Service work order classification method, electronic device, storage medium and program product

Abstract

The present application provides a service work order classification method, an electronic device, a storage medium, and a program product, belonging to the field of artificial intelligence. The method includes: performing a clustering operation on at least one sample service work order to obtain at least one service work order set; based on a first large language model and first prompt word information, performing scenario recognition processing on the sample service work orders included in the service work order set to obtain at least one work order problem scenario and the sample service work orders corresponding to each work order problem scenario in the service work order set, and each work order problem scenario includes a problem phenomenon field, a key error reporting field, a troubleshooting solution field, a processing solution field, and the field values of these self-fields. The present application can automatically determine the work order problem scenario to which the service work order belongs, and the work order problem scenario includes multiple fields and field values, so the classification result is more accurate and the usability is stronger.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and particularly to a method for classifying service work orders, an electronic device, a storage medium, and a program product. Background Art

[0002] A service work order refers to a record created by an enterprise through a customer service management system or other similar tools for tracking and managing customer problems or needs. As a service request management tool, service work orders play an important role in the enterprise operation process. They can not only effectively manage customer problems and needs, improve user satisfaction, but also improve the work efficiency of users and help enterprises improve products and services. In order to better process service work orders, it is necessary to classify service work orders.

[0003] Related technologies classify service work orders manually. The classification results highly depend on the experience of users, and the information content of the classification results is less, resulting in inaccurate and unusable classification results. Summary of the Invention

[0004] Embodiments of this application provide a method for classifying service work orders, an electronic device, a storage medium, and a program product, which can automatically determine the work order problem scenarios to which the service work orders belong. Each work order problem scenario includes multiple fields and field values, and the classification results are more accurate and have higher usability. The technical solutions are as follows:

[0005] In a first aspect, a method for classifying service work orders is provided. The method includes:

[0006] Performing a clustering operation on at least one sample service work order to obtain at least one service work order set, where any service work order set included in the at least one service work order set includes sample service work orders with the same work order category;

[0007] For any service work order set, based on a first large language model and first prompt information, performing scenario recognition processing on the sample service work orders included in the any service work order set to obtain at least one work order problem scenario and the sample service work orders corresponding to each work order problem scenario in the any service work order set. Each work order problem scenario includes a problem scenario description field and its field value, and the problem scenario description field includes a problem phenomenon field, a key error field, a troubleshooting plan field, and a handling plan field.

[0008] In a second aspect, a device for classifying service work orders is provided. The device includes:

[0009] A clustering module, configured to perform a clustering operation on at least one sample service work order to obtain at least one set of service work orders, where each set of service work orders in the at least one set of service work orders includes sample service work orders of the same work order category;

[0010] An identification module, configured to, for any set of service work orders, perform scenario identification processing on the sample service work orders included in the any set of service work orders based on a first large language model and first prompt information, to obtain at least one work order problem scenario and the sample service work orders corresponding to each work order problem scenario in the any set of service work orders, where each work order problem scenario includes a problem scenario description field and its field value, and the problem scenario description field includes a problem phenomenon field, a key error field, a troubleshooting solution field, and a handling solution field.

[0011] In a third aspect, an electronic device is provided, including a processor and a memory; the memory stores at least one piece of program code; the at least one piece of program code is used to be called and executed by the processor to implement the service work order classification method described in the first aspect.

[0012] In a fourth aspect, a computer-readable storage medium is provided, in which at least one computer program is stored, and when the at least one computer program is executed by a processor, it can implement the service work order classification method described in the first aspect.

[0013] In a fifth aspect, a computer program product is provided, the computer program product includes a computer program, and when the computer program is executed by a processor, it can implement the service work order classification method described in the first aspect.

[0014] The beneficial effects brought by the technical solution provided in the embodiments of the present application are:

[0015] In the embodiments of the present application, by clustering at least one sample service work order, at least one service work order set can be obtained. Each service work order set in the at least one service work order set includes sample service work orders of the same work order category. Then, using the first large language model plus the first prompt information, scene recognition is performed on the sample service work orders included in each service work order set, obtaining at least one work order problem scene and the sample service work orders corresponding to each work order problem scene in any service work order set. The work order problem scene includes a problem scene description field and its field values. The problem scene description field includes a problem phenomenon field, a key error field, a troubleshooting plan field, a handling plan field, etc., and can comprehensively describe the situation of the service work order. Since there is no need to classify manually and with the help of the large language model, the work order problem scene to which the sample service work order belongs can be automatically determined, and the determined work order problem scene is summarized from the sample service work orders of the actual scene by the large language model, which can accurately reflect the category of the sample service work order. Therefore, the classification result is more accurate, the classified category is more universal, and the usability is stronger. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0017] Figure 1 It is a flowchart of a service work order classification method provided by an embodiment of the present application;

[0018] Figure 2 It is a flowchart of a training method of a vector conversion model provided by an embodiment of the present application;

[0019] Figure 3 It is a schematic diagram of the index structure of the work order knowledge base constructed in the embodiments of the present application;

[0020] Figure 4 It is an overall flowchart of a service work order classification provided by an embodiment of the present application;

[0021] Figure 5 It is a flowchart of another service work order classification method provided by an embodiment of the present application;

[0022] Figure 6 It is a schematic diagram of the structure of a service work order classification device provided by an embodiment of the present application;

[0023] Figure 7 It shows a structural block diagram of an electronic device provided by an exemplary embodiment of the present application. Detailed implementation manners

[0024] To make the objectives, technical solutions and advantages of this application clearer, the following will further describe the implementation manners of this application in detail with reference to the accompanying drawings.

[0025] It can be understood that the terms "each", "multiple" and "any one" used in the embodiments of this application, etc., multiple includes two or more, each refers to each one of the corresponding multiple, and any one refers to any one of the corresponding multiple. For example, multiple words include 10 words, and each word refers to each of these 10 words, and any one word refers to any one of the 10 words.

[0026] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with the relevant laws, regulations and standards of relevant countries and regions, and corresponding operation entrances are provided for users to choose to authorize or refuse.

[0027] Artificial intelligence is to use a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, a theory, method, technology and application system that perceives the environment, acquires knowledge and uses knowledge to obtain better results. In other words, artificial intelligence is a comprehensive technology in computer science. It attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a way similar to human intelligence. Artificial intelligence also studies the design principles and implementation methods of various intelligent machines, so that the machines have the functions of perception, reasoning and decision-making.

[0028] Artificial intelligence technology is an interdisciplinary subject, involving a wide range of fields, including both hardware-level technologies and software-level technologies. The basic technologies of artificial intelligence generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. The software technologies of artificial intelligence mainly include several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0029] With the research and progress of artificial intelligence technology, artificial intelligence technology has been studied and applied in multiple fields. For example, common ones include smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, driverless, autonomous driving, robots, smart healthcare, smart customer service, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role. The solution provided by the embodiments of this application involves technologies such as natural language processing and machine learning / deep learning of artificial intelligence.

[0030] As described above, service work orders play an important role in the operation of an enterprise. Therefore, how to process service work orders is crucial. Different service work orders have different problems, and the processing methods adopted are also different. In order to better process service work orders, before processing service work orders, it is necessary to classify service work orders, so as to timely process the problems existing in service work orders based on the classification results, improve user satisfaction, and enhance the service experience of products.

