Work order processing method, device and equipment

Abstract

The invention provides a work order processing method, device and equipment, and relates to the technical field of cloud computing, according to the technical scheme of the invention, a work order chain comprising a plurality of first work orders is generated according to received user demand information, and a plurality of resources included in the plurality of first work orders are clustered according to semantic similarity, so that a plurality of field clusters are obtained; therefore, according to the work order information and the field cluster, the number of the N approval nodes corresponding to the plurality of work order chains is merged to obtain the M approval nodes, the plurality of first work orders are processed according to the M approval nodes to obtain the processing result, the approval process of the work order is simplified, and the work order processing efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cloud computing, in particular to a work order processing method, device and equipment. BACKGROUND

[0002] The private cloud resource delivery scenario is generally a process in which a large enterprise builds a private cloud platform by itself to provide computing, storage, network and other private cloud resources for each tenant within the enterprise. In related technologies, when applying for a private cloud resource, a user initiates a resource application through a work order system, such as an independent work order of "technical solution review", "KVM application", "MySQL cluster application", etc. The work order system initiates an approval by calling a workflow engine based on pre-set approval rules. However, in related technologies, there are many approval nodes corresponding to a work order, and multiple work orders are approved one by one, thereby causing the problem of low work order processing efficiency in the private cloud resource scenario. SUMMARY

[0003] The present application provides a work order processing method, device and related equipment, which solves the problem of low work order processing efficiency in the private cloud resource scenario in related technologies.

[0004] To solve the above problems, the present application is implemented as follows:

[0005] In a first aspect, the present application provides a work order processing method applied to a service platform, comprising:

[0006] generating a work order chain according to received user demand information, the work order chain comprising a plurality of first work orders and work order information, the first work orders being used to request application for use of private cloud resources in the service platform, the work order information being used to indicate a dependency relationship between the plurality of first work orders, the dependency relationship being used to indicate an execution sequence of the plurality of first work orders, the work order chain corresponding to N approval nodes in the service platform, N being a positive integer greater than 1;

[0007] clustering a plurality of fields included in the plurality of first work orders according to semantic similarity to obtain a plurality of field clusters, each field cluster comprising at least two fields;

[0008] merging the N approval nodes based on the work order information and the plurality of field clusters to obtain M approval nodes, M being a positive integer less than N;

[0009] processing the plurality of first work orders based on the M approval nodes to generate a processing result.

[0010] Optionally, the clustering a plurality of fields included in the plurality of first work orders according to semantic similarity to obtain a plurality of field clusters comprises:

[0011] extracting text content included in the plurality of first work orders to obtain a plurality of fields;

[0012] converting the plurality of fields into a plurality of field vectors, the plurality of fields corresponding to the plurality of field vectors one by one;

[0013] performing semantic similarity calculation on any two field vectors in the plurality of field vectors to obtain similarity information, the similarity information including semantic similarity between any two field vectors in the plurality of field vectors;

[0014] performing clustering on the plurality of fields according to the similarity information to obtain a plurality of field clusters.

[0015] Optionally, the performing clustering on the plurality of fields according to the similarity information to obtain a plurality of field clusters includes:

[0016] filtering a plurality of semantic similarities included in the similarity information based on a preset threshold to obtain a plurality of target semantic similarities, the target semantic similarity being greater than or equal to the preset threshold, and each target semantic similarity corresponding to two fields;

[0017] aggregating the two fields corresponding to each target semantic similarity into the same field cluster to obtain the plurality of field clusters.

[0018] Optionally, the merging the N approval nodes based on the work order information and the plurality of field clusters to obtain M approval nodes includes:

[0019] generating an approval node relationship graph based on the work order information and the N approval nodes, the approval node relationship graph being used to represent a dependency relationship between the N approval nodes;

[0020] inputting the approval node relationship graph into a pre-trained graph neural network model to perform feature extraction to obtain N graph features, the N approval nodes corresponding to the N graph features one by one;

[0021] performing clustering on the N graph features based on a density adaptive hierarchical clustering algorithm to obtain a plurality of clustering results, each clustering result including at least two graph features;

[0022] merging at least two approval nodes corresponding to the at least two graph features included in each clustering result in the N approval nodes to obtain the M approval nodes.

[0023] Optionally, the processing the plurality of first work orders based on the M approval nodes to generate a processing result includes:

[0024] determine, based on a preset mapping relationship table, a plurality of target text contents corresponding to the plurality of field clusters in a database, wherein the database includes a plurality of preset text contents, the preset mapping relationship table includes a plurality of mapping relationships, and each mapping relationship is used to indicate, in the plurality of preset text contents, a field cluster corresponding to each preset text content; and perform information filling on the plurality of first work orders based on the plurality of target text contents to obtain a plurality of second work orders, wherein the plurality of second work orders correspond to the plurality of first work orders one by one.

[0025] perform processing on the plurality of second work orders based on the M approval nodes to generate a processing result.

[0026] Optionally, the generating a work order chain according to the received user demand information comprises:

[0027] converting the received user demand information into a target text, wherein the target text is a natural language text;

[0028] performing semantic disintegration on the target text to obtain a plurality of feature contents, wherein different feature contents correspond to different feature types;

[0029] filling the plurality of feature contents into a preset prompt word template to obtain a target prompt word;

[0030] inputting the target prompt word into a pre-trained large language model to generate the work order chain.

[0031] In a second aspect, an embodiment of the present application provides a work order processing apparatus applied to a service platform, and the apparatus comprises:

[0032] a receiving module configured to generate a work order chain according to received user demand information, wherein the work order chain includes a plurality of first work orders and work order information, the first work order is used to request application for use of private cloud resources in the service platform, the work order information is used to indicate a dependency relationship between the plurality of first work orders, the dependency relationship is used to indicate an execution sequence of the plurality of first work orders, the work order chain corresponds to N approval nodes in the service platform, and N is a positive integer greater than 1;

[0033] a clustering module configured to cluster a plurality of fields included in the plurality of first work orders according to semantic similarity to obtain a plurality of field clusters, and each field cluster includes at least two fields;

[0034] a merging module configured to merge the N approval nodes based on the work order information and the plurality of field clusters to obtain M approval nodes, and M is a positive integer less than N;

[0035] The processing module is configured to process the plurality of first work orders based on the M approval nodes to generate a processing result.

