Task scheme determination method and device, electronic equipment, program product and medium

Abstract

The invention relates to the technical field of task allocation, and discloses a task scheme determination method and device, electronic equipment, a program product and a medium. The method comprises the steps of obtaining task demand information; constructing an initial task atlas according to the task demand information; the initial task atlas comprises a plurality of task nodes; deleting redundant nodes in the initial task atlas according to the task attributes of the task nodes to obtain a target task atlas; the task attribute is an index for evaluating the historical execution process of the task node; and extracting all task schemes in the target task atlas. Thus, by deleting the redundant nodes, redundant task items can be reduced under the condition that key links of the formed target task graph are not lost, and then a simplified task scheme is obtained under the condition that implementation of the whole scheme is not affected.

Description

Technical Field

[0001] The present application relates to the technical field of task allocation, and in particular to a method and device for determining a task plan, an electronic device, a program product, and a medium. Background Art

[0002] In the work process, it is inevitable that after a task requirement is proposed, a complete task plan needs to be designed according to the task requirement, and then the subtasks in the task plan are distributed to the corresponding users to facilitate the subsequent completion of the task requirement. At present, there is a problem of redundant task plans.

[0003] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present application, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0004] In order to provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. The summary is not an extensive review, nor is it intended to identify key / critical elements or delineate the scope of protection of these embodiments, but rather serves as a prelude to the detailed description that follows.

[0005] The embodiments of the present application provide a method and device for determining a task plan, an electronic device, a program product, and a medium, so as to provide a streamlined task plan without affecting the implementation of the overall plan.

[0006] An embodiment of the present application provides a method for determining a task plan, including: obtaining task requirement information; constructing an initial task map based on the task requirement information; the initial task map includes multiple task nodes; deleting redundant nodes in the initial task map based on the task attributes of each task node to obtain a target task map; the task attributes are indicators for evaluating the historical execution process of the task nodes; extracting all task plans in the target task map; wherein each path formed by the task nodes in the target task map is a task plan.

[0007] In the above implementation, considering that the task attributes of the task node can reflect the behavior of the task node in the past, the task attributes of the task node can be used to determine the unnecessary task nodes, that is, the redundant nodes. By deleting the redundant nodes, the target task map formed can reduce the redundant task items without losing the key links, and thus obtain a streamlined task plan without affecting the implementation of the overall plan.

[0008] Furthermore, redundant nodes in the initial task graph are deleted according to the task attributes of each of the task nodes to obtain a target task graph, including: determining the nodes to be deleted according to the task attributes, deleting the nodes to be deleted from the initial task graph to obtain an intermediate graph; determining whether necessary connection nodes are missing in the intermediate graph; and in the case of the lack of necessary connection nodes, completing the missing necessary connection nodes in the intermediate graph to obtain the target task graph.

[0009] In the above implementation, the probability that the task node is a redundant node can be inferred based on the historical execution status of the task node. By completing the necessary connection nodes, the probability of the task node from which the non-redundant nodes have been deleted can be reduced, thereby increasing the probability that the task plan can be implemented smoothly, thereby obtaining a streamlined task plan without affecting the implementation of the overall plan.

[0010] Furthermore, the task attributes include historical execution time; determining the nodes to be deleted based on the task attributes includes: obtaining the node out-degree of each of the task nodes; determining the to-be-judged score corresponding to each of the task nodes based on the historical execution time and node out-degree corresponding to each of the task nodes; if the to-be-judged score corresponding to the task node is lower than a preset score, determining the task node as a node to be deleted.

[0011] In the above implementation, it is considered that the node out-degree of the task node can reflect the degree of association between the task node and other task nodes. At the same time, the historical execution time of the task node can reflect the execution efficiency of the task node. Combining the node out-degree and historical execution time of the task node can comprehensively reflect the importance of the task node, and then it is convenient to delete redundant nodes to streamline the task plan.

[0012] Furthermore, determining whether necessary connection nodes are missing in the intermediate graph includes: obtaining a node completion model; the node completion model is trained using a preset historical scheme; the preset historical scheme has sample task nodes, necessary connection nodes of the sample task nodes, and node attributes of the sample task nodes; inputting each of the task nodes and the node attributes of each of the task nodes into the node completion model to obtain necessary connection nodes corresponding to each of the task nodes; determining whether the task node lacks necessary connection nodes based on the necessary connection nodes and actual connection nodes of the task node; the actual connection nodes are other task nodes connected to the task node in the target task graph.

[0013] In the above implementation, the historical scheme can reflect the connection relationship between each task node. By training the node completion model through the historical scheme, the completion model can capture the dependency relationship between different task nodes, so as to automatically identify the necessary connection nodes of each task node. And due to the generalization ability of the model itself, compared with other methods of identifying the necessary connection nodes of task nodes, the model can be more flexible.

[0014] Furthermore, each of the task nodes corresponds to a task item execution subject; the task item execution subject is used to execute the corresponding task node; the node attributes of each of the task nodes in the target task map include: time constraints. The method also includes: obtaining the estimated time corresponding to the task item execution subject of each of the task nodes in the target task map; the estimated time reflects the efficiency of the task item execution subject in processing the task node; determining the priority score of each of the task nodes according to the time constraints and the estimated time of each of the task nodes in the target task map; determining the priority level of each of the task solutions according to the priority score of each of the task nodes.

[0015] In the above implementation, the time constraint of the task node is taken into consideration, which can reflect the urgency of the task node, and the estimated time consumption reflects the efficiency of the task item execution subject in processing the task node. Therefore, through the time constraint of the task node and the estimated time consumption, it is convenient to arrange the priority level of the task scheme according to the urgency and processing speed, so as to give priority to the task scheme that can be processed quickly and is urgent.

[0016] Furthermore, the estimated time consumption corresponding to the task item execution subject of each task node in the target task map is obtained, including: inputting the task item execution subject of each task node in the target task map into a preset sorting model to obtain the efficiency ranking of each task item execution subject; and determining the estimated time consumption of each task item execution subject according to the efficiency ranking.

[0017] In the above implementation, considering that the model can reasonably infer the efficiency ranking of different task item execution entities through learning, the efficiency ranking of the task item execution entities can be obtained conveniently and accurately by inputting the task item execution entities of each task node in the target task map into a preset ranking model.

[0018] Furthermore, the sorting model is obtained in the following manner: obtaining first training data and scores and modification records for historical plans within a historical preset time; the modification records are used to describe the efficiency ranking of multiple task item execution entities after correction; the first training data is data located before the historical preset time for training the sorting model; obtaining second training data based on the scores and the modification records; training the preset model using the first training data and the second training data to obtain a reference sorting model; determining target model parameters based on first model parameters of the reference sorting model and second model parameters of the historical sorting model; adjusting the model parameters of the historical sorting model to target model parameters to obtain the sorting model.

