Task execution department determination method and device, equipment, medium and product

Abstract

Embodiments of the invention disclose a task execution department determination method and apparatus, a device, a medium and a product. The method comprises the steps of obtaining description text information of a to-be-processed work order; processing the input description text information and the target vector corresponding to each to-be-executed department based on a target work order processing model, and outputting a department evaluation attribute between each to-be-executed department and the to-be-processed work order; wherein the target vector comprises a department relation vector between a current to-be-executed department and other to-be-executed departments and a responsibility description vector of the current to-be-executed department; and determining a target execution department corresponding to the to-be-processed work order according to the department evaluation attribute. According to the technical scheme provided by the embodiment of the invention, the efficiency of determining the execution department corresponding to the work order to be processed is improved, and the accuracy of determining the execution department is also improved.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of data processing technology, and in particular to a method, device, equipment, medium and product for determining a task execution department. Background Art

[0002] In order to respond to various demands of users, some platforms can analyze and process the demands reported by users to determine the corresponding functional departments, so that the functional departments can perform the corresponding duties. The functional departments corresponding to the demands are mainly determined by the experience of the staff.

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

[0004] When determining the functional departments based on manual experience, not only does it require a high level of knowledge accumulation from the staff and have strong limitations, but also when some demands may involve multiple functional departments, it is impossible to accurately determine the functional departments, which in turn leads to the problem that the demands cannot be handled in a timely and effective manner. Summary of the invention

[0005] The embodiments of the present invention provide a method, device, equipment, medium and product for determining a task execution department, so as to realize intelligent determination of the execution department corresponding to the work order to be processed, which not only improves the accuracy of determining the execution department, but also improves the efficiency of determining the execution department.

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

[0007] Get the description text information of the pending work order;

[0008] Based on the target work order processing model, the input description text information and the target vector corresponding to each to-be-executed department are processed to output the department evaluation attribute between each to-be-executed department and the to-be-processed work order; wherein the target vector includes the department relationship vector between the current to-be-executed department and other to-be-executed departments and the responsibility description vector of the current to-be-executed department;

[0009] According to the department evaluation attribute, a target execution department corresponding to the work order to be processed is determined.

[0010] Optionally, the target work order processing model includes a disposal department recommendation sub-model, and the target work order processing model processes the input description text information and the target vector corresponding to each to-be-executed department, and outputs the department evaluation attribute between each to-be-executed department and the to-be-processed work order, including:

[0011] Determine an event description vector of the event description text in the description text information based on the disposal department recommendation sub-model, and weight the event description vector based on a predetermined attention weight to obtain a semantic vector to be spliced; wherein the attention weight includes a first weight and a second weight, the first weight is determined based on the work order title and the work order event description in the historical work order, and the second weight is determined based on the work order event category and the work order event description in the historical work order;

[0012] Based on the handling department recommendation sub-model, the semantic vector to be spliced ​​is spliced ​​with the target vector corresponding to each department to be executed respectively and then further processed to obtain the department evaluation attribute of the work order to be processed corresponding to each department to be executed.

[0013] Optionally, determining a target execution department corresponding to the to-be-processed work order according to the department evaluation attribute includes:

[0014] At least one target execution department is determined according to the department evaluation attribute corresponding to each to-be-executed department and a preset attribute threshold and / or a preset number of execution departments.

[0015] Optionally, the method further includes:

[0016] The work order processing model to be trained is trained based on historical training samples to obtain a trained work order processing model to be used; wherein the work order processing model to be trained includes a department association learning sub-model for determining the department relationship vector between the current department to be executed and other departments to be executed, a responsibility description sub-model for determining the responsibility description vector corresponding to each department to be executed, and a disposal department recommendation sub-model for processing input text, department relationship vector and responsibility description vector;

[0017] Based on the work order processing model to be used, the target work order processing model is determined.

[0018] Optionally, determining the target work order processing model based on the work order processing model to be used includes:

[0019] Obtaining the duty description vector corresponding to each department to be executed determined by the duty description sub-model in the work order processing model to be used, and the department relationship vector corresponding to each department to be executed determined by the department relevance learning sub-model;

[0020] Based on the department relationship vector and the responsibility description vector corresponding to the same department to be executed, determine the target vector corresponding to each department to be executed, encapsulate the target vector as the input parameter of the disposal department recommendation sub-model, and obtain the updated disposal department recommendation sub-model;

[0021] The updated disposal department recommendation sub-model is retained to obtain the target work order processing model.

[0022] Optionally, before training the work order processing model to be trained based on the historical training samples to obtain the trained work order processing model to be used, the method further includes:

[0023] Acquire a plurality of first historical work orders, wherein the first historical work orders include work order event categories;

[0024] Determine an undirected graph including all departments to be executed according to the work order event categories of the plurality of first historical work orders, wherein the undirected graph includes department identifiers of all departments to be executed and department edge weights between any two departments to be executed;

[0025] The historical training samples are determined based on each second historical work order, the undirected graph, and the responsibilities description text corresponding to each to-be-executed department.

[0026] Optionally, determining an undirected graph including all to-be-executed departments according to the work order event categories of the plurality of first historical work orders comprises:

[0027] For the first historical work orders of the same work order event category, the first edge weight between any two to-be-executed departments is determined according to the number of first historical work orders corresponding to each to-be-executed department and the total number of the first historical work orders;

[0028] Determine the department edge weight between any two departments to be executed according to the first edge weight between any two departments to be executed corresponding to all work order event categories;

[0029] According to the department to be executed and the department edge weights between any two departments to be executed, an undirected graph corresponding to all departments to be executed is determined.

[0030] Optionally, the training of the work order processing model to be trained based on the historical training samples to obtain the trained work order processing model to be used includes:

[0031] For each historical training sample, the second historical work order in the current training sample, the undirected graph, and the responsibilities description corresponding to each department to be executed are input into the work order processing model to be trained;

[0032] Based on the department association learning sub-model in the work order processing model to be trained, the undirected graph is processed to obtain the predicted department relationship vector corresponding to each department to be executed, the responsibility description vector of each department to be executed is vectorized based on the responsibility description sub-model to obtain the predicted description vector, and the predicted department relationship vector, the predicted description vector and the work order description text of the second historical work order of the same department to be executed are processed based on the disposal department recommendation sub-model, and the department prediction result of each department to be executed is output;

[0033] Based on the department prediction result, determine the predicted execution department, and determine the loss value according to the predicted execution department and the actual execution department of the second historical work order, so as to modify the model parameters in the work order processing model to be trained based on the loss value;

[0034] The convergence of the loss function in the work order processing model to be trained is taken as a training goal to obtain the work order processing model to be used.

[0035] Optionally, the department relevance learning sub-model is a graph neural network, and the undirected graph is processed based on the department relevance learning sub-model to obtain a predicted department relationship vector corresponding to each department to be executed, including:

[0036] Based on the grid nodes in the graph neural network, the departments to be executed and the department edge weights in the undirected graph are processed in sequence to obtain a first eigenvector;

[0037] The first feature vector is processed based on the graph convolution layer in the graph neural network to obtain a predicted department relationship vector corresponding to each to-be-executed department and other to-be-executed departments.

[0038] Optionally, the duty description includes at least one duty text; the duty description of each department to be executed is vectorized based on the duty description sub-model to obtain a prediction description vector, including:

[0039] Determine the semantic vector corresponding to at least one responsibility text corresponding to each department to be executed based on the responsibility description sub-model;

[0040] According to the preset text weight of the responsibility text of each department to be executed and the semantic vector corresponding to the responsibility text, the prediction description vector corresponding to the department to be executed is determined.

