statements belonging to recall event C, and 1 target department responsibility statement belonging to recall event E. There is no target department responsibility statement belonging to recall event B and recall event D. Then, the reliability of recall event A can be assigned a value of 0.3, the reliability of recall event B can be assigned a value of 0, the reliability of recall event C can be assigned a value of 0.2, the reliability of recall event D can be assigned a value of 0, and the reliability of recall event E can be assigned a value of 0.1. The first confidences corresponding to recall event A, recall event B, recall event C, recall event D and recall event E are: y1, y2, y3, y4, y5 respectively. When the first weight value corresponding to the first confidence is α1 and the second weight value corresponding to the reliability is α2, the event similarity attribute value corresponding to the recall event A is S1=α1×y1+α2×0.3. Based on the same processing method, the event similarity attribute value corresponding to each recall event can be determined, and then, one or more similar events corresponding to the event to be processed can be determined according to the numerical value of each event similarity attribute value.

[0104] On the basis of the above embodiment, optionally, a target processing result corresponding to the event to be processed can also be determined based on at least one similar event. The target processing result is a method and manner in which the event to be processed needs to be processed. Exemplarily, the target processing result is an execution department corresponding to the event to be processed. In this embodiment, the execution department corresponding to the event to be processed can be determined based on the execution departments involved in similar events.

[0105] It should be noted that the acquisition, storage, use, and processing of data in the technical solution of the present invention are in compliance with the relevant provisions of national laws and regulations. For example, network requests are intercepted and processed with the authorization of the user.

[0106] The technical solution of the embodiment of the present invention is to obtain multiple recall events associated with the pending event, wherein the pending event and the recall event both include multiple event description statements, and then, according to the multiple event description statements of the pending event and the pre-generated content relationship graph, determine at least one target department responsibility statement corresponding to each event description statement in the pending event, wherein the content relationship graph is generated based on historical work order events and at least part of the department responsibility statements of all execution departments associated with the historical work order events. Since the content relationship graph is determined based on a large number of historical work order events, it has captured the potential causal relationship between the event description statement and the department responsibility statement. Through the event description statement of the pending event and the content relationship graph, the target department responsibility statement involved in the pending event can be determined. Further, according to the event description statements of the pending event and the recall event, and at least one target department responsibility statement of the pending event, determine the first confidence of each recall event relative to the pending event, thereby, according to the first confidence of each recall event and the recall event to which the target responsibility department belongs, determine at least one similar event from multiple recall events. The technical solution provided in this embodiment can automatically determine multiple similar events that are close to the event to be processed from a large number of events. Through similar events, relevant event processing experience can be quickly obtained, which not only reduces labor costs but also improves event processing efficiency. Figure 3 This is a schematic diagram of another method for determining similar events provided by an embodiment of the present invention. Based on the above embodiment, the specific method for determining the content relationship diagram is described, and the steps S110-S130 are described in detail. The specific implementation method can refer to the technical solution of this embodiment. Among them, the technical terms that are the same or corresponding to the above embodiment are not repeated here.

[0107] like Figure 3 As shown, the method specifically comprises the following steps:

[0108] S210 . Process multiple event description statements of the event to be processed and at least some historical description statements of the historical work order events based on a similarity matching algorithm to obtain event similarities of the historical work order events with respect to the event to be processed.

[0109] Among them, the similarity matching algorithm is a processing method used to measure the similarity between two objects (such as text). The historical description sentence is a sentence in the historical work order event used to describe the process of reporting the problem. Event similarity is used to characterize the similarity between the historical work order event and the event to be processed. For example, the larger the event similarity value, the closer the historical work order event is to the event to be processed.

[0110] In this embodiment, the event library may include a large number of historical work order events. Similar departments and / or recent events may be used as the basis for recall, and a preliminary screening of the large number of historical work order events may be performed to obtain a portion of historical work order events. For this portion of historical work order events, a text matching process may be performed on multiple event description statements of the event to be processed and the historical description statements of these historical work order events using a similarity matching algorithm to obtain the event similarity of each historical work order event relative to the event to be processed.

[0111] S220: Determine multiple recall events associated with the event to be processed based on the event similarities corresponding to the historical work order events.

[0112] Among them, the recall event is an event in the historical work order event.

[0113] In this embodiment, a similarity threshold or a recall event quantity threshold may be pre-set, and determining multiple recall events associated with the event to be processed may include at least two methods. The first method may be to arrange the event similarities corresponding to each historical work order event in descending order, and determine the historical work order events corresponding to the event similarities of the recall event quantity threshold ranked top as the multiple recall events associated with the event to be processed. The second method may be to determine the historical work order events whose event similarity is greater than or equal to the similarity threshold as the multiple recall events associated with the event to be processed. The purpose of this setting is to determine the recall events from some historical work order events through the similarity matching algorithm. Since the processing process of the similarity matching algorithm does not need to consume a large amount of computing resources, it can efficiently determine multiple recall events that are relatively close to the event to be processed.

[0114] Next, it is explained in detail how to generate a content relationship graph based on historical work order events and at least part of the department responsibility statements of all execution departments associated with the historical work order events. The content relationship graph can be generated in the following manner: obtain multiple historical work order events; for multiple historical work order events, input the historical work order events into a pre-trained sentence embedding model to obtain sentence embeddings for each historical description sentence in the historical work order events; and, based on the sentence embedding model, process the department responsibility statements of the execution departments corresponding to the historical work order events to obtain the responsibility sentence embeddings corresponding to each department responsibility statement; determine the content relationship graph through the sentence embeddings of all historical work order events and the corresponding responsibility sentence embeddings.

[0115] Among them, the historical work order events include multiple historical description sentences and department responsibility sentences of the executing department. The historical description sentence is the historical problem description text. The sentence embedding model is used to convert sentences into vectors of fixed length. The main purpose of the sentence embedding model is to convert text data (i.e., sentences) into numerical form so that machine learning algorithms can process and analyze them. The sentence embedding model greatly improves the processing performance of text data by encoding sentences into density vectors. Sentence embedding is the mathematical form of the text that can be understood by the computer. Sentence embedding is the sentence embedding corresponding to the historical description sentence. Responsibility sentence embedding is the sentence embedding corresponding to the department responsibility sentence.

