Operation and maintenance work order quality inspection methods, devices, equipment and media

Abstract

The present invention discloses a method, device, equipment and medium for quality inspection of operation and maintenance work orders, and relates to the field of big data technology. The method comprises: inputting prompt information and the operation and maintenance work order into a trained large language model, and outputting a set of entity information of multiple entity types in the operation and maintenance work order, wherein the prompt information is used to prompt the trained large language model to extract entity information of multiple entity types from text sequence information of the operation and maintenance work order, and perform feature splicing on the set of entity information and the operation and maintenance work order to obtain work order splicing features, and inputting the work order splicing features into a trained pre-trained language model for processing, and outputting a high-level semantic feature representation tensor, and inputting the high-level semantic feature representation tensor into a trained quality inspection model to obtain a predicted quality inspection score, so as to perform intelligent quality inspection on the operation and maintenance work order, improve quality inspection efficiency, and alleviate the manpower demand and cost of quality inspectors.

Description

Background Art

[0002] With the development of communication technology, the scale of operators' networks is constantly expanding, especially the popularization and application of 5G communications, which has continuously increased the complexity of networks. As business gradually accumulates, the number of network operation and maintenance work orders continues to increase. Strictly controlling the work order quality inspection process is of vital importance to operators' network optimization.

[0003] However, in the relevant technologies, in the current network operation and maintenance work, the quality inspection of work orders mainly adopts the manual review mode, which has high requirements on the professional experience of personnel and has low quality inspection efficiency.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0005] The present disclosure provides a method, device, equipment and medium for quality inspection of operation and maintenance work orders, which at least to a certain extent overcome the problems in related technologies that manual review relies on professional experience and has low quality inspection efficiency.

[0006] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0007] In a first aspect, an embodiment of the present disclosure provides a method for quality inspection of an operation and maintenance work order, the method comprising:

[0008] Inputting the prompt information and the operation and maintenance work order into the trained large language model to obtain a set of entity information of multiple entity types in the operation and maintenance work order; the prompt information is used to prompt the trained large language model to extract entity information of multiple entity types from the text sequence information of the operation and maintenance work order;

[0009] Perform feature splicing on the set of entity information and the operation and maintenance work order to obtain a work order splicing feature;

[0010] Inputting the work order concatenation features into a trained pre-trained language model for processing to obtain a high-level semantic feature representation tensor;

[0011] The high-level semantic feature representation tensor is input into the trained quality inspection model to obtain a predicted quality inspection score.

[0012] In a second aspect, an embodiment of the present disclosure provides a data processing method, the method comprising:

[0013] Inputting the prompt information and the data to be processed into the large language model to obtain auxiliary reference information; wherein the large language model extracts the directional auxiliary reference information from the data to be processed according to the prompt information, so that the large language model can understand the semantic information in the data to be processed;

[0014] Perform feature splicing on the auxiliary reference information and the data to be processed to obtain a splicing feature to be processed;

[0015] The to-be-processed splicing features are input into the to-be-executed tasks to obtain the task execution results.

[0016] In a third aspect, an embodiment of the present disclosure provides an operation and maintenance work order quality inspection device, including:

[0017] An entity information acquisition unit, used to input prompt information and an operation and maintenance work order into a trained large language model to obtain a set of entity information of multiple entity types in the operation and maintenance work order; the prompt information is used to prompt the trained large language model to extract entity information of multiple entity types from the text sequence information of the operation and maintenance work order;

[0018] A splicing unit, used for performing feature splicing on the set of entity information and the operation and maintenance work order to obtain a work order splicing feature;

[0019] A processing unit, used for inputting the work order concatenation features into a trained pre-trained language model for processing to obtain a high-level semantic feature representation tensor;

[0020] The quality inspection unit is used to input the high-level semantic feature representation tensor into the trained quality inspection model to obtain a predicted quality inspection score.

[0021] In a fourth aspect, an embodiment of the present disclosure provides a data processing device, including:

[0022] An auxiliary information acquisition unit is used to input the prompt information and the data to be processed into the large language model to obtain auxiliary reference information; wherein the large language model extracts the directional auxiliary reference information from the data to be processed according to the prompt information, so that the large language model understands the semantic information in the data to be processed;

[0023] A feature splicing unit, used for performing feature splicing on the auxiliary reference information and the data to be processed to obtain a splicing feature to be processed;

[0024] The execution unit is used to input the splicing features to be processed into the task to be executed to obtain the task execution result.

[0025] In a fifth aspect, an embodiment of the present disclosure provides an electronic device, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute the method described in the first aspect above by executing the executable instructions.

[0026] In a sixth aspect, an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method described in the first aspect above.

[0027] In a seventh aspect, according to another aspect of the present disclosure, a computer program product or a computer program is also provided, the computer program product or the computer program comprising computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes any of the methods described above.

[0028] The embodiments of the present disclosure provide a method, apparatus, device and medium for quality inspection of operation and maintenance work orders. The method includes: inputting prompt information and operation and maintenance work orders into a trained large language model, and outputting a set of entity information of multiple entity types in the operation and maintenance work orders. The prompt information is used to prompt the trained large language model to extract entity information of multiple entity types from the text sequence information of the operation and maintenance work orders, perform feature splicing on the set of entity information and the operation and maintenance work orders to obtain work order splicing features, input the work order splicing features into a trained pre-trained language model for processing, output a high-level semantic feature representation tensor, and input the high-level semantic feature representation tensor into a trained quality inspection model to obtain a predicted quality inspection score. Through prompt information with directional information, the large language model is assisted to mine the information of the operation and maintenance work order, and a collection of entity information is obtained to improve the efficiency of knowledge acquisition; the collection of entity information is spliced ​​into the operation and maintenance work order and input into the pre-trained language model. The input feature granularity is higher, which enables the pre-trained language model to more accurately determine the logical relationship between entities in the operation and maintenance work order, and then predict the quality inspection score through the quality inspection model. Not only is the whole process completed automatically, the quality inspection efficiency is high, and the output quality inspection score is highly accurate. For predictive maintenance of the network, potential faults and performance problems can be discovered in advance, which can significantly improve the operation and maintenance efficiency while effectively alleviating the manpower demand and cost of quality inspectors, which helps operators reduce costs and increase efficiency.