[0036] In a third aspect, the present application provides an electronic device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, and the computer program, when executed by the processor, implements the steps in the method of the first aspect.

[0037] In a fourth aspect, the present application provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program, and the computer program, when executed by a processor, implements the steps in the method of the first aspect.

[0038] In a fifth aspect, the present application provides a computer program product, comprising computer instructions, and the computer instructions, when executed by a processor, implement the steps in the method of the first aspect.

[0039] The present application provides a work order processing method, device and equipment, and relates to the technical field of cloud computing. The method comprises the following steps: generating a work order chain according to received user demand information, the work order chain comprising a plurality of first work orders and work order information, the first work orders being used to request the application for using private cloud resources in a service platform, the work order information being used to indicate a dependency relationship between the plurality of first work orders, the dependency relationship being used to indicate the execution sequence of the plurality of first work orders, the work order chain corresponding to N approval nodes in the service platform, N being a positive integer greater than 1; clustering a plurality of fields included in the plurality of first work orders according to semantic similarity to obtain a plurality of field clusters, each field cluster comprising at least two fields; merging the N approval nodes based on the work order information and the plurality of field clusters to obtain M approval nodes, M being a positive integer less than N; and processing the plurality of first work orders based on the M approval nodes to generate a processing result. The technical scheme of the present application generates a work order chain comprising a plurality of first work orders according to received user demand information, and clusters a plurality of resources included in the plurality of first work orders according to semantic similarity to obtain a plurality of field clusters, thereby merging the number of N approval nodes corresponding to the plurality of work order chains based on the work order information and the field clusters to obtain M approval nodes, and processing the plurality of first work orders based on the M approval nodes to obtain a processing result, so that the approval process of the work order is simplified, and the work order processing efficiency is improved. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the present application, the drawings required to be used in the description of the present application will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0041] Figure 1 A flowchart of a work order processing method provided by an embodiment of the present application is shown in FIG. 1.

[0042] Figure 2 A system architecture diagram of a service platform provided by an embodiment of the present application is shown in FIG. 2.

[0043] Figure 3 A structural diagram of a work order processing device provided by an embodiment of the present application is shown in FIG. 3.

[0044] Figure 4 A structural diagram of an electronic device provided by an embodiment of the present application is shown in FIG. 4. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0046] The terms "first", "second", and the like in the embodiments of the present application are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device including a series of steps or units does not necessarily have to include only those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to the process, method, product, or device. In addition, "and / or" is used in the present application to represent at least one of the connected objects, for example, A and / or B and / or C represents 7 cases including A alone, B alone, C alone, A and B both exist, B and C both exist, A and C both exist, and A, B and C all exist.

[0047] Referring to Figure 1 , Figure 1 A flowchart of a work order processing method provided by an embodiment of the present application is shown in FIG. 1. As shown in FIG. 1, the work order processing method is applied to a service platform and can include the following steps: Figure 1

[0048] ​Step 101, generating a work order chain according to the received user demand information, the work order chain including a plurality of first work orders and work order information, the first work orders being used to request application for use of private cloud resources in the service platform, the work order information being used to indicate a dependency relationship between the plurality of first work orders, the dependency relationship being used to indicate an execution sequence of the plurality of first work orders, the work order chain corresponding to N approval nodes in the service platform, N being a positive integer greater than 1.

[0049] In the embodiment, the user demand information is a request generated by a user requesting application for use of private cloud resources in the service platform. It needs to be noted that the user demand information in the embodiment is natural language. For example, the user's natural language demand can be input through the Copilot page, such as "create a new development and test environment for a research and development personnel". At this time, the Copilot calls the intelligent work order aggregation engine to automatically generate a scenario-based work order chain access portal for the user.

[0050] The work order chain includes a plurality of first work orders and work order information. The work order information is used to indicate a dependency relationship between the first work orders. For example, the work order chain can be a scenario-based work order chain with a directed acyclic graph (DAG) dependency relationship. The dependency relationship is used to indicate an execution sequence of the plurality of first work orders. For example, some work orders need to be executed after other work orders. Therefore, the execution sequence between the work orders needs to be considered when executing the work orders.

[0051] After generating the work order chain, N approval nodes included in the work order chain are determined. The approval nodes are used to approve the first work orders. Specifically, each first work order can need one or more approval nodes for approval. All the first work orders correspond to N approval nodes. It needs to be described that the N approval nodes can include the same approval nodes. For example, A work order needs C approval node for approval. After A work order is completed, B work order also needs C approval node for approval, that is, C approval node needs to approve different work orders.

[0052] Step 102, clustering a plurality of fields included in the plurality of first work orders according to semantic similarity, obtaining a plurality of field clusters, each field cluster including at least two fields.

[0053] In this embodiment, semantic similarity refers to determining the similarity of texts by analyzing their meanings. Specifically, the fields of multiple work orders are evaluated by semantic similarity. Each work order may contain several fields, such as work order number, submitter, problem description, priority, etc. Clustering is a process of grouping similar data items into a class. The clustering result in this embodiment is to form multiple field clusters, each containing at least two fields that are semantically related or similar.

[0054] By clustering fields, data management can be optimized, more valuable information can be analyzed and mined, and the efficiency of work order processing can be improved. For example, in report generation, statistical analysis, and automated processing scenarios, the aggregated field clusters are prioritized.

[0055] Step 103: Based on the work order information and the multiple field clusters, the N approval nodes are merged to obtain M approval nodes, where M is a positive integer less than N.

[0056] In this embodiment, the N approval nodes are merged based on the determined work order information and the clustered multiple field clusters, thereby obtaining M approval nodes, where M is less than N, for example, 100 approval nodes are reduced to 50 approval nodes.

[0057] Specifically, for example, when N is 100, the 100 approval nodes are determined to be the same or similar based on the work order information and the multiple field clusters. These same or similar approval nodes are merged to achieve simultaneous approval of multiple first work orders by the same approval node, thereby improving the efficiency of approval.

[0058] Step 104: Based on the M approval nodes, the multiple first work orders are processed to generate a processing result.

[0059] In this embodiment, the process of processing multiple work orders based on approval nodes includes reviewing work order content, verifying information integrity and accuracy, and making decisions (such as processing results usually refer to the status after approval (such as "approved", "rejected", "to be supplemented", etc.) or subsequent operations.