[0116] Specifically, each time a city event is solved, it can be regarded as a historical work order event, and the historical work order event can be saved in the event library, so that a large number of historical work order events can be obtained from the event library. The processing process for each historical work order event is the same, and any one of the historical work order events is used as an example for explanation. The input of the historical work order event can be input into the pre-trained sentence embedding model, and the sentence embedding model can encode each historical description sentence in the historical work order event to obtain the sentence embedding corresponding to each historical description sentence. At the same time, the sentence embedding model can encode each department responsibility statement of the execution department corresponding to the historical work order event to obtain the responsibility sentence embedding corresponding to each department responsibility statement. Based on the same processing method, one or more sentence embeddings corresponding to each historical work order event can be obtained, as well as one or more responsibility sentence embeddings. Furthermore, for each sentence embedding, the target responsibility sentence embedding corresponding to the sentence embedding can be determined by the degree of association between the sentence embedding and the corresponding multiple responsibility sentence embeddings. The sentence embedding and the corresponding target responsibility sentence embedding are respectively represented as nodes, and a connection edge between the sentence embedding node and the target responsibility sentence embedding node can be established. Based on the same processing method, the connection edges between all sentence embedding nodes and the corresponding target responsibility sentence embedding nodes can be obtained, and these nodes and connection edges together constitute a content relationship graph. The purpose of this setting is that since the pre-created content relationship graph has captured the potential mapping relationship between the event description sentence and the department responsibility sentence, the target department responsibility sentence corresponding to the event description sentence can be quickly and conveniently determined through the event description sentence of the event to be processed and the content relationship graph, thereby improving the efficiency of determining the target department responsibility sentence.

[0117] Optionally, a specific implementation method for determining the content relationship graph through the sentence embeddings of all historical work order events and the corresponding responsibility sentence embeddings may include: obtaining multiple cluster sets by clustering all sentence embeddings; for the sentence embeddings in each cluster set, determining the target responsibility sentence embedding of the sentence embedding based on the distance information between the sentence embedding and all responsibility sentence embeddings of the execution department to which the sentence embedding belongs; determining the content relationship graph based on all sentence embeddings in each cluster set and the corresponding target responsibility sentence embeddings.

[0118] Among them, each cluster set includes at least one sentence embedding. The distance information is used to characterize the correlation between the sentence embedding and all the responsibility sentence embeddings of the executive department to which the sentence embedding belongs. For example, the distance information can be a Euclidean distance value. The smaller the Euclidean distance value between the sentence embedding and a certain responsibility sentence embedding, the higher the correlation between the two. The target responsibility sentence embedding can be understood as the responsibility sentence embedding corresponding to one or more distance information that meets the preset conditions. The preset conditions may include but are not limited to: the distance information is less than or equal to the preset distance threshold; each distance information is arranged in ascending order, and a preset number of responsibility sentence embeddings are selected before the distance information ranking.

[0119] In this embodiment, all sentence embeddings can be clustered based on a preset clustering algorithm to obtain multiple clustering sets, wherein the preset clustering algorithm may include but is not limited to K-Means clustering, hierarchical clustering, density clustering, spectral clustering, etc. Next, the processing method for each clustering set is consistent, and any one of the clustering sets is used as an example for explanation. The clustering set may include multiple sentence embeddings, and for each sentence embedding, the distance information between the sentence embedding and all responsibility sentence embeddings of the executive department to which the sentence embedding belongs is calculated, and multiple distance information can be obtained, so that the responsibility sentence embedding corresponding to the distance information that meets the preset conditions is determined as the target responsibility sentence embedding of the sentence embedding, and one or more target responsibility sentence embeddings corresponding to each sentence embedding in the clustering set can be obtained based on the same determination method. Furthermore, for each cluster set, each sentence embedding in the cluster set is taken as a node, and the target responsibility sentence embedding corresponding to each sentence embedding is also taken as a node, and the connecting edges between the sentence embedding nodes and the target responsibility sentence embedding nodes are established. The processing method of each cluster set is the same. Finally, a content relationship graph including a large number of sentence embedding nodes, corresponding target responsibility sentence embedding nodes and corresponding connecting edges is obtained. The purpose of this setting is to cluster the sentence embeddings and construct a content relationship graph based on all the sentence embeddings and the corresponding target responsibility sentence embeddings in the cluster set. Since the sentences can be converted into the vector form of sentence embeddings, it is easier to capture the potential associations between different sentence expressions. When the content relationship graph is subsequently applied to determine the target department responsibility sentence corresponding to the event description sentence, the target department responsibility sentence can be determined based on the multiple target responsibility sentence embeddings corresponding to the cluster set to which the event description sentence belongs, thereby improving the matching degree between the target department responsibility sentence and the event description sentence.

[0120] S230: For each event description sentence, vectorize the event description sentence based on the pre-trained sentence embedding model to obtain the sentence embedding to be used of the event description sentence.

[0121] Among them, the sentence embedding to be used is the vectorized representation form of the event description sentence.

[0122] In this embodiment, in order to convert text data (i.e., sentences) into numerical form so that the machine learning algorithm can process and analyze, a sentence embedding model for encoding sentences can be pre-trained. Multiple event description sentences of the event to be processed can be input into the sentence embedding model to obtain the sentence embedding to be used corresponding to each event description sentence.

[0123] S240, determining the target cluster set to which the sentence embedding to be used belongs, and determining at least one target department responsibility sentence of the event description sentence according to the number of associations of sentence embeddings associated with each responsibility sentence embedding in the target cluster set.

[0124] The target department responsibility sentences are sentences in the department responsibility sentences associated with all sentences in the target cluster set. The target cluster set is some cluster sets in the content relationship graph.

[0125] In this embodiment, since the cluster center of each cluster set in the content relationship graph is determined, the target cluster set to which the sentence embedding to be used belongs can be determined by calculating the distance information between the sentence embedding to be used and each cluster center. For example, if the distance value between the sentence embedding to be used and a certain cluster center is preset, then the distance set is the target cluster set. It should be noted that a sentence embedding to be used can correspond to multiple target cluster sets. In these target cluster sets, there is a situation where a certain responsibility sentence embedding corresponds to multiple sentence embeddings. Based on this, the number of sentence embeddings associated with each responsibility sentence embedding can be used as the number of associations. Thus, some responsibility sentence embeddings with a large number of associations can be used as target responsibility sentence embeddings, and the department responsibility statements corresponding to each target responsibility sentence embedding can be determined as at least one target department responsibility statement of the event description statement. Based on the same processing method, at least one target department responsibility statement corresponding to each event description statement can be determined.

[0126] In this embodiment, by sentence embedding and determining the target cluster set, at least one target department responsibility statement corresponding to each event description statement in the event to be processed is determined, which can improve the richness of the determined target department responsibility statements and improve the matching of the target department responsibility statements with the event description statements.

[0127] S250. Adjust at least one target department responsibility statement corresponding to the event to be processed to a responsibility description statement corresponding to the execution department in the recall event, and determine a first confidence matrix of the event description statement relative to the department responsibility statement.

[0128] In this embodiment, for a certain recall event, the execution department corresponding to the recall event can be determined according to the recorded text description information corresponding to the recall event. Based on this, multiple duty description statements corresponding to the execution department are easily obtained. Furthermore, at least one target department duty statement of the event to be processed can be adjusted to multiple duty description statements corresponding to the execution department. This is because, in general, similar events often correspond to similar duty statements, so exchanging the corresponding duty statements of similar events will not produce large causal differences. Therefore, after adjusting the duty statement, further calculating the first confidence of the event description statement relative to the department duty statement can characterize the degree of similarity between the event to be processed and the adjusted recall event.