[0029] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0031] Figure 1 A schematic diagram showing a flow chart of a method for quality inspection of an operation and maintenance work order in an embodiment of the present disclosure is shown;

[0032] Figure 2 A schematic diagram showing an operation and maintenance work order in an embodiment of the present disclosure is shown;

[0033] Figure 3 A schematic diagram of a process for obtaining work order splicing features in an embodiment of the present disclosure is shown;

[0034] Figure 4 A flow chart showing a quality inspection model training process in an embodiment of the present disclosure is shown;

[0035] Figure 5 A schematic diagram of a process for obtaining a true quality inspection score after encoding in an embodiment of the present disclosure is shown;

[0036] Figure 6 A schematic diagram showing a modeling process of intelligent quality inspection of operation and maintenance work orders in an embodiment of the present disclosure;

[0037] Figure 7 A schematic diagram showing a flow chart of a data processing method in an embodiment of the present disclosure;

[0038] Figure 8 A schematic diagram of the structure of an operation and maintenance work order quality inspection device in an embodiment of the present disclosure is shown;

[0039] Fig. 9 A schematic structural diagram of an electronic device in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0040] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0041] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0042] Glossary:

[0043] Pre-trained Language Model (PLM): is a deep learning model that uses a large amount of corpus for unsupervised pre-training, and can then be used for various downstream tasks such as text classification, named entity recognition, sentiment analysis, etc. Currently, popular pre-trained language models include many, such as: Bidirectional Encoder Representations from Transformers (BERT), Generative Pre-trained Transformer (GPT), and Robustly Optimized BERT Pretraining Approach (RoBERTa).

[0044] Large Language Model (LLM): An artificial intelligence model designed to understand and generate human language. It can reach billions or trillions of parameter scales, and has learned a lot of language knowledge during the training process, so it is highly flexible and scalable, and has excellent reasoning ability. Currently, popular large language models include: Chat Generative Pre-trained Transformer (chat-GPT), Chat Generative Language Model (chat-GLM), Victorian Christmas Unicorn Accumulation Network (Victorian Christmas Unicorn Accumulation Network, Vicuna), etc.

[0045] Because in the relevant technologies, in the current network operation and maintenance work, the quality inspection of work orders mainly adopts the manual review mode, which has high requirements on the professional experience of personnel and has low quality inspection efficiency.

[0046] Based on this, the present application provides an operation and maintenance work order quality inspection method, which includes: inputting prompt information and the operation and maintenance work order into a trained large language model, and outputting a set of entity information of multiple entity types in the operation and maintenance work order, wherein the prompt information is used to prompt the trained large language model to extract entity information of multiple entity types from the text sequence information of the operation and maintenance work order, and perform feature splicing on the set of entity information and the operation and maintenance work order to obtain work order splicing features, and input the work order splicing features into a trained pre-trained language model for processing, and outputting a high-level semantic feature representation tensor, and inputting the high-level semantic feature representation tensor into the trained quality inspection model to obtain a predicted quality inspection score. Through prompt information with directional information, the large language model is assisted to mine the information of the operation and maintenance work order, and a collection of entity information is obtained to improve the efficiency of knowledge acquisition; the collection of entity information is spliced ​​into the operation and maintenance work order and input into the pre-trained language model. The input features are more granular, allowing the pre-trained language model to more accurately determine the logical relationship between entities in the operation and maintenance work order, and then the quality inspection score is predicted through the quality inspection model. Not only is the entire process completed automatically, the quality inspection efficiency is high, and the output quality inspection score is highly accurate.

[0047] The present exemplary implementation is described in detail below with reference to the accompanying drawings and embodiments.

[0048] First, an operation and maintenance work order quality inspection method is provided in an embodiment of the present disclosure. The method can be executed by any electronic device with computing and processing capabilities. In the following process, the electronic device is taken as a terminal device as an example.

[0049] Figure 1 A flow chart of a method for quality inspection of operation and maintenance work orders in an embodiment of the present disclosure is shown as follows: Figure 1 As shown, the operation and maintenance work order quality inspection method provided in the embodiment of the present disclosure includes the following steps:

[0050] S102: Input the prompt information and the operation and maintenance work order into the trained large language model to obtain a set of entity information of multiple entity types in the operation and maintenance work order. The prompt information is used to prompt the trained large language model to extract entity information of multiple entity types from the text sequence information of the operation and maintenance work order.

[0051] Entities are specific things that exist in the real world, such as people, places, objects, organizations, etc. In the fields of computer science and natural language processing, entities usually refer to specific nouns or noun phrases that appear in texts, such as names of people, places, and institutions. Entity information recognition is to automatically identify entities with specific meanings from texts, and classify and label them.

[0052] In a possible embodiment, the large language model trained in the present application can use conventional large language models that have been trained, that is, some large language models are directly taken for use, such as conventional chat-GPT, etc., without additional training, which can improve efficiency.