[0060] The process of generating a processing result takes into account the input and decision of each approval node, while recording approval opinions and related information (such as approval, rejection, supplementary information, etc.) and feeding back the processing result to relevant personnel.

[0061] The technical solution of the present application generates a work order chain including a plurality of first work orders according to received user demand information, and clusters a plurality of resources included in the plurality of first work orders according to semantic similarity, thereby obtaining a plurality of field clusters. Thus, the number of N approval nodes corresponding to the plurality of work order chains is merged according to the work order information and the field clusters, thereby obtaining M approval nodes. The plurality of first work orders are processed according to the M approval nodes, thereby obtaining a processing result, which realizes a simplified approval process of work orders and improves work order processing efficiency.

[0062] Optionally, the clustering of the plurality of fields included in the plurality of first work orders according to the semantic similarity comprises:

[0063] extracting text content included in the plurality of first work orders to obtain a plurality of fields;

[0064] converting the plurality of fields into a plurality of field vectors, the plurality of fields corresponding to the plurality of field vectors one by one;

[0065] performing semantic similarity calculation on any two field vectors in the plurality of field vectors to obtain similarity information, the similarity information including semantic similarity between any two field vectors in the plurality of field vectors;

[0066] clustering the plurality of fields according to the similarity information to obtain the plurality of field clusters.

[0067] In this embodiment, after obtaining the plurality of first work orders, the text content included in the plurality of first work orders needs to be extracted to generate a plurality of fields. Specifically, after obtaining the DAG work order chain, a shared field pool is constructed based on a shared field pool management unit. The shared field pool management unit first obtains all work order ids corresponding to the DAG work order chain, and then obtains all form fields corresponding to the work order based on the work order id.

[0068] There are N first work orders on the DAG work order chain, and all field names of the N first work orders need to be cleaned and preprocessed. The shared field pool management unit calls a Large Language Model (LLM) + Retrieval Augmented Generation (RAG) capability, and based on the knowledge base (historical work order data, work order operation guidelines, etc.) mounted by the LLM, all fields are retrieved, matched and preprocessed. The preprocessing result is that, taking the work order id as the shared field pool table index, the classification sub_pool is completed according to the data type (such as string, numerical value, enumeration, etc.) and business attribute (such as organization, resource pool, network configuration, security policy related, etc.) of the field name, and stored in the specified table structure of the database for subsequent query and calling.

[0069] After obtaining the plurality of fields, the plurality of fields are converted into a plurality of field vectors, and semantic similarities between any two field vectors are calculated to obtain similarity information, the similarity information including the semantic similarities between any two field vectors in the plurality of field vectors. Specifically, the semantic analysis unit of the intelligent work order aggregation engine performs semantic vectorization on the field names in the shared field pool and performs semantic similarity calculation.

[0070] The semantic analysis unit of the intelligent work order aggregation engine calls the Sentence-BERT model to encode all field name texts in the shared field pool of the first work order to the Nth work order in the DAG work order chain into 768-dimensional vectors. To improve clustering efficiency, the semantic analysis unit first finds the classification of the shared field pool, calculates the cosine similarity between two field name text vectors under the same classification sub_pool (organization, data type, etc. related), and obtains a similarity value between 0 and 1, which is used to measure the semantic similarity between the two.

[0071] According to the generated similarity information, the plurality of fields are clustered to obtain the plurality of field clusters. Specifically, in the case where the semantic similarity meets a preset threshold, the corresponding two fields can be clustered to generate a plurality of field clusters.

[0072] Optionally, the clustering of the plurality of fields according to the similarity information to obtain the plurality of field clusters comprises:

[0073] Based on a preset threshold, a plurality of target semantic similarities are obtained by screening the plurality of semantic similarities included in the similarity information, the target semantic similarity being greater than or equal to the preset threshold, and each target semantic similarity corresponding to two fields.

[0074] The two fields corresponding to each target semantic similarity are aggregated into the same field cluster to obtain the plurality of field clusters.

[0075] In this embodiment, by the preset threshold, a plurality of target semantic similarities are determined by screening the plurality of semantic similarities, each target semantic similarity being greater than or equal to the preset threshold. Thus, the two fields corresponding to the target semantic similarity are aggregated into the same field cluster, thereby generating a plurality of field clusters.

[0076] Specifically, the work order fields are clustered, and a mapping rule between similar fields is constructed. In the intelligent work order aggregation engine, a rule engine is configured, and the threshold of the semantic similarity is set to 0.85. When the similarity value calculated by the semantic analysis unit is greater than or equal to the threshold, the field mapping management unit is triggered.

[0077] The field mapping management unit establishes a mapping relationship (for example, "machine room name" and "IDC_location" are considered as the same field) according to the similarity calculation result of the field names in the shared field pool with a similarity >= 0.85, and records the name, position (belonging to the work order id, the DAG work order chain sequence), and the like of the field (the similar field in the shared field pool), and stores the mapping relationship in the configuration file or the database table of the intelligent work order aggregation engine, so as to be used by the subsequent automatic filling operation. It should be noted that there can be two or more fields with a similarity > 0.85, and therefore the number of work order fields involved in the mapping relationship can also be >= 2.

[0078] By clustering the fields, not only can the data management be optimized, more valuable information can be analyzed and mined, and the efficiency of work order processing can be improved.

[0079] Optionally, the N approval nodes are merged based on the work order information and the plurality of field clusters to obtain M approval nodes, including:

[0080] An approval node relationship graph is generated based on the work order information and the N approval nodes, and the approval node relationship graph is used to represent the dependency relationship between the N approval nodes;

[0081] The approval node relationship graph is input into a pre-trained graph neural network model for feature extraction to obtain N graph features, and the N approval nodes and the N graph features are one-to-one corresponding;

[0082] The N graph features are clustered based on a density adaptive hierarchical clustering algorithm to obtain a plurality of clustering results, and each clustering result includes at least two graph features;

[0083] In the N approval nodes, at least two approval nodes corresponding to at least two graph features included in each clustering result are merged to obtain the M approval nodes.