[0019] In the above implementation, the modification record can reflect the efficiency ranking of the task item execution subject that meets the user's expectations, and the new ranking model can be trained by comprehensively modifying the record, which can improve the quality of the training data for training the new ranking model, thereby facilitating the improvement of the accuracy of the new ranking model. The target model parameters are determined by comprehensively referring to the first model parameters of the ranking model and the second model parameters of the historical ranking model, and a part of the historical data is retained, so that the parameter update of the ranking model will not produce abrupt changes, thereby improving the stability of the ranking model.

[0020] Furthermore, the target model parameters are determined according to the first model parameters of the reference sorting model and the second model parameters of the historical sorting model, including: obtaining the quality score of the historical solution; the quality score is used to reflect the logical correctness of the historical solution; determining the weight parameter according to the quality score; weighting the first model parameter and the second model parameter according to the weight parameter to obtain the target model parameter.

[0021] In the above implementation, the quality score of the historical scheme can reflect the quality of the historical training data used to train the sorting model. The weight parameters of the first model parameter and the second model parameter are determined in combination with the quality score. This can facilitate increasing the proportion of the second model parameter of the historical sorting model when the quality of the historical sorting model is good, so that the target model parameters obtained can make the sorting model more accurate.

[0022] Furthermore, the node attributes of each task node in the target task map include a preset node score; the task attributes include a historical execution success rate; the priority score of each task node is determined based on the estimated time consumption and the urgency coefficient, including: for each task node: the priority score of the task node is determined based on the estimated time consumption, the time constraint, the historical execution success rate and the preset node score corresponding to the task node.

[0023] In the above implementation, it is considered that the historical execution success rate of the task node can reflect the probability of whether the task node can be executed normally. By increasing the execution success rate of the task node to determine the priority score of the task node, the probability that the task plan cannot be executed due to a problem with a task node during execution can be reduced, thereby improving the stability of the task plan.

[0024] Furthermore, the method also includes: inputting the task plan into a preset risk prediction model to obtain the risk type corresponding to the task plan and the risk probability of the risk type; when the risk probability is greater than the preset probability, determining a risk management strategy according to the risk type, and adjusting the task plan according to the risk management strategy.

[0025] In the above implementation scheme, by predicting the risk type of the task plan and the risk probability of the risk type, it is possible to predict whether there are problems with the task plan, and then adjust the task plan in advance to improve the stability of the task plan.

[0026] Furthermore, the method also includes: when the risk probability is less than or equal to a preset probability, obtaining the execution result of the task plan; performing quality verification on the execution result; when the quality verification fails, determining a risk management strategy based on the risk type, and adjusting the task plan according to the risk management strategy.

[0027] In the above implementation scheme, by verifying the quality of the execution results of the task plan and adjusting the task plan, a task plan with qualified execution results can be obtained after continuous adjustment, so that the subsequent task plan formed based on the task requirement information is more reasonable.

[0028] Furthermore, the execution result is quality verified, including: determining the semantic similarity between the execution result and a preset historical execution result, and determining that the quality verification has failed if the semantic similarity is less than a preset similarity; the historical execution result and the execution result belong to the same task scenario; and / or determining whether the execution result meets preset indicators, and determining that the quality verification has failed if the execution result does not meet the preset indicators; the preset indicators are compliance conditions for the task scenario to which the execution result belongs.

[0029] In the above implementation scheme, whether the execution result is compliant is determined by comparing the semantic similarity between the execution result and the preset historical execution result, or by determining whether the execution result meets the preset indicators. The process is simple and relatively accurate, and can easily determine whether the execution result is compliant, thereby reflecting whether the task plan is logical, so as to improve the determination of the task plan.

[0030] Furthermore, an initial task graph is constructed based on the task requirement information, including: converting the task requirement information into structured data; the structured data includes task items; determining the dependency relationship between each of the task items; taking the task items as the task nodes and connecting them according to the dependency relationship to obtain the initial task graph.

[0031] In the above implementation scheme, considering that task requirement information may come from multiple sources, by converting the task requirement information into structured data, data from different sources can be unified, and there is no need to use different parsing methods for task requirement information from different sources, thereby more conveniently constructing a task map.

[0032] An embodiment of the present application provides a task plan determination device, including: an acquisition module, used to obtain task requirement information; a construction module, used to construct an initial task map according to the task requirement information; the initial task map includes multiple task nodes; an adjustment module, used to delete redundant nodes in the initial task map according to the task attributes of each task node, and obtain a target task map; the task attribute is an indicator for evaluating the historical execution process of the task node; a task determination module, used to extract all task plans in the target task map; wherein each path formed by the task nodes in the target task map is a task plan.

[0033] An embodiment of the present application provides an electronic device, including a processor and a memory, wherein the memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the above-mentioned task plan determination method.

[0034] An embodiment of the present application provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the above-mentioned task scheme determination method is implemented.

[0035] An embodiment of the present application provides a storage medium storing computer executable instructions. When the computer executable instructions are called and executed by a processor, the computer executable instructions enable the processor to perform the above-mentioned task scheme determination method.

[0036] The above general description and the following description are exemplary and explanatory only and are not intended to limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] One or more embodiments are exemplarily described by corresponding drawings, which do not limit the embodiments. Elements with the same reference numerals in the drawings are shown as similar elements, and the drawings do not constitute a scale limitation, and wherein: Figure 1It is a schematic diagram of a method for determining a task plan provided in an embodiment of the present application; Figure 2 It is a schematic diagram of a task scheme provided in an embodiment of the present application; Figure 3 is a flowchart of another method for determining a task solution provided in an embodiment of the present application; Figure 4 is a schematic diagram of a task solution determination device provided in an embodiment of the present application; Figure 5 It is a schematic diagram of an electronic device provided in an embodiment of the present application.

[0038] Reference numerals: 1: Acquisition module; 2: Construction module; 3: Adjustment module; 4: Task determination module; 5: Processor; 6: Bus; 7: Memory; 8: Communication interface. DETAILED DESCRIPTION

[0039] In order to be able to understand the features and technical contents of the embodiments of the present application in more detail, the implementation of the embodiments of the present application is described in detail below in conjunction with the accompanying drawings. The attached drawings are for reference only and are not used to limit the embodiments of the present application. In the following technical description, for the convenience of explanation, a full understanding of the disclosed embodiments is provided through multiple details. However, one or more embodiments can still be implemented without these details. In other cases, to simplify the drawings, well-known structures and devices can be simplified for display.