[0031] To improve the processing efficiency of service work orders, currently, related technologies classify historical service work orders based on user experience or traditional categories, and then build a work order knowledge base based on the classification results of historical service work orders. However, user experience depends on users' understanding of product problems, and traditional categories depend on the understanding of previous products. Whether it is the classification method based on user experience or the classification method based on traditional categories, the determined classification results may deviate from the actual application scenarios of products, resulting in inaccurate classification results, lack of universality and usability. Furthermore, since the work order knowledge base is built based on historical service work orders and their classification results, in the case of inaccurate classification results, the built work order knowledge base will not be accurate either. And because the classification of the work order knowledge base is not universal and usable, a large number of processing solutions in the work order knowledge base cannot be applied and become invalid solutions.

[0032] In addition, when expanding the work order knowledge base, the related technology adopts a stacking method. The so-called stacking method means that when optimizing and merging similar solutions, there may be a large number of different solutions for the same problem. Users write down a solution as they find one, and quickly increase the number of documents in this classification by means of exhaustive enumeration. However, when searching using this work order knowledge base, often a single problem will hit multiple processing solutions, and users cannot distinguish which processing solution is truly effective. They may piece together the hit processing solutions to obtain the final processing solution, and this pieced-together final processing solution may not be able to solve the problem. Facing multiple processing solutions for the same problem, users may also nest multiple processing solutions. The so-called nesting means that when there are multiple processing solutions for the same problem and there are differences among multiple processing solutions, when adding a new processing solution, users will reference the links of similar processing solutions and then add new steps, resulting in multiple nesting of some documents. In actual problem handling, it often cannot solve the problem, and a large number of out-of-scheme operations are required for adjustment, increasing the probability of triggering faults and prolonging the problem handling duration.

[0033] Currently, in the scenario of service work order processing, users are more eager to know whether the processing solution is suitable for the current problems that occur, and then conduct adaptation verification. This makes it more important to determine the problems of service work orders than simply obtaining the processing solutions. However, the classification results determined by the methods of related technologies are inaccurate, and the classification results are usually given based on the problem phenomena, so the determined problem phenomena are inaccurate. Moreover, the current service work order processing methods all give the processing solutions on the premise of clarifying the problem phenomena. When the determined problem phenomena are inaccurate, the processing solutions given based on the problem phenomena are also inaccurate, thus increasing the risk of secondary failures.

[0034] In addition, in the real service work order processing scenario, determining the root cause of the problems in the service work order helps to associate the problem phenomena, troubleshooting solutions, and processing solutions according to this anchor point. Then, based on this association result, regularly troubleshoot and solve the problems existing in the service work order to eliminate the influence of human judgment factors and improve the processing efficiency of service work orders. However, the classification results determined by related technologies only give the problem phenomena and processing solutions, and do not give information in other dimensions, nor do they associate the information in these dimensions, resulting in less information in the determined classification results. Based on this work order knowledge base, the processing efficiency of service work orders is low and the accuracy is not high.

[0035] To solve the problems of low accuracy and lack of usability in classifying service work orders by related technologies, and at the same time meet the needs of users in the service work order processing scenario, the embodiments of this application propose the concept of 4K. The 4K specifically includes: K1: Problem Phenomena (description of on-site problem manifestations and root causes), K2: Key Error Reports (referring to the core logs, error messages, or API call failures, etc. generated along with the problem root cause), K3: Troubleshooting Solutions (a set of methods for locating the problem root cause), and K4: Processing Solutions (implementation methods for specifically solving the above problems). Then, at the 4K level, by leveraging the powerful processing capabilities of large language models, automatically generate scenario-based 4K tags for sample service work orders in real scenarios, associate the scenario-based 4K tags, avoid the interference of human factors, improve the accuracy and usability of the classification results, and increase the probability of the processing solution hitting the problem.

[0036] Based on the proposed 4K concept and by leveraging large language models, the embodiments of this application identify the work order problem scenarios (scenario-based 4K tags) for sample service work orders through multiple operation steps such as text preprocessing (including data cleaning, formatting processing, etc.), vectorization training (including training vector conversion models), refined classification (including clustering to obtain service work order sets and generating work order problem tags for service work order sets), and scenario intelligent generation.

[0037] An embodiment of the present application provides a method for classifying service work orders. Taking the execution of the embodiment of the present application by an electronic device as an example, the electronic device can be a terminal with strong computing power, such as a laptop, a desktop computer, etc., or a server. The server can be a single physical server or a cluster or distributed system composed of multiple physical servers. See Figure 1 , the method flow provided by the embodiment of the present application includes:

[0038] 101. Determine the target feature vector corresponding to each of at least one sample service work order.

[0039] Among them, the sample service work orders include historical service work orders and newly added service work orders, etc. The historical service work order refers to a service work order generated before a certain moment and labeled with a work order category label by a person. This moment can be the moment for review after the processing of product-related service work orders is completed. The work order category label marked on the historical service work order can be marked manually, and the marked work order category label is an N-level label, where N is greater than or equal to 1. Different levels of labels can reflect the abstraction degree of the problem causes and processing methods of service work orders, and are obtained by gradually refining the problem causes and processing methods of service work orders from the product level. In the embodiment of the present application, the N-level labels marked on the historical service work orders can include first-level work order category labels and second-level work order category labels, etc. The newly added service work order in the embodiment of the present application refers to a service work order generated before a certain moment but not labeled with a work order category label.

[0040] Optionally, there may be some sensitive data and some useless data in the generally obtained sample service work orders. To avoid the leakage of sensitive information and reduce the resources consumed in the process of processing sample service work orders, data cleaning can be performed on at least one sample service work order. When performing data cleaning on at least one sample service work order, the cleaning content can be determined first, and then the cleaning order can be determined for the determined cleaning content. Furthermore, a regular expression written by the user based on the determined cleaning content and cleaning order is obtained, and by executing this regular expression, data cleaning is performed on each sample service work order to obtain at least one cleaned sample service work order. Among them, the cleaning content includes sensitive fields, Htttp links, solution codes, etc. The cleaning order can be determined by the user according to actual cleaning requirements. For example, clean the Htttp link first, then clean the solution code, and finally determine the sensitive fields, etc.

[0041] Furthermore, after at least one cleaned sample service work order is obtained, a preset format may be used to format the at least one cleaned sample service work order to unify the data format in the at least one cleaned sample service work order, thereby obtaining at least one structured sample service work order. The preset format may be a structured data format including specified fields and field values. For example, when the specified fields include fields such as work order title, troubleshooting results, and troubleshooting process, the preset format may be "work order title: troubleshooting results: troubleshooting process:".

[0042] The embodiment of the present application can avoid sensitive data leakage, reduce the computing resources consumed in the service work order processing process, and improve the processing efficiency of the service work order by performing data cleaning, formatting and other processing on at least one sample service work order.

[0043] Among them, the target feature vector can update the distance in the vector space of sample service work orders belonging to the same work order category. Specifically, the target feature vectors corresponding to the sample service work orders of the same category have similarities in the high-dimensional vector space, and the target feature vectors corresponding to the sample service work orders of different categories have differences in the high-dimensional vector space. When determining the target feature vector corresponding to at least one sample service work order, the embodiments of the present application may adopt some algorithms that can determine the target feature vector, and may also use the trained model. Taking the use of the trained model to determine the target feature vector corresponding to at least one sample service work order as an example, the determination process may include the following steps:

[0044] 1011. Call a vector initialization model to perform feature extraction on at least one sample service work order to obtain an initial feature vector corresponding to the at least one sample service work order.

[0045] The vector initialization model is used to convert each sample service ticket into a high-dimensional vector form. The vector initialization model is a model that can convert text into computer-recognizable characters, such as the BGE model. When each sample service ticket is input into the vector initialization model, it is processed by the vector initialization model to obtain the initial feature vector corresponding to each sample service ticket.