[0084] In this embodiment, it is assumed that the DAG work order chain contains N work orders, and the i-th work order contains M approval nodes. The set of approval nodes of the DAG work order chain V = {[v11, v12,..., v1M],..., [vN1, vN2,..., vNM]}. The goal of this step is to cluster the approval nodes with similar semantics and business associations into K clusters (K << M) to compress the approval flow. Since pure text features may ignore business associations, for example, "cloud team_storage_approval" and "cloud team_storage_approval" can be considered as duplicates from the text; but it cannot identify that "cloud team_KVM_approval" and "container team_container_approval" belong to the same "computing resource domain" organization, which is an implicit equivalent relationship. Therefore, the density adaptive hierarchical clustering algorithm HDBSCAN needs to capture such semantic equivalent relationships in a high-dimensional latent space by fusing graph neural networks and text embeddings.

[0085] First, the approval node information of the DAG work order chain is obtained. The API interface provided by the workflow engine is called, and the work order id of the DAG work order chain is taken as the input parameter to obtain all the approval node set V = [v1, v2,..., vM] corresponding to the work order id. The approval node viM includes approval node name, approver, and other information. The data collection unit extracts the text features of the approval node triplets, including role role_type, resource type resource_type, and operation type action_type, such as ["cloud team", "storage", "approval"], and extracts 768-dimensional text feature vectors through Sentence-BERT.

[0086] After data standardization preprocessing, the approval node relationship graph G = (V, E) is further constructed to form graph structure data. The node entity V corresponds to each approval task, and the edge relationship E is divided into three dimensions: first, the sequential dependence edge seq_weight, which is generated by analyzing the explicit dependence relationship between nodes; second, the role association edge role_sim, which automatically identifies different nodes with the same approval role and establishes a connection; and third, the resource association edge res_sim, which associates nodes that process the same type of resources. The weight of the sequential dependence edge depends on the number of hops between nodes, the weight of the role association edge is based on the historical approval similarity, and the weight of the resource association edge is calculated according to the resource coupling degree. The dynamic weight of the edge is shown in the following formula:

[0087] .

[0088] where: α = 0.6, β = 0.3, γ = 0.1.

[0089] The double-layer GAT graph attention mechanism is adopted to extract deep features of the graph neural network. The first layer focuses on local features and analyzes the neighbor relationship within a 3-hop range of each node in the relationship graph G. The edge_weight is used as an attention bias to maintain gradient stability. Given the neighbor set V(i) of node i, the attention coefficient of GAT is obtained:

[0090]

[0091] where LeakyReLU is a neural network activation function, and are the feature vectors of nodes and , W is a learnable weight matrix, is the attention vector, denotes the concatenation operation. After obtaining , the attention coefficient is then normalized using softmax:

[0092] .

[0093] The new feature of the final node i is the weighted sum of its neighbor nodes:

[0094] .

[0095] The second layer focuses on global features and compresses the features of the entire graph through graph pooling to obtain a graph-level representation, thereby extracting the overall pattern of the work order chain. The multi-layer perceptron (MLP) is used to implement the pooling, and the global-level feature can be obtained:

[0096] .

[0097] Finally, the global feature and the local feature of each node are concatenated to form a 128-dimensional graph feature vector to enhance the global context information of each node:

[0098] .

[0099] The improved HDBSCAN algorithm is executed by fusing the graph features and text features of the nodes. The hybrid distance measurement method is innovatively adopted in this proposal, which combines the graph structure distance (accounting for 60%) and the text semantic distance (accounting for 40%). The improved HDBSCAN can identify explicit duplicate nodes (such as the same triple) and implicit equivalent nodes (such as "cloud team_KVM" and "container team_container" belong to the same computing resource domain in business). The algorithm running process contains three-stage optimization:

[0100] Phase one, calculate mutual distance. Before calculating mutual distance, first need to get mixed distance. Set the node i, j corresponding to the 128-dimensional graph feature vector is z i, z j, 768-dimensional text feature vector is t i, t j, graph structure distance uses normalized hamming distance, text semantic distance uses cosine distance, both are added after weighting to get mixed distance:

[0101] + .

[0102] Among them, the graph structure distance formula, R represents the length of the graph feature vector (here 128), then

[0103] .

[0104] Text semantic distance formula is as follows:

[0105] .

[0106] At this time, the mutual distance From the core distance of node i, j , And the mixed distance of the two Take the maximum value, need to note that, here , Node i, j to its kth neighbor distance, since Normalized processing, so , Also need to do the corresponding normalization processing. Mutual distance calculation formula is as follows:

[0107] .

[0108] Phase two, generate hierarchical tree based on mutual distance matrix. Set N as the number of approval nodes, use Prim algorithm to construct minimum spanning tree MST:

[0109] .

[0110] Phase three, based on stability threshold, execute clustering on MST to extract merged cluster. Set λ as stability threshold =1 / , that is, the reciprocal of mutual distance, use λ to cut MST, and select the cluster with high stability. Assuming n is the node of MST, p is the leaf node of node n, , Stability threshold of child node p and parent node n respectively. Calculate for each node n in MST, define stability as:

[0111] .

[0112] First, all nodes of the MST are traversed from bottom to top to calculate S(n), if S(n) > 0 and S(p)≤ 0 of the child node of n, n is reserved as an initial cluster, otherwise pruning is performed. Finally, the initial cluster Clusters = {C1, C2, …, Ck} is obtained. Here Ck is the compressed approval node.

[0113] In this embodiment, the graph structure features and text semantic features of the DAG work order chain approval nodes are adaptively weighted and fused, and an improved HDBSCAN algorithm is used to realize intelligent compression of the approval nodes.

[0114] Optionally, based on the M approval nodes, the plurality of first work orders are processed to generate a processing result, including:

[0115] Based on a preset mapping relationship table, a plurality of target text contents corresponding one-to-one to the plurality of field clusters are determined in a database, wherein the database includes a plurality of preset text contents, and the preset mapping relationship table includes a plurality of mapping relationships, and each mapping relationship is used to indicate that each preset text content corresponds to a field cluster in the plurality of preset text contents; based on the plurality of target text contents, information filling is performed on the plurality of first work orders to obtain a plurality of second work orders, and the plurality of second work orders correspond one-to-one to the plurality of first work orders;

[0116] Based on the M approval nodes, the plurality of second work orders are processed to generate a processing result.

[0117] In this embodiment, by using the preset mapping relationship table including a plurality of mapping relationships, a plurality of target text contents corresponding one-to-one to a plurality of field clusters are determined in a database, so that the plurality of first work orders are information-filled according to the plurality of target text contents to obtain a plurality of second work orders, and the plurality of second work orders are processed according to the M approval nodes to generate a processing result.