[0040] The terms "first", "second", etc. in the specification and claims of the embodiments of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the embodiments of the present application described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions.

[0041] Unless otherwise stated, the term "plurality" means two or more.

[0042] The term "correspondence" may refer to an association relationship or a binding relationship. The correspondence between A and B means that there is an association relationship or a binding relationship between A and B.

[0043] Embodiment 1 The present application provides a method for determining a task plan. Figure 1 As shown, Figure 1 A basic flow chart of the method for determining a task solution provided in an embodiment of the present application includes: Step S101, obtaining task requirement information.

[0044] In some embodiments, the task requirement information is used to characterize the task requirements that are expected to be achieved.

[0045] Exemplarily, the task requirement information may include: task items, such as reimbursement approval. The task requirement information may also include: time constraints. The time constraints are used to describe the expected completion time. The task requirement information may also include: expected resource requirements. The resource requirements are used to describe the resources needed to complete the task, such as the need for a financial system interface.

[0046] For example, assume that the task requirement information is that the reimbursement approval process needs to be completed today, including application submission, departmental preliminary review, financial review, system entry, and email notification to the applicant. Figure 2 As shown in the figure, a task plan can be obtained based on the task requirement information, which includes multiple task items, such as "submitting reimbursement application", "department head approval" and "financial system entry". At the same time, the time constraint is "today" and the resource requirement is "financial system interface".

[0047] Step S102, constructing an initial task map according to task requirement information.

[0048] Among them, the initial task map includes multiple task nodes.

[0049] Exemplarily, each task node reflects a task item, and each task node in the initial task map also carries node attributes, which may include: time constraints, preset node scores, and the like.

[0050] In some embodiments, step S102 may include: converting the task requirement information into structured data; the structured data includes task items; determining the dependency relationship between the task items; taking the task items as task nodes and connecting them according to the dependency relationship to obtain an initial task graph.

[0051] Exemplarily, the task requirement information is input into the preset BERT-Whisper (multimodal encoder) to convert the task requirement information into a 256-dimensional vector. The 256-dimensional vector is then converted into a set task triple to obtain structured data including task items. The task requirement information is then analyzed through dependency syntax to obtain the dependency relationship between the task items.

[0052] Among them, the task triplet can be G=(T, C, R). G is the task triplet, T is the task item, C is the time constraint, and R is the expected resource requirement. A 256-dimensional vector can be obtained through V=Transformer_Encoder([Text_Emb;Audio_Emb;Graph_Emb]). Among them, Text_Emb represents text embedding representation, Audio_Emb represents audio embedding representation, Graph_Emb represents graph embedding representation, and V represents a 256-dimensional vector. This formula describes the concatenation of the embedded representations of text, audio, and graph, and inputting them into Transformer_Encoder to generate a unified 256-dimensional vector V. In this way, by first converting the task requirement information into a 256-dimensional vector, the problem of multi-source data alignment can be solved, so that the information in the task requirement information can be better extracted.

[0053] For example, taking "product online promotion" as an example, the task requirement information can be described as: "This month, new product market promotion needs to be completed, including online advertising, offline event planning, social media operation, KOL (referring to the cooperation between enterprises or brands and online opinion leaders with great influence, professional knowledge or fan base in a certain field) and effect monitoring." The initial task map constructed may include the following task nodes: market research, budget allocation, online advertising design, offline event venue reservation, social media content production, KOL screening and signing, advertising execution, event execution, data collection and analysis, and effect report generation. The dependency relationship of each task node may be: market research points to budget allocation. Budget allocation points to online advertising design, offline event venue reservation, social media content production, KOL screening and signing. Among them, "online advertising design" points to "advertising execution". "Offline event venue reservation" points to "event execution". "Social media content production" points to "advertising execution" and "event execution". "KOL screening and signing" points to "advertising execution" and "event execution". "Advertising execution" points to "data collection and analysis". "Event execution" points to "data collection and analysis". "Data Collection and Analysis" points to "Performance Report Generation".

[0054] Step S103, deleting redundant nodes in the initial task graph according to the task attributes of each task node to obtain a target task graph.

[0055] Optionally, the task attribute is an indicator for evaluating the historical execution process of the task node, for example, historical execution time, that is, the time consumed by the task node in the historical execution process.

[0056] In some embodiments, step S103 may include: determining the node to be deleted according to the task attribute, deleting the node to be deleted from the initial task map, and obtaining an intermediate map. Determining whether the intermediate map lacks necessary connection nodes, and if necessary connection nodes are missing, completing the missing necessary connection nodes in the intermediate map to obtain the target task map.

[0057] In an optional manner of the above embodiment, the task attribute includes historical execution time. Determining the node to be deleted according to the task attribute may be: obtaining the node out-degree of each task node; determining the to-be-judged score corresponding to each task node according to the historical execution time and node out-degree corresponding to each task node; if the to-be-judged score corresponding to the task node is lower than a preset score, determining the task node as a node to be deleted.

[0058] Among them, the redundant node is a node to be deleted that is not a necessary connection node.

[0059] Among them, the node out-degree is the number of other task nodes connected to the task node.

[0060] In the above optional manner, according to the historical execution time and node out-degree corresponding to each task node, the to-be-judged score corresponding to each task node is determined, which may be: obtaining a first weight set. For each task node: weighting the historical execution time of the task node and the negative first power of the node out-degree of the task node according to the first weight set to obtain the to-be-judged score of the task node.

[0061] The first weight set records the first weight corresponding to the historical execution time and the second weight corresponding to the node out-degree.

[0062] Exemplarily, the score to be judged = the second weight × the node out-degree -1 + first weight × historical execution time.

[0063] Optionally, the first weight and the second weight may be optimized by a PPO (Proximal Policy Optimization) algorithm.

[0064] In an optional manner of the above embodiment, determining whether necessary connection nodes are missing in the intermediate graph includes: obtaining a node completion model. Inputting each task node and the node attributes of each task node into the node completion model to obtain necessary connection nodes corresponding to each task node. Determining whether the task node lacks necessary connection nodes based on the necessary connection nodes and actual connection nodes of the task node.

[0065] Among them, the actual connection nodes are other task nodes connected to the task node in the target task graph.

[0066] Among them, the node completion model is used to determine the necessary connection nodes that the task nodes need to connect to.

[0067] In the above optional method, the node completion model can be obtained by using the preset historical scheme training in the following way: the historical scheme with the necessary connection node labels of the sample task nodes is input into the preset first training model for training to obtain the node completion model.

[0068] Among them, the historical plan can have sample task nodes, necessary connection nodes of the sample task nodes, and node attributes of the sample task nodes.