[0046] 1012. Call a vector conversion model to convert an initial feature vector of at least one sample service work order to obtain a target feature vector corresponding to the at least one sample service work order.

[0047] Among them, the vector conversion model is used to perform conversion processing on the initial feature vectors of sample service work orders, change the positions of the initial feature vectors in the high-dimensional space, and obtain the target feature vectors corresponding to each sample service work order. The target feature vectors corresponding to sample service work orders of the same category have similarity in the high-dimensional vector space, and the target feature vectors corresponding to sample service work orders of different categories have differences in the high-dimensional vector space. Based on the geometric properties of the target feature vectors in the high-dimensional vector space, when clustering sample service work orders, sample service work orders with the same semantic information can be clustered into one category. The vector conversion model has a neural network structure, which can be a ResNet model, etc. The vector conversion model can be trained using historical service work orders, and the historical service work orders are labeled with N-level work order category labels, such as the first-level work order category label and the second-level work order category label, etc. Figure 2 The training process of the vector conversion model is shown, and the training process includes the following steps:

[0048] 201. Invoke the vector initialization model to determine the initial feature vectors corresponding to multiple historical service work orders.

[0049] 202. Invoke the vector conversion model to be trained to perform conversion processing on the initial feature vectors corresponding to multiple historical service work orders, and obtain the target feature vectors corresponding to multiple historical service work orders.

[0050] 203. According to the N-level work order category labels and target feature vectors corresponding to multiple historical service work orders respectively, determine N loss function values corresponding to the N-level work order category labels.

[0051] For the N-level work order category labels marked on the historical service work orders, N classification tasks can be set. Each classification task corresponds to a first-level classification label, and each classification task corresponds to a loss function. Each loss function can be a cross-entropy function. The N loss function values respectively reflect that the target feature vectors corresponding to historical service work orders with the same-level work order category labels are closer. Taking the N-level work order category labels as the first-level work order category label and the second-level work order category label as an example, the constructed loss function includes the first loss function and the second loss function. The first loss function corresponds to the first-level work order category label, and the first loss function value reflects that the target feature vectors corresponding to historical service work orders with the same first-level work order category label are closer; the second loss function corresponds to the second-level work order category label, and the second loss function value reflects that the target feature vectors corresponding to historical service work orders with the same second-level work order category label are closer.

[0052] 204. Based on the N loss function values, determine the target loss function value to train the vector conversion model to be trained according to the target loss function value.

[0053] Using the weight values corresponding to the N loss function values respectively, the N loss function values are weighted and summed to obtain the target loss function value. If the obtained target loss function value is greater than the preset threshold, the model parameters of the vector conversion model to be trained are adjusted, and the above steps are executed based on the vector conversion model with adjusted parameters until the training cutoff condition is met. The training cutoff condition includes that the number of adjustments reaches the set number, or the target loss function value is less than the preset threshold, etc. The vector conversion model corresponding to when the training cutoff condition is met is obtained, and this vector conversion model is used as the trained vector conversion model. The historical service work orders in the embodiments of the present application have multi-level labels, and the vector conversion model trained for the classification task set for the multi-level labels can enable the target feature vector to have good classification results on different-level labels.

[0054] Based on the trained vector conversion model, for the historical service work orders in at least one sample service work order, the initial feature vector corresponding to the historical service work order can be input into the trained vector conversion model to output the work order target feature vector corresponding to the historical service work order; for the new service work orders in at least one sample service work order, the initial feature vector corresponding to the new service work order can be input into the trained vector conversion model to output the work order target feature vector corresponding to the new service work order. By using the trained feature conversion model to process the historical service work orders and new service work orders in the embodiments of the present application, the accuracy of the determined target feature vector is improved.

[0055] 102. Based on the target feature vectors corresponding to at least one sample service work order respectively, perform a clustering operation on at least one sample service work order to obtain at least one service work order set.

[0056] In order to be able to extract the common attributes of the sample service work orders of the same category, after obtaining the target feature vectors corresponding to at least one sample service work order respectively, a clustering operation will also be performed on at least one sample service work order, so as to obtain at least one service work order set. Each service work order set includes at least one sample service work order, and the work order categories of the sample service work orders included in the same service work order set are the same. When performing a clustering operation on at least one sample service work order based on the target feature vectors corresponding to at least one sample service work order respectively, a preset clustering algorithm (such as the K-means clustering algorithm, etc.) can be used for clustering, or clustering can also be performed based on the trained clustering model. Of course, other methods can also be used for clustering, which will not be elaborated here one by one. Taking the use of a preset clustering algorithm for clustering as an example, the clustering process includes the following steps:

[0057] 1021. Using a preset clustering algorithm, perform a clustering operation on at least one target feature vector corresponding to at least one sample service work order to obtain at least one cluster.

[0058] When clustering at least one sample service work order using a preset clustering algorithm, a preset number of categories can be set in advance, and then by calculating the distances (such as Euclidean distance, cosine similarity, etc.) between the target feature vectors of at least one work order, the target feature vectors of work orders with distances less than the distance threshold are grouped into a cluster. After clustering, at least one cluster can be finally obtained, and each cluster includes at least one target feature vector.

[0059] Optionally, the multiple historical service work orders included in at least one sample service work order are labeled with N-level work order category labels. To improve the speed of the clustering operation, according to the N-level work order category labels labeled on the multiple historical service work orders, the historical service work orders with the same N-level work order category labels are divided into an initial cluster to obtain at least one initial cluster. Then, calculate the distances between the target feature vectors of the new service work orders in at least one sample service work order and the target feature vectors of each historical service work order in each initial cluster. If the distances between the target feature vectors of any new service work order and the target feature vectors of each historical service work order in any initial cluster are all less than the distance threshold, add this new service work order to this initial cluster. This method is used to process other new service work orders in at least one sample service work order, and finally at least one cluster can be obtained.

[0060] 1022. Combine the sample service work orders corresponding to the target feature vectors in each cluster to form an initial service work order set, and obtain at least one initial service work order set.

[0061] 1023. In response to the user's adjustment operation, adjust at least one initial service work order set to obtain at least one service work order set.

[0062] The adjustment process for at least one initial service work order combination includes: judging whether the preset number of categories set in advance needs to be adjusted. If it does not need to be adjusted, manually check whether at least one initial service work order set needs to be adjusted. If no manual adjustment is required, use at least one initial service work order set as at least one service work order; if manual adjustment of at least one initial service work order set is required, judge whether any initial service work order set in at least one initial service work order set needs to be split. If any service work order set in at least one initial service work order set needs to be split, re-use the preset clustering algorithm to split this service work order set. If no service work order set in at least one initial service work order set needs to be split, merge the same or similar initial service work order sets in at least one initial service work order set, and finally obtain at least one service work order set.

[0063] The target feature vectors in the embodiments of the present application can enable the target feature vectors corresponding to sample service work orders of the same category to have similarity in the high-dimensional vector space, and the target feature vectors corresponding to sample service work orders of different categories to have differences in the high-dimensional vector space. By performing a clustering operation on at least one sample service work order based on the target feature vectors corresponding to each of them, the sample service work orders with the same work order category can be clustered into a service work order set, so as to perform scenario recognition processing on the service work order set with the same work order category in subsequent operations, ensuring the accuracy of the recognition result.