[0129] Next, how to determine the first confidence matrix of the event description statement relative to the department responsibility statement is described in detail. The specific implementation method of determining the first confidence matrix includes: determining a first vector matrix based on the event description statement, and creating a second vector matrix based on the department responsibility statement; determining the matching coefficient matrix of the event description statement relative to the department responsibility statement based on the first vector matrix and the second vector matrix; obtaining the first unidirectional attention matrix by normalizing the matching coefficient matrix columns; and determining the first confidence matrix based on the first unidirectional attention matrix.

[0130] Among them, each row vector in the first vector matrix can represent an event description statement. Each row vector in the second vector matrix can represent a department responsibility statement. The element values ​​in the matching coefficient matrix are used to represent the degree of matching between the event description statement and the department responsibility statement. The first unidirectional attention matrix is ​​a matrix expression obtained by processing the matching coefficient matrix using a unidirectional attention mechanism. Each element in the first row of the first unidirectional attention matrix represents the degree of matching between the first event description statement in the event to be processed and each target department responsibility statement. Similarly, the meaning of each element in other rows can be known.

[0131] In this embodiment, each adjusted target department responsibility statement is converted into a corresponding first eigenvector, and each first eigenvector constitutes a first vector matrix, and each event description statement of the event to be processed is converted into a corresponding second eigenvector, and each second eigenvector constitutes a second vector matrix; and the transpose of the first vector matrix and the second vector matrix is ​​multiplied by a dot multiplication operation to obtain a matching coefficient matrix of the event description statement relative to the target department responsibility statement, and then the matching coefficient matrix is ​​column normalized to obtain a first unidirectional attention matrix, and then the first unidirectional attention matrix and the transpose of the first unidirectional attention matrix are multiplied by a dot multiplication operation to obtain a first confidence matrix. In this embodiment, through the matrix operation method and the attention mechanism, the first confidence matrix of a recall event relative to the event to be processed can be accurately determined.

[0132] S260: Determine at least one similar event corresponding to the event to be processed based on the first confidence level of each recall event and the recall event to which the target department responsibility statement belongs.

[0133] The technical solution of the embodiment of the present invention predetermines a content relationship graph. Since the predetermined content relationship graph has captured the potential mapping relationship between the event description statement and the department responsibility statement, the target department responsibility statement corresponding to the event description statement can be quickly and conveniently determined in the subsequent event description statement and the content relationship graph of the event to be processed, thereby improving the efficiency of determining the target department responsibility statement. In specific applications, by processing multiple event description statements of the event to be processed and historical description statements of at least part of the historical work order events based on a similarity matching algorithm, the event similarity of the historical work order event relative to the event to be processed is obtained, and then, based on the event similarity corresponding to the historical work order event, multiple recall events associated with the event to be processed are determined. Since the processing process of the similarity matching algorithm does not need to consume a large amount of computing resources, multiple recall events that are relatively close to the event to be processed can be efficiently determined. Further, for each event description sentence, the event description sentence is vectorized based on the pre-trained sentence embedding model to obtain the sentence embedding to be used of the event description sentence, thereby determining the target cluster set to which the sentence embedding to be used belongs, and determining at least one target department responsibility sentence of the event description sentence according to the number of associated sentence embeddings associated with each responsibility sentence embedding in the target cluster set. By sentence embedding and determining the target cluster set, at least one target department responsibility sentence corresponding to each event description sentence in the event to be processed is determined, which can improve the richness of the target department responsibility sentence determined, and improve the matching of the target department responsibility sentence with the event description sentence. When determining the first confidence matrix, at least one target department responsibility sentence corresponding to the event to be processed is adjusted to the responsibility description sentence corresponding to the execution department in the recall event, and the first confidence matrix of the event description sentence relative to the department responsibility sentence is determined, and then the first confidence is determined according to the first confidence matrix, which improves the accuracy of the first confidence, so that similar events that are closer to the event to be processed can be determined according to the first confidence, so that the staff can obtain event processing experience with greater utilization value, and further improve the processing efficiency of the event.

[0134] Figure 4 This is a schematic diagram of another method for determining similar events provided by an embodiment of the present invention. On the basis of the above-mentioned embodiment, the first confidence level can also be updated according to the processing opinion statement in the recall event to determine the at least one similar event based on the updated first confidence level. The specific implementation method can refer to the technical solution of this embodiment. Among them, the technical terms that are the same or corresponding to the above-mentioned embodiment are not repeated here.

[0135] like Figure 4 As shown, the method specifically comprises the following steps:

[0136] S310: Acquire multiple recall events associated with the event to be processed.

[0137] Among them, both pending events and recalled events include multiple event description statements.

[0138] S320: Determine at least one target department responsibility statement corresponding to each event description statement in the event to be processed based on multiple event description statements of the event to be processed and a pre-generated content relationship diagram.

[0139] The content relationship graph is generated based on historical work order events and at least part of the department responsibility statements of all execution departments associated with the historical work order events.

[0140] S330: Determine a first confidence level of each recall event relative to the event to be processed based on the event description statements of the event to be processed and the recall event, and at least one target department responsibility statement of the event to be processed.

[0141] S340: Determine at least one similar event corresponding to the event to be processed based on the first confidence level of each recall event and the recall events to which the target responsible department belongs.

[0142] Among them, similar incidents are incidents in recall events.

[0143] S350. Determine a statement matching matrix based on at least one target department responsibility statement and multiple event description statements, and determine an opinion matching matrix based on multiple event description statements and handling opinions in the recall event.

[0144] The handling opinions are the recorded text of the problem solving results in the recall event. The element values ​​in the sentence matching matrix are used to represent the similarity between the description sentence and the target department's responsibility sentence, and the element values ​​in the opinion matching matrix are used to represent the matching degree between the description sentence and the handling opinions.

[0145] In this embodiment, the processing process for each recall event is consistent, and any one of them is used as an example for illustrative description. The recall event includes at least one historical event description statement, an execution department statement, and at least one handling opinion statement. The pending event includes multiple event description statements and at least one target department responsibility statement determined in step S320.