[0069] Among them, the first training model can be a graph neural network (GNN).

[0070] In the above optional method, whether the task node lacks necessary connection nodes is determined based on the necessary connection nodes and actual connection nodes of the task node, including: for each task node: comparing whether the necessary connection nodes of the task node exist in the actual connection nodes of the task node; if not, the task node lacks necessary connection nodes; if so, the task node does not lack necessary connection nodes.

[0071] In another optional method of the above embodiment, determining whether necessary connection nodes are missing in the intermediate graph includes: for each task node in the intermediate graph: searching a preset connection node database for necessary connection nodes that correspond to both the task node and the node attributes of the task node.

[0072] The connection node database stores the corresponding relationships among the task nodes, the node attributes of the task nodes and the necessary connection nodes.

[0073] In some embodiments, each task node corresponds to a task item execution subject, and the task item execution subject is used to execute the corresponding task node. And the node attributes of each task node in the target task map include: time constraints. The task plan determination method also includes: obtaining the estimated time corresponding to the task item execution subject of each task node in the target task map; determining the priority score of each task node according to the time constraints and estimated time of each task node in the target task map; determining the priority level of each task plan according to the priority score of each task node.

[0074] Optionally, the estimated time consumption reflects the efficiency of the task item execution body in processing the task nodes.

[0075] In an optional manner of the above embodiment, obtaining the estimated time corresponding to the task item execution subject of each task node in the target task map may include: inputting the task item execution subject of each task node in the target task map into a preset sorting model to obtain the efficiency ranking of each task item execution subject; and determining the estimated time of each task item execution subject based on the efficiency ranking.

[0076] Among them, the ranking model is used to judge the efficiency of the task item execution subject of each task node, so as to output the ranking of the task item execution subject of each task node according to the efficiency.

[0077] In the above optional method, the ranking model can be obtained in the following way: a sample graph with a ranking label of the task item execution subject efficiency is input into a preset second training model for training to obtain a ranking model.

[0078] Among them, the second training model can be: a multi-armed bandit model.

[0079] In the above optional method, before the current moment, there will be a historical plan determined according to the historical task requirement information, as well as the historical execution results after executing the historical plan, and feedback information received from the user on the overall process from the formation of the historical plan to the execution of the historical plan. The historical sorting model can be adjusted according to the user's feedback information. That is, the sorting model can also be obtained in the following ways: obtain the first training data and the scoring and modification records for the historical plan within the historical preset time; obtain the second training data according to the scoring and modification records; use the first training data and the second training data to train the preset model to obtain a reference sorting model; determine the target model parameters according to the first model parameters of the reference sorting model and the second model parameters of the historical sorting model; adjust the model parameters of the historical sorting model to the target model parameters to obtain the sorting model.

[0080] The first training data is data used to train the sorting model before the historical preset time. The modification record is used to describe the efficiency ranking of the multiple task item execution entities after correction. The historical sorting model refers to the sorting model obtained before the current moment. Exemplarily, the sorting model obtained by the last update before the current moment can be used as the historical sorting model.

[0081] Optionally, obtaining the second training data according to the score and modification record may be: obtaining the efficiency ranking of each task item execution subject with a score higher than a preset score, and obtaining the efficiency ranking of each task item execution subject modified by the user as the efficiency ranking to be trained. The task item execution subjects corresponding to the efficiency ranking to be trained, the task nodes corresponding to each task item execution subject, and the node attributes of each task node are used as the second training data.

[0082] Before obtaining the second training data according to the scoring and modification records, the scoring and modification records may be processed using differential privacy technology to protect user privacy. The differential privacy noise may be set to 0.1.

[0083] Optionally, the preset model is trained using the first training data and the second training data to obtain a reference ranking model, which may be: obtaining the quality scores of historical solutions. As the loss function, the preset model is trained using the first training data and the second training data to obtain a reference sorting model. Wherein, L represents the loss function. In this way, by designing a special loss function, it is possible to force the model to give priority to reducing the error of low-quality tasks when optimizing.

[0084] Optionally, determining the target model parameters according to the first model parameters of the reference sorting model and the second model parameters of the historical sorting model can be: obtaining the quality score of the historical scheme; determining the weight parameter according to the quality score; weighting the first model parameter and the second model parameter according to the weight parameter to obtain the target model parameter.

[0085] The weight parameters include a sixth weight corresponding to the first model parameter and a seventh weight corresponding to the second model parameter.

[0086] Exemplarily, determining the weight parameter according to the quality score may be performed by searching a preset weight database for the weight parameter corresponding to the quality score. The weight database stores the weight parameter corresponding to the quality score.

[0087] Optionally, the quality score is used to reflect the logical correctness of the historical plan, and the quality score can be obtained in the following way: for each historical plan: determine the structural similarity between the task plan and the historical plan, obtain the semantic similarity and rule pass rate corresponding to the historical plan, weight the structural similarity, semantic similarity and rule pass rate according to a preset third weight set, and obtain the alternative score of the historical plan; calculate the average value of the alternative scores of each historical plan to obtain the quality score.

[0088] Among them, the third weight set records the eighth weight corresponding to the structural similarity, the ninth weight corresponding to the semantic similarity, and the tenth weight corresponding to the rule passing rate.

[0089] In one implementation, determining the structural similarity between the task solution and the historical solution may include: calculating a first similarity between a task node in the task solution and a task node in the historical solution as the structural similarity.

[0090] In the above implementation, determining the first similarity between the task nodes in the task solution and the task nodes in the historical solution may be: obtaining the equivalent number of the task nodes in the task solution and the historical solution, calculating the equivalent number and dividing it by the total number to obtain the first similarity. The total number is the number of task nodes in the task solution.

[0091] In another embodiment, determining the structural similarity between the task plan and the historical plan can be: determining a first similarity between the task nodes in the task plan and the task nodes in the historical plan; determining a second similarity between the dependency relationships of each task node in the task plan and the dependency relationships of each task node in the historical plan; and determining the structural similarity by combining the first similarity and the second similarity.

[0092] The dependency relationship of a task node refers to the pointing relationship between the task node and other task nodes. For example, the dependency relationship is: the task node "activity execution" points to "data collection and analysis".

[0093] In the above-mentioned embodiment, determining the second similarity between the dependency of each task node in the task scheme and the dependency of each task node in the historical scheme can be: converting the dependency of each task node in the task scheme into a first dependency vector. Converting the dependency of each task node in the historical scheme into a second dependency vector. For the first dependency vector of each task node in the task scheme: calculating the vector similarity between the first dependency vector and each second dependency vector in turn, if there is a vector similarity greater than a preset vector similarity, it is considered that the same dependency exists in the historical scheme. Accumulate the number of equivalent dependencies, calculate the number of equivalent dependencies divided by the total number of dependencies, and obtain the second similarity.