[0053] If the number of operation and maintenance tickets in certain vertical fields is large, but conventional chat-GPT cannot recognize the technical language in the professional field, then training can be carried out for the vertical field. For the operation and maintenance tickets generated by different technologies, the large language model of the corresponding field can be trained, and the reasoning ability of the large language model can be used to generate a collection of entity information.

[0054] Among them, the text sequence information of the operation and maintenance work order can be: X = [x1, x2, ..., x n ]; The entity information in the preset format includes: entity description, entity type, and at least one or more of the following information: entity probability, entity starting position index and end position index.

[0055] In a possible embodiment, the obtained entity probability, entity starting position index and end position index are standardized to obtain the entity probability, entity starting position index and end position index in a numerical format after standardization. The content of the specific numerical format can retain the number of digits after the decimal point as required.

[0056] In a possible embodiment, taking entity information in a target format as an example, the entity information includes: entity description, entity type, entity probability, entity start position index, and end position index.

[0057] It can be represented using a tuple of the form, in which Represents the text description of the i-th entity information, i.e., entity description; Represents the category label of the i-th entity information, that is, the entity type; Represents the probability value of the i-th entity information, that is, the entity probability, s i and e i It represents the starting position index and the ending position index of the entity in the text sequence.

[0058] Among them, the entity start position index and the entity end position index can be used to represent the start position and end position of the entity information retrieved when the entity information is indexed. The entity start position index and the entity end position index after standardization are numerical and can be used to represent the position information encoding of the entity information in the entire work order.

[0059] The entity probability, entity start position index, and entity end position index can be expressed in numerical form.

[0060] In a possible embodiment, if the large language model is desired to output a set of entity information in a target format, the prompt information input into the large language model can be set to: "Please extract the entity information of the fault phenomenon description, fault cause location, and fault solution from the following conversation content of the original operation and maintenance work order, and format and output the entity information in the original operation and maintenance work order text according to the pattern of '(entity description, entity type, entity probability, entity starting position index, entity ending position index)'."

[0061] Fault phenomenon description, fault cause location, and fault solution are entity types. For each entity type, enter a corresponding entity information. If one of the above contents hopes that the large language model will output three entity types, then a set of entity information with three entity types is formed. The specific content corresponding to the fault phenomenon description is the specific description content of the entity, that is, the entity description.

[0062] For example, for Figure 2 The operation and maintenance work order shown in the figure inputs the content of the operation and maintenance work order and the prompt information in the above embodiment into the large language model. The specific set of entity information outputted may be:

[0063] (Wavelength division multiplexing system interruption alarm, fault phenomenon description, 0.98213, 0, 0.06896); (Equipment system disk failure, fault cause location, 0.97962, 0.36207, 0.42241); (Replace system disk, fault solution, 0.93171, 0.45690, 0.50000); (Replace system disk, fault solution, 0.94692, 0.62931, 0.67241).

[0064] The above entity probabilities, entity start position indexes, and entity end position indexes have been standardized, and the specific values ​​are rounded to five decimal places.

[0065] because Figure 2 The text content of changing the system disk appears twice, so in the process of large language model processing, the two occurrences of the content are output as entity descriptions of the fault solution, but the entity probabilities and position indexes behind them are different.

[0066] In a possible embodiment, if other content needs to be output in the operation and maintenance work order, for example, entity information of other entity types, or entity information of other content, guiding content can be added to the prompt information so that the large language model can understand the content of the prompt information and process the operation and maintenance work order, guiding the large language model to obtain entity information in the context of the text sequence information corresponding to the operation and maintenance work order.

[0067] In the related art, traditional knowledge enhancement means are usually targeted at specific task models, using additional logical reasoning models or knowledge retrieval means to integrate external knowledge from various data sources, while the present application uses a large language model to process the operation and maintenance work order combined with prompt information, and obtain entity information from the original operation and maintenance work order to assist the subsequent processing process. The essence is to assist in information mining of the original text based on the guiding content in the prompt information. Compared with the prior art, the advantage of the present application is that there is no need to supervise the training of the algorithm, thereby eliminating the pre-dependence on a large amount of annotated data; in addition, there is no need to build an external knowledge base, let alone involve knowledge base updates and cross-domain knowledge fusion issues, avoiding additional manpower and time investment. Directly using the downstream reasoning capabilities of the large language model to achieve the extraction of entity information in the text of the operation and maintenance work order, significantly improving the efficiency of knowledge acquisition.

[0068] S104: Perform feature splicing on the set of entity information and the operation and maintenance work order to obtain work order splicing features.

[0069] In a possible embodiment, the set of entity information serves as additional domain knowledge, and feature splicing of the set of entity information and the operation and maintenance work order is performed. Although it is only a simple splicing, since the set of entity information is extracted from the text of the operation and maintenance work order, the features contained in the set of entity information are more fine-grained than the external knowledge involved in previous semantic enhancement technologies.

[0070] Typical external knowledge, such as vocabulary, grammatical rules, contextual semantics and events, usually exists in the form of text after being acquired. In contrast, entity information not only contains rich directional text, such as entity description, entity type and location index, but also quantitative numerical priors, such as standardized entity probability and location index, which makes it have more complete potential for expressing domain knowledge features.

[0071] In one possible embodiment, Figure 3 A flow chart for obtaining work order splicing features is shown, such as Figure 3 As shown, the following steps are included:

[0072] S302: Add a label to entity information of each entity type in the set of entity information to obtain labeled entity information.

[0073] Among them, you can choose any label to mark.

[0074] S304: Perform feature splicing on the marked entity information and the operation and maintenance work order to obtain the work order splicing feature.