[0064] 103. For any service work order set, based on the first large language model and the first prompt word information, perform scenario recognition processing on the sample service work orders included in any service work order set to obtain at least one work order problem scenario and the sample service work orders corresponding to each work order problem scenario in any service work order set.

[0065] Among them, at least one sample service work order belonging to the service work order set is mounted under each work order problem scenario. Each work order problem scenario includes a problem scenario description field and its field values. The problem scenario description field includes a problem phenomenon field, a key error field, a troubleshooting solution field, a handling solution field, etc. For any service work order set, based on the first large language model and the first prompt word information, performing scenario recognition processing on the sample service work orders included in any service work order set can obtain at least one work order problem scenario and the sample service work orders corresponding to each work order problem scenario in any service work order set. Specifically, when processing, the first prompt word information can be generated based on each sample service work order included in any service work order set, and the first prompt word information is input into the first large language model to enable the first large language model to generate at least one work order problem scenario corresponding to any service work order set; it is also possible to batch process the sample service work orders included in any service work order set to obtain at least one batch of sample service work orders. For each batch of sample service work orders, generate the first prompt word information corresponding to at least one batch of sample service work orders, and then input the first prompt word information corresponding to at least one batch of sample service work orders into the first large language model to enable the first large language model to generate at least one work order problem scenario corresponding to at least one batch of sample service work orders. Of course, other processing methods can also be used, and the embodiments of the present application will not elaborate one by one. Taking the batch processing of the sample service work orders included in any service work order set as an example, the following steps can be included:

[0066] 1031. Batch process the sample service work orders included in any service work order set to obtain at least one batch of sample service work orders.

[0067] Limited by the input token number of the large language model, when the number of sample service work orders included in the service work order set is large, it is necessary to process the sample service work orders included in the service work order set in batches to obtain at least one batch of sample service work orders. The number of sample service work orders in each batch can be 5, 10, etc., and the embodiments of the present application do not make specific limitations on this.

[0068] 1032. Generate at least one batch of first prompt word information corresponding to the sample service work orders in sequence.

[0069] Fill at least one batch of sample service work orders into the corresponding positions of the first prompt template to obtain at least one batch of first prompt word information corresponding to the sample service work orders. Among them, the first prompt template is used to instruct the first large language model to extract the field values of the problem phenomenon field and the key error field from the filled sample service work orders. The content of the first prompt template can be: You are a work order processing expert and are very good at summarizing work order scenarios. I will give you the specific information of several work orders, including problem descriptions and solutions, and possibly key errors. Please summarize several combinations of problem phenomena + key error information, and hang the work order ids that meet this combination under this combination. Try to hang multiple work orders under the same combination. In the case of no key error, please summarize. The answer format is returned in json, for example: {{problem phenomenon: ***, key error information: ***, work order id list: {x,x,…},{…}}. Note that only give json and no other information.}.

[0070] 1033. Input the first prompt word information corresponding to at least one batch of sample service work orders into the first large language model respectively, so that the first large language model generates at least one batch of work order problem scenarios corresponding to the sample service work orders.

[0071] Input the obtained first prompt word information corresponding to at least one batch of sample service work orders into the first large language model. The first large language model identifies the sample service work orders in the input batches, so as to generate at least one batch of work order problem scenarios corresponding to the sample service work orders. Among them, each work order problem scenario includes the field values of the problem phenomenon field and the key error field corresponding to the sample service work orders in the corresponding batch.

[0072] 1034. Based on the sample service work orders belonging to the target work order problem scenario in any service work order set, determine the field values of the troubleshooting plan field and the handling plan field corresponding to the target work order problem scenario.

[0073] Among them, the target work order problem scenario is any one of at least one work order problem scenario. In the embodiments of the present application, by means of the first large language model and the first prompt information, each batch of sample service work orders is processed, which not only improves the processing speed of the model, but also generates the corresponding target work order problem scenario for each batch of sample service work orders by virtue of the powerful recognition ability of the large language model. The target work order problem scenario includes the field values of the troubleshooting plan field and the processing plan field, and contains more information and more accurate classification results than the related technologies.

[0074] Specifically, based on the sample service work orders belonging to the target work order problem scenario in any service work order set, determining the field values of the troubleshooting plan field and the processing plan field corresponding to the target work order problem scenario includes, but is not limited to, the following two cases:

[0075] In a possible implementation manner, if the sample service work orders belonging to the target work order problem scenario contain the corpus related to the troubleshooting plan and the corpus related to the processing plan, then the field values of the troubleshooting plan field and the processing plan field corresponding to the target work order problem scenario are extracted from the corpus related to the troubleshooting plan and the corpus related to the processing plan.

[0076] In this case, since the sample service work orders in the target work order problem scenario contain the corpus related to the troubleshooting plan, the field value of the troubleshooting plan field can be extracted from the corpus related to the troubleshooting plan. Similarly, since the sample service work orders in the target work order problem scenario contain the corpus related to the solution plan, the field value of the solution plan field can be extracted from the corpus related to the solution plan. If there is only one sample service work order in the target work order problem scenario that contains the corpus related to the troubleshooting plan, the field value of the troubleshooting plan field is directly extracted from the corpus related to the troubleshooting plan contained in this sample service work order. If there is only one sample service work order in the target work order problem scenario that contains the corpus related to the solution plan, the field value of the solution plan field is directly extracted from the corpus related to the solution plan. If multiple sample service work orders in the target work order problem scenario contain the corpus related to the troubleshooting plan, and the corpus related to the troubleshooting plan contained in the multiple sample service work orders is the same or similar, the field value of the troubleshooting plan field is directly extracted from the corpus related to the troubleshooting plan contained in the multiple sample service work orders. If multiple sample service work orders in the target work order problem scenario contain the corpus related to the solution plan, and the corpus related to the solution plan contained in the multiple sample service work orders is the same or similar, the field value of the solution plan field is directly extracted from the corpus related to the solution plan contained in the multiple sample service work orders. If multiple sample service work orders in the target work order problem scenario contain the corpus related to the troubleshooting plan, but the corpus related to the troubleshooting plan contained in the sample service work orders is different, then a vote can be conducted on the multiple different corpora. According to the voting results, the field value of the troubleshooting plan field is extracted from the corpus related to the troubleshooting plan contained in the sample service work order with the highest number of votes. If multiple sample service work orders in the target work order problem scenario contain the corpus related to the solution plan, but the corpus related to the solution plan contained in the sample service work orders is different, then a vote can be conducted on the multiple different corpora. According to the voting results, the field value of the solution plan field is extracted from the corpus related to the solution plan contained in the sample service work order with the highest number of votes. Among them, when extracting the field value of the solution plan field from the corpus related to the solution plan, a large language model can be called or natural language processing technology can be used to perform semantic analysis on the corpus related to the solution plan, and through semantic analysis, the field value of the solution plan field is extracted from it. Similarly, when extracting the field value of the troubleshooting plan field from the corpus related to the troubleshooting plan, a large language model can be called or natural language processing technology can be used to perform semantic analysis on the corpus related to the troubleshooting plan, and through semantic analysis, the field value of the troubleshooting plan field is extracted from it.

[0077] In another possible implementation, if the sample service work orders belonging to the target work order problem scenario do not contain the corpus related to the troubleshooting plan and the corpus related to the processing plan, the field values of the troubleshooting plan field and the field values of the processing plan field corresponding to the target work order problem scenario can be generated by invoking the second large language model. For example, the sample service work orders belonging to the target work order problem scenario can be filled into the specified positions of the second prompt template to obtain the prompt information, and then the prompt information can be input into the second large language model to instruct the second large language model to analyze and summarize the field values of the troubleshooting plan field and the field values of the processing plan field from the information filled in the specified positions of the third prompt template.