[0094] The first dependency vector represents a single dependency of a task node in the task solution.

[0095] The second dependency vector represents a single dependency of a task node in the historical solution.

[0096] The number of equivalent dependencies is the number of identical dependencies between the historical solution and the task solution. The total number of dependencies is the total number of dependencies in the task solution.

[0097] In the above implementation, the structural similarity is determined by combining the first similarity and the second similarity, which may be: weighting the first similarity and the second similarity according to a preset fourth weight set to obtain the structural similarity.

[0098] The fourth weight set records the eleventh weight corresponding to the first similarity and the twelfth weight corresponding to the second similarity.

[0099] In one implementation, the structural similarity, semantic similarity and rule passing rate are weighted according to a preset third weight set to obtain the candidate score of the historical solution, including: calculating the quality score = α×structural similarity+β×semantic similarity+γ×rule passing rate. Wherein, α is the eighth weight, β is the ninth weight, and γ is the tenth weight.

[0100] Alternatively, the quality score can be obtained in the following manner: determine the structural similarity between the task solution and the historical solution; if there is a case where the structural similarity is greater than the preset structural similarity, for each historical solution with a structural similarity greater than the preset structural similarity: obtain the semantic similarity and rule pass rate of the historical solution, and determine the score to be used based on the calculated semantic similarity and rule pass rate. Calculate the average of the scores to be used for each historical solution with a structural similarity greater than the preset structural similarity as the quality score. If there is no case where the structural similarity is greater than the preset structural similarity, the set quality score is used as the quality score.

[0101] In the above optional method, determining the estimated time consumption of each task item execution entity according to the efficiency ranking may include: for each task item execution entity: obtaining the sequence number of the task item execution entity in the efficiency ranking; searching for the sequence number in a preset time consumption evaluation database to obtain the estimated time consumption corresponding to the task item execution entity.

[0102] In an optional method of the above embodiment, determining the priority score of each task node based on the time constraints and estimated time consumption of each task node in the target task map may include: for each task node: searching for the task node, the estimated time consumption corresponding to the task node, and the time constraints corresponding to the task node in a preset first score database to obtain the priority score corresponding to the task node.

[0103] The first score database stores the corresponding relationship between the task nodes, the estimated time consumption and the time constraint conditions.

[0104] In an optional manner of the above embodiment, the node attributes of each task node in the target task map include a preset node score; the task attributes include a historical execution success rate. The priority score of each task node is determined according to the time constraint and estimated time consumption of each task node in the target task map, including: for each task node: determining the priority score of the task node according to the estimated time consumption, time constraint, historical execution success rate and preset node score corresponding to the task node.

[0105] The preset node score is used to reflect the initial priority of the task node. Engineers can set the initial priority of different task nodes based on experience.

[0106] In the above optional method, the priority score of the task node is determined according to the estimated time consumption, time constraints, historical execution success rate and preset node score corresponding to the task node, which can be: determining the efficiency score according to the time constraints and estimated time consumption corresponding to the task node; determining the alternative score according to the historical execution success rate; and obtaining the priority score corresponding to the task node by combining the preset node score, efficiency score and alternative score.

[0107] In the above optional manner, the efficiency score is determined according to the time constraint and the estimated time consumption, which may be: searching the time constraint of the task node in a preset coefficient database to obtain the urgency coefficient of the task node. Get the efficiency score. Among them, q is the efficiency score, λ is the urgency coefficient, and t is the estimated time.

[0108] In the above optional manner, determining the candidate score according to the historical execution success rate may be: searching the historical execution success rate corresponding to the task node in a preset second score database to obtain the candidate score corresponding to the task node.

[0109] The second score database stores the corresponding relationship between the historical execution success rate and the candidate score.

[0110] In the above optional manner, the preset node score, efficiency score and alternative score are comprehensively considered to obtain the priority score corresponding to the task node, which may be: taking the sum of the preset node score, efficiency score and alternative score as the priority score.

[0111] In the above optional manner, the priority score corresponding to the task node is obtained by combining the preset node score, efficiency score and alternative score, which may be: obtaining a second weight set. The preset node score, efficiency score and alternative score are weighted according to the second weight set to obtain the priority score corresponding to the task node.

[0112] Among them, the second weight set records the third weight corresponding to the preset node score, the fourth weight corresponding to the efficiency score and the fifth weight corresponding to the alternative score.

[0113] For example, assume that there are task nodes A and task nodes B. The task item executor of task node A is financial specialist A, and the task item executor of task node B is financial specialist B. At the same time, the time constraints of task nodes A and task node B are both to be completed today. The node attributes of task node A include a preset node score of 0.8, and the node attributes of task node B include a preset node score of 0.5. The efficiency ranking of the task item executors output by the sorting model is financial specialist A first and financial specialist B last. According to the efficiency ranking, it is matched that the estimated time for financial specialist A is 1 hour, and the estimated time for financial specialist B is 2 hours. The time constraint condition "Complete today" of task nodes A and task node B is searched in the preset coefficient database, and the urgency coefficients of task nodes A and task node B are both 0.1. The efficiency of task node A can be calculated to be divided into The efficiency of task node B can be calculated as Assume that the candidate score of task node A is 0.9, the third weight is 1, the fourth weight is 0.3, and the fifth weight is 1. The priority score of task node A = 0.8 + 0.3 × 0.904 + 0.9 = 1.9712.

[0114] In the above embodiment, the priority level of each task plan is determined according to the priority score of each task node, which can be: for each task plan: calculate the average of the priority scores of each task node in the task plan as the target score; search for the target score in a preset priority level database to obtain the priority level corresponding to the target score.

[0115] The priority processing level database stores the corresponding relationship between the target scores and the priority processing levels.

[0116] Step S104, extracting all task solutions in the target task map.

[0117] Among them, each path composed of task nodes in the target task graph is a task plan.

[0118] In some embodiments, the task plan determination method also includes: inputting the task plan into a preset risk prediction model to obtain the risk type corresponding to the task plan and the risk probability of the risk type; when the risk probability is greater than the preset probability, determining the risk management strategy according to the risk type, and adjusting the task plan according to the risk management strategy.

[0119] Among them, the task plan determination method may also include: when the risk probability is less than or equal to the preset probability, obtaining the execution result of the task plan; performing quality verification on the execution result; when the quality verification fails, determining the risk management strategy according to the risk type, and adjusting the task plan according to the risk management strategy.