[0118] Specifically, first, the fields in the work order page are automatically filled based on the mapping relationship, and the shared field pool is dynamically updated. When the user enters the i th work order filling page of the DAG work order chain through the Copilot front-end interface, the front end uses WebSocket to monitor the changes of the field values. When the user completes the field value filling of the current page and enters the i+1 th work order filling page, at this time, WebSocket synchronously updates the field values of the i th work order page to the field values of all work order pages from i+1 th to N th which have established mapping relationship. After entering the i+1 th work order filling page, when focusing on a field that can be automatically filled, the front-end interface sends a request to the intelligent work order aggregation engine.

[0119] After receiving the request, the pre-established mapping relationship table is queried to obtain the field value in the shared field pool corresponding to the current field, and the field value is fed back to the front-end interface in a pre-filled form. The front-end displays the filled content in the corresponding form field position. The user can view the automatically filled content, choose to accept (directly submit the form) or modify it (such as adjusting the specific expression of the machine room name), to meet the actual business needs, and then continue to fill in the next work order page.

[0120] After completing the DAG work order chain field filling and executing the compressed work flow approval node, the non-standard resource automation adaptation layer needs to collect the field content filled in the work order as an input parameter, and retrieve the historical AnsiblePlaybook / API interface address of the same type of work order using RAG. The LLM generates a complete script through the Prompt template, automatically fills in the parameters (such as machine room, security domain, and quota), and finally outputs a script chain (Playbook or cloud resource management platform API interface call list) arranged in DAG order. Through the resource automatic delivery unit of the non-standard resource automation adaptation layer, the script chain is injected into the Ansible Runner / cloud management platform to execute the resource delivery instruction.

[0121] In some embodiments, to adapt to the changing business process and avoid security risks caused by the execution of the intelligently generated script, the embodiment also adds a closed-loop feedback and security control function. Specifically, it includes dynamic effect verification and closed-loop optimization. The execution result of each time is written back to the system, and the successful execution automatically supplements the case and adds it to the RAG knowledge base, and the abnormal failure is sent to the RLHF fine-tuning queue together with the manual correction. The seq_weight, role_sim, and res_sim are recalculated periodically with new data to refresh the graph weight, so that the work order chain, script, and clustering weight models are simultaneously online incremental learning, and the Prompt template of the LLM is continuously adjusted to ensure that the more it is used, the more accurate it is. The scene-based work order chain execution process is visualized. The dynamic DAG is rendered in real time on the Web, the color of the node changes from green to yellow to red according to the execution state, and the user can click on the node to drill down to view the work order filling content, approval process, execution script, log, etc. When a node is timed out or failed, it will automatically trigger WeChat Work / email alarm and highlight the path, and at the same time provide a management background for the operation personnel to grasp the overall delivery rhythm at a glance. The security and audit reinforcement link is added. The script is scanned statically by the strategy engine before execution to intercept unauthorized or high-risk instructions, and the key resource operation is automatically snapped before execution. When an abnormality or manual trigger rollback is triggered, a rollback Playbook can be pulled up in seconds, and the whole process of approval and script execution is packaged into a compliance audit report, which supports one-key download.

[0122] Optionally, the generating the work order chain according to the received user demand information comprises:

[0123] convert the received user demand information into target text, the target text being natural language text;

[0124] perform semantic disassembly on the target text to obtain a plurality of feature contents, different feature contents corresponding to different feature types;

[0125] fill the plurality of feature contents into a preset prompt word template to obtain a target prompt word;

[0126] input the target prompt word into a pre-trained large language model to generate a work order, to obtain the work order chain.

[0127] In this embodiment, after receiving the input user demand information, the target text is subjected to semantic disassembly to obtain a plurality of feature contents, and the plurality of feature contents are filled into a preset prompt word template to generate a corresponding prompt word. The prompt word is input into a pre-trained large language model to generate a work order, to obtain the work order chain.

[0128] Specifically, first, a structured Prompt template and a workflow are preset, the user demand is disassembled by LLM and industry knowledge base is called, and is mapped to a DAG dependency relationship scenario-based work order chain. It is assumed here that the length of the scenario-based work order chain is N, that is, there are N work orders in total. Among them, the industry knowledge base is derived from historical work order data, private cloud resource specification documents, approval rule library, operation guide, user access behavior trajectory records indexed by IT systems, etc.

[0129] When the user inputs a natural language demand such as “newly build a development test environment for R&D personnel”, in this process, the preset structured Prompt template is used to guide the LLM to disassemble the user demand. The Prompt template includes inquiries and guidance on delivery scenarios (such as “new IT system” and “create development test environment”), resource types (such as storage, network, and computing resources), operation types (such as new, configuration, and expansion), and specific requirements (such as security level and performance parameters), so that the LLM can convert the user demand into specific work order tasks.

[0130] The preset Prompt template is as follows:

[0131] # Role

[0132] You are an experienced private cloud resource delivery expert, and now you need to convert the natural language demand proposed by the user, such as “newly build a development test environment for R&D personnel”, into a specific scenario-based work order chain.

[0133] ## Skills

[0134] When the user inputs a request, you must invoke the 'getScenOrders' workflow, pass the user's input content to the input of the 'getScenOrders' workflow unchanged, and return the workflow call result.

[0135] ##Limitations

[0136] - Only discuss content related to private cloud resource delivery scenarios and refuse to answer topics unrelated to this.

[0137] - The output content must be organized according to the given format and cannot deviate from the framework requirements.