[0078] Further, as the work order knowledge base is used, when the sample service work orders belonging to the target work order problem scenario contain the corpus related to the troubleshooting plan and the corpus related to the processing plan, the field values of the troubleshooting plan field can be extracted from the corpus related to the troubleshooting plan included in the sample service work orders, and the field values of the processing plan field can be extracted from the corpus related to the processing plan included in the sample service work orders. Then, the extracted field values of the troubleshooting plan field and the field values of the processing plan field are replaced with the field values of the troubleshooting plan field and the field values of the processing plan field generated for the target work order problem scenario to improve the accuracy of the 4K scenario tags.

[0079] In another embodiment of the present application, since the first large language model performs scenario recognition on the sample service work orders included in the service work order set in batches, and the sample service work orders included in the same service work order set have the same category, this may result in the same or similar scenarios in the work order problem scenarios generated by the first large language model for at least one batch of sample service work orders included in the service work order set. Therefore, before determining the field values of the troubleshooting plan field and the field values of the processing plan field corresponding to the target work order problem scenario based on the sample service work orders belonging to the target work order problem scenario in any of the service work order sets, the work order problem scenarios corresponding to at least one batch of sample service work orders included in the same service work order set can be merged.

[0080] For any set of service work orders, when merging the work order problem scenarios corresponding to at least one batch of sample service work orders included in the set of service work orders, a third large language model can be called to merge the at least one work order problem scenario corresponding to the at least one batch of sample service work orders, and the merged work order problem scenario corresponding to the set of service work orders can be obtained. When calling the fourth large language model for merging, the work order problem scenarios corresponding to at least one batch of sample service work orders included in the set of service work orders can be filled into the third prompt template to obtain prompt information, and then the prompt information can be input into the fourth large language model, so that the third large language model can merge the work order problem scenarios corresponding to at least one batch of sample service work orders included in the set of service work orders, and the merged work order problem scenario corresponding to the set of service work orders can be obtained. Among them, the third prompt template is used to instruct the third large language model to merge the same or similar work order problem scenarios in the work order problem scenarios corresponding to at least one batch of sample service work orders included in any set of service work orders. The content of the third prompt template can be:

[0081] #Role

[0082] Now you are a scenario analysis expert, very good at analyzing work order scenarios. I will give you several simple work order scenarios. The scenarios contain scenario IDs and problems.

[0083] #Task

[0084] You need to merge similar work order scenarios. Your output should tell me the problem phenomenon after merging and the included scenario IDs.

[0085] #Requirements

[0086] Answer format: New problem phenomenon: ****** Included scenario IDs: *****

[0087] Note: Only give the information of the merged scenario, no other information. Try to include as many similar simple work order scenarios as possible in the merged work order scenario, but do not arrange the simple work order scenarios. The new problem description should be as simple as possible, not exceeding 20 characters. Here are several simple work order scenarios I give you

[0088] #Data

[0089] Scenario ID: {}; Problem phenomenon: {}.

[0090] Accordingly, when determining the field value of the troubleshooting solution field and the field value of the handling solution field corresponding to the target work order problem scenario based on the sample service work orders belonging to the target work order problem scenario in any of the service work order sets, the field value of the troubleshooting solution field and the field value of the handling solution field corresponding to the same merged work order problem scenario can be determined based on the sample service work orders belonging to the same merged work order problem scenario in any of the service work order sets.

[0091] Furthermore, since at least one work order problem scenario generated by invoking the first large language model for each service work order set only includes the field value of the problem phenomenon field and the field value of the key error field, after obtaining the field value of the troubleshooting solution field and the field value of the handling solution field corresponding to each service work order set under at least one work order problem scenario, the troubleshooting solution field and its field value and the handling solution field and its field value corresponding to each work order problem scenario can be mounted under the corresponding work order problem scenario, so as to obtain at least one work order problem scenario corresponding to each service work order set. This work order problem scenario is a 4K scenario, including a problem phenomenon field, a key error field, a troubleshooting solution field, and a handling solution field, and also including the field values of these fields.

[0092] In another embodiment of the present application, each service work order set obtained through the clustering operation is labeled with an N-level work order category label, and there may be at least two service work order sets with the same N-level label in the at least one service work order set. In order to distinguish different service work order sets, the embodiment of the present application can also generate corresponding (N + 1)-level work order category labels for each of the at least one service work order set, and then, based on the corresponding (N + 1)-level work order category labels generated for each of the at least one service work order set and at least one work order problem scenario corresponding to the at least one service work order set, construct a work order knowledge base. Among them, when generating corresponding (N + 1)-level work order category labels for each of the at least one service work order set, it can be generated by means of the fourth large language model, or can be manually labeled. Taking generating corresponding (N + 1)-level work order category labels for each of the at least one service work order set by means of the fourth large language model as an example, this process includes the following steps:

[0093] First step, generate at least one piece of second prompt information corresponding to the service work order set according to the work order problem information described in the sample service work orders included in the at least one service work order set.

[0094] Based on the work order problem information described in the sample work orders included in each set of service work orders, in accordance with the requirements of the fourth prompt template, fill the information that meets the requirements in the sample work orders included in each set of service work orders into the corresponding positions of the fourth prompt template to obtain the second prompt word information corresponding to each set of service work orders. Among them, the fourth large language model can be the Qwen 2.5 large model, etc. The fourth prompt template is used to instruct the fourth large language model to process the sample work orders included in each set of service work orders filled, and summarize the work order category labels corresponding to each set of service work orders from them. The content of this fourth prompt template can be:

[0095] # Role

[0096] Now you are an expert in problem summarization, very good at summarizing problems. I will give you the problem descriptions of the work orders that have been classified. Please name these categories.

[0097] # Data

[0098] There are a total of {} categories below

[0099] The {}th category: {}

[0100] # Task

[0101] Please name this category according to the work order problem description in each category.

[0102] The name taken cannot be the same as the names of other categories and should be clearly distinguishable. Using numerical suffixes is not allowed.

[0103] # Requirements

[0104] Output format: The name of the {}th category: ***; Explanation: ***

[0105] Second step, input at least one piece of the second prompt word information into the fourth large language model respectively, so that the fourth large language model generates the corresponding work order category labels for at least one set of service work orders based on at least one piece of the second prompt word information.

[0106] After obtaining the second prompt word information corresponding to each set of service work orders, input the second prompt word information into the second large language model respectively. The second large language model, based on the indication of the second prompt word information, identifies the information filled in the corresponding positions of the second prompt information, so as to generate the corresponding N + 1 level work order category labels for each set of service work orders respectively.

[0107] Based on the generated level N+1 work order category labels for each set of at least one service work order, the work order knowledge base can be constructed using the level N+1 work order category labels of each set of service work orders as indexes, and using different work order problem scenarios and the included sample service work orders under the same level N+1 work order category label as index content. This work order knowledge base can be used to review service work orders, and can also determine the work order problem scenario to which a service work order belongs after obtaining a new service work order, so as to provide processing suggestions based on the determined work order problem scenario. The constructed work order knowledge base presents a tree structure according to the label hierarchy, and this tree structure is as shown in Figure 3 .