[0120] In the above embodiment, the risk prediction model can be determined in the following manner: historical scenarios with risk type labels are input into a preset third training model for training to obtain a risk prediction model.

[0121] Among them, the third training model can be a Transformer-XL model (a deep learning model), a temporal convolutional network (TCN), etc.

[0122] In an optional manner of the above embodiment, performing quality verification on the execution result may include: determining the semantic similarity between the execution result and a preset historical execution result, and determining that the quality verification has failed if the semantic similarity is less than the preset similarity; and / or determining whether the execution result meets preset indicators, and determining that the quality verification has failed if the execution result does not meet the preset indicators.

[0123] Wherein, the historical execution result and the execution result belong to the same task scenario. Exemplarily, the historical schemes for obtaining the historical execution results can be stored, and the structural similarity between the task scheme for obtaining the execution result and the historical schemes can be compared. When the structural similarity is higher than a preset threshold, it is determined that the historical execution result and the execution result belong to the same task scenario.

[0124] In the above optional manner, determining the semantic similarity between the execution result and the preset historical execution result may be: calculating the similarity between the first vector representing the execution result and the second vector representing the historical execution result as the semantic similarity.

[0125] The preset indicator is the compliance condition of the task scenario to which the execution result belongs. For example, if the task scenario is the compliance condition of the reimbursement amount, the amount displayed in the execution result can be less than the set budget. If the task scenario is the compliance condition of leave application, the leave duration displayed in the execution result can be less than the set duration.

[0126] In the above optional method, determining whether the execution result meets the preset indicator may include: comparing the content of the execution result with the corresponding preset indicator; if the content of the execution result does not meet the preset indicator, the preset indicator is not met; otherwise, the preset indicator is met.

[0127] The preset indicator may include multiple sub-indicators. For each sub-indicator, the content of the execution result is compared with the sub-indicator; if the content of the execution result does not meet the sub-indicator, the sub-indicator is not satisfied; otherwise, the sub-indicator is satisfied. If the number of unsatisfied sub-indicators is greater than the preset number, the execution result is considered to not meet the preset indicator.

[0128] Correspondingly, in the case of multiple sub-indicators, if the execution result does not meet some of the sub-indicators, the rule pass rate can be obtained by dividing the number of satisfied categories by the total number of categories. The number of satisfied categories means how many sub-indicators are met, and the total number of categories means the number of sub-indicators included in the preset indicator.

[0129] In the above embodiment, determining the risk management strategy according to the risk type may include: searching for the risk management strategy corresponding to the risk type in a preset risk management database.

[0130] Among them, the preset risk management and control database stores the corresponding relationship between risk types and risk management and control strategies.

[0131] Optionally, the risk management strategy is a strategy for reducing the risk of the task plan, for example: adding a backup approver.

[0132] Exemplarily, the risk management strategy may specifically be to add a task node for approval by a backup approver after the set task node when the risk type is set.

[0133] For example, it can be set to directly reject when the rule pass rate is less than 80%; and trigger manual review when 80%≤rule pass rate<90% and semantic similarity<85%. In this way, it can better determine whether there is a problem with the task plan.

[0134] Exemplarily, it can be set that when the rule pass rate is greater than 90% and the semantic similarity is greater than 90%, it is determined that the quality verification has passed.

[0135] Exemplary, combined Figure 3As shown, the overall process of the present application generally goes through multimodal task understanding, dynamic task map construction, resource optimization matching, risk prediction and mitigation, quality closed-loop verification, and incremental strategy optimization. For example, task requirement information can be obtained, and the task requirement information can be in the form of voice, text, or flowchart. Then the task requirement information is converted into the form of a structured task tuple G=(T, C, R). According to the extracted structured task tuple G=(T, C, R), an initial task map is constructed. Then the nodes to be deleted in the initial task map are deleted to obtain an intermediate map. In the case where the intermediate map lacks necessary connection nodes, the missing necessary connection nodes are completed in the intermediate map to obtain the target task map (i.e., the optimized task map). The task item execution subject is assigned to the task nodes in the target task map, and the priority score of each task node is determined according to the time constraints, estimated time consumption, historical execution success rate, and preset node scores of each task node in the target task map. For the target task map, one or more task plans are extracted from the target task map, and the priority level of each task plan is determined according to the priority score of each task node in each task plan, so as to obtain a task plan with a priority level (i.e., a resource allocation plan). Each task node of a single task plan forms a task execution state sequence in order, and the task execution state sequence is input into a preset risk prediction model to obtain the risk type corresponding to the task plan and the risk probability of the risk type (i.e., the risk assessment result), and the task plan is adjusted according to the risk type corresponding to the task plan and the risk probability of the risk type. The adjusted task plan is the mitigation strategy. After that, the task plan is executed to obtain the execution result of the task plan (i.e., the task execution result). The execution result can be quality verified. Determine whether the execution result is compliant. If it is compliant with the rules, it will be verified. If it is not compliant, the execution result will be rejected and the execution result will be manually reviewed. At the same time, user feedback on the task plan and the execution result can also be obtained. Finally, the sorting model can be optimized according to the user's feedback (i.e., user feedback data) to obtain the model weight (model parameter) of the updated sorting model.

[0136] In this way, the implementation method of the present application can automatically set the overall dialogue goal to generate contextually coherent user actions. And it has sufficient generalization ability to generate reasonable behaviors for dialogue situations that do not appear in the corpus. And quantitative feedback scores can be given to guide model learning optimization. As shown in Table 1, the test environment is 1,000 cross-departmental collaborative task records, deployed on an NVIDIA A100 GPU cluster. Compared with traditional office automation systems, such as traditional AI (artificial intelligence) assistants, this application has higher task processing accuracy, shorter emergency task response time, and shorter training cycle for new employees.

[0137] Table 1

[0138] Embodiment 2 Combination Figure 4 As shown, based on the same inventive concept, the embodiment of the present application provides a task solution determination device, including: an acquisition module 1, a construction module 2, an adjustment module 3 and a task determination module 4. Among them, the acquisition module 1 is used to obtain task requirement information; the construction module 2 is used to construct an initial task map according to the task requirement information; the initial task map includes multiple task nodes; the adjustment module 3 is used to delete redundant nodes in the initial task map according to the task attributes of each task node to obtain a target task map; the task attribute is an indicator for evaluating the historical execution process of the task node; the task determination module 4 is used to extract all task solutions in the target task map; wherein each path composed of task nodes in the target task map is a task solution.

[0139] In some embodiments, construction module 2 is used to construct an initial task map based on task requirement information in the following manner: converting the task requirement information into structured data; the structured data includes task items; determining the dependency relationship between the task items; using the task items as task nodes and connecting them according to the dependency relationship to obtain an initial task map.