[0138] The large language model processing process is as follows: semantic decomposition of requirements, extraction of key information, including delivery scenarios delivery_scen ("new IT system", "create development test environment", etc.), resource types resource_type (storage, network, computing, etc.), operation types operation_type (new, expansion, etc.), specific requirements specific_requirements (security level, performance parameters, etc.). Map the decomposed tasks to atomized work orders. By calling the industry knowledge base, match the operation process and work order template related to user demand. For example, creating a development test environment involves "technical scheme review", "MySQL cluster application", "network policy opening" and other atomic work orders. Build a scene-based work order chain with DAG dependency relationship. Clearly define the execution order and dependency relationship between each work order, such as 'technical scheme review → MySQL cluster application work order → network policy opening work order', etc., and output the following structured JSON:

[0139] {

[0140] "work order chain": [

[0141] {"id": 1, "type": "technical scheme review", "specific_requirements": []},

[0142] {"id": 2, "type": "MySQL cluster application", "specific_requirements": [1]},

[0143] {"id": 3, "type": "network policy opening", "specific_requirements": [2]} ]

[0145] }

[0146] In addition, for ambiguous requirements that users may raise, the Copilot front-end interface further increases the requirement clarification interaction mechanism. To ensure that the generated work order chain is consistent with the context and has operability, when the LLM detects ambiguity in the user's requirements, it actively asks questions (such as: "The 'development test environment' you mentioned refers to creating a new development environment or a new test environment?"), and generates different options in a multi-round dialogue manner, allowing the user to confirm:

[0147] Please confirm your scenario requirements:

[0148] [1] Development environment (KVM maximum specification 2C4G)

[0149] [2] Production environment (KVM maximum specification 32C64G)

[0150] The front-end feeds back the user's selection results to the back-end system, so that the system can learn and optimize itself based on user feedback. The back-end system can also store new user requirements and corresponding work order chain generation and verification results in the industry knowledge base, constantly enriching and perfecting the content of the knowledge base, and improving the efficiency and accuracy of subsequent processing of similar requirements.

[0151] As shown in Figure 2 , the system architecture diagram of the service platform in this embodiment includes three core modules: Figure 2

[0152] Intelligent work order aggregation engine: This module has two main functions. One is the generation of scenario-based work order chains. This module uses the LLM+RAG capabilities provided by the AI application development platform to break down user requirements (such as "create database resources") into atomic work order chains (technical scheme review → apply Mysql cluster → apply Mysql service → MySQL program user management), and generates a DAG dependency relationship to avoid manually linking multiple independent work orders. The second is cross-work order field intelligent filling. By establishing a shared field pool for the DAG work order chain that has been constructed, and using Sentence-BERT to calculate the semantic similarity of the fields, a mapping relationship is established for fields that are the same or similar. When the user enters the filling page of the DAG work order chain, the system automatically queries the shared field pool and fills in according to the mapping rules.

[0153] ​The HDBSCAN density clustering module's main function is to vectorize the approval nodes of the DAG work order chain using Sentence-BERT and automatically compress them based on the HDBSCAN algorithm. Sentence-BERT encodes approval nodes (including role + resource type + operation type, such as "Cloud Team_Storage_Approval") into a 768-dimensional vector. Then, it uses the HDBSCAN algorithm's reachability formula to merge similar approvals: dmreach=max{corek(a), corek(b), d(a, b)}, ultimately compressing approval nodes that appear multiple times in the work order chain, such as merging four instances of "Cloud Team Approval" into one.

[0154] Non-standard Resource Automation Adaptation Layer: This module is primarily used for Ansible script generation and automated resource delivery. After users complete work order filling and process approval through the "Intelligent Work Order Aggregation Engine" and "HDBSCAN Density Clustering Module," this module uses the shared field pool established by the DAG work order chain and the filled content to generate an Ansible Playbook or a list of associated cloud resource management platform API interfaces using the LLM and the mounted industry knowledge base. Subsequently, the Ansible Playbook is executed or the list of cloud resource management platform API interfaces is called according to the DAG work order chain to execute batch non-standard resource delivery instructions (such as storage expansion or network policy activation).

[0155] The technical solution of this application generates a work order chain including multiple first work orders based on the received user demand information, and clusters multiple resources included in the multiple first work orders according to semantic similarity to obtain multiple field clusters. Then, based on the work order information and field clusters, the number of N approval nodes corresponding to the multiple work order chains is merged to obtain M approval nodes. The multiple first work orders are then processed according to the M approval nodes to obtain the processing results, thereby simplifying the work order approval process and improving the work order processing efficiency.

[0156] See Figure 3 , Figure 3 This is a structural diagram of the work order processing device provided in the embodiments of this application. For example... Figure 3 As shown, the work order processing device 300 includes:

[0157] The receiving module 310 is configured to generate a work order chain according to the received user demand information, the work order chain comprising a plurality of first work orders and work order information, the first work orders being used to request application for use of private cloud resources in the service platform, the work order information being used to indicate a dependency relationship between the plurality of first work orders, the dependency relationship being used to indicate an execution sequence of the plurality of first work orders, the work order chain corresponding to N approval nodes in the service platform, N being a positive integer greater than 1;

[0158] The clustering module 320 is configured to cluster a plurality of fields included in the plurality of first work orders according to semantic similarity, to obtain a plurality of field clusters, each of the field clusters comprising at least two fields.

[0159] The merging module 330 is configured to merge the N approval nodes based on the work order information and the plurality of field clusters, to obtain M approval nodes, M being a positive integer less than N.

[0160] The processing module 340 is configured to process the plurality of first work orders based on the M approval nodes, to generate a processing result.

[0161] Optionally, the clustering module 320 comprises:

[0162] The first extracting submodule is configured to extract text content included in the plurality of first work orders, to obtain a plurality of fields.

[0163] The first converting submodule is configured to convert the plurality of fields into a plurality of field vectors, the plurality of fields and the plurality of field vectors corresponding one-to-one.

[0164] The computing submodule is configured to perform semantic similarity calculation on any two field vectors in the plurality of field vectors, to obtain similarity information, the similarity information comprising semantic similarity between any two field vectors in the plurality of field vectors.

[0165] The first clustering submodule is configured to cluster the plurality of fields according to the similarity information, to obtain the plurality of field clusters.

[0166] Optionally, the clustering submodule comprises:

[0167] The screening unit is configured to screen a plurality of semantic similarities included in the similarity information based on a preset threshold, to obtain a plurality of target semantic similarities, the target semantic similarities being greater than or equal to the preset threshold, each of the target semantic similarities corresponding to two fields.

[0168] The aggregation unit is configured to aggregate the two fields corresponding to each of the target semantic similarities into a same field cluster, to obtain the plurality of field clusters.