[0041] Optionally, the work order description text includes a work order title, a work order event description, and a work order event category. The disposal department recommendation sub-model processes the predicted department relationship vector, the predicted description vector, and the work order description text of the second historical work order of the same department to be executed, and outputs a department prediction result for each department to be executed, including:

[0042] Based on the processing of the work order title and the work order event description by the handling department recommendation sub-model, a first weight in the attention weight is updated; and based on the processing of the work order event category and the work order event description by the handling department recommendation sub-model, a second weight in the attention weight is updated;

[0043] Attention processing is performed on the work order event description based on the updated first weight and the updated second weight to obtain a predicted concatenated semantic vector after attention processing;

[0044] The predicted splicing semantic vector is spliced ​​with the predicted department relationship vector and the predicted description vector corresponding to each department to be executed, respectively, to obtain a to-be-processed vector corresponding to each department to be executed;

[0045] The vectors to be processed are classified and processed based on the disposal department recommendation sub-model, and a department prediction result of each department to be executed is output.

[0046] In a second aspect, an embodiment of the present invention further provides a device for determining a task execution department, the device comprising:

[0047] The information acquisition module is used to obtain the descriptive text information of the work order to be processed;

[0048] The attribute output module is used to process the input description text information and the target vector corresponding to each to-be-executed department based on the target work order processing model, and output the department evaluation attribute between each to-be-executed department and the to-be-processed work order; wherein the target vector includes the department relationship vector between the current to-be-executed department and other to-be-executed departments and the responsibility description vector of the current to-be-executed department;

[0049] The department determination module is used to determine the target execution department corresponding to the to-be-processed work order according to the department evaluation attribute.

[0050] In a third aspect, an embodiment of the present invention provides a device, the device comprising:

[0051] one or more processors;

[0052] A memory for storing one or more programs;

[0053] When the one or more programs are executed by the one or more processors, the one or more processors implement the method for determining the task execution department provided by any embodiment of the present invention.

[0054] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for determining a task execution department as provided in any embodiment of the present invention.

[0055] In a fifth aspect, an embodiment of the present invention further provides a computer program product, including a computer program, characterized in that when the computer program is executed by a processor, the computer program implements the method for determining a task execution department as described in any one of the embodiments of the present invention.

[0056] The technical solution provided by the embodiment of the present invention can input the descriptive text information into the target work order processing model after obtaining the descriptive text information of the work order to be processed, so that the target work order processing model analyzes and processes the descriptive text information and the department relationship vector including the relationship between departments and the responsibility description vector corresponding to the department responsibility, thereby determining the degree of fit between each department to be executed and the work order to be processed, and then determining the target execution department that can execute the work order to be processed from all departments to be executed based on the degree of fit, thereby solving the problem of low efficiency and low accuracy in the prior art when the execution department is determined by relying on user experience, and realizing the intelligent determination of the target execution department corresponding to the work order to be processed. Furthermore, the target work order processing model is obtained based on big data training, and therefore, the accuracy of determining the target execution department is also improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] 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.

[0058] Figure 1 A flowchart of a method for determining a task execution department provided by an embodiment of the present invention;

[0059] Figure 2 A flowchart of a method for determining a task execution department provided by an embodiment of the present invention;

[0060] Figure 3 A flowchart of a method for determining a task execution department provided by an embodiment of the present invention;

[0061] Figure 4 A relationship diagram between the department relevance learning sub-model, the responsibility description sub-model and the disposal department recommendation sub-model provided in an embodiment of the present invention;

[0062] Figure 5 A flowchart of a method for determining a task execution department provided by an embodiment of the present invention;

[0063] Figure 6 A schematic diagram of the structure of a device for determining a task execution department provided by an embodiment of the present invention;

[0064] Figure 7 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0065] 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.

[0066] 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.

[0067] Before introducing the technical solution provided by the embodiment of the present invention, an example description of the application scenario can be given first. The embodiment of the present invention can be applied in any scenario where it is necessary to determine the functional department corresponding to a certain event. For example, when a convenience hotline is received, the content of the appeal corresponding to the convenience hotline can be recorded, and a work order to be processed can be generated. Based on the technical solution provided by the embodiment of the present invention, the work order to be processed is analyzed and processed, and the corresponding target functional department can be determined. The target execution department is the target functional department. The functional department and the execution department have the same meaning, and both are government departments that handle a certain event or specific execution units under a certain department.

[0068] The corresponding application can be developed based on the technical solution provided in the embodiment of the present invention. The technical solution provided in the embodiment of the present invention can also be integrated into a functional module, which can be integrated into any application, that is, the technical solution provided in the embodiment of the present invention is displayed as a functional module in the display interface of the application. The user can directly edit the corresponding appeal content in the application, so that after clicking to confirm, the background corresponding to the application can receive the appeal content and generate a pending work order. Of course, the application or function can also be staff-oriented, and the staff can record the call content when receiving the convenience service hotline. During the call or when the call is completed, the call content is edited in the content editing interface corresponding to the function, and after detecting the confirmation control of clicking the editing completion, the background can generate a pending work order. Based on the solution provided in the embodiment of the present invention, the pending work order is analyzed and processed to determine the corresponding execution department. The above method achieves the effect of intelligently determining the target execution department.

[0069] Accordingly, the content recorded on the content editing page or during the call is used as the descriptive text information of the event.

[0070] Figure 1 A flowchart of a method for determining a task execution department provided in an embodiment of the present invention. This embodiment can be applied to any scenario where it is necessary to determine the task execution department corresponding to an event. The method for determining the task execution department provided in this embodiment can be executed by a device for determining the task execution department. The device can be implemented by software and / or hardware and integrated in an electronic device, which can be a mobile terminal or a PC, etc.

[0071] like Figure 1 As shown, the method specifically comprises the following steps:

[0072] S110. Obtain the descriptive text information of the work order to be processed.

[0073] Among them, the pending work order refers to the processing task generated by the event that needs to be further processed or resolved. The pending work order includes multiple event description statements. The event description statement refers to a statement 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 to record the ins and outs of the event in detail. The purpose 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. The collection of all the above event description statements can be used as descriptive text information. The method of generating the pending work order can be referred to the detailed description of the above application scenario, which will not be repeated here.

[0074] Specifically, the work order to be processed can be processed when it is generated in the background to obtain the description text information of the work order to be processed, or the background can be checked every preset time to see whether the work order to be processed is generated to obtain the description text information in the work order to be processed, or the background can be checked at a preset time every day to see whether the work order to be processed is generated to obtain its description text information when the work order to be processed is included. In practical applications, in order to improve the timeliness of processing the work order to be processed, the description text information can be obtained as soon as the generation of the work order to be processed is detected to determine the target execution department corresponding to the work order to be processed.

[0075] Exemplarily, the acquisition of pending work orders is instant, and users can directly upload the content of the appeal in the application. The content of the appeal can be: the tenant defaults on the rent from xx time to xx time, and the landlord refuses to pay the rent after multiple communications, and the house is cut off from power, and the tenant hopes to restore the power. After the appeal content is edited and confirmed, the background can generate a pending work order. Because it is an instant processing setting, the description text information of the pending work order can be obtained after the pending work order is generated. At this time, the description text information of the pending work order includes: the tenant defaults on the rent from xx time to xx time, and the landlord refuses to pay the rent after multiple communications, and the house is cut off from power, and the tenant hopes to restore the power.

[0076] S120 , processing the input description text information and the target vector corresponding to each to-be-executed department based on the target work order processing model, and outputting the department evaluation attribute between each to-be-executed department and the to-be-processed work order.

[0077] Among them, the target work order processing model is a pre-trained neural network model, which is used to analyze and process the input data to output the department evaluation attributes of each department to be executed relative to the work order to be processed. The department evaluation attribute is used to characterize the degree of fit between the department to be executed and the work order to be processed. Optionally, the higher the department evaluation attribute, the higher the degree of fit between the department to be executed and the work order to be processed. The target vector can be predetermined or generated when processing the work order to be processed. The target vector includes the department relationship vector between the current department to be executed and other departments to be executed and the responsibility description vector of the current department to be executed. The number of target vectors is adapted to the number of departments to be executed. For example, there are three departments to be executed, A, B and C, and department A has five responsibility statements. The target vector corresponding to department A includes the department relationship vector A1 of the relationship between department A and other departments B and C, and the responsibility description vector A2 composed of the responsibility vector of each responsibility statement in department A. For the responsibility department B and the responsibility department C, the meaning of the corresponding target vector is the same, which will not be repeated here.