[0075] Among them, the specific splicing method can be serial splicing, and the feature splicing process of entity information and operation and maintenance work order text sequence can be expressed as:

[0076]

[0077] in, represents the work order concatenation feature; X represents the text sequence in the operation and maintenance work order; H represents the set of entity information; H1, H i Indicates entity information of different entity types; [EXT] indicates a special token tag.

[0078] Marking entity information with labels can help language models better understand the structure and meaning of input text and improve model performance.

[0079] In a possible embodiment, feature splicing is performed with the operation and maintenance work order based on a set of entity information in a target format including entity description, entity type, entity probability, entity start position index and end position index to obtain a work order splicing feature.

[0080] S106: Input the work order concatenation features into the trained pre-trained language model for processing, and output a high-level semantic feature representation tensor.

[0081] In a possible embodiment, it can be expressed by the following formula:

[0082]

[0083] Among them, F h Represents a high-level semantic feature representation tensor; represents the work order concatenation feature; PLM represents the pre-trained language model.

[0084] In a possible embodiment, the work order splicing features are input into a trained pre-trained language model, so that the pre-trained language model can process the operation and maintenance work order according to a set of entity information in a target format, and output a high-level semantic feature representation tensor for characterizing entity types, entity descriptions, relationships between entities, and areas where entities are located.

[0085] It should be noted that, in the entity information of the present application, except for the entity description, the remaining type semantics, type description and probability values ​​in the entity information are all important prompt information, which can provide obvious guiding prompts for the quality inspection of operation and maintenance work orders throughout the whole process, and be used to increase the accuracy of the model in processing the input content.

[0086] Generally speaking, there are logical associations such as semantic causality, transitions, and conditions between different entities. Based on entities, we can better mine such associations with the help of contextual semantics. The entity probability value can be used as a measure of such logical associations to assist in quantifying the degree of uncertainty in the relationship between entities. Entity probability can be used to characterize the association between entities and entity contexts in operation and maintenance work orders. Through entity probability, the pre-trained language model can clarify the association between entities and entity contexts in operation and maintenance work orders, and better understand semantics. Entity probability and entity relationship have a potential positive correlation.

[0087] In addition, as the entity start position index and end position index in this application, since text sequences often have certain long-range dependencies, the entity start and end position indexes can also be considered as an explicit position information encoding, which emphasizes the region of interest (ROI) in a natural and intuitive form, and can help the model to deduplicate redundant information and understand the essential characteristics of the data. The generalization ability of the model is further improved, and it also provides a certain degree of explainability for high-level semantic features from a spatial perspective.

[0088] Exemplarily, the pre-trained language model is used to process the work order splicing features, which can be understood as text encoding. When marked by labels, the token vector embedding dimension can be set to d=768, and the token number limit can be set to n=512. Finally, the set of entity information such as (wavelength division multiplexing system interruption alarm, fault phenomenon description, 0.98213, 0, 0.06896), (equipment system disk failure, fault cause location, 0.97962, 0.36207, 0.42241), (replace system disk, fault solution, 0.93171, 0.45690, 0.50000), (replace system disk, fault solution, 0.94692, 0.62931, 0.67241) is encoded once to obtain a high-level semantic feature representation tensor.

[0089] S108: Input the high-level semantic feature representation tensor into the trained quality inspection model to obtain a predicted quality inspection score.

[0090] In a possible embodiment, the structure of the quality inspection model may be a combination of a linear neural network layer, a normalization processing layer, and a nonlinear activation function, wherein the quality inspection model may use a multilayer perceptron (MLP).

[0091] In a possible embodiment, the predicted quality inspection score is decoded according to the average value of the actual quality inspection scores of multiple operation and maintenance work order samples stored in the trained quality inspection model to obtain the target quality inspection score of the operation and maintenance work order.

[0092] Exemplarily, the predicted quality inspection score may be subjected to exponential calculation, i.e., exp, and multiplied by the average of the actual quality inspection scores of multiple operation and maintenance work order samples to obtain the target quality inspection score.

[0093] For example, it can be processed by the following formula:

[0094] S pred-gt =exp(S pred )·S mean

[0095] Among them, S pred-gt Represents the target quality inspection score after decoding; S pred represents the predicted quality inspection score; S mean Indicates the average value of the actual quality inspection scores of multiple operation and maintenance work order samples.

[0096] In a possible embodiment, the decoding process of the predicted quality inspection score can be integrated into the quality inspection model, and the structure of the quality inspection model can be a linear neural network layer, a normalization processing layer, a nonlinear activation function and a decoding layer. The decoding layer is mainly used to encrypt the predicted quality inspection score to obtain the target quality inspection score.

[0097] For example, for Figure 2 The final target quality inspection score was 96.12 points, and the actual quality inspection score after review by manual quality inspectors and experts was 96 points. The absolute error is only 0.12 points. If the decimal point is not retained, the target quality inspection score is extremely accurate.

[0098] The operation and maintenance work orders are automatically processed in the above manner. The inventor of the present application considers using a model to perform quality inspection on the information in the operation and maintenance work orders. However, in the related technology, once information is processed through an artificial intelligence model, such as classification, prediction, feature extraction, etc., the information is generally processed through a model. In order to further improve the generalization ability of the model and enhance its interpretability, the inventor of the present application naturally extracts key semantic information, which becomes the most critical step in the intelligent quality inspection modeling process of the work order.

[0099] But how can it be accurate without the need for additional training, reliance on a large amount of annotated data, or the construction of an external knowledge base, let alone knowledge base updates and cross-domain substantive integration? The inventors noticed that in daily communication, teachers can point out key paragraphs and verses in the article to provide reference tips independent of the original text, and on this basis, emphasize and interpret them to help enhance the semantic expression of the original text.