[0108] See Figure 3 , for a certain product, first-level classification, second-level classification and third-level classification can be carried out. Through the first-level classification, the first-level work order category labels 1, 1, 1, 3, … can be obtained. For the first-level work order category label 1, through the second-level classification, the second-level work order category labels 1, 2, 3, … under the first-level work order category label 1 can be obtained. Among them, the labels obtained by the first-level classification and the labels obtained by the second-level classification can be manually maintained. For the second-level work order category label 1 under the first-level work order category label 1, through the third-level classification, the third-level work order category labels 1, 2, 3, … included in the second-level work order category label 1 under the first-level work order category label 1 can be obtained. Among them, the third-level work order category label 1 includes 4K scenarios 1, 2, 3, …, and each 4K scenario includes fields such as problem phenomenon, key error, troubleshooting plan and processing plan and the field values of these fields, and each 4K scenario includes multiple service work orders. After obtaining the third-level work order category labels of each set of service work orders, different work order problem scenarios and the included sample service work orders under the same third-level work order category label, the work order knowledge base as shown in Figure 3 can be constructed using the third-level work order category label as the index and different work order problem scenarios and the included sample service work orders under the same third-level work order category label as the index content.

[0109] Further, to improve the accuracy of the constructed work order knowledge base, after obtaining at least one work order problem scenario corresponding to each service work order set, the at least one work order problem scenario corresponding to each service work order set can be calibrated manually to obtain at least one calibrated work order problem scenario corresponding to each service work order set. Correspondingly, when constructing the work order knowledge base, the N+1-level work order category label of each service work order set can be used as the index, and different calibrated work order problem scenarios and the included sample service work orders under the N+1-level work order category label can be used as the index content to construct the work order knowledge base. By calibrating the constructed service work order database, the accuracy of the constructed work order knowledge base is improved.

[0110] For the above service work order classification process, the following will be described in detail in combination with Figure 4 See Figure 4 , this process includes the following steps:

[0111] Step1. Data cleaning.

[0112] For the provided original data, data cleaning is performed, and at the same time, tools such as regular expressions are used to remove http links, solution codes, etc. from the original data, and the data format is unified as the input of the 4key large model.

[0113] Step2. Vectorization model.

[0114] Two vectorization processes are designed for historical data and new data. For historical data with labels, first, the BGE model is used to initialize the vectorization of the work order data, and then the initialized vectors are used to train the feature transformer ResNet, so that the work order data under the same classification has similarity in the high-dimensional space, and the work order data of different classifications has differences in the high-dimensional space. For new data without labels, first, the BGE model is used to initialize the vectorization of the work order data, and then the trained feature transformer ResNet neural network is used to perform feature transformation on it.

[0115] Step3. Clustering model.

[0116] By virtue of the property of "similarity within the class and difference between classes" of the vector data after feature transformation in the high-dimensional space, the K-means clustering algorithm is used to cluster the data, and the classes are obtained by minimizing the distance between each data and the clustering center. Then, the large language model is called to generate a class name for each class.

[0117] Step4. Summary and refinement model.

[0118] 1). For 4key scenario generation: Use prompt engineering to generate 4key scenarios for the data;

[0119] 2) Scene merging: Due to the limitation of the number of tokens input to the large model, work orders under the same third-level classification need to be batch-generated according to the number of work orders, resulting in the generation of the same or similar scenes and causing 4key scene duplication. With the help of the language understanding ability of the large model, the same or similar scenes are merged.

[0120] 3) Solution mounting: Match the solutions used in the work order processing process and mount them to the corresponding 4key scenes;

[0121] 4) Troubleshooting suggestion generation: For 4key scenes and work order data with missing information, with the help of the language generation ability of the large model, use information such as work order summaries and solution processes to generate troubleshooting suggestions.

[0122] Step 5, Evaluation and calibration: For the generated 4key scenes, support manual error correction and calibration to ensure data accuracy.

[0123] In the embodiment of the present application, the computing and evaluation capabilities of the large model are utilized to re-learn the historical service work orders in the actual scenario, automatically generate problem phenomena and root causes of problems, and associate the troubleshooting methods and solution plans corresponding to the phenomena, gradually improving the coverage rate of the scenarios, avoiding the understanding errors caused by manual maintenance, reducing the labor cost of summary and review, and gradually moving towards the intelligent operation and maintenance direction of automatic association, automatic tagging, and automatic push.

[0124] All the above optional technical solutions can be combined arbitrarily to form optional embodiments of the present application, which will not be elaborated here one by one.

[0125] The embodiment of the present application provides a method for classifying service work orders. This method applies the work order knowledge base constructed in the above embodiment. Taking an electronic device executing the embodiment of the present application as an example, see Figure 5 , the method flow provided by the embodiment of the present application includes:

[0126] 501. Obtain the target service work order to be processed.

[0127] Among them, the target service work order is an unprocessed service work order obtained by the user in the actual application scenario. After obtaining the target service work order, the user can assign the first-level work order category label and the second-level work order category label to the target service work order, and then provide the target service work order to the electronic device, so that the electronic device determines the target work order problem scenario to which the target service work order belongs, and then based on the problem phenomenon field, key error field, troubleshooting plan field, and solution plan field included in the target work order problem scenario and the field values of these fields, provide processing guidance information for the operation and maintenance personnel.

[0128] 502. Determine the target feature vector corresponding to the target service work order.

[0129] Optionally, for the obtained target service work order, a regularization formula written by the user can be used to clean the data of the target service work order to obtain the cleaned target service work order, and then a preset format can be used to format the cleaned target service work order.

[0130] In this step, a vector initialization model can be called to extract features from the target service work order to obtain the initial feature vector corresponding to the target service work order, and then a vector conversion model can be called to perform conversion processing on the initial feature vector of the target service work order to obtain the target feature vector corresponding to the target service work order.

[0131] 503. Based on the target feature vector corresponding to the target service work order and the work order knowledge base, determine the target work order problem scenario to which the target service work order belongs.

[0132] Taking the work order knowledge base including three-level work order category labels as an example, after obtaining the target feature vector corresponding to the target service work order, the index of the first-level work order category label of the target service work order can be used to query the first-level work order category label identical to the first-level work order category label from the work order knowledge base, and then query the second-level work order category label identical to the second-level work order category label corresponding to the target service work order under the first-level work order category label, obtain each third-level work order category label included in the second-level work order category label in the work order knowledge base, and then calculate the similarity between the target feature vector corresponding to the target service work order and the target feature vectors of each service work order corresponding to each third-level work order category label. According to the similarity calculation result, the work order problem scenario corresponding to the third-level work order category label with the largest similarity is determined as the target work order problem scenario, and this target work order problem scenario is used to provide processing suggestions for the target service work order. The above takes the work order knowledge base including three-level work order category labels as an example. Of course, if the work order knowledge base includes more than three levels of labels, the above method can be used to match the labels layer by layer to find the index with the longest match with the work order category label marked by the target service work order, and then further find the target work order problem scenario through similarity calculation based on the target feature vector corresponding to the target service work order.

[0133] Further, after determining the target work order problem scenario, the problem phenomenon field, key error field, troubleshooting plan field, processing plan field included in the target work order problem scenario and the field values of these fields can be obtained, and then these fields and their field values can be provided to the user, so as to provide guiding suggestions for the processing of the target service work order.

[0134] All the above optional technical solutions can be combined arbitrarily to form optional embodiments of the present application, which will not be elaborated one by one here.