[0146] Specifically, each event description statement of the pending event can be converted into a feature vector, each target department responsibility statement can be converted into a feature vector, and each handling opinion statement can be converted into a feature vector. The pending event can be represented as E(x) = {X = [x1, x2, ..., x n ], {D=[d1,d2,...,d m ]}, where x1 represents the feature vector of the first event description statement in the pending event, d1 represents the feature vector of the first target department responsibility statement corresponding to the pending event, and the pending event E(x) corresponds to n event description statements and m target department responsibility statements. Any recall event can be represented by E(y), and the event description statement corresponding to the recall event is E(y)={Y=[y1,y2,...,y u ],V,F=[f1,f2,...,f b ]}, where y1 represents the feature vector of the first historical event description statement of the reported problem in the recall event, V represents the executive department corresponding to the recall event, F represents the handling opinions for the reported problems in the recall event, and f1 represents the feature vector of the first handling opinion statement for the reported problems in the recall event. The recall event E(y) corresponds to u historical event description statements and b handling opinion statements. Based on this, the feature vector corresponding to the responsibility statement of each target department and the transpose of the feature vector corresponding to each event description statement can be multiplied to obtain the statement matching matrix Perform a dot multiplication operation on the transpose of the feature vector corresponding to each handling opinion statement and the feature vector corresponding to each event description statement to obtain the opinion matching matrix

[0147] S360: Determine a statement set according to the statement matching matrix and the opinion matching matrix.

[0148] Among them, the statement set includes at least one first statement, and the first statement is an event description statement whose correlation with the target department's responsibility statement and handling opinions is lower than a preset threshold.

[0149] In this embodiment, the statement matching matrix and the opinion matching matrix can be transformed respectively. According to the transformed statement matching matrix and the opinion matching matrix, one or several event description statements that are weakly associated with the target department's responsibility statement and the handling opinion statement can be determined from multiple event description statements, and the set of these event description statements can be used as a statement set.

[0150] Specifically, according to the sentence matching matrix and the opinion matching matrix, a specific implementation method of determining the sentence set may include:

[0151] First, by normalizing the columns of the sentence matching matrix, we obtain the sentence unidirectional attention matrix of the event description sentence relative to the target department responsibility sentence; and by normalizing the rows of the sentence matching matrix, we obtain the responsibility unidirectional attention matrix.

[0152] In this embodiment, the sentence matching matrix After column normalization, we can obtain the sentence unidirectional attention matrix of the event description sentence relative to the target department responsibility sentence. The Pth row in the sentence unidirectional attention matrix can be expressed as:

[0153] Hα(p)=softmax(H(1,p),H(2,p),...H(n,p))

[0154] Where Hα(p) represents x p The degree of match with each target department's responsibility statement, x p Represents the feature vector corresponding to the p-th event description statement of the event to be processed.

[0155] Sentence matching matrix Perform row normalization to obtain the responsible unidirectional attention matrix. The Qth row of the responsible unidirectional attention matrix can be expressed as:

[0156] Hβ(q)=softmax(H(q,1),H(q,2),…H(q,m))

[0157] Where Hβ(q) represents d p The degree of match with each event description statement, d p Represents the feature vector corresponding to the qth target department responsibility statement.

[0158] Secondly, by normalizing the columns of the opinion matching matrix, we obtain the first opinion one-way attention matrix of the event description sentence relative to the handling opinion, and by normalizing the rows of the opinion matching matrix, we obtain the second opinion one-way attention matrix.

[0159] In this embodiment, the opinion matching matrix After column normalization, we can obtain the first opinion one-way attention matrix of the event description sentence relative to the handling opinion sentence. The Pth row in the first opinion one-way attention matrix can be expressed as:

[0160] Hα(p)=softmax(G(1,p),G(2,p),...G(n,p))

[0161] In the formula, Gα(p) represents x p The degree of match with each handling opinion statement, x pRepresents the feature vector corresponding to the p-th event description statement of the event to be processed.

[0162] Sentence matching matrix Perform row normalization to obtain the second opinion unidirectional attention matrix. The Qth row of the second opinion unidirectional attention matrix can be expressed as:

[0163] Gβ(q)=softmax(G(q,1),G(q,2),…G(q,m))

[0164] In the formula, Gβ(q) represents f q The degree of match with each event description statement, f q Represents the feature vector corresponding to the qth handling opinion statement.

[0165] For example, a diagram of the recorded text description information of a recall event can be found in Figure 5 ,like Figure 5 As shown in the figure, the text description information of the recall event includes three aspects, namely, the reported problem (i.e., the historical event description statement), the dispatched department (i.e., the execution department), and the reply of the Automobile Transportation Group (the handling opinion statement). After analyzing the causal logic of the dispatch process of the recall event, it can be seen that the reported problem is the cause of the dispatched department, and the reported problem and the dispatched department are the causes of the reply of the Automobile Transportation Group. Figure 5 As shown, the text content corresponding to the black solid underline is the causal relationship between the reported problem and the dispatched department; the text content corresponding to the black dotted underline is the causal relationship between the reported problem and the handling opinion. For this reason, in this embodiment, a bidirectional attention mechanism can be used to analyze the causal relationship, and the attention mechanism is used to enable the model to have the ability to capture the focus of attention between data.

[0166] Furthermore, based on the sentence unidirectional attention matrix and the responsibility unidirectional attention matrix, the second sentence is determined from the event description sentence; and based on the first opinion unidirectional attention matrix and the second opinion unidirectional attention matrix, the third sentence is determined from the event description sentence.

[0167] The second statement is an event description statement that is highly relevant to the target department's responsibility statement, and the third statement is an event description statement that is highly relevant to the handling opinion statement.

[0168] In this embodiment, the specific implementation method of determining the second statement from the event description statement is:

[0169] The responsibility unidirectional attention matrix is ​​averaged in the column direction to obtain the first weight coefficient corresponding to each event description sentence. The first weight coefficient β corresponding to any event description sentence have The determination method can be expressed as:

[0170]

[0171] Based on the same processing method, the first weight coefficient corresponding to each event description statement can be obtained. Furthermore, the first weight coefficients can be screened according to the first weight threshold, and the event description statements whose first weight coefficients are greater than or equal to the first weight threshold are determined as the second statement set D.

[0172] Similarly, the specific implementation method of determining the third statement from the event description statement is:

[0173] The second opinion unidirectional attention matrix is ​​averaged in the column direction to obtain the second weight coefficient corresponding to each event description sentence. The second weight coefficient β corresponding to any event description sentence is gave The determination method can be expressed as:

[0174]

[0175] Based on the same processing method, the second weight coefficient corresponding to each event description statement can be obtained. Furthermore, the second weight coefficients can be screened according to the second weight threshold, and the event description statements whose second weight coefficients are greater than or equal to the second weight threshold are determined as the third statement H.

[0176] Finally, the statements in the event description statement except the second statement and the third statement are taken as the first statement in the statement set.

[0177] In this embodiment, each statement in the event description statement except the second statement set D and the third statement set H is taken as the first statement, and each first statement constitutes the statement set R. Based on this method, the first statement whose correlation with the target department's responsibility statement and handling opinion is lower than the preset threshold can be quickly and efficiently determined, providing a data basis for the subsequent calculation of the second confidence.

[0178] Exemplarily, the event to be processed includes 6 event description statements, namely statement 1, statement 2, statement 3, statement 4, statement 5 and statement 6. The second statement set D includes statement 1, statement 2 and statement 3. The third statement set D includes statement 2, statement 3 and statement 4. Then the statement set R includes statement 5 and statement 6.