[0078] Specifically, the descriptive text information may be input into the target work order processing model, and the target work order processing model may process the target vector of each department to be executed with the vector corresponding to the descriptive text information, and output the department evaluation attribute corresponding to each department to be executed.

[0079] For example, there are three departments to be executed, A, B and C. Department A to be executed includes five responsibilities, Department B to be executed has four responsibilities, and Department C to be executed has three responsibilities. The target vector corresponding to Department A to be executed includes A1 and A2, the target vector corresponding to Department B to be executed includes B1 and B2, and the target vector corresponding to Department C to be executed includes C1 and C2, wherein subscript 1 represents the department relationship vector of the relationship between departments, and subscript 2 represents the responsibility description vector composed of the responsibility statements of the departments to be executed. The target work order processing model can convert the input description text information into a text vector and splice it with each target vector respectively to obtain the final vector processed by the target work order processing model. The target work order processing model can output the department evaluation attributes of each department to be executed relative to the work order to be processed. That is, the department evaluation attribute A3 corresponding to Department A to be executed, the department evaluation attribute B3 corresponding to Department B to be executed, and the department evaluation attribute C3 corresponding to Department C to be executed are obtained.

[0080] S130. Determine the target execution department corresponding to the work order to be processed based on the department evaluation attributes.

[0081] The target execution department is the functional department that is finally determined from all the departments to be executed to execute the pending work order. That is, the functional department that is finally selected from the departments to be executed A, the departments to be executed B and the departments to be executed C to execute the pending work order is used as the target execution department.

[0082] Specifically, at least one target execution department can be determined based on the department evaluation attributes corresponding to each department to be executed. Optionally, the departments to be executed whose department evaluation attributes are higher than a preset evaluation attribute threshold can be used as target execution departments. Alternatively, a preset number of departments to be executed can be selected as target execution departments in order of department evaluation attributes from high to low.

[0083] The technical solution provided by the embodiment of the present invention can input the descriptive text information into the target work order processing model after obtaining the descriptive text information of the work order to be processed, so that the target work order processing model analyzes and processes the descriptive text information and the department relationship vector including the relationship between departments and the responsibility description vector corresponding to the department responsibility, thereby determining the degree of fit between each department to be executed and the work order to be processed, and then determining the target execution department that can execute the work order to be processed from all departments to be executed based on the degree of fit, thereby solving the problem of low efficiency and low accuracy in the prior art when the execution department is determined by relying on user experience, and realizing the intelligent determination of the target execution department corresponding to the work order to be processed. Furthermore, the target work order processing model is obtained based on big data training, and therefore, the accuracy of determining the target execution department is also improved.

[0084] Figure 2A flowchart of a method for determining a task execution department is provided in an embodiment of the present invention. Based on the above-mentioned embodiment, the target work order processing model may include a disposal department recommendation sub-model. The specific method for processing the descriptive text information and the target vector based on the disposal department recommendation sub-model can be found in the detailed description of this embodiment, wherein the technical terms that are the same as or corresponding to the above-mentioned embodiment are not repeated in this embodiment.

[0085] like Figure 2 As shown, the method includes:

[0086] S210: Obtain the descriptive text information of the work order to be processed.

[0087] S220, determining an event description vector of the event description text in the description text information based on the disposal department recommendation sub-model, and weighting the event description vector based on a predetermined attention weight to obtain a semantic vector to be spliced.

[0088] Among them, the disposal department recommendation sub-model can be a multi-label classification model. That is, the disposal department recommendation sub-model can analyze and process the event description vector and the target vector in the description text information, so as to determine the degree of fitness of the department to be executed and the corresponding department. The description text information includes multiple event description texts. The event description text can be any one of the title, content and event category. In other words, the event description text describes the overview of the event from different angles and levels. Therefore, the content in the event description text can be processed based on the attention fusion mechanism in the disposal department recommendation sub-model to obtain the semantic vector to be spliced ​​that takes into account different contents. The semantic vector to be spliced ​​is the vector obtained after processing each vector based on the attention weight.

[0089] It should be noted that the event description content, event category and event title in the event description text have different degrees of completeness corresponding to different aspects of the content. Correspondingly, the weights corresponding to different aspects of the content also vary to a certain extent. Therefore, the event description text in the descriptive text information can be processed based on the attention weight to obtain the semantic vector to be spliced ​​corresponding to the descriptive text information after comprehensive consideration of all aspects.

[0090] Specifically, the disposal department recommendation sub-model includes a first weight and a second weight, wherein the first weight is determined based on the work order title and the work order event description in the historical work order, and the second weight is determined based on the work order event category and the work order event description in the historical work order. Based on the first weight and the second weight, the descriptive text information in the work order to be processed can be processed to obtain a weighted semantic vector to be spliced.

[0091] It should also be noted that when training the disposal department recommendation sub-model, the following method can be used to determine the attention weight in the disposal department recommendation sub-model. That is to say, in order to understand the descriptive text information in the pending work order, that is, the content of the appeal uploaded by the user, it is necessary to enable the disposal department recommendation department to learn the semantic information corresponding to the appeal content in the descriptive text information. Both historical work orders and pending work orders include event titles, event description texts, and event categories. That is, the content in the descriptive text information describes the situation of the event from different angles and levels. Therefore, the work order appeal analysis module based on the attention fusion mechanism is used to fuse them and learn the semantic information of the event. Generally, since the information provided by the event description is the most complete among the above three aspects, the attention weight can be determined based on the event description.

[0092] Considering the above, we can use the event title and event category to learn the attention weight of each word in the event description text, so as to capture the keywords of the event description text. Here, we take the content analysis of one of the historical work orders as an example. The exemplary description can be found in the following description:

[0093] Vectorize the work order title and work order event description in the historical work order to obtain the semantic vector S of the work order title title , and the semantic vector S corresponding to the work order event description des , the attention weight is learned by determining the correlation between the work order title and the work order event description based on the following formula:

[0094]

[0095] Among them, W A and W B is a pre-parameter used to characterize the trainable matrix of the same size, S title is the semantic vector of the work order title, S des is the semantic vector describing the work order event, d k is the pre-set feature dimension, softmax is the activation layer function, A title,des The weight is learned for the correlation between the work order title and the work order event description, i.e., the first weight to be trained as indicated above.

[0096] At the same time, the work order event category and work order event description in the historical work order are vectorized to obtain the semantic vector S of the work order event category. category , and the semantic vector S corresponding to the work order event description des , the attention weight is learned by determining the correlation between the work order event category and the work order event description based on the following formula:

[0097]

[0098] Among them, W A and W B is a pre-parameter used to characterize the trainable matrix of the same size, S category is the semantic vector of the work order event category, S des is the semantic vector describing the work order event, d k is the pre-set feature dimension, softmax is the activation layer function, A category,des The weight is learned for the correlation between the work order event category and the work order event description, that is, the second weight to be trained as mentioned above.

[0099] The first weight to be trained and the second weight to be trained in the disposal department recommendation sub-model are trained based on historical work order events to obtain the first weight and the second weight that can be used.

[0100] On the basis of the existence of the first weight and the second weight, the semantic vector to be spliced ​​is obtained by performing attention processing on the description text information based on the attention weight, which can be:

[0101] B′ des =A category,des A title,des B des

[0102] Among them, B′ des is the semantic vector to be concatenated corresponding to the description text information of the work order to be processed, B des An event description vector that describes the event in the text information.

[0103] S230. Based on the disposal department recommendation sub-model, the semantic vector to be spliced ​​is spliced ​​with the target vector corresponding to each department to be executed, and then the processing is continued to obtain the department evaluation attribute of the work order to be processed corresponding to each department to be executed.