[0100] Based on this, the inventors considered that reference prompts can be constructed as an auxiliary means to enhance the directionality of semantic expression in the process of building the model. By constructing prompt information, it is used to prompt the large language model to extract entity information of various entity types from the text sequence information of the operation and maintenance work order, and output it according to the preset format, and the set of output entity information continues to be used as the input of subsequent downstream tasks. The set of entity information includes a large amount of information in the original operation and maintenance work order, which is also conducive to the subsequent pre-trained language model to understand the semantics. The first-layer large language model is guided by the prompt information, and the obtained set of entity information can guide the subsequent quality inspection process, helping the model in the entire process to understand the semantic information in the original operation and maintenance work order, and improve the accuracy of the quality inspection score predicted by the model.

[0101] Therefore, the entities in this application can also be called reference entities (Referential Entity). As a priori hints containing fine-grained domain knowledge, they enrich the text semantic expression paradigm while highlighting the direction of learning contextual associations, assisting the work order quality inspection process to perform accurate automated scoring to ensure the reliability of the quality inspection level results.

[0102] The solution of this application is compared with the current language model processing information. When the current language model is migrated to downstream tasks, this method of building directional reference information to help understand the original semantics is missing. The specific application of this method in the quality inspection direction of operation and maintenance work orders is also missing.

[0103] The inventors also considered that during the quality inspection of operation and maintenance work orders in the prior art, the information in the operation and maintenance work orders that manual quality inspectors pay special attention to is, for example, information such as fault phenomenon description, fault cause location, and fault solution. In this case, this information can be used as entity types and as guiding vocabulary through prompt information to assist model processing. In the overall quality inspection process, the model can effectively understand the semantic information of the work order, accurately quantify the quality inspection level score, and improve the accuracy of the quality inspection score, which will significantly improve the efficiency of periodic, repetitive, and regular quality inspection work and effectively alleviate the manpower requirements and maintenance costs of quality inspectors.

[0104] The quality inspection model used in the above process is a trained model. The following content describes the training process of the quality inspection model. Figure 4 A flow chart of the quality inspection model training process is shown in FIG. Figure 4 As shown, the following steps are included:

[0105] S402: Obtain a training data set, where the training data set includes multiple operation and maintenance work order samples.

[0106] S404: Prompt information and multiple operation and maintenance work order samples are passed through the trained large language model to obtain a set of entity information of multiple entity types in each operation and maintenance work order sample.

[0107] In a possible embodiment, the processing method of the large language model for multiple operation and maintenance work order samples and prompt information can be expressed by the following formula:

[0108] H (m) =LLM(p m ,X m )

[0109] Among them, H (m) represents the set of entity information generated by the mth operation and maintenance work order sample; p m Indicates the prompt information that motivates the large language model to generate entity information; X m Represents the text sequence information of the mth operation and maintenance work order sample.

[0110] The set of entity information of the mth operation and maintenance work order sample can be expressed as:

[0111]

[0112] The composition of the i-th entity information can be expressed as:

[0113]

[0114] S406: Perform feature splicing on the set of entity information corresponding to each operation and maintenance work order sample and each operation and maintenance work order sample to obtain a work order splicing feature sample.

[0115] In a possible embodiment, the work order splicing feature sample can be expressed as:

[0116]

[0117] in, Represents the work order splicing feature sample of the mth operation and maintenance work order sample.

[0118] S408: Input the work order concatenation feature samples into the trained pre-trained language model to obtain a high-level semantic feature representation tensor training sample corresponding to each operation and maintenance work order sample.

[0119] In a possible embodiment, the high-level semantic feature representation tensor training sample corresponding to the mth operation and maintenance work order sample can be expressed as

[0120] S410: Extracting samples to be trained from the high-level semantic feature representation tensor training samples and inputting them into an untrained quality inspection model to obtain predicted quality inspection score samples.

[0121] In a possible embodiment, some high-level semantic feature representation tensor training samples are extracted as training samples and input into an untrained quality inspection model for training, where the quality inspection model takes MLP as an example. The training process can be expressed by the following formula:

[0122]

[0123] in, Represents the predicted quality inspection score sample corresponding to the mth operation and maintenance work order sample.

[0124] S412: Determine the quality inspection loss function value according to the encoded real quality inspection score corresponding to each operation and maintenance work order sample stored in the quality inspection model and the corresponding predicted quality inspection score sample.

[0125] In a possible embodiment, it is necessary to use the actual quality inspection score as a reference to train the predicted quality inspection score samples and determine the quality inspection loss function value.

[0126] To determine the value of the quality inspection loss function, a loss function needs to be used for supervision. The specific formula can be:

[0127]

[0128] Among them, l s Represents the value of the quality inspection loss function; Indicates the actual quality inspection score after encoding.

[0129] In a possible embodiment, the encoded true quality inspection score pre-stored in the untrained quality inspection model is obtained in the following manner: Figure 5 A flowchart for processing the real quality inspection score is shown. Figure 5 As shown, specifically including:

[0130] S502: Add random noise information to the real quality inspection score corresponding to each operation and maintenance work order sample to obtain a noise real quality inspection score.

[0131] In a possible embodiment, slight random noise information is added to the real quality inspection score to alleviate the overfitting phenomenon and improve the robustness of the neural network regression capability. The random noise information is specifically some score values ​​randomly generated by comparing with the real quality inspection score.

[0132] Among them, random noise information can be generated according to the Gaussian distribution law, and its probability density function is as follows:

[0133]

[0134] In this application, random noise information is generated by Gaussian distribution law, and the meanings of the letters in the formula are no longer redundantly described.