[0179] S370: Determine a second confidence level of the recalled event relative to the event to be processed based on the statement set, the recalled event, and the event description statements of the event to be processed.

[0180] In this embodiment, for each recall event, the statement set and the handling opinion statement in the recall event can be reconstructed to describe the event to be processed, and then, based on the reconstructed event description statement, the second confidence of each recall event relative to the event to be processed can be determined.

[0181] In this embodiment, the specific implementation methods for determining the second confidence level of each recall event relative to the event to be processed are consistent, and any one of them is taken as an example for description.

[0182] Specifically, based on the statement set, the recalled event, and the event description statements of the event to be processed, a specific implementation method of determining the second confidence level of the recalled event relative to the event to be processed may include:

[0183] First, based on the handling opinions of the recall event, the event description statement of the pending event is updated, and at least part of the first statement in the statement set in the pending event is replaced with the event description statement in the recall event to obtain an updated pending event.

[0184] In this embodiment, since the pending event has not been processed, there is no processing opinion for the pending event, and the processing opinion of the recall event can be used as part of the event description statement of the pending event. And a part of the first statement in the statement set in the pending event can be replaced with the event description statement in the recall event (here refers to the historical event description statement in the recall event), so as to complete the reconstruction process of the event description statement of the pending event and obtain the updated pending event.

[0185] On the basis of the above examples, the historical event description statements in the recall event may include: problem description A, problem description B, problem description C, and problem description D, and the handling opinion statements of the recall event may include: opinion statement E, opinion statement F, opinion statement G, and opinion statement H. The opinion statements E, opinion statement F, opinion statement G, and opinion statement H of the recall event may be used as part of the event description statements of the event to be processed, and the first statement in the statement set (i.e., statement 5 and statement 6) is replaced with the problem description B and the problem description C, and the problem description B and the problem description C in the recall event are replaced with statements 5 and 6. Based on this, the updated event description statements corresponding to the event to be processed include: statement 1, statement 2, statement 3, statement 4, problem description B, problem description C, opinion statement E, opinion statement F, opinion statement G, and opinion statement H, and the problem description B and the problem description C in the recall event are changed to statements 5 and 6.

[0186] Subsequently, an updated opinion matrix regarding handling opinions between pending events and recalled events is determined.

[0187] On the basis of the above examples, the feature vectors corresponding to each statement are determined respectively, and the feature vectors corresponding to statement 1, statement 2, statement 3, statement 4, problem description B, and problem description C are concatenated with the feature vectors corresponding to the responsibility statements of each target department to obtain the feature matrix to be processed. The feature vectors of opinion statement E, opinion statement F, opinion statement G, and opinion statement H constitute the opinion feature matrix. The counterfactual inference coefficient matrix L is obtained by performing dot multiplication on the transpose of the feature matrix to be processed and the opinion feature matrix. Then, the counterfactual inference coefficient matrix L is column-normalized to obtain the updated opinion matrix γ(p) regarding the handling opinions between the events to be processed and the recall events.

[0188] Finally, based on the opinion matrix, a second confidence level of the updated event to be processed corresponding to the recalled event is determined.

[0189] In this embodiment, the covariance matrix C can be calculated z =γ(p)·γ(p) T , the final covariance matrix C z The confidence value obtained by performing the partial correlation CI-test is the second confidence of the updated pending event corresponding to the recalled event. If the two events are similar, then exchanging partial information of the two events will not produce a large causal difference. The calculation basis of the second confidence is to exchange partial information content of the pending event and the recalled event. Therefore, the second confidence can characterize the similarity between the pending event and the recalled event.

[0190] S380: Update the first confidence of the recalled event relative to the event to be processed according to the second confidence, so as to determine at least one similar event based on the updated first confidence.

[0191] In this embodiment, the specific implementation of the new recall event based on the second confidence level relative to the first confidence level of the event to be processed may include: updating the first confidence level of the recall event relative to the event to be processed based on the weighted sum of the first confidence level and the second confidence level; or replacing the value of the first confidence level with the value of the second confidence level. Further, based on the updated value of the first confidence level, the recall event whose first confidence level is greater than a preset confidence threshold is determined as at least one similar event.

[0192] In this embodiment, the first confidence level is updated by the second confidence level, thereby improving the reliability of the confidence level and further improving the matching degree between similar events and the event to be processed.

[0193] On the basis of the above embodiments, a third confidence level can be further determined according to the text similarity between the event description statement of the event to be processed and the event description statement of the recalled event. Based on this, the method for determining similar events provided in this embodiment can also include: for each recalled event, determining the third confidence level between the event description statement of the event to be processed and the event description statement of the recalled event; updating the first confidence level or the second confidence level based on the third confidence level to determine at least one similar event based on the updated first confidence level.

[0194] In this embodiment, for each recall event, the text similarity calculation can be performed on the event description statement of the event to be processed and the event description statement of the recall event by a text similarity algorithm, and the text similarity corresponding to the event description statements of the two events is obtained as the third confidence. Furthermore, the first confidence can be updated according to the weighted sum of the third confidence, the second confidence and the first confidence; or, the first confidence can be updated by replacing the second confidence with the third confidence and then calculating the weighted sum of the second confidence and the first confidence. The first confidence is updated by the confidence determined by multiple methods, which further improves the reliability of the confidence.

[0195] Based on the above embodiment, the method further includes: determining a preset number of recall events in descending order according to the first confidence as similar events to the event to be processed.

[0196] In this embodiment, the larger the first confidence value is, the higher the similarity between the recalled event and the event to be processed is. The first confidences can be arranged in descending order, and then the recalled events corresponding to the first confidences of the first preset number of rankings are used as similar events to the event to be processed. Recalled events with high first confidences are preferentially selected as similar events to the event to be processed, which ensures the availability of the determined similar events, so that the determined similar events can provide a reference basis for the processing process of the event to be processed to a greater extent.

[0197] The technical solution of the embodiment of the present invention can also determine a statement matching matrix based on at least one target department responsibility statement and multiple event description statements, and determine an opinion matching matrix based on multiple event description statements and handling opinions in recall events, wherein the element values ​​in the statement matching matrix are used to characterize the similarity between the description statement and the target department responsibility statement, and the element values ​​in the opinion matching matrix are used to characterize the matching degree between the description statement and the handling opinions; determine a statement set based on the statement matching matrix and the opinion matching matrix; wherein the first statement in the statement set is a statement whose correlation with the target responsibility department and the handling opinion is lower than a preset threshold; determine a second confidence of the recall event relative to the to-be-processed event based on the statement set, the recall event, and the event description statement of the to-be-processed event; and update the first confidence of the recall event relative to the to-be-processed event based on the second confidence, so as to determine at least one similar event based on the updated first confidence. The technical solution provided in this embodiment can update the first confidence level through confidence levels determined in a variety of ways, further improving the reliability of the confidence level, thereby improving the matching degree between similar events and events to be processed, so that similar events that are closer to the events to be processed can be determined based on the updated first confidence level. This enables staff to gain more valuable event processing experience, which helps to further improve event processing efficiency.