[0104] It can be understood that after obtaining the semantic vector to be spliced, the target vector corresponding to each department to be executed can be obtained. For each department to be executed, the target vector of the current department to be executed can be spliced ​​with the semantic vector to be spliced ​​together to obtain the vector to be reprocessed corresponding to each department to be executed. Furthermore, based on the disposal department recommendation sub-model, the vector to be reprocessed corresponding to each department to be executed is classified and processed to obtain the department evaluation attribute corresponding to the department to be executed.

[0105] Exemplarily, the process of concatenating the target vector corresponding to each department to be executed with the semantic vector to be concatenated may be:

[0106] X d =B′ des +B′ des Zd

[0107] Among them, X d Represents the vector to be processed again corresponding to the department to be executed, Z d is the target vector corresponding to the department to be executed, and d represents a department to be executed.

[0108] After obtaining the vector to be processed again corresponding to the department to be executed, it can be classified and processed based on the fully connected network in the disposal department recommendation sub-model to obtain the department evaluation attribute between the work order to be processed and the department to be executed, where the department evaluation attribute y d The value range is y d ∈[0,1].y d The closer the value is to 1, the higher the degree of fit between the department to be executed and the work order to be processed. Conversely, the lower the degree of fit between the department to be executed and the work order to be processed.

[0109] S240: Determine at least one target execution department according to the department evaluation attribute corresponding to each to-be-executed department and a preset attribute threshold and / or a preset number of execution departments.

[0110] The preset attribute threshold is a preset threshold range, which is used to screen the departments to be executed that meet the conditions. The preset number of execution departments is a preset constraint condition used to constrain and determine several target execution departments.

[0111] Specifically, the departments to be executed whose department evaluation attributes are higher than the preset attribute threshold are selected as target execution departments; or, the departments to be executed whose number is consistent with the preset execution department number are selected from high to low according to the department evaluation attributes as target execution departments; or, the departments to be executed whose department evaluation attributes are higher than the preset attribute threshold and whose number is less than the preset execution department number are selected as target execution departments.

[0112] For example, after obtaining the department evaluation attribute y between the work orders to be processed and the departments to be executed d Afterwards, the department evaluation attribute can be compared with a predefined threshold to determine the target execution department that needs to be involved in the process of handling the pending work order.

[0113] After obtaining at least one target execution department, at least one target execution department and prompt information requiring collaboration among multiple departments may be displayed on the display interface.

[0114] The technical solution provided by the embodiment of the present invention can perform attention processing on the event description, event title and event category in the description text information based on the predetermined attention weight after the disposal department recommendation sub-model receives the descriptive text information, so as to obtain the weighted semantic vector to be spliced, that is, considering the impact of different contents in the event description text on the department evaluation attributes, so as to obtain a more focused semantic vector to be spliced, and improve the accuracy of the determination of the semantic vector to be spliced. Furthermore, based on the disposal department recommendation sub-model, the semantic vector to be spliced ​​and the target vector are further processed to obtain the department evaluation attributes corresponding to each department to be executed, so as to achieve the effect of improving the accuracy of the department evaluation attributes under the premise that the semantic vector to be spliced ​​is more accurate, and further improve the adaptability between the target execution department and the work order to be processed.

[0115] On the basis of the foregoing embodiment, before processing the work order to be processed based on the target work order processing model, it is necessary to first obtain the training samples of the target work order processing model, and then train the target work order processing model based on the training samples, wherein the method for determining the training samples can be referred to the detailed description of this embodiment, wherein the technical terms that are the same or corresponding to the above embodiment are not repeated in this embodiment.

[0116] Step 1: Obtain multiple first historical work orders, wherein the first historical work orders include work order event categories.

[0117] Specifically, the work orders generated and processed before the current moment can be regarded as historical work orders. Each historical work order will be marked with an event category. The event category is used to characterize the related fields of the historical work order. Optionally, the event category can be road traffic, public facilities, welfare security, etc. The corresponding department to be executed can be determined based on the event category. Therefore, the execution departments that have handled the same event category in historical events have certain correlations in their departmental responsibilities. For any two departments to be executed, the more events of the same event category they handle, the stronger the correlation between the two departments to be executed. The co-occurrence frequency of handling similar events can be used as an indicator of responsibility correlation.

[0118] Specifically, a certain amount of historical work orders can be obtained from all historical work orders generated and processed before the current moment, and used as the first historical work order. At this time, the first historical work order includes the work order event category.

[0119] Step 2: Determine an undirected graph including all to-be-executed departments according to the work order event categories of the plurality of first historical work orders.

[0120] The undirected graph includes the department identifiers of all departments to be executed and the department edge weights between any two departments to be executed. The department identifier corresponds to the department to be executed, and the department edge weight is used to characterize the co-occurrence frequency between two departments to be executed. Based on the above, it can be seen that the co-occurrence frequency is used to characterize the frequency of two departments to be executed jointly handling the same event category.

[0121] In this embodiment, the method of determining the undirected graph based on the first historical work order can be: for the first historical work order of the same work order event category, the first edge weight between any two departments to be executed is determined according to the number of first historical work orders corresponding to each department to be executed and the total number of the first historical work orders; the department edge weight between any two departments to be executed is determined according to the first edge weights between all any two departments to be executed corresponding to all work order event categories; the undirected graph corresponding to all departments to be executed is determined according to the department edge weights between the department to be executed and any two departments to be executed.

[0122] It can be understood as follows: first, according to the preset event category and the event categories of all first historical work orders, the first historical work order under each event category is determined. Next, the first historical work order under each event category is analyzed and processed to obtain an undirected graph.

[0123] It should be noted that the processing method for the first historical work order under each event category is the same, and the processing of one event category is taken as an example for explanation.

[0124] Specifically, the number of first historical work orders corresponding to each department to be executed under the current event category and the total number of all first historical work orders under the current event category are counted. For any two departments to be executed, the first edge weight between any two departments to be executed can be determined based on the number and total number of first historical work orders of each department to be executed under the current event category. In other words, the first edge weight is the co-occurrence frequency between the two departments to be executed under the current event category. Next, the first edge weights between the same two departments to be executed in all event categories are weighted to obtain the department edge weight between the two departments to be executed. In other words, the department edge weight is the co-occurrence frequency between the two departments to be executed after comprehensively considering all event categories. In this embodiment, the undirected graph intuitively shows the association relationship between the departments to be executed.

[0125] Next, we will use a specific example to explain how to determine an undirected graph. In order to quantify the relevance of responsibilities between departments to be executed, we can create an undirected graph G = {D, E} including all departments to be executed, where the node set D is the department ID of the department to be executed, and the edge set E is the event co-occurrence frequency between departments to be executed, that is, the department edge weight. The edge set E is determined in the following way:

[0126]

[0127] Among them, M c,i The number of events in which the first historical work order of event category c is assigned to the department to be executed, M c,j The number of events that represent the first historical work order of event category c assigned to the department j to be executed. Category c includes multiple categories from c1 to cn. M c is the total number of events of category c, e i,j It is the department edge weight between department node i and department node j.

[0128] Based on the above method, the departmental association relationship between the departments to be executed can be determined, thereby realizing the inference of the potential association between departments in the absence of clear cross-departmental event labels.

[0129] Optionally, for two event categories c1 and c2, the department edge weight between pending execution department i and pending execution department j is

[0130]

[0131] Among them, M c1,i The number of events in which the first historical work order of event category c1 is assigned to the department to be executed, M c1,j The number of events in which the first historical work order of event category c1 is assigned to the department to be executed, M c1 is the total number of events of category c1, M c2,i The number of events in which the first historical work order of event category c2 is assigned to the department to be executed, M c2,j The number of events in which the first historical work order of event category c2 is assigned to the department to be executed, M c2 is the total number of events of category c2, e i,j It is the department edge weight between the department to be executed i and the department to be executed j.

[0132] Step three: determine historical training samples based on each second historical work order, the undirected graph, and the responsibilities description text corresponding to each department to be executed.