[0135] S504: Standardize the actual quality inspection score of the noise to obtain a standardized score.

[0136] In a possible embodiment, the scores are generally obtained by quality inspection by quality inspectors and expert review. The true quality inspection score interval is usually between [0, 100], so the noise true quality inspection score needs to be standardized. On the one hand, it ensures that data of different scales can be processed, and the predicted quality inspection score samples can be calculated and compared within the same range, while ensuring that the neural network parameters will not produce large gradient values ​​after error back propagation, thereby improving the stability of network training.

[0137] S506: Determine an average value of the real quality inspection scores according to the real quality inspection scores corresponding to each operation and maintenance work order sample.

[0138] S508: Perform a logarithmic operation on the ratio of the average value of the real quality inspection score to the real quality inspection score corresponding to each operation and maintenance work order sample to determine the encoded real quality inspection score corresponding to each operation and maintenance work order sample.

[0139] In a possible embodiment, after calculating the average of the real quality inspection scores using the real quality inspection scores of multiple operation and maintenance work order samples, the real quality inspection score of each operation and maintenance work order sample is encoded using the following formula:

[0140]

[0141] in, represents the actual quality inspection score of the mth operation and maintenance work order sample; Represents the actual quality inspection score of the mth operation and maintenance work order sample after encoding.

[0142] The encoded true quality inspection score is obtained in the above manner and is saved in the untrained quality inspection model before training for use.

[0143] S414: Determine whether the quality inspection loss function value meets the preset value; if so, execute step S416; if not, adjust the network parameters and return to step S410.

[0144] S416: Save the network parameters of the current quality inspection model to obtain the trained quality inspection model.

[0145] In a possible embodiment, when modeling the quality inspection model, the operation and maintenance work order samples in the training process may include structured work orders and unstructured work orders. The degree of unstructuredness is very high in the technical fields related to network operation and maintenance. Network operation and maintenance work orders involve a lot of information, including information on network equipment, links, services, faults, etc. This information has different values ​​and functions in different scenarios and backgrounds. Unstructured work orders can better store and express this rich information.

[0146] Therefore, in the process of training, for the technical field of network operation and maintenance, we can mainly extract unstructured operation and maintenance work orders from the domain knowledge base for training, such as Figure 6 As shown, Figure 6 The schematic diagram of the modeling process of intelligent quality inspection of operation and maintenance work orders is shown. In the modeling process, the original text in the unstructured work order, that is, the unstructured operation and maintenance work order sample, is input into the LLM large language model, and the collection of entity information of multiple entity types in each operation and maintenance work order sample can be obtained. The collection of entity information corresponding to each operation and maintenance work order sample is Figure 6 It can be represented by REi, and the set of entity information corresponding to each operation and maintenance work order sample is concatenated with the original text of each operation and maintenance work order sample, and input into the PLM pre-trained language model to obtain the high-level semantic feature representation tensor training sample corresponding to each operation and maintenance work order sample, and then continue to input into the MLP multi-layer perceptron for training. After the MLP is trained, it can be put into use.

[0147] In the entire process of automated quality inspection of operation and maintenance work orders, large language models, pre-trained language models and multi-layer perceptrons are used, among which multi-layer perceptrons are trained specifically. The whole process can be divided into three parts. The first part: through prompt information, entity information is generated with the help of the reasoning ability of the large language model to migrate to downstream tasks. The second part: fine-grained domain knowledge enhancement is achieved by integrating prior entity information, and domain knowledge is enhanced through information such as entity description, type, probability and index position in the entity information. The third part: intelligent quality inspection modeling of work orders under the domain knowledge enhancement strategy, that is, the model training and construction process of MLP.

[0148] It should be noted that Figure 6The content of the input large language model is specially marked. In the entire modeling process, the model is built through the domain knowledge enhancement mode with directional information. The collection of entity information is generated through prompt information. The entity information includes entity description with directional information, entity type, entity start position index and entity end position index. Numerical prior refers to the numerical entity probability, entity start position index and entity end position index. Based on these contents in the entity information, association mining and uncertainty measurement of entity relationships can be realized, and the information mining of the original text can be assisted, so that it has a more complete domain knowledge representation potential, so that the quality inspection process of the operation and maintenance work order can realize the automatic scoring of the work order quality inspection intelligently, efficiently and accurately.

[0149] When the current language model is transferred to downstream tasks, this method of constructing guiding reference information to help understand the original semantics is missing. Figure 7 A flow chart of a data processing method is shown; Figure 7 As shown, the following steps are included:

[0150] S702: Input the prompt information and the data to be processed into the large language model to obtain auxiliary reference information, wherein the large language model extracts directional auxiliary reference information from the data to be processed according to the prompt information, so that the large language model can understand the semantic information in the data to be processed.

[0151] S704: Perform feature splicing on the auxiliary reference information and the data to be processed to obtain the splicing features to be processed.

[0152] S706: Input the splicing features to be processed into the task to be executed to obtain the task execution result.

[0153] In a possible embodiment, the prompt information is input into the large language model to assist the large language model in understanding the semantics of the text sequence and other contents in the processed data, and to extract auxiliary reference information, and the auxiliary reference information is used in downstream tasks to be executed.

[0154] Before being used in the task to be executed, the auxiliary reference information is feature spliced ​​with the data to be processed to obtain the spliced ​​features to be processed, so that the spliced ​​features to be processed can further assist in understanding the information data that is the focus of the data to be processed in the subsequent task to be processed, and obtain the task execution result.