[0198] Figure 6 A schematic diagram of the structure of a similar event determination device provided in an embodiment of the present invention, the device comprises: a recall event determination module 410, a responsibility statement determination module 420, a confidence determination module 430 and a similar event determination module 440.

[0199] The recall event determination module 410 is used to obtain multiple recall events associated with the pending event, wherein both the pending event and the recall event include multiple event description statements;

[0200] The responsibility statement determination module 420 is used to determine at least one target department responsibility statement corresponding to each event description statement in the event to be processed based on multiple event description statements of the event to be processed and a pre-generated content relationship graph; wherein the content relationship graph is generated based on historical work order events and at least part of the department responsibility statements of all execution departments associated with the historical work order events;

[0201] A confidence determination module 430, configured to determine a first confidence of each recall event relative to the event to be processed based on event description statements of the event to be processed and the recall event, and at least one target department responsibility statement of the event to be processed;

[0202] The similar event determination module 440 is used to determine at least one similar event corresponding to the event to be processed based on the first confidence of each recall event and the recall event to which the target responsible department belongs; wherein the similar event is an event in the recall event.

[0203] Based on the above device, optionally, the recall event determination module 410 includes:

[0204] A similarity determination unit is used to process multiple event description statements of the event to be processed and historical description statements of at least part of the historical work order events based on a similarity matching algorithm to obtain event similarity of the historical work order events with respect to the event to be processed;

[0205] The recall event determination unit is used to determine multiple recall events associated with the event to be processed based on the event similarity corresponding to the historical work order events; wherein the recall event is an event in the historical work order events.

[0206] Based on the above device, optionally, the similar event determination device further includes: a content relationship graph determination module; the content relationship graph determination module specifically includes:

[0207] A historical event acquisition unit, used to acquire a plurality of historical work order events, wherein the historical work order events include a plurality of historical description statements and departmental responsibility statements of the executing department;

[0208] A sentence embedding determination unit is used to input the historical work order events into a pre-trained sentence embedding model for multiple historical work order events to obtain the sentence embedding of each historical description sentence in the historical work order events; and to process the department responsibility sentences of the execution department corresponding to the historical work order events based on the sentence embedding model to obtain the responsibility sentence embedding corresponding to each department responsibility sentence

[0209] The content relationship graph determination unit is used to determine the content relationship graph through the sentence embedding of all historical work order events and the corresponding responsibility sentence embedding.

[0210] Based on the above device, optionally, the content relationship graph determining unit includes:

[0211] The sentence embedding clustering subunit is used to obtain a plurality of clustering sets by clustering all sentence embeddings, wherein each clustering set includes at least one sentence embedding;

[0212] A responsibility sentence embedding determination subunit is used to determine, for each sentence embedding in each cluster set, a target responsibility sentence embedding of the sentence embedding according to distance information between the sentence embedding and all responsibility sentence embeddings of the execution department to which the sentence embedding belongs;

[0213] The content relationship graph determination subunit is used to determine the content relationship graph based on all sentence embeddings in each cluster set and the corresponding target responsibility sentence embeddings.

[0214] Based on the above device, optionally, the responsibility statement determination module 420 includes:

[0215] A sentence vectorization unit, for vectorizing each event description sentence based on a pre-trained sentence embedding model to obtain a sentence embedding to be used for the event description sentence;

[0216] A responsibility statement determination unit, used to determine a target cluster set to which the sentence embedding to be used belongs, and determine at least one target department responsibility statement of the event description statement according to the number of associations of sentence embeddings associated with each responsibility sentence embedding in the target cluster set;

[0217] The target department responsibility sentences are sentences in the department responsibility sentences associated with all the sentence embeddings in the target cluster set.

[0218] Based on the above-mentioned device, optionally, the confidence determination module 430 is specifically used to adjust at least one target department responsibility statement corresponding to the event to be processed to a responsibility description statement corresponding to the execution department in the recall event, and determine a first confidence matrix of the event description statement relative to the department responsibility statement.

[0219] Based on the above device, optionally, the confidence determination module 430 specifically includes:

[0220] A vector matrix determination unit, configured to determine a first vector matrix based on the event description statement, and to create a second vector matrix based on the departmental responsibility statement;

[0221] A matching coefficient matrix determining unit, used to determine a matching coefficient matrix of the event description statement relative to the department responsibility statement based on the first vector matrix and the second vector matrix;

[0222] An attention matrix determination unit, configured to obtain a first unidirectional attention matrix by normalizing the matching coefficient matrix columns;

[0223] The confidence matrix determination unit is used to determine a first confidence matrix based on the first unidirectional attention matrix.

[0224] Based on the above device, optionally, the similar event determination device further includes: a similar event update module; the similar event update module specifically includes:

[0225] An opinion matching matrix determining unit is used to determine a statement matching matrix based on at least one target department responsibility statement and a plurality of event description statements, and to determine an opinion matching matrix based on a plurality of event description statements and handling opinions in the recall events, wherein the element values ​​in the statement matching matrix are used to characterize the similarity between the description statement and the target department responsibility statement, and the element values ​​in the opinion matching matrix are used to characterize the matching degree between the description statement and the handling opinions;

[0226] A statement set determination unit, used to determine a statement set according to a statement matching matrix and an opinion matching matrix; wherein the first statement in the statement set is a statement whose relevance to the target responsible department and the handling opinion is lower than a preset threshold;

[0227] A confidence determination unit, configured to determine a second confidence of the recalled event relative to the event to be processed based on the statement set, the recalled event, and the event description statement of the event to be processed;

[0228] The similar event graph updating unit is used to update the first confidence of the recalled event relative to the event to be processed according to the second confidence, so as to determine at least one similar event based on the updated first confidence.

[0229] Based on the above device, optionally, the statement set determination unit further includes:

[0230] The sentence matrix determination subunit is used to obtain a sentence unidirectional attention matrix of the event description sentence relative to the target department responsibility sentence by normalizing the sentence matching matrix columns; and to obtain a responsibility unidirectional attention matrix by normalizing the sentence matching matrix rows;

[0231] The opinion matrix determination subunit is used to obtain a first opinion one-way attention matrix of the event description sentence relative to the handling opinion by normalizing the columns of the opinion matching matrix, and to obtain a second opinion one-way attention matrix by normalizing the rows of the opinion matching matrix;

[0232] a sentence determination subunit, configured to determine a second sentence from the event description sentence according to the sentence unidirectional attention matrix and the responsibility unidirectional attention matrix; and to determine a third sentence from the event description sentence based on the first opinion unidirectional attention matrix and the second opinion unidirectional attention matrix;

[0233] The statement set determining subunit is used for the statement set determining subunit, and is used for taking the statements in the event description statement except the second statement and the third statement as the first statement in the statement set.