[0135] Please refer to Figure 6 which shows a schematic structural diagram of a service work order classification device provided by an embodiment of the present application. The device can be implemented through software, hardware, or a combination of both, and becomes all or part of an electronic device. The device includes:

[0136] A clustering module 601, configured to perform a clustering operation on at least one sample service work order to obtain at least one service work order set, and each service work order set included in the at least one service work order set includes sample service work orders of the same work order category;

[0137] An identification module 602, configured to, for any service work order set, perform scenario identification processing on the sample service work orders included in the any service work order set based on a first large language model and first prompt information, to obtain at least one work order problem scenario and the sample service work orders corresponding to each work order problem scenario in the any service work order set. Each work order problem scenario includes a problem scenario description field and its field value, and the problem scenario description field includes a problem phenomenon field, a key error reporting field, a troubleshooting plan field, and a handling plan field.

[0138] In another embodiment of the present application, the device further includes:

[0139] A data cleaning module, configured to perform data cleaning on the at least one sample service work order to obtain at least one cleaned sample service work order;

[0140] A formatting module, configured to perform formatting processing on the at least one cleaned sample service work order in a preset format.

[0141] In another embodiment of the present application, the clustering module 601 is configured to determine target feature vectors corresponding to the at least one service work order respectively, where the target feature vectors can make the sample service work orders belonging to the same work order category closer in distance in the vector space; based on the target feature vectors corresponding to the at least one service sample work order respectively, perform a clustering operation on the at least one sample service work order to obtain the at least one service work order set.

[0142] In another embodiment of the present application, a determination module is configured to call a vector initialization model to perform feature extraction on the at least one sample service work order to obtain initial feature vectors corresponding to the at least one sample service work order; call a vector conversion model to perform conversion processing on the initial feature vectors of the at least one sample service work order to obtain target feature vectors corresponding to the at least one sample service work order.

[0143] In another embodiment of the present application, the at least one sample service work order includes a plurality of historical service work orders, and the plurality of historical service work orders are respectively labeled with N-level work order category labels, where N is greater than or equal to 1; the apparatus further includes:

[0144] The determination module is further configured to call a vector initialization model to determine initial feature vectors corresponding to the plurality of historical service work orders;

[0145] The conversion module is configured to call the to-be-trained vector conversion model to perform conversion processing on the initial feature vectors corresponding to the plurality of historical service work orders to obtain target feature vectors corresponding to the plurality of historical service work orders;

[0146] The determination module is configured to determine N loss function values corresponding to the N-level work order category labels according to the N-level work order category labels and the target feature vectors respectively corresponding to the plurality of historical service work orders; wherein, the N loss function values respectively reflect that the target feature vectors corresponding to the historical service work orders with the same-level work order category labels have closer distances;

[0147] The determination module is further configured to determine a target loss function value based on the N loss function values to train the to-be-trained vector conversion model according to the target loss function value.

[0148] In another embodiment of the present application, the clustering module is configured to perform a clustering operation on at least one target feature vector corresponding to the at least one sample service work order by using a preset clustering algorithm to obtain at least one cluster; form an initial service work order set by using the sample service work orders corresponding to the target feature vectors in any one of the clusters to obtain at least one initial service work order set; and in response to an adjustment operation of a user, adjust the at least one initial service work order set to obtain the at least one service work order set.

[0149] In another embodiment of the present application, an identification module is configured to batch process the sample service work orders included in any one of the service work order sets to obtain at least one batch of sample service work orders; sequentially generate first prompt word information corresponding to the at least one batch of sample service work orders; respectively input the first prompt word information corresponding to the at least one batch of sample service work orders into the first large language model, so that the first large language model generates at least one work order problem scenario corresponding to the at least one batch of sample service work orders; wherein, any work order problem scenario includes the field values of the problem phenomenon field and the key error field corresponding to the sample service work orders of the corresponding batch; based on the sample service work orders belonging to the target work order problem scenario in any one of the service work order sets, determine the field values of the troubleshooting plan field and the handling plan field corresponding to the target work order problem scenario, and the target work order problem scenario is any one of the at least one work order problem scenarios.

[0150] In another embodiment of the present application, the identification module is configured to extract from the troubleshooting plan-related corpus and the handling plan-related corpus if the sample service work orders belonging to the target work order problem scenario contain the troubleshooting plan-related corpus and the handling plan-related corpus; if the sample service work orders belonging to the target work order problem scenario do not contain the troubleshooting plan-related corpus and the handling plan-related corpus, call the second large language model to generate the field values of the troubleshooting plan field and the handling plan field corresponding to the target work order problem scenario.

[0151] In another embodiment of the present application, the device further includes:

[0152] A merging module is configured to call a third large language model to merge the at least one work order problem scenario corresponding to the at least one batch of sample service work orders to obtain a merged work order problem scenario corresponding to any one of the service work order sets;

[0153] The identification module is configured to determine the field values of the troubleshooting plan field and the handling plan field corresponding to the same merged work order problem scenario based on the sample service work orders belonging to the same merged work order problem scenario in any one of the service work order sets.

[0154] In another embodiment of the present application, each of the at least one service work order sets is labeled with an N-level work order category label, and the device further includes:

[0155] A generation module is configured to generate corresponding (N + 1)-level work order category labels for each of the at least one service work order sets;

[0156] A construction module is used to construct a work order knowledge base by indexing with the work order category labels of the N+1 level of at least one set of service work orders, and using different work order problem scenarios and the included sample service work orders under the same N+1 level work order category label as the indexing content.

[0157] In another embodiment of the present application, a generation module is used to generate at least one second prompt word information corresponding to the at least one set of service work orders according to the work order problem information described in the sample service work orders included in the at least one set of service work orders; and respectively input the at least one second prompt word information into the fourth large language model, so that the fourth large language model generates corresponding N+1 level work order category labels for the at least one set of service work orders based on the at least one second prompt word information.

[0158] In another embodiment of the present application, the device further includes:

[0159] An acquisition module is used to acquire a target service work order to be processed;

[0160] A determination module is used to determine a target feature vector corresponding to the target service work order;

[0161] The determination module is further used to determine a target work order problem scenario to which the target service work order belongs based on the target feature vector corresponding to the target service work order and the work order knowledge base, and the target work order problem scenario is used to provide processing suggestions for the target service work order.

[0162] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0163] Figure 7 The block diagram of a structure of an electronic device 700 provided by an exemplary embodiment of the present application is shown. Generally, the electronic device 700 includes: a processor 701 and a memory 702.

[0164] The processor 701 can be implemented in at least one of the hardware forms of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 701 can also include a main processor and a coprocessor. The main processor is a processor for processing data in the wake state; the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 701 can be integrated with a GPU (Graphics Processing Unit), and the GPU is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 701 can also include an artificial intelligence processor, which is used to process computational operations related to machine learning.

[0165] The memory 702 can include one or more computer-readable storage media, and the computer-readable storage media can be non-transitory computer-readable storage media. For example, the non-transitory computer-readable storage media can be CD-ROM (Compact Disc Read-Only Memory), ROM, RAM (Random Access Memory), magnetic tape, floppy disk, and optical data storage devices, etc. At least one computer program is stored in the computer-readable storage media, and when the at least one computer program is executed, it can implement the above-mentioned method for constructing a work order knowledge base or the method for processing service work orders.

[0166] Of course, the above-mentioned electronic device may necessarily further include other components, such as an input / output interface, a communication component, etc. The input / output interface provides an interface between the processor and the peripheral interface module, and the above-mentioned peripheral interface module can be an output device, an input device, etc. The communication component is configured to facilitate communication between the electronic device and other devices in a wired or wireless manner.

[0167] Those skilled in the art can understand that Figure 7 the structure shown in does not constitute a limitation on the electronic device 700, and it may include more or fewer components than shown in the figure, or combine certain components, or adopt different component arrangements.