[0169] Optionally, the merging module 330 comprises:

[0170] a first generating submodule, configured to generate an approval node relationship graph based on the work order information and the N approval nodes, the approval node relationship graph being used to represent a dependency relationship between the N approval nodes;

[0171] a second extracting submodule, configured to input the approval node relationship graph into a pre-trained graph neural network model to perform feature extraction, to obtain N graph features, the N approval nodes and the N graph features corresponding one by one;

[0172] a second clustering submodule, configured to perform clustering on the N graph features based on a density-adaptive hierarchical clustering algorithm, to obtain a plurality of clustering results, each of the clustering results comprising at least two graph features;

[0173] a merging submodule, configured to merge, in the N approval nodes, at least two approval nodes corresponding to at least two graph features included in each of the clustering results, to obtain the M approval nodes.

[0174] Optionally, the processing module 340 comprises:

[0175] a determining submodule, configured to determine, based on a preset mapping relationship table, a plurality of target text contents corresponding one by one to the plurality of field clusters in a database, wherein the database comprises a plurality of preset text contents, and the preset mapping relationship table comprises a plurality of mapping relationships, each of the mapping relationships being used to indicate, in the plurality of preset text contents, a field cluster corresponding to each preset text content; and perform information filling on the plurality of first work orders based on the plurality of target text contents, to obtain a plurality of second work orders, the plurality of second work orders corresponding one by one to the plurality of first work orders;

[0176] a processing submodule, configured to perform processing on the plurality of second work orders based on the M approval nodes, to generate a processing result.

[0177] Optionally, the receiving module 310 comprises:

[0178] a second converting submodule, configured to convert the received user demand information into a target text, the target text being a natural language text;

[0179] a disassembling submodule, configured to perform semantic disassembly on the target text, to obtain a plurality of feature contents, different feature contents corresponding to different feature types;

[0180] a filling-in submodule, configured to fill the plurality of feature contents into a preset prompt word template, to obtain a target prompt word;

[0181] A second human generation sub-module is configured to input the target prompt word into a pre-trained large language model to generate a work order chain.

[0182] According to the technical solution of the present application, the work order chain including a plurality of first work orders is generated according to the received user demand information, and the plurality of resources included in the plurality of first work orders are clustered according to semantic similarity, thereby obtaining a plurality of field clusters. Thus, the number of N approval nodes corresponding to the plurality of work order chains is merged according to the work order information and the field clusters, thereby obtaining M approval nodes. Thus, the plurality of first work orders are processed according to the M approval nodes, thereby obtaining a processing result. The technical solution of the present application realizes the simplification of the approval process of the work order and improves the work order processing efficiency.

[0183] The present application also provides an electronic device. Please refer to Figure 4 The electronic device can include a processor 401, a memory 402, and a program 4021 stored in the memory 402 and executable on the processor 401.

[0184] When the program 4021 is executed by the processor 401, it can implement Figure 1 any step in the corresponding method embodiment:

[0185] According to the received user demand information, a work order chain is generated, the work order chain including a plurality of first work orders and work order information, the first work order being used to request the application for using private cloud resources in the service platform, the work order information being used to indicate the dependency relationship between the plurality of first work orders, the dependency relationship being used to indicate the execution order of the plurality of first work orders, the work order chain corresponding to N approval nodes in the service platform, N being a positive integer greater than 1;

[0186] According to the semantic similarity, a plurality of fields included in the plurality of first work orders are clustered to obtain a plurality of field clusters, each field cluster including at least two fields;

[0187] Based on the work order information and the plurality of field clusters, the N approval nodes are merged to obtain M approval nodes, M being a positive integer less than N;

[0188] Based on the M approval nodes, the plurality of first work orders are processed to generate a processing result.

[0189] Optionally, the clustering of the plurality of fields included in the plurality of first work orders according to the semantic similarity to obtain a plurality of field clusters includes:

[0190] The text content included in the plurality of first work orders is extracted to obtain a plurality of fields;

[0191] convert the plurality of fields into a plurality of field vectors, the plurality of fields corresponding to the plurality of field vectors one by one;

[0192] perform semantic similarity calculation on any two field vectors in the plurality of field vectors to obtain similarity information, the similarity information including semantic similarities between any two field vectors in the plurality of field vectors;

[0193] cluster the plurality of fields according to the similarity information to obtain a plurality of field clusters.

[0194] Optionally, the clustering the plurality of fields according to the similarity information to obtain the plurality of field clusters includes:

[0195] filter a plurality of semantic similarities included in the similarity information based on a preset threshold to obtain a plurality of target semantic similarities, the target semantic similarity being greater than or equal to the preset threshold, and each target semantic similarity corresponding to two fields;

[0196] aggregate the two fields corresponding to each target semantic similarity into a same field cluster to obtain the plurality of field clusters.

[0197] Optionally, the merging the N approval nodes based on the work order information and the plurality of field clusters to obtain M approval nodes includes:

[0198] generating an approval node relationship graph based on the work order information and the N approval nodes, the approval node relationship graph being used to represent a dependency relationship between the N approval nodes;

[0199] inputting the approval node relationship graph into a pre-trained graph neural network model to perform feature extraction to obtain N graph features, the N approval nodes corresponding to the N graph features one by one;

[0200] performing clustering on the N graph features based on a density adaptive hierarchical clustering algorithm to obtain a plurality of clustering results, each clustering result including at least two graph features;

[0201] merging at least two approval nodes corresponding to the at least two graph features included in each clustering result in the N approval nodes to obtain the M approval nodes.

[0202] Optionally, the processing the plurality of first work orders based on the M approval nodes to generate a processing result includes:

[0203] determine a plurality of target text contents corresponding to the plurality of field clusters one by one based on a preset mapping relationship table in a database, wherein the database includes a plurality of preset text contents, the preset mapping relationship table includes a plurality of mapping relationships, and each mapping relationship is used to indicate, in the plurality of preset text contents, a field cluster corresponding to each preset text content; perform information filling on the plurality of first work orders based on the plurality of target text contents to obtain a plurality of second work orders, the plurality of second work orders corresponding to the plurality of first work orders one by one;

[0204] perform processing on the plurality of second work orders based on the M approval nodes to generate a processing result.

[0205] Optionally, the generating a work order chain according to the received user demand information comprises:

[0206] converting the received user demand information into a target text, the target text being a natural language text;

[0207] performing semantic disintegration on the target text to obtain a plurality of feature contents, different feature contents corresponding to different feature types;

[0208] filling the plurality of feature contents into a preset prompt word template to obtain a target prompt word;

[0209] inputting the target prompt word into a pre-trained large language model to generate a work order, and obtaining the work order chain.