[0234] Based on the above device, optionally, the confidence determination unit further includes:

[0235] The pending event updating subunit is used to update the event description statement of the pending event based on the handling opinions of the recall event for each recall event, and replace at least part of the first statement in the statement set in the pending event with the event description statement in the recall event to obtain an updated pending event;

[0236] An opinion matrix updating subunit, used to determine an updated opinion matrix on handling opinions between pending events and recalled events;

[0237] The confidence determination subunit is used to determine the second confidence of the updated event to be processed corresponding to the recalled event based on the opinion matrix.

[0238] Based on the above-mentioned device, optionally, a similar event updating module is further used to determine, for each recall event, a third confidence between the event description statement of the event to be processed and the event description statement of the recall event; update the first confidence or the second confidence based on the third confidence to determine at least one similar event based on the updated first confidence.

[0239] On the basis of the above device, optionally, the similar event updating module is further used to determine a preset number of recall events as similar events to the event to be processed in descending order according to the first confidence level.

[0240] On the basis of the above-mentioned device, optionally, the similar event determination module is further used to determine the target processing result corresponding to the event to be processed based on the at least one similar event.

[0241] The technical solution of the embodiment of the present invention is to obtain multiple recall events associated with the pending event, wherein the pending event and the recall event both include multiple event description statements, and then, according to the multiple event description statements of the pending event and the pre-generated content relationship graph, determine at least one target department responsibility statement corresponding to each event description statement in the pending event, wherein the content relationship graph is generated based on historical work order events and at least part of the department responsibility statements of all execution departments associated with the historical work order events. Since the content relationship graph is determined based on a large number of historical work order events, it has captured the potential causal relationship between the event description statement and the department responsibility statement. Through the event description statement of the pending event and the content relationship graph, the target department responsibility statement involved in the pending event can be determined. Further, according to the event description statements of the pending event and the recall event, and at least one target department responsibility statement of the pending event, determine the first confidence of each recall event relative to the pending event, thereby, according to the first confidence of each recall event and the recall event to which the target responsibility department belongs, determine at least one similar event from multiple recall events. The technical solution provided by this embodiment can automatically determine multiple similar events that are close to the event to be processed from a large number of events, and can quickly obtain relevant event processing experience through similar events, which not only reduces labor costs but also improves event processing efficiency. The similar event determination device provided by the embodiment of the present invention can execute the similar event determination method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0242] It is worth noting that the various units and modules included in the above system are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the protection scope of the embodiments of the present invention.

[0243] Figure 7 The present invention is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. Figure 7 A block diagram of an exemplary electronic device 50 suitable for implementing exemplary embodiments of the present invention is shown. Figure 7 The electronic device 50 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0244] like Figure 7 As shown, the electronic device 50 is in the form of a general computing device. The components of the electronic device 50 may include but are not limited to: one or more processors or processing units 501, a system memory 502, and a bus 503 connecting different system components (including the system memory 502 and the processing unit 501).

[0245] Bus 503 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus structures. For example, these architectures include but are not limited to Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MAC) bus, Enhanced ISA bus, Video Electronics Standards Association (VESA) local bus and Peripheral Component Interconnect (PCI) bus.

[0246] The electronic device 50 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the electronic device 50, including volatile and non-volatile media, removable and non-removable media.

[0247] System memory 502 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 504 and / or cache memory 505. Electronic device 50 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 506 may be used to read and write non-removable, non-volatile magnetic media ( Figure 7 not shown, usually called a "hard drive"). Although Figure 7 Not shown in the figure, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 503 via one or more data medium interfaces. The memory 502 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the embodiments of the present invention.

[0248] A program / utility 508 having a set (at least one) of program modules 507 may be stored, for example, in the memory 502, such program modules 507 including but not limited to an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. The program modules 507 generally perform the functions and / or methods of the embodiments described herein.

[0249] The electronic device 50 may also communicate with one or more external devices 509 (e.g., keyboards, pointing devices, displays 510, etc.), may communicate with one or more devices that enable a user to interact with the electronic device 50, and / or communicate with any device that enables the electronic device 50 to communicate with one or more other computing devices (e.g., a network card, a modem, etc.). Such communication may be performed through an input / output (I / O) interface 511. Furthermore, the electronic device 50 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter 512. As shown, the network adapter 512 communicates with other modules of the electronic device 50 through the bus 503. It should be understood that although Figure 7 Not shown, other hardware and / or software modules may be used in conjunction with the electronic device 50, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0250] The processing unit 501 executes various functional applications and page processing by running the programs stored in the system memory 502, for example, implementing the method for determining similar events provided in the embodiment of the present invention.

[0251] An embodiment of the present invention further provides a storage medium containing computer executable instructions. When the computer executable instructions are executed by a computer processor, they are used to perform a method for determining similar events. The method includes:

[0252] Acquire multiple recall events associated with the pending event, wherein both the pending event and the recall event include multiple event description statements;

[0253] Determine at least one target department responsibility statement corresponding to each event description statement in the event to be processed based on multiple event description statements of the event to be processed and a pre-generated content relationship graph; wherein the content relationship graph is generated based on historical work order events and at least part of the department responsibility statements of all execution departments associated with the historical work order events;

[0254] Determine a first confidence level of each recalled event relative to the event to be processed based on event description statements of the event to be processed and the recalled event, and at least one target department responsibility statement of the event to be processed;

[0255] At least one similar event corresponding to the event to be processed is determined based on the first confidence of each recall event and the recall event to which the target responsible department belongs; wherein the similar event is an event in the recall event.

[0256] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0257] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0258] Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0259] Computer program code for performing the operation of embodiments of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0260] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for determining similar events, characterized in that: include: Acquire multiple recall events associated with the pending event, wherein the pending event and the recall event both include multiple event description statements; According to the multiple event description statements of the event to be processed and the pre-generated content relationship graph, at least one target department responsibility statement corresponding to each event description statement in the event to be processed is determined; wherein the content relationship graph is generated based on the historical work order event and at least part of the department responsibility statements of all execution departments associated with the historical work order event; Determine a first confidence level of each recall event relative to the event to be processed according to the event description statements of the event to be processed and the recall event, and at least one target department responsibility statement of the event to be processed; According to the first confidence of each recall event and the recall event to which the target department responsibility statement belongs, at least one similar event corresponding to the event to be processed is determined; wherein the similar event is an event in the recall event.

2. The method according to claim 1, characterized in that The obtaining of multiple recall events associated with the event to be processed includes: Processing multiple event description statements of the event to be processed and historical description statements of at least part of historical work order events based on a similarity matching algorithm to obtain event similarity of the historical work order event with respect to the event to be processed; Based on the event similarity corresponding to the historical work order event, a plurality of recall events associated with the event to be processed are determined; wherein the recall event is an event in the historical work order event.