[0133] The second historical work order is also at least a part of all the historical work orders. The first historical work order and the second historical work order can be the same or different, as long as they can be used.

[0134] Specifically, corresponding training samples may be generated based on each second historical work order, the undirected graph, and the responsibilities description statement of each department to be executed.

[0135] It should be noted that in order to improve the accuracy of the target work order processing model, as many and rich training samples as possible can be obtained.

[0136] It should also be noted that for each department to be executed, there are corresponding regulations, which may include the responsibilities of the department, and its responsibilities are mainly presented through job description statements.

[0137] The technical solution provided by the embodiment of the present invention can process the acquired first historical work order to obtain an undirected graph for training the participating department association learning sub-model, so as to obtain the department relationship vector between each department to be executed. At the same time, based on the second historical work order, the undirected graph and the responsibility description of each department to be executed, the training samples can be determined, thereby facilitating the subsequent training of the responsibility description sub-model and the handling department recommendation sub-model.

[0138] Figure 3 A flowchart of a method for determining a task execution department is provided in an embodiment of the present invention. Based on the above-mentioned embodiment, a work order processing model to be trained can be trained based on training samples. The specific implementation method can refer to the detailed description of this embodiment, wherein the technical terms that are the same as or corresponding to the above-mentioned embodiment are not repeated in this embodiment.

[0139] like Figure 3 As shown, the method includes:

[0140] S310: For each historical training sample, the second historical work order, the undirected graph, and the responsibility description corresponding to each department to be executed in the current training sample are input into the work order processing model to be trained.

[0141] The work order processing model to be trained is a model with default parameters. The work order processing model to be trained includes a department relevance learning sub-model, a responsibility description sub-model, and a disposal department recommendation sub-model.

[0142] S320, based on the department association learning sub-model in the work order processing model to be trained, the undirected graph is processed to obtain the predicted department relationship vector corresponding to each department to be executed, the responsibility description vector of each department to be executed is vectorized based on the responsibility description sub-model to obtain the predicted description vector, and the predicted department relationship vector, predicted description vector and the work order description text of the second historical work order of the same department to be executed are processed based on the disposal department recommendation sub-model, and the department prediction result of each department to be executed is output. Figure 4 To understand the work order processing model to be trained provided by the embodiment of the present invention.

[0143] The department relevance learning sub-model can be a graph convolutional network, which iteratively performs convolution operations on the nodes in the graph convolutional network to capture the relevance between the departments to be executed. In this process, each node of the department to be executed will be assigned a coding representation h d, this convolution process is completed by weighted average of all neighboring node representations:

[0144]

[0145] Among them, the initial w is e i,j , w p-1 is the departmental edge weight of the adjacent node, h i is the encoding representation of adjacent nodes, N is the number of adjacent departments to be executed, and σ represents the nonlinear activation function ReLU. The graph convolution output h of the last layer d As a representation of the department d ,These node vector representation information not only contains the characteristics of the department to be executed itself, but also incorporates the relationship information with other departments to be executed.

[0146] It can be understood that the training of the above graph neural network is based on the department identification of the department to be executed in the undirected graph, the department edge weight between any two departments to be executed, and the number of other departments to be executed associated with the department to be executed.

[0147] That is to say, based on the grid nodes in the graph neural network, the departments to be executed and the department edge weights in the undirected graph are processed in turn to obtain the first eigenvector; based on the graph convolution layer in the graph neural network, the first eigenvector is processed to obtain the predicted department relationship vector corresponding to each department to be executed and other departments to be executed.

[0148] In the training phase, the vector output by the department relevance learning sub-model is used as the predicted department relationship vector.

[0149] Usually, the responsibility text stipulated in government documents is the direct source of the responsibilities of government departments. In government documents, the responsibilities of each department are divided into several items, each of which is a relatively independent responsibility and is related to each other. In order to extract the core responsibilities and characteristics of each department, the semantics of a single responsibility can be learned first to obtain a responsibility description sub-model that can process all responsibility statements. Each responsibility statement is a responsibility description. The responsibility description sub-model is mainly a model that converts responsibility statements into semantic vectors. Therefore, the responsibility description sub-model can be but is not limited to the BERT network. During the training phase, the semantic vector produced by the responsibility description sub-model is a predicted description vector.

[0150] In this embodiment, the responsibility description includes at least one responsibility text; based on the responsibility description sub-model, the responsibility description of each department to be executed is vectorized to obtain a predicted description vector, including: determining the semantic vector corresponding to at least one responsibility text corresponding to each department to be executed based on the responsibility description sub-model; determining the predicted description vector corresponding to the department to be executed according to the pre-set text weight of the responsibility text of each department to be executed and the semantic vector corresponding to the responsibility text.

[0151] It can be understood that the duty description includes at least one duty text. When there are multiple duty texts, the duty description of each department to be executed can be vectorized based on the duty description sub-model to obtain the semantic vector of each duty text. According to the pre-set text weight corresponding to each duty text, all semantic vectors of the department to be executed are weighted to obtain the predicted description vector corresponding to the department to be executed.

[0152] For example, the responsibility description sub-model is a BERT network. Each responsibility text of the department to be executed d is converted into a semantic vector r d,i Considering the different importance of different responsibilities in actual work, different text weights can be assigned to different responsibility statements to obtain the overall responsibility representation of the department to be executed:

[0153]

[0154] Among them, R d is the responsibility description vector of the department to be executed, w d,i represents the weight corresponding to the i-th duty description in the department to be executed, r d,i Represents the semantic vector corresponding to the i-th duty description in the department to be executed d.

[0155] The above prediction description vector and prediction department relationship vector can be concatenated to obtain a department representation vector that integrates the internal correlation and specified responsibility information between departments:

[0156] Z d =[R d ,H d ]

[0157] Among them, Z d is the department representation vector corresponding to the department to be executed, R d is the responsibility description vector of the department to be executed, H d is the department relationship vector of the department to be executed d relative to other departments to be executed.

[0158] It should be noted that the above department relationship vector and responsibility description vector can be a simple vector concatenation, or the weights between different vectors can be dynamically adjusted based on the attention mechanism to reflect the relative importance of each vector in a specific demand. In this way, not only can we get a vector representation of the basic responsibilities of the department, but also a vector that reflects the interaction and collaboration between departments.

[0159] It should be noted that the representation of the departmental responsibilities of the departments to be executed (government departments) needs to capture and express the internal connections and external characteristics of government departments when performing public functions. The internal connections are reflected by learning the correlation between departments from the execution departments corresponding to the historical work order events; the external characteristics are learned from the responsibility texts stipulated in government documents. The above method combines structured government document regulations with historical work order events to enhance the understanding and expression of government department responsibilities.

[0160] The disposal department recommendation sub-model can process the above-mentioned prediction description vector, the prediction department relationship vector and the work order description text in the second historical work order, so as to output the department prediction result corresponding to the department to be executed. At this time, the department prediction result mainly includes the degree of adaptation between the predicted department to be executed and the second historical work order.

[0161] In this embodiment, the predicted department relationship vector, predicted description vector and work order description text of the same department to be executed are processed based on the disposal department recommendation sub-model, and the department prediction result corresponding to the department to be executed is output, which can be: based on the disposal department recommendation sub-model, the work order title and work order event description are processed to update the first weight in the attention weight, and based on the disposal department recommendation sub-model, the work order event category and work order event description are processed to update the second weight in the attention weight; based on the updated first weight and the updated second weight, the work order event description is processed for attention processing to obtain the predicted splicing semantic vector after attention processing; the predicted splicing semantic vector is spliced ​​with the predicted department relationship vector and the predicted description vector corresponding to each department to be executed, respectively, to obtain the to-be-processed vector corresponding to each department to be executed; based on the disposal department recommendation sub-model, the to-be-processed vector is classified and processed, and the department prediction result of each department to be executed is output.