[0210] The technical solution of the present application generates a work order chain including a plurality of first work orders according to received user demand information, and clusters a plurality of resources included in the plurality of first work orders according to semantic similarity, thereby obtaining a plurality of field clusters. Thereby, the number of N approval nodes corresponding to the plurality of work order chains is merged according to the work order information and the field clusters, thereby obtaining M approval nodes. Thereby, the plurality of first work orders are processed according to the M approval nodes to obtain a processing result, which realizes a simplified approval process of work orders and improves the work order processing efficiency.

[0211] The embodiment of the present application also provides a computer readable storage medium, and the computer readable storage medium stores a computer program. The computer program is executed by a processor to implement each process of the above-mentioned work order processing embodiment and achieve the same technical effect. To avoid repetition, details are not described here. The computer readable storage medium includes a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0212] The embodiment of the application further provides a computer program product stored in a storage medium, which is executed by at least one processor to implement each process of the work order processing method embodiment, and can achieve the same technical effects. To avoid repetition, details are not described herein.

[0213] It should be noted that in this document, the terms "comprising", "including", or any other variant thereof are intended to cover non-exclusive inclusions, so that processes, methods, articles or devices including a series of elements not only include those elements, but also include other elements not explicitly listed, or include elements inherent to such processes, methods, articles or devices. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0214] From the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment method can be realized by means of software and necessary general hardware platform, of course, it can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes a plurality of instructions for making a communication device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) execute the methods described in various embodiments of the present application.

[0215] The embodiments of the application are described above in combination with the drawings, but the application is not limited to the above specific embodiments, and the above specific embodiments are only illustrative, not restrictive, and those skilled in the art can make many forms under the inspiration of the application without departing from the scope of the application and the protection scope of the claims.

Claims

1. A method for processing work orders, characterized in that, Applied to a service platform, the method includes: Based on the received user request information, a work order chain is generated. The work order chain includes multiple first work orders and work order information. The first work order is used to request the use of private cloud resources in the service platform. The work order information is used to indicate the dependency relationship between the multiple first work orders. The dependency relationship is used to indicate the execution order of the multiple first work orders. The work order chain corresponds to N approval nodes in the service platform, where N is a positive integer greater than 1. Based on semantic similarity, the multiple fields included in the multiple first work orders are clustered to obtain multiple field clusters, and each field cluster includes at least two fields; Based on the work order information and the multiple field clusters, the N approval nodes are merged to obtain M approval nodes, where M is a positive integer less than N; Based on the M approval nodes, the multiple first work orders are processed to generate processing results.

2. The method according to claim 1, characterized in that, The method of clustering multiple fields included in the multiple first work orders based on semantic similarity yields multiple field clusters, including: The text content of the multiple first work orders is extracted to obtain multiple fields; The multiple fields are converted into multiple field vectors, and the multiple fields correspond one-to-one with the multiple field vectors; Semantic similarity is calculated for any two field vectors among the plurality of field vectors to obtain similarity information, wherein the similarity information includes the semantic similarity between any two field vectors among the plurality of field vectors; Based on the similarity information, the multiple fields are clustered to obtain the multiple field clusters.

3. The method according to claim 2, characterized in that, The step of clustering the multiple fields based on the similarity information to obtain the multiple field clusters includes: Based on a preset threshold, multiple semantic similarities included in the similarity information are filtered to obtain multiple target semantic similarities. The target semantic similarities are greater than or equal to the preset threshold, and each target semantic similarity corresponds to two fields. The two fields corresponding to each target semantic similarity are aggregated into the same field cluster to obtain the multiple field clusters.

4. The method according to claim 1, characterized in that, Based on the work order information and the multiple field clusters, the N approval nodes are merged to obtain M approval nodes, including: An approval node relationship diagram is generated based on the work order information and the N approval nodes. The approval node relationship diagram is used to represent the dependency relationship between the N approval nodes. The approval node relationship graph is input into a pre-trained graph neural network model for feature extraction, resulting in N graph features. The N approval nodes correspond one-to-one with the N graph features. A density-adaptive hierarchical clustering algorithm is used to cluster the N graph features to obtain multiple clustering results, each of which includes at least two graph features. In the N approval nodes, at least two approval nodes corresponding to at least two graph features included in each clustering result are merged to obtain the M approval nodes.

5. The method according to claim 1, characterized in that, The process of processing the multiple first work orders based on the M approval nodes and generating processing results includes: Based on a preset mapping table, multiple target text contents corresponding one-to-one with the multiple field clusters are determined in the database. The database includes multiple preset text contents, and the preset mapping table includes multiple mapping relationships. Each mapping relationship is used to indicate the field cluster corresponding to each preset text content. Based on the multiple target text contents, information is filled into the multiple first work orders to obtain multiple second work orders, and the multiple second work orders correspond one-to-one with the multiple first work orders. Based on the M approval nodes, the multiple second work orders are processed to generate processing results.

6. The method according to claim 1, characterized in that, The step of generating a work order chain based on the received user request information includes: The received user request information is converted into target text, which is natural language text. The target text is semantically decomposed to obtain multiple feature contents, and different feature contents correspond to different feature types; The multiple feature contents are filled into a preset prompt word template to obtain the target prompt word; The target prompt word is input into a pre-trained large language model to generate work orders, thus obtaining the work order chain.

7. A work order processing device, characterized in that, The device, applied to a service platform, includes: The receiving module is used to generate a work order chain based on the received user request information. The work order chain includes multiple first work orders and work order information. The first work orders are used to request the use of private cloud resources in the service platform. The work order information is used to indicate the dependency relationship between the multiple first work orders. The dependency relationship is used to indicate the execution order of the multiple first work orders. The work order chain corresponds to N approval nodes in the service platform, where N is a positive integer greater than 1. The clustering module is used to cluster multiple fields included in the multiple first work orders according to semantic similarity to obtain multiple field clusters, each of the field clusters including at least two fields; The merging module is used to merge the N approval nodes based on the work order information and the multiple field clusters to obtain M approval nodes, where M is a positive integer less than N; The processing module is used to process the multiple first work orders based on the M approval nodes and generate processing results.

8. An electronic device, characterized in that, include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, Includes computer instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1 to 6.

Images