3. The method according to claim 1, characterized in that The content relationship diagram is generated in the following manner: Acquire multiple historical work order events, wherein the historical work order events include multiple historical description statements and department responsibility statements of the execution department; For the multiple historical work order events, the historical work order events are input into the pre-trained sentence embedding model to obtain the sentence embedding of each historical description sentence in the historical work order event; and based on the sentence embedding model, the department responsibility statements of the execution department corresponding to the historical work order event are processed to obtain the responsibility sentence embedding corresponding to each department responsibility statement. The content relationship graph is determined by embedding the sentences of all historical work order events and embedding the corresponding responsibility sentences.

4. The method according to claim 3, characterized in that Determining the content relationship graph by embedding the sentences of all historical work order events and the corresponding responsibility sentences includes: By clustering all sentence embeddings, a plurality of cluster sets are obtained, wherein each cluster set includes at least one sentence embedding; For each sentence embedding in each cluster set, determine the target responsibility sentence embedding of the sentence embedding according to the distance information between the sentence embedding and all responsibility sentence embeddings of the execution department to which the sentence embedding belongs; The content relationship graph is determined based on all sentence embeddings in each cluster set and the corresponding target responsibility sentence embeddings.

5. The method according to claim 1, characterized in that The step of determining at least one target department responsibility statement corresponding to each event description statement in the event to be processed based on the multiple event description statements of the event to be processed and the pre-generated content relationship diagram includes: For each event description sentence, vectorize the event description sentence based on a pre-trained sentence embedding model to obtain a sentence embedding to be used for the event description sentence; Determine a target cluster set to which the sentence embedding to be used belongs, and determine at least one target department responsibility sentence of the event description sentence according to the number of associations of sentence embeddings associated with each responsibility sentence embedding in the target cluster set; The target department responsibility statements are statements in which all statements in the target cluster set are embedded in the associated department responsibility statements.

6. The method according to claim 1, characterized in that The determining of a first confidence matrix according to the event description statements of the pending event and the recall event, and at least one target department responsibility statement of the pending event, comprises: At least one target department responsibility statement corresponding to the event to be processed is adjusted to a responsibility description statement corresponding to the execution department in the recall event, and a first confidence matrix of the event description statement relative to the department responsibility statement is determined.

7. The method according to claim 6, characterized in that The step of determining a first confidence matrix of the event description statement relative to the department responsibility statement includes: Creating a first vector matrix based on the departmental responsibility statements, and determining a second vector matrix based on the event description statements; Based on the first vector matrix and the second vector matrix, determining a matching coefficient matrix of the event description statement relative to the target department responsibility statement; By normalizing the matching coefficient matrix columns, a first unidirectional attention matrix is ​​obtained; Determine the first confidence matrix based on the first unidirectional attention matrix.

8. The method according to claim 1, characterized in that: The method further comprises: Determine a sentence matching matrix based on the at least one target department responsibility statement and the multiple event description statements, and determine an opinion matching matrix based on the multiple event description statements and the handling opinions in the recall event, wherein the element values ​​in the sentence matching matrix are used to characterize the similarity between the description statement and the target department responsibility statement, and the element values ​​in the opinion matching matrix are used to characterize the matching degree between the description statement and the handling opinions; Determine a statement set according to the statement matching matrix and the opinion matching matrix; wherein the first statement in the statement set is a statement whose relevance to the target department's responsibility statement and handling opinion is lower than a preset threshold; Determining a second confidence level of the recalled event relative to the event to be processed based on the statement set, the recalled event, and the event description statement of the event to be processed; According to the second confidence, the first confidence of the recalled event relative to the event to be processed is updated to determine the at least one similar event based on the updated first confidence.

9. The method according to claim 8, characterized in that The step of determining a statement set according to the statement matching matrix and the opinion matching matrix includes: By normalizing the columns of the sentence matching matrix, a sentence unidirectional attention matrix of the event description sentence relative to the target department responsibility sentence is obtained; and by normalizing the rows of the sentence matching matrix, a responsibility unidirectional attention matrix is ​​obtained; By normalizing the columns of the opinion matching matrix, a first opinion one-way attention matrix of the event description sentence relative to the handling opinion is obtained, and by normalizing the rows of the opinion matching matrix, a second opinion one-way attention matrix is ​​obtained; Determining a second sentence from the event description sentence based on the sentence unidirectional attention matrix and the responsibility unidirectional attention matrix; and determining a third sentence from the event description sentence based on the first opinion unidirectional attention matrix and the second opinion unidirectional attention matrix; The statements in the event description statements except the second statement and the third statement are used as the first statement in the statement set.

10. The method according to claim 8, characterized in that The determining, based on the statement set, the recalled event, and the event description statement of the event to be processed, a second confidence level of the recalled event relative to the event to be processed includes: For each recall event, based on the handling opinion of the recall event, the event description statement of the pending event is updated, and at least part of the first statement in the statement set in the pending event is replaced with the event description statement in the recall event, to obtain an updated pending event; Determine an opinion matrix regarding handling opinions between the updated pending events and the recalled events; Based on the opinion matrix, a second confidence level of the updated event to be processed corresponding to the recalled event is determined.

11. The method according to claim 1 or 8, characterized in that: The method further comprises: For each recall event, determining a third confidence level between the event description statement of the event to be processed and the event description statement of the recall event; The first confidence level or the second confidence level is updated based on the third confidence level to determine the at least one similar event based on the updated first confidence level.

12. The method according to claim 11, characterized in that The method further comprises: A preset number of recalled events are determined in descending order according to the first confidence as similar events to the event to be processed.

13. The method according to claim 1, characterized in that The method further comprises: Based on the at least one similar event, a target processing result corresponding to the event to be processed is determined.

14. A device for determining similar events, characterized in that: include: A recall event determination module, used for acquiring a plurality of recall events associated with an event to be processed, wherein both the event to be processed and the recall event include a plurality of event description statements; A responsibility statement determination module is used to determine at least one target department responsibility statement corresponding to each event description statement in the event to be processed based on multiple event description statements of the event to be processed and a pre-generated content relationship graph; wherein the content relationship graph is generated based on historical work order events and at least part of the department responsibility statements of all execution departments associated with the historical work order events; A confidence determination module, configured to determine a first confidence of each recall event relative to the event to be processed based on event description statements of the event to be processed and the recall event, and at least one target department responsibility statement of the event to be processed; A similar event determination module is used to determine at least one similar event corresponding to the event to be processed based on the first confidence of each recall event and the recall event to which the target responsible department belongs; wherein the similar event is an event in the recall event.

15. An electronic device, characterized in that: The electronic device comprises: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method for determining similar events as described in any one of claims 1-13.

16. A storage medium comprising computer executable instructions, wherein the computer executable instructions are used to perform the method for determining similar events according to any one of claims 1 to 13 when executed by a computer processor.

17. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the computer program implements the method for determining similar events according to any one of claims 1-13.

Images