[0140] In some embodiments, the adjustment module 3 is used to delete redundant nodes in the initial task graph according to the task attributes of each task node in the following manner to obtain a target task graph: determine the nodes to be deleted according to the task attributes, delete the nodes to be deleted from the initial task graph to obtain an intermediate graph; determine whether necessary connection nodes are missing in the intermediate graph, and if necessary connection nodes are missing, complete the missing necessary connection nodes in the intermediate graph to obtain the target task graph.

[0141] In some embodiments, the task attributes include historical execution time; the adjustment module 3 is used to determine the nodes to be deleted according to the task attributes in the following manner: obtaining the node out-degree of each task node; determining the to-be-judged score corresponding to each task node according to the historical execution time and node out-degree corresponding to each task node; if the to-be-judged score corresponding to the task node is lower than the preset score, the task node is determined to be a node to be deleted.

[0142] In some embodiments, the adjustment module 3 is used to determine whether the intermediate graph lacks necessary connection nodes in the following manner: obtaining a node completion model; the node completion model is obtained by training using a preset historical scheme; the preset historical scheme has sample task nodes, necessary connection nodes of the sample task nodes, and node attributes of the sample task nodes; each task node and the node attributes of each task node are input into the node completion model to obtain the necessary connection nodes corresponding to each task node; determining whether the task node lacks necessary connection nodes based on the necessary connection nodes and actual connection nodes of the task node; the actual connection nodes are other task nodes connected to the task node in the target task graph.

[0143] In some embodiments, each task node corresponds to a task item execution subject; the task item execution subject is used to execute the corresponding task node; the node attributes of each task node in the target task map include: time constraints. The task scheme determination device also includes: a priority processing level determination module, which is used to obtain the estimated time consumption corresponding to the task item execution subject of each task node in the target task map; the estimated time consumption reflects the efficiency of the task item execution subject in processing the task node; the priority score of each task node is determined according to the time constraint condition and the estimated time consumption of each task node in the target task map; the priority processing level of each task scheme is determined according to the priority score of each task node.

[0144] In some embodiments, a priority processing level determination module is used to obtain the estimated time consumption corresponding to the task item execution subject of each task node in the target task map: the task item execution subject of each task node in the target task map is input into a preset sorting model to obtain the efficiency ranking of each task item execution subject; and the estimated time consumption of each task item execution subject is determined according to the efficiency ranking.

[0145] In some embodiments, the task plan determination device also includes: a model determination module, which is used to obtain first training data and scores and modification records for historical plans within a historical preset time period; the first training data is data located before the historical preset time period and used to train the sorting model; the second training data is obtained according to the scores and modification records; the preset model is trained using the first training data and the second training data to obtain a reference sorting model; the target model parameters are determined according to the first model parameters of the reference sorting model and the second model parameters of the historical sorting model; the model parameters of the historical sorting model are adjusted to the target model parameters to obtain the sorting model.

[0146] In some embodiments, the model determination module is used to determine the target model parameters based on the first model parameters of the reference sorting model and the second model parameters of the historical sorting model in the following manner: obtaining the quality score of the historical solution; the quality score is used to reflect the logical correctness of the historical solution; determining the weight parameter based on the quality score; weighting the first model parameter and the second model parameter according to the weight parameter to obtain the target model parameter.

[0147] In some embodiments, the node attributes of each task node in the target task map include a preset node score; the task attributes include a historical execution success rate; the priority processing level determination module is used to determine the priority score of each task node according to the time constraints and estimated time consumption of each task node in the target task map in the following manner: For each task node: determine the priority score of the task node according to the estimated time consumption, time constraints, historical execution success rate and preset node score corresponding to the task node.

[0148] In some embodiments, the task plan determination device also includes: a risk determination module, which is used to input the task plan into a preset risk prediction model to obtain the risk type corresponding to the task plan and the risk probability of the risk type; when the risk probability is greater than the preset probability, determine the risk management strategy according to the risk type, and adjust the task plan according to the risk management strategy.

[0149] In some embodiments, the task plan determination device also includes: a compliance determination module, which is used to obtain the execution result of the task plan when the risk probability is less than or equal to the preset probability; perform quality verification on the execution result; if the quality verification fails, determine the risk management strategy according to the risk type, and adjust the task plan according to the risk management strategy.

[0150] In some embodiments, the compliance determination module is used to perform quality verification on the execution results in the following manner: determining the semantic similarity between the execution results and the preset historical execution results, and determining that the quality verification has failed if the semantic similarity is less than the preset similarity; the historical execution results and the execution results belong to the same task scenario; and / or determining whether the execution results meet preset indicators, and determining that the quality verification has failed if the execution results do not meet the preset indicators; the preset indicators are the compliance conditions of the task scenario to which the execution results belong.

[0151] Combination Figure 5 As shown, an embodiment of the present application provides an electronic device, including a processor 5 and a memory 7. Optionally, the device may also include a communication interface 8 and a bus 6. The processor 5, the communication interface 8, and the memory 7 may communicate with each other through the bus 6. The communication interface 8 may be used for information transmission. The processor 5 may call the logic instructions in the memory 7 to execute the above-mentioned task scheme determination method.

[0152] In addition, the logic instructions in the above-mentioned memory 7 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product.

[0153] The memory 7 is a computer-readable storage medium that can be used to store software programs and computer executable programs, such as program instructions / modules corresponding to the method in the embodiment of the present application. The processor 5 executes the functional application and data processing by running the program instructions / modules stored in the memory 7, that is, realizing the above-mentioned task solution determination method.

[0154] The memory 7 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required for at least one function; the data storage area may store data created according to the use of the terminal device, etc. In addition, the memory 7 may include a high-speed random access memory and may also include a non-volatile memory.

[0155] The electronic device may be a computer or a server, etc.

[0156] An embodiment of the present application provides a storage medium storing computer executable instructions, wherein the computer executable instructions are configured to execute the above-mentioned task solution determination method.

[0157] An embodiment of the present application provides a computer program product, which includes a computer program stored on a storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes the above-mentioned task plan determination method.

[0158] The computer-readable storage medium mentioned above may be a transient computer-readable storage medium or a non-transitory computer-readable storage medium.

[0159] The technical solution of the embodiment of the present application can be embodied in the form of a software product, which is stored in a storage medium and includes one or more instructions for a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the embodiment of the present application. The aforementioned storage medium may be a non-transient storage medium, including: a USB flash drive, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk or an optical disk, and other media that can store program codes, or a transient storage medium.