[0162] Among them, the work order description text includes the work order title, the work order event description and the work order event category. The disposal department recommendation submodel can be a work order title and a work order event description vector in the second historical work order, and obtain a work order title vector and a work order description vector. Based on the work order title vector, the work order description vector and the predefined attention learning function, the first weight in the attention weight is determined, that is, the first weight is mainly determined based on the work order title and the work order event description. The work order event category and the work order event description are processed in the same way to update the second weight in the attention weight. Based on the updated first and second weights, the work order event description is processed to obtain a predicted splicing semantic vector after attention processing. The predicted splicing semantic vector is spliced ​​with the predicted department relationship vector and the predicted description vector of each department to be executed, respectively, to obtain a to-be-processed vector corresponding to each department to be executed. Since the disposal department recommendation submodel is a classification model, the to-be-processed vector can be classified and processed based on the disposal department recommendation submodel to obtain the department prediction result of the department to be executed.

[0163] S330. Determine the predicted execution department based on the department prediction result, and determine the loss value according to the predicted execution department and the actual execution department of the second historical work order, so as to correct the model parameters in the work order processing model to be trained based on the loss value.

[0164] The department prediction result is used to characterize the degree of adaptation between the second historical work order and the department to be executed. The department to be executed determined based on the department prediction result is used as the predicted execution department. Optionally, the two departments to be executed with the highest department prediction results are used as the predicted execution departments. The second historical work order includes the actual execution department. The loss value can be determined based on the actual execution department and the predicted execution department, so as to modify the model parameters in the work order processing model to be trained based on the loss value.

[0165] S340. Taking the convergence of the loss function in the work order processing model to be trained as a training target, and obtaining the work order processing model to be used.

[0166] The loss function can be a function between the model prediction result and the actual result, which is used to quantify the difference between the model prediction. The main purpose of the loss function is to minimize the loss. Convergence can be that during the model training process, the value of the loss function gradually stabilizes and tends to a value. Convergence indicates that the model is optimizing its parameters. The model obtained when the loss function converges is used as the work order processing model to be used.

[0167] The solution provided by the embodiment of the present invention can train the work order processing model to be trained based on the above-mentioned method to obtain the work order processing model to be used that integrates department association relationships and department responsibility descriptions, so that when the work order to be processed is subsequently processed based on the work order processing model to be used, the target execution department that is compatible with it can be accurately determined.

[0168] Figure 5 A flowchart of a method for determining a task execution department is provided in an embodiment of the present invention. Based on the above-mentioned embodiment, after the work order processing model to be used is obtained, the target work order processing model can be determined based on the work order processing model to be used. The specific implementation method can refer to the detailed description of this embodiment, wherein the technical terms that are the same as or corresponding to the above-mentioned embodiment are not repeated in this embodiment.

[0169] S510: Train the work order processing model to be trained based on historical training samples to obtain a trained work order processing model to be used.

[0170] It should be noted that the method of training the to-be-trained work order processing model based on historical training samples can be referred to the detailed description of the above embodiment, which will not be repeated in this embodiment.

[0171] S520: Determine a target work order processing model based on the work order processing model to be used.

[0172] It can be understood that the target work order processing model can include a responsibility description sub-model, a department relevance learning sub-model and a disposal department recommendation sub-model. Alternatively, in order to simplify the model structure, the target work order processing model can only include a disposal department recommendation sub-model.

[0173] If the target work order processing model only includes the disposal department recommendation sub-model, determining the target work order processing model based on the work order processing model to be used can be: obtaining the responsibility description vector corresponding to each to-be-executed department determined by the responsibility description sub-model in the work order processing model to be used, and the department relationship vector corresponding to each to-be-executed department determined by the department association learning sub-model; determining the target vector corresponding to each to-be-executed department based on the department relationship vector and responsibility description vector corresponding to the same to-be-executed department, encapsulating the target vector as the input parameter of the disposal department recommendation sub-model to obtain the updated disposal department recommendation sub-model; retaining the updated disposal department recommendation sub-model to obtain the target work order processing model.

[0174] It can be understood as obtaining the department relationship vector of each department to be executed determined by the department association learning sub-model in the work order processing model to be used, and at the same time, obtaining the responsibility description vector of each department to be executed determined by the responsibility description sub-model in the work order processing model to be used. The department relationship vector and responsibility description vector of the same department to be executed are concatenated to obtain the target vector corresponding to the department to be executed. The target vector corresponding to each department to be executed is bound as the input parameter of the disposal department recommendation sub-model to obtain the updated disposal department recommendation sub-model. The department relationship learning sub-model and responsibility description sub-model in the work order processing model to be used are removed to obtain the target work order processing model that only includes the updated disposal department recommendation sub-model.

[0175] It should be noted that if the target work order processing model includes three sub-models, you can only obtain the department relationship vector output by the department relationship learning sub-model and the responsibility description vector output by the responsibility description sub-model, and repeat the above steps to obtain the department evaluation attributes corresponding to each department to be executed.

[0176] The solution provided by the embodiment of the present invention can simplify the model structure of the target work order processing model, so that it can be deployed on any application or server that needs to determine the target execution department, so as to achieve the effect of strong versatility in the deployment of the target work order processing model.

[0177] The following is an embodiment of a device for determining a task execution department provided in an embodiment of the present invention. The device and the method for determining a task execution department in the above-mentioned embodiments belong to the same inventive concept. For details not fully described in the embodiment of the device for determining a task execution department, reference can be made to the embodiment of the method for determining a task execution department in the above-mentioned embodiment.

[0178] Figure 6 A schematic diagram of the structure of a device for determining a task execution department provided in an embodiment of the present invention, the device specifically includes: an information acquisition module 610, an attribute output module 620, and a department determination module 630. The information acquisition module 610 is used to obtain the descriptive text information of the work order to be processed; the attribute output module 620 is used to process the input descriptive text information and the target vector corresponding to each department to be executed based on the target work order processing model, and output the department evaluation attribute between each department to be executed and the work order to be processed; wherein the target vector includes the department relationship vector between the current department to be executed and other departments to be executed and the responsibility description vector of the current department to be executed; the department determination module 630 is used to determine the target execution department corresponding to the work order to be processed based on the department evaluation attribute.

[0179] The technical solution provided by the embodiment of the present invention can input the descriptive text information into the target work order processing model after obtaining the descriptive text information of the work order to be processed, so that the target work order processing model analyzes and processes the descriptive text information and the department relationship vector including the relationship between departments and the responsibility description vector corresponding to the department responsibility, thereby determining the degree of fit between each department to be executed and the work order to be processed, and then determining the target execution department that can execute the work order to be processed from all departments to be executed based on the degree of fit, thereby solving the problem of low efficiency and low accuracy in the prior art when the execution department is determined by relying on user experience, and realizing the intelligent determination of the target execution department corresponding to the work order to be processed. Furthermore, the target work order processing model is obtained based on big data training, and therefore, the accuracy of determining the target execution department is also improved.

[0180] On the basis of the above technical solution, the attribute output module 620 also includes: a semantic vector acquisition unit, which is used to determine the event description vector of the event description text in the descriptive text information based on the handling department recommendation sub-model, and weight the event description vector based on a predetermined attention weight to obtain a semantic vector to be spliced; wherein the attention weight includes a first weight and a second weight, the first weight is determined based on the work order title and work order event description in the historical work order, and the second weight is determined based on the work order event category and work order event description in the historical work order; an attribute acquisition unit, which is used to splice the semantic vector to be spliced ​​with the target vector corresponding to each department to be executed based on the handling department recommendation sub-model and then continue processing to obtain the department evaluation attribute of the work order to be processed corresponding to each department to be executed.

[0181] On the basis of the above technical solutions, the department determination module 630 is further used to determine at least one target execution department according to the department evaluation attribute corresponding to each to-be-executed department and the preset attribute threshold and / or the preset number of execution departments.