[0155] In a possible embodiment, feature splicing of the auxiliary reference information with the data to be processed to obtain the feature of the splicing to be processed may include: adding tags to the auxiliary reference information to obtain marked reference information, and feature splicing of the marked reference information with the operation and maintenance work order to obtain the feature of the splicing to be processed.

[0156] In one possible embodiment, Figure 7 The method in the above can be used in the specific application scenario of network operation and maintenance work orders, and can also be used in other application scenarios, such as the inspection of operation tickets in power plants, etc. Through the above method, during the processing process, the model can have a more accurate semantic understanding of the data to be processed, acquire knowledge more efficiently, and process faster.

[0157] Based on the same inventive concept, the disclosed embodiment also provides an operation and maintenance work order quality inspection device, such as the following embodiment. Since the principle of solving the problem in the device embodiment is similar to that in the above method embodiment, the implementation of the device embodiment can refer to the implementation of the above method embodiment, and the repeated parts will not be repeated.

[0158] Figure 8 A schematic diagram of the structure of a maintenance work order quality inspection device in an embodiment of the present disclosure is shown. Figure 8 As shown, the operation and maintenance work order quality inspection device 80 includes: an entity information acquisition unit 801, which is used to input prompt information and the operation and maintenance work order into a trained large language model to obtain a set of entity information of multiple entity types in the operation and maintenance work order; the prompt information is used to prompt the trained large language model to extract entity information of multiple entity types from the text sequence information of the operation and maintenance work order; a splicing unit 802, which is used to perform feature splicing on the set of entity information and the operation and maintenance work order to obtain a work order splicing feature; a processing unit 803, which is used to input the work order splicing feature into a trained pre-trained language model for processing to obtain a high-level semantic feature representation tensor; a quality inspection unit 804, which is used to input the high-level semantic feature representation tensor into the trained quality inspection model to obtain a predicted quality inspection score.

[0159] Those skilled in the art will appreciate that various aspects of the present disclosure may be implemented as systems, methods or program products. Therefore, various aspects of the present disclosure may be specifically implemented in the following forms, namely: complete hardware implementation, complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to herein as "circuits", "modules" or "systems".

[0160] Refer to the following Fig. 9 The electronic device 900 according to this embodiment of the present disclosure is described. Fig. 9 The electronic device 900 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0161] like Fig. 9As shown, the electronic device 900 is in the form of a general computing device. The components of the electronic device 900 may include but are not limited to: at least one processing unit 910, at least one storage unit 920, and a bus 930 connecting different system components (including the storage unit 920 and the processing unit 910).

[0162] The storage unit stores a program code, which can be executed by the processing unit 910, so that the processing unit 910 performs the steps of various exemplary embodiments of the present disclosure described in the above “Exemplary Method” section of this specification. For example, the processing unit 910 can perform any step in the above method embodiments.

[0163] The storage unit 920 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 9201 and / or a cache storage unit 9202 , and may further include a read-only storage unit (ROM) 9203 .

[0164] The storage unit 920 may also include a program / utility 9204 having a set (at least one) of program modules 9205, such program modules 9205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0165] Bus 930 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0166] The electronic device 900 may also communicate with one or more external devices 940 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 900, and / or may communicate with any device that enables the electronic device 900 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 950. Furthermore, the electronic device 900 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 960. As shown, the network adapter 960 communicates with other modules of the electronic device 900 via a bus 930. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 900, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0167] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0168] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer program product or a computer program, which includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the method in the above embodiment.

[0169] In an exemplary embodiment of the present disclosure, a computer-readable storage medium is also provided, which may be a readable signal medium or a readable storage medium. A program product capable of implementing the above method of the present disclosure is stored thereon. In some possible implementations, various aspects of the present disclosure may also be implemented in the form of a program product, which includes a program code, and when the program product is run on a terminal device, the program code is used to cause the terminal device to execute the steps according to various exemplary implementations of the present disclosure described in the above “Exemplary Method” section of this specification.

[0170] More specific examples of computer-readable storage media in the present disclosure may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0171] In the present disclosure, a computer readable storage medium may include a data signal propagated in baseband or as part of a carrier wave, wherein a readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A readable signal medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0172] Alternatively, the program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0173] In a specific implementation, the program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., using an Internet service provider to connect through the Internet).

[0174] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0175] In addition, although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.

[0176] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0177] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the appended claims.

Claims

1. A method for quality inspection of operation and maintenance work orders, characterized in that: The method comprises: Input the prompt information and the operation and maintenance work order into the trained large language model to obtain a set of entity information of multiple entity types in the operation and maintenance work order; the prompt information is used to prompt the trained large language model to extract entity information of multiple entity types from the text sequence information of the operation and maintenance work order; the entity information of each entity type in the set of entity information output by the large language model is output according to a preset format; wherein the entity information in the preset format includes: entity description, entity type, and at least one or more of the following information: entity probability, entity start position index, and end position index; Perform feature splicing on the set of entity information and the operation and maintenance work order to obtain a work order splicing feature; Inputting the work order concatenation features into a trained pre-trained language model for processing to obtain a high-level semantic feature representation tensor; Inputting the high-level semantic feature representation tensor into the trained quality inspection model to obtain a predicted quality inspection score; The work order concatenation features are input into the trained pre-trained language model for processing to obtain a high-level semantic feature representation tensor, including: The work order splicing features are input into a trained pre-trained language model so that the pre-trained language model processes the operation and maintenance work order according to a set of entity information in a target format to obtain a high-level semantic feature representation tensor for characterizing entity types, entity descriptions, relationships between entities, and areas where entities are located; the target format includes entity descriptions, entity types, entity probabilities, entity start position indexes, and end position indexes.