[0168] An embodiment of the present application provides a computer-readable storage medium, and at least one computer program is stored in the computer-readable storage medium. When the at least one computer program is executed by a processor, it can implement the above-mentioned method for constructing a work order knowledge base or the above-mentioned method for processing service work orders.

[0169] An embodiment of the present application provides a computer program product, which includes a computer program that can implement the above-mentioned method for constructing a work order knowledge base or the above-mentioned service work order processing method when executed by a processor.

[0170] The above embodiments are only used to illustrate the technical solutions of the present application, rather than limiting them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A service work order classification method, characterized in that: The method comprises: Performing a clustering operation on at least one sample service work order to obtain at least one service work order set, wherein the sample service work orders included in each service work order set in the at least one service work order set have the same work order category, and each service work order set in the at least one service work order set is annotated with an N-level work order category label; For any service ticket set, based on the first language model and the first prompt word information, scene recognition processing is performed on the sample service tickets included in the any service ticket set to obtain at least one ticket problem scene and the sample service tickets corresponding to each ticket problem scene in the any service ticket set, each ticket problem scene includes a problem scene description field and its field value, and the problem scene description field includes a problem phenomenon field, a key error field, a troubleshooting solution field, and a processing solution field; Generating a corresponding N+1th level work order category label for each of the at least one service work order set; A work order knowledge base is constructed using the N+1-level work order category label of at least one service work order set as an index, and using different work order problem scenarios and the sample service work orders included under the same N+1-level work order category label as index content. The work order knowledge base has a tree structure according to the label hierarchy, and is used to match the labels layer by layer after obtaining a new service work order, so as to determine the work order problem scenario to which the service work order belongs.

2. The method according to claim 1, characterized in that The clustering operation is performed on at least one sample service work order to obtain at least one service work order set, including: Determine a target feature vector corresponding to each of the at least one sample service work order, wherein the target feature vector can make the sample service work orders belonging to the same work order category closer in the vector space; Based on the target feature vector corresponding to each of the at least one sample service work orders, a clustering operation is performed on the at least one sample service work order to obtain the at least one service work order set.

3. The method according to claim 2, characterized in that The determining of the target feature vector corresponding to each of the at least one sample service work orders includes: Calling a vector initialization model to perform feature extraction on the at least one sample service work order to obtain an initial feature vector corresponding to the at least one sample service work order; The vector conversion model is called to convert the initial feature vector of the at least one sample service work order to obtain a target feature vector corresponding to the at least one sample service work order.

4. The method according to claim 3, characterized in that The at least one sample service work order includes a plurality of historical service work orders, each of which is marked with an N-level work order category label, where N is greater than or equal to 1; the method further includes: Calling a vector initialization model to determine initial feature vectors corresponding to the multiple historical service work orders; Calling the vector conversion model to be trained, converting the initial feature vectors corresponding to the multiple historical service work orders, and obtaining target feature vectors corresponding to the multiple historical service work orders; According to the N-level work order category labels and target feature vectors corresponding to the multiple historical service work orders, N loss function values ​​corresponding to the N-level work order category labels are determined; wherein the N loss function values ​​respectively reflect that the target feature vectors corresponding to the historical service work orders with the same level of work order category labels have a closer distance; Based on the N loss function values, a target loss function value is determined to train the vector conversion model to be trained according to the target loss function value.

5. The method according to claim 2, characterized in that: The step of performing a clustering operation on the at least one sample service work order based on the target feature vector corresponding to each of the at least one sample service work order to obtain the at least one service work order set includes: Using a preset clustering algorithm, clustering the at least one target feature vector corresponding to the at least one sample service work order to obtain at least one cluster; The sample service orders corresponding to the target feature vector in each cluster are combined into an initial service order set to obtain at least one initial service order set; In response to the user's adjustment operation, the at least one initial service ticket set is adjusted to obtain the at least one service ticket set.

6. The method according to claim 1, characterized in that The method of performing scene recognition processing on the sample service work orders included in any one of the service work order sets based on the first language model and the first prompt word information to obtain at least one work order problem scene includes: Processing the sample service work orders included in any one of the service work order sets in batches to obtain at least one batch of sample service work orders; Sequentially generate first prompt word information corresponding to the at least one batch of sample service work orders; Inputting the first prompt word information corresponding to the at least one batch of sample service work orders into the first large language model respectively, so that the first large language model generates at least one work order problem scenario corresponding to the at least one batch of sample service work orders; wherein each work order problem scenario includes a field value of a problem phenomenon field and a field value of a key error field corresponding to the corresponding batch of sample service work orders; Based on a sample service ticket in any one of the service ticket sets that belongs to a target ticket problem scenario, determine a field value of a troubleshooting solution field and a field value of a processing solution field corresponding to the target ticket problem scenario, wherein the target ticket problem scenario is any one of the at least one ticket problem scenario.

7. The method according to claim 6, characterized in that The determining, based on a sample service ticket belonging to a target service ticket problem scenario in any service ticket set, a field value of a troubleshooting solution field and a field value of a processing solution field corresponding to the target service ticket problem scenario includes: If the sample service ticket belonging to the target ticket problem scenario contains corpus related to the troubleshooting solution and corpus related to the processing solution, extract the field value of the troubleshooting solution field and the field value of the processing solution field corresponding to the target ticket problem scenario from the corpus related to the troubleshooting solution and the corpus related to the processing solution; If the sample service work order belonging to the target work order problem scenario does not contain corpus related to the troubleshooting solution and corpus related to the processing solution, the second largest language model is called to generate the field value of the troubleshooting solution field and the field value of the processing solution field corresponding to the target work order problem scenario.

8. The method according to claim 6, characterized in that Before determining the field value of the troubleshooting solution field and the field value of the processing solution field corresponding to the target work order problem scenario based on the sample service work order belonging to the target work order problem scenario in any service work order set, the method further includes: Invoking the third language model to merge at least one work order problem scenario corresponding to the at least one batch of sample service work orders to obtain a merged work order problem scenario corresponding to any one of the service work order sets; The determining, based on a sample service ticket belonging to a target service ticket problem scenario in any service ticket set, a field value of a troubleshooting solution field and a field value of a processing solution field corresponding to the target service ticket problem scenario includes: Based on the sample service tickets in any one of the service ticket sets that belong to the same merged ticket problem scenario, the field value of the troubleshooting solution field and the field value of the processing solution field corresponding to the same merged ticket problem scenario are determined.

9. The method according to claim 1, characterized in that: Generating a corresponding N+1th level work order category label for each of the at least one service work order set includes: Generate at least one second prompt word information corresponding to the at least one service work order set according to the work order problem information described in the sample service work order included in the at least one service work order set; The at least one second prompt word information is respectively input into the fourth largest language model, so that the fourth largest language model generates a corresponding N+1th level work order category label for the at least one service work order set based on the at least one second prompt word information.

10. The method according to claim 1, characterized in that The method further comprises: Get the target service work order to be processed; Determine a target feature vector corresponding to the target service work order; Based on the target feature vector corresponding to the target service work order and the work order knowledge base, a target work order problem scenario to which the target service work order belongs is determined.

11. An electronic device, characterized in that: It comprises a processor and a memory; the memory stores at least one program code; the at least one program code is used to be called and executed by the processor to implement the service work order classification method as described in any one of claims 1 to 10.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one computer program, and when the at least one computer program is executed by a processor, the service ticket classification method according to any one of claims 1 to 10 can be implemented.

13. 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 service ticket classification method according to any one of claims 1 to 10 can be implemented.

Images