[0182] On the basis of the above-mentioned technical solutions, the device also includes: a model acquisition module, which is used to train the work order processing model to be trained based on historical training samples to obtain a trained work order processing model to be used; wherein the work order processing model to be trained includes a department association learning sub-model for determining the department relationship vector between the current department to be executed and other departments to be executed, a responsibility description sub-model for determining the responsibility description vector corresponding to each department to be executed, and a disposal department recommendation sub-model for processing input text, department relationship vector and responsibility description vector; a model determination module, which is used to determine the target work order processing model based on the work order processing model to be used.

[0183] On the basis of the above-mentioned technical solutions, the model determination module includes: a vector acquisition unit, which is used to obtain the responsibility description vector corresponding to each department to be executed determined by the responsibility description submodel in the work order processing model to be used, and the department relationship vector corresponding to each department to be executed determined by the department association learning submodel; a submodel acquisition unit, which is used to determine the target vector corresponding to each department to be executed based on the department relationship vector and responsibility description vector corresponding to the same department to be executed, and encapsulate the target vector as the input parameter of the disposal department recommendation submodel to obtain the updated disposal department recommendation submodel; a target model acquisition unit, which is used to retain the updated disposal department recommendation submodel to obtain the target work order processing model.

[0184] On the basis of the above-mentioned technical solutions, the device also includes: a first historical work order acquisition module, used to acquire multiple first historical work orders, wherein the first historical work orders include work order event categories; an undirected graph determination module, used to determine an undirected graph including all departments to be executed according to the work order event categories of the multiple first historical work orders, wherein the undirected graph includes department identifiers of all departments to be executed and department edge weights between any two departments to be executed; a historical training sample determination module, used to determine the historical training samples based on each second historical work order, the undirected graph and the responsibility description text corresponding to each department to be executed.

[0185] On the basis of the above-mentioned technical solutions, the undirected graph determination module includes: a first edge weight determination unit, which is used to determine the first edge weight between any two departments to be executed for the first historical work order of the same work order event category according to the number of first historical work orders corresponding to each department to be executed and the total number of the first historical work orders; a department edge weight determination unit, which is used to determine the department edge weight between any two departments to be executed according to the first edge weights between all any two departments to be executed corresponding to all work order event categories; and an undirected graph determination unit, which is used to determine the undirected graph corresponding to all departments to be executed according to the department edge weights between the department to be executed and any two departments to be executed.

[0186] On the basis of the above-mentioned technical solutions, the model acquisition module includes: a sample input unit, which is used to input the second historical work order in the current training sample, the undirected graph and the responsibility description corresponding to each department to be executed into the work order processing model to be trained for each historical training sample; a result output unit, which is used to process the undirected graph based on the department association learning sub-model in the work order processing model to be trained to obtain the predicted department relationship vector corresponding to each department to be executed, to vectorize the responsibility description of each department to be executed based on the responsibility description sub-model to obtain the predicted description vector, and to recommend the disposal department based on the disposal department recommendation sub-model. The predicted department relationship vector, predicted description vector and work order description text of the same department to be executed are processed to output the department prediction results of each department to be executed; a loss value determination unit is used to determine the predicted execution department based on the department prediction results, and determine the loss value according to the predicted execution department and the actual execution department of the second historical work order, so as to correct the model parameters in the work order processing model to be trained based on the loss value; a model acquisition unit to be used is used to take the convergence of the loss function in the work order processing model to be trained as the training target to obtain the work order processing model to be used.

[0187] On the basis of the above technical solutions, the result output unit includes: a department relationship vector obtaining subunit, a description vector obtaining subunit and a result output subunit.

[0188] The department correlation learning submodel in the department relationship vector acquisition subunit is a graph neural network, which includes: a first feature vector acquisition subunit, which is used to process the departments to be executed and the department edge weights in the undirected graph in sequence based on the grid nodes in the graph neural network to obtain a first feature vector; a department relationship vector acquisition subunit, which is used to process the first feature vector based on the graph convolution layer in the graph neural network to obtain a predicted department relationship vector corresponding to each department to be executed and other departments to be executed.

[0189] The description vector obtains at least one responsibility text in the responsibility description subunit, and the subunit includes: a semantic vector determination subunit, which is used to determine the semantic vector corresponding to at least one responsibility text corresponding to each department to be executed based on the responsibility description submodel; a description vector determination subunit, which is used to determine the predicted description vector corresponding to the department to be executed based on the pre-set text weight of the responsibility text of each department to be executed and the semantic vector corresponding to the responsibility text.

[0190] The work order description text in the result output subunit includes the work order title, work order event description and work order event category, and the subunit includes: a weight updating subunit, which is used to process the work order title and the work order event description based on the disposal department recommendation submodel, and update the first weight in the attention weight, and process the work order event category and the work order event description based on the disposal department recommendation submodel, and update the second weight in the attention weight; a semantic vector acquisition subunit, which is used to process the work order event description based on the updated first weight and the updated second weight to obtain the predicted spliced ​​semantic vector after the attention processing; a to-be-processed vector acquisition subunit, which is used to splice the predicted spliced ​​semantic vector with the predicted department relationship vector and the predicted description vector corresponding to each to-be-executed department, respectively, to obtain the to-be-processed vector corresponding to each to-be-executed department; a result output subunit, which is used to classify the to-be-processed vector based on the disposal department recommendation submodel, and output the department prediction result of each to-be-executed department.

[0191] The device for determining a task execution department provided in the embodiment of the present invention can execute the method for determining a task execution department provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method for determining a task execution department.

[0192] It is worth noting that in the embodiment of the above-mentioned device for determining the task execution department, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned 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 scope of protection of the present invention.

[0193] Figure 7 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 12 suitable for use in implementing embodiments of the present invention is shown. Figure 7 The electronic device 12 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.

[0194] like Figure 7 As shown, the electronic device 12 is in the form of a general purpose computing device. The components of the electronic device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28, and a bus 18 that connects various system components (including the system memory 28 and the processing unit 16).

[0195] Bus 18 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 architectures. By way of example, these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.

[0196] The electronic device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the electronic device 12, including volatile and non-volatile media, removable and non-removable media.

[0197] The system memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The electronic device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 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 18 via one or more data medium interfaces. The system memory 28 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 various embodiments of the present invention.

[0198] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in system memory 28, such program modules 42 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. Program modules 42 generally perform the functions and / or methods of the embodiments described herein.

[0199] The electronic device 12 may also communicate with one or more external devices 14 (e.g., keyboards, pointing devices, displays 24, etc.), may communicate with one or more devices that enable a user to interact with the electronic device 12, and / or may communicate with any device that enables the electronic device 12 to communicate with one or more other computing devices (e.g., network cards, modems, etc.). Such communication may be performed via an input / output (I / O) interface 22. Furthermore, the electronic device 12 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 20. As shown, the network adapter 20 communicates with other modules of the electronic device 12 via a bus 18. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0200] The processing unit 16 executes various functional applications and data processing by running the program stored in the system memory 28, for example, implementing the steps of the method for determining the task execution department provided in any embodiment of the present invention, the method comprising:

[0201] Get the description text information of the pending work order;

[0202] Based on the target work order processing model, the input description text information and the target vector corresponding to each department to be executed are processed to output the department evaluation attribute between each department to be executed and the work order to be processed; wherein the target vector includes the department relationship vector between the current department to be executed and other departments to be executed and the responsibility description vector of the current department to be executed;

[0203] According to the department evaluation attribute, a target execution department corresponding to the work order to be processed is determined.

[0204] Of course, those skilled in the art can understand that the processor can also implement the technical solution of the method for determining the task execution department provided in any embodiment of the present invention.

[0205] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the method steps for determining a task execution department provided in the above embodiment of the present invention are implemented. The method includes:

[0206] Get the description text information of the pending work order;

[0207] Based on the target work order processing model, the input description text information and the target vector corresponding to each department to be executed are processed to output the department evaluation attribute between each department to be executed and the work order to be processed; wherein the target vector includes the department relationship vector between the current department to be executed and other departments to be executed and the responsibility description vector of the current department to be executed;

[0208] According to the department evaluation attribute, a target execution department corresponding to the work order to be processed is determined.