2. The method according to claim 1, characterized in that The step of performing feature splicing on the set of entity information and the operation and maintenance work order to obtain a work order splicing feature includes: Adding labels to entity information of each entity type in the set of entity information to obtain labeled entity information; The marked entity information is feature spliced ​​with the operation and maintenance work order to obtain a work order splicing feature.

3. The method according to claim 1, characterized in that The step of performing feature splicing on the set of entity information and the operation and maintenance work order to obtain a work order splicing feature includes: According to a set of entity information in a target format including entity description, entity type, entity probability, entity start position index and end position index, feature splicing is performed with the operation and maintenance work order to obtain the work order splicing feature.

4. The method according to claim 1, characterized in that: After the high-level semantic feature representation tensor is input into the trained quality inspection model for processing to obtain a predicted quality inspection score, the method further includes: The predicted quality inspection score is decoded according to the average value of the real quality inspection scores of multiple operation and maintenance work order samples saved in the trained quality inspection model to obtain the target quality inspection score of the operation and maintenance work order.

5. The method according to claim 1, characterized in that The training process of the quality inspection model includes: Obtain a training data set; the training data set includes multiple operation and maintenance work order samples; Obtaining a high-level semantic feature representation tensor training sample corresponding to each operation and maintenance work order sample according to the plurality of operation and maintenance work order samples by using the trained large language model and the trained pre-trained language model; Extracting samples to be trained from the high-level semantic feature representation tensor training samples and inputting them into an untrained quality inspection model to obtain predicted quality inspection score samples; Determine the quality inspection loss function value based on the encoded real quality inspection score corresponding to each operation and maintenance work order sample saved in the untrained quality inspection model and the corresponding predicted quality inspection score sample; Until the quality inspection loss function value meets the preset value, the network parameters of the current quality inspection model are saved to obtain the trained quality inspection model.

6. The method according to claim 5, characterized in that The method of obtaining a high-level semantic feature representation tensor training sample corresponding to each operation and maintenance work order sample by using the trained large language model and the trained pre-trained language model according to the multiple operation and maintenance work order samples includes: The prompt information and the multiple operation and maintenance work order samples are used in the trained large language model to obtain a set of entity information of multiple entity types in each operation and maintenance work order sample; Perform feature splicing on the set of entity information corresponding to each operation and maintenance work order sample and each operation and maintenance work order sample to obtain a work order splicing feature sample; The work order concatenation feature samples are input into the trained pre-trained language model to obtain the high-level semantic feature representation tensor training samples corresponding to each operation and maintenance work order sample.

7. The method according to claim 5, characterized in that The processing process of the encoded real quality inspection score corresponding to each operation and maintenance work order sample includes: Add random noise information to the actual quality inspection score corresponding to each operation and maintenance work order sample to obtain the noise actual quality inspection score; Standardizing the true quality inspection score of the noise to obtain a standardized score; Determine an average value of the real quality inspection scores according to the real quality inspection scores corresponding to each operation and maintenance work order sample; A logarithmic operation is performed on the ratio of the average value of the real quality inspection score to the real quality inspection score corresponding to each operation and maintenance work order sample to determine the encoded real quality inspection score corresponding to each operation and maintenance work order sample.

8. A data processing method, characterized in that: The method comprises: Inputting prompt information and data to be processed into a large language model to obtain auxiliary reference information; wherein the large language model extracts directional auxiliary reference information from the data to be processed according to the prompt information, so that the large language model understands the semantic information in the data to be processed; the data to be processed is an operation and maintenance work order; the auxiliary reference information includes a collection of entity information of multiple entity types in the operation and maintenance work order; The entity information of each entity type in the set of entity information output by the large language model is output in a preset format; wherein the entity information in the preset format includes: entity description, entity type, and at least one or more of the following information: entity probability, entity start position index, and end position index; Perform feature splicing on the auxiliary reference information and the data to be processed to obtain a splicing feature to be processed; Input the to-be-processed splicing features into the to-be-executed task to obtain the task execution result; Inputting the splicing features to be processed into the task to be executed, and obtaining the task execution result, including: The spliced ​​features are input into a trained pre-trained language model so that the pre-trained language model processes the operation and maintenance work order according to a set of entity information in a target format, obtains a high-level semantic feature representation tensor for characterizing entity types, entity descriptions, relationships between entities, and areas where entities are located, and obtains the task execution result; the target format includes entity descriptions, entity types, entity probabilities, entity start position indexes, and end position indexes.

9. An operation and maintenance work order quality inspection device, characterized in that: include: An entity information acquisition unit, used to input prompt information and an operation and maintenance work order into a trained large language model to obtain a set of entity information of multiple entity types in the operation and maintenance work order; the prompt information is used to prompt the trained large language model to extract entity information of multiple entity types from the text sequence information of the operation and maintenance work order; The entity information of each entity type in the set of entity information output by the large language model is output in a preset format; wherein the entity information in the preset format includes: entity description, entity type, and at least one or more of the following information: entity probability, entity start position index, and end position index; A splicing unit, used for performing feature splicing on the set of entity information and the operation and maintenance work order to obtain a work order splicing feature; A processing unit, used for inputting the work order concatenation features into a trained pre-trained language model for processing to obtain a high-level semantic feature representation tensor; A quality inspection unit, used for inputting the high-level semantic feature representation tensor into a trained quality inspection model to obtain a predicted quality inspection score; The processing unit is also used to input the work order splicing features into a trained pre-trained language model, so that the pre-trained language model processes the operation and maintenance work order according to a set of entity information in a target format, and obtains a high-