[0160] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0161] The above description is only an embodiment of the present application and is not intended to limit the protection scope of the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application. At the same time, the above embodiments can be combined with each other to form new embodiments without conflict.

Claims

1. A method for determining a task plan, characterized in that: include: Obtaining task requirement information; Constructing an initial task map according to the task requirement information; The initial task graph includes multiple task nodes; Deleting redundant nodes in the initial task graph according to the task attributes of each of the task nodes to obtain a target task graph; the task attributes are indicators for evaluating the historical execution process of the task nodes; All task solutions in the target task map are extracted; wherein each path formed by the task nodes in the target task map is a task solution.

2. The method according to claim 1, characterized in that Deleting redundant nodes in the initial task graph according to the task attributes of each of the task nodes to obtain a target task graph includes: Determine a node to be deleted according to the task attribute, delete the node to be deleted from the initial task graph, and obtain an intermediate graph; Determining whether necessary connection nodes are missing in the intermediate graph; In the case where the necessary connection nodes are missing, the missing necessary connection nodes are completed in the intermediate graph to obtain the target task graph.

3. The method according to claim 2, characterized in that The task attributes include historical execution time; determining the node to be deleted according to the task attributes includes: Obtaining the node out-degree of each of the task nodes; Determine the to-be-judged score corresponding to each task node according to the historical execution time and node out-degree corresponding to each task node; If the to-be-determined score corresponding to the task node is lower than a preset score, the task node is determined to be a node to be deleted.

4. The method according to claim 2, characterized in that: Determining whether necessary connection nodes are missing in the intermediate graph includes: Obtaining a node completion model; the node completion model is obtained by training using a preset historical scheme; the preset historical scheme includes a sample task node, necessary connection nodes of the sample task node, and node attributes of the sample task node; Inputting each of the task nodes and the node attributes of each of the task nodes into the node completion model to obtain the necessary connection nodes corresponding to each of the task nodes; Determine whether the task node lacks necessary connection nodes based on the necessary connection nodes and actual connection nodes of the task node; the actual connection nodes are other task nodes connected to the task node in the target task graph.

5. The method according to claim 1, characterized in that Each of the task nodes corresponds to a task item execution subject; the task item execution subject is used to execute the corresponding task node; The node attributes of each task node in the target task graph include: time constraints; the method further includes: Obtaining the estimated time consumption corresponding to the task item execution subject of each task node in the target task map; the estimated time consumption reflects the efficiency of the task item execution subject in processing the task node; Determine the priority score of each task node according to the time constraint condition and the estimated time consumption of each task node in the target task map; The priority level of each task solution is determined according to the priority score of each task node.

6. The method according to claim 5, characterized in that Obtaining the estimated time consumption corresponding to the task item execution subject of each task node in the target task map includes: Inputting the task item execution subject of each task node in the target task map into a preset sorting model to obtain the efficiency sorting of each task item execution subject; The estimated time required for executing each task item is determined according to the efficiency ranking.

7. The method according to claim 6, characterized in that The ranking model is obtained in the following way: Obtaining first training data and the scoring and modification records of historical solutions within a historical preset time; the first training data is data located before the historical preset time for training the sorting model; the modification records are used to describe the efficiency ranking of the multiple task item execution entities after correction; Acquire second training data according to the score and the modification record; Using the first training data and the second training data to train a preset model to obtain a reference sorting model; Determining target model parameters according to first model parameters of the reference sorting model and second model parameters of the historical sorting model; The model parameters of the historical sorting model are adjusted to target model parameters to obtain the sorting model.

8. The method according to claim 7, characterized in that Determining a target model parameter according to a first model parameter of the reference sorting model and a second model parameter of the historical sorting model includes: Obtaining a quality score of the historical solution; the quality score is used to reflect the logical correctness of the historical solution; Determining a weight parameter according to the quality score; The first model parameter and the second model parameter are weighted according to the weight parameter to obtain a target model parameter.

9. The method according to claim 5, characterized in that The node attribute of each task node in the target task map includes a preset node score; the task attribute includes a historical execution success rate; and the priority score of each task node is determined according to the time constraint condition and the estimated time consumption of each task node in the target task map, including: For each of the task nodes: determine the priority score of the task node according to the estimated time consumption, the time constraint, the historical execution success rate and the preset node score corresponding to the task node.

10. The method according to claim 1, characterized in that The method further comprises: Inputting the task plan into a preset risk prediction model to obtain the risk type corresponding to the task plan and the risk probability of the risk type; When the risk probability is greater than a preset probability, a risk management strategy is determined according to the risk type, and the task plan is adjusted according to the risk management strategy.

11. The method according to claim 10, characterized in that The method further comprises: When the risk probability is less than or equal to a preset probability, obtaining the execution result of the task plan; Performing quality verification on the execution results; In the event that the quality verification fails, a risk management strategy is determined based on the risk type, and the task plan is adjusted according to the risk management strategy.

12. The method according to claim 11, characterized in that Performing quality verification on the execution results, including: Determine the semantic similarity between the execution result and a preset historical execution result, and if the semantic similarity is less than the preset similarity, determine that the quality verification fails; the historical execution result and the execution result belong to the same task scenario; and / or, Determine whether the execution result meets the preset indicators. If the execution result does not meet the preset indicators, determine that the quality verification has failed; the preset indicators are the compliance conditions of the task scenario to which the execution result belongs.

13. The method according to any one of claims 1 to 12, characterized in that: Constructing an initial task map according to the task requirement information, including: Converting the task requirement information into structured data; the structured data includes task items; Determining dependencies between the task items; The task items are used as the task nodes and connected according to the dependency relationships to obtain the initial task graph.

14. A task plan determination device, characterized in that: include: An acquisition module is used to obtain task requirement information; A construction module, used to construct an initial task map according to the task requirement information; The initial task graph includes multiple task nodes; An adjustment module, used to delete redundant nodes in the initial task map according to the task attributes of each task node to obtain a target task map; the task attributes are indicators for evaluating the historical execution process of the task nodes; The task determination module is used to extract all task solutions in the target task map; wherein each path formed by the task nodes in the target task map is a task solution.

15. An electronic device, characterized in that: It includes a processor and a memory, wherein the memory stores computer executable instructions that can be executed by the processor, and the processor executes the computer executable instructions to implement the task plan determination method described in any one of claims 1 to 13.

16. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the task scheme determination method according to any one of claims 1 to 13 is implemented.

17. A storage medium, characterized in that: The storage medium stores computer executable instructions, and when the computer executable instructions are called and executed by the processor, the computer executable instructions enable the processor to implement the task solution determination method described in any one of claims 1 to 13.

Images