[0209] 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.

[0210] 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.

[0211] 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.

[0212] 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 a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or entirely on a remote computer or electronic device. 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).

[0213] 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 a task execution department, characterized in that: include: Get the description text information of the pending work order; Based on the target work order processing model, the input description text information and the target vector corresponding to each to-be-executed department are processed to output the department evaluation attribute between each to-be-executed department and the to-be-processed work order; wherein the target vector includes the department relationship vector between the current to-be-executed department and other to-be-executed departments and the responsibility description vector of the current to-be-executed department; According to the department evaluation attribute, a target execution department corresponding to the work order to be processed is determined.

2. The method according to claim 1, characterized in that: The target work order processing model includes a disposal department recommendation sub-model, and the target work order processing model processes the input description text information and the target vector corresponding to each to-be-executed department, and outputs the department evaluation attribute between each to-be-executed department and the to-be-processed work order, including: Determine an event description vector of the event description text in the description text information based on the disposal department recommendation sub-model, and weight the event description vector based on a predetermined attention weight to obtain a semantic vector to be spliced; wherein the attention weight includes a first weight and a second weight, the first weight is determined based on the work order title and the work order event description in the historical work order, and the second weight is determined based on the work order event category and the work order event description in the historical work order; Based on the handling department recommendation sub-model, the semantic vector to be spliced ​​is spliced ​​with the target vector corresponding to each department to be executed respectively, and then further processed to obtain the department evaluation attribute of the work order to be processed corresponding to each department to be executed.

3. The method according to claim 1, characterized in that Determining the target execution department corresponding to the to-be-processed work order according to the department evaluation attribute includes: At least one target execution department is determined according to the department evaluation attribute corresponding to each to-be-executed department and a preset attribute threshold and / or a preset number of execution departments.

4. The method according to claim 1, characterized in that The method further comprises: The work order processing model to be trained is trained based on historical training samples to obtain a trained work order processing model to be used; wherein the work order processing model to be trained includes a department association learning sub-model for determining the department relationship vector between the current department to be executed and other departments to be executed, a responsibility description sub-model for determining the responsibility description vector corresponding to each department to be executed, and a disposal department recommendation sub-model for processing input text, department relationship vector and responsibility description vector; Based on the work order processing model to be used, the target work order processing model is determined.

5. The method according to claim 4, characterized in that The determining the target work order processing model based on the work order processing model to be used includes: Obtaining the duty description vector corresponding to each department to be executed determined by the duty description submodel in the work order processing model to be used, and the department relationship vector corresponding to each department to be executed determined by the department relevance learning submodel; Based on the department relationship vector and the responsibility description vector corresponding to the same department to be executed, determine the target vector corresponding to each department to be executed, encapsulate the target vector as the input parameter of the disposal department recommendation sub-model, and obtain the updated disposal department recommendation sub-model; The updated disposal department recommendation sub-model is retained to obtain the target work order processing model.

6. The method according to claim 4, characterized in that Before training the work order processing model to be trained based on the historical training samples to obtain the trained work order processing model to be used, the method further includes: Acquire a plurality of first historical work orders, wherein the first historical work orders include work order event categories; According to the work order event categories of the plurality of first historical work orders, an undirected graph including all departments to be executed is determined, wherein the undirected graph includes department identifiers of all departments to be executed and department edge weights between any two departments to be executed; The historical training samples are determined based on each second historical work order, the undirected graph, and the responsibilities description text corresponding to each to-be-executed department.

7. The method according to claim 6, characterized in that The step of determining an undirected graph including all to-be-executed departments according to the work order event categories of the plurality of first historical work orders comprises: For the first historical work orders of the same work order event category, the first edge weight between any two to-be-executed departments is determined according to the number of first historical work orders corresponding to each to-be-executed department and the total number of the first historical work orders; Determine the department edge weight between any two departments to be executed according to the first edge weight between any two departments to be executed corresponding to all work order event categories; According to the department to be executed and the department edge weights between any two departments to be executed, an undirected graph corresponding to all departments to be executed is determined.

8. The method according to claim 6, characterized in that The training of the work order processing model to be trained based on the historical training samples to obtain the trained work order processing model to be used includes: For each historical training sample, the second historical work order in the current training sample, the undirected graph, and the responsibilities description corresponding to each department to be executed are input into the work order processing model to be trained; Based on the department association learning sub-model in the work order processing model to be trained, the undirected graph is processed to obtain the predicted department relationship vector corresponding to each department to be executed, the responsibility description vector of each department to be executed is vectorized based on the responsibility description sub-model to obtain the predicted description vector, and the predicted department relationship vector, the predicted description vector and the work order description text of the second historical work order of the same department to be executed are processed based on the disposal department recommendation sub-model, and the department prediction result of each department to be executed is output; Based on the department prediction result, determine the predicted execution department, and determine the loss value according to the predicted execution department and the actual execution department of the second historical work order, so as to modify the model parameters in the work order processing model to be trained based on the loss value; The convergence of the loss function in the work order processing model to be trained is taken as a training goal to obtain the work order processing model to be used.

9. The method according to claim 8, characterized in that The department relevance learning sub-model is a graph neural network. Based on the department relevance learning sub-model, the undirected graph is processed to obtain a predicted department relationship vector corresponding to each department to be executed, including: Based on the grid nodes in the graph neural network, the departments to be executed and the department edge weights in the undirected graph are processed in sequence to obtain a first eigenvector; The first feature vector is processed based on the graph convolution layer in the graph neural network to obtain a predicted department relationship vector corresponding to each to-be-executed department and other to-be-executed departments.

10. The method according to claim 8, characterized in that The responsibilities description comprises at least one responsibilities text; the responsibilities description of each department to be executed is vectorized based on the responsibilities description sub-model to obtain a prediction description vector, including: Determine the semantic vector corresponding to at least one responsibility text corresponding to each department to be executed based on the responsibility description sub-model; According to the preset text weight of the responsibility text of each department to be executed and the semantic vector corresponding to the responsibility text, the prediction description vector corresponding to the department to be executed is determined.

11. The method according to claim 8, characterized in that The work order description text includes a work order title, a work order event description, and a work order event category. The disposal department recommendation sub-model processes the predicted department relationship vector, the predicted description vector, and the work order description text of the same to-be-executed department, and outputs a department prediction result for each to-be-executed department, including: Based on the processing of the work order title and the work order event description by the handling department recommendation sub-model, a first weight in the attention weight is updated; and based on the processing of the work order event category and the work order event description by the handling department recommendation sub-model, a second weight in the attention weight is updated; Attention processing is performed on the work order event description based on the updated first weight and the updated second weight to obtain a predicted concatenated semantic vector after attention processing; The predicted splicing semantic vector is spliced ​​with the predicted department relationship vector and the predicted description vector corresponding to each department to be executed, respectively, to obtain a to-be-processed vector corresponding to each department to be executed; The vectors to be processed are classified and processed based on the disposal department recommendation sub-model, and a department prediction result of each department to be executed is output.

12. A device for determining a task execution department, characterized in that: include: The information acquisition module is used to obtain the descriptive text information of the work order to be processed; The attribute output module is used to process the input description text information and the target vector corresponding to each to-be-executed department based on the target work order processing model, and output the department evaluation attribute between each to-be-executed department and the to-be-processed work order; wherein the target vector includes the department relationship vector between the current to-be-executed department and other to-be-executed departments and the responsibility description vector of the current to-be-executed department; The department determination module is used to determine the target execution department corresponding to the to-be-processed work order according to the department evaluation attribute.

13. A device, characterized in that: The device comprises: one or more processors; A memory 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 a task execution department as described in any one of claims 1 to 11.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method for determining a task execution department as described in any one of claims 1 to 11 is implemented.

15. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the computer program implements the method for determining a task execution department as described in any one of claims 1 to 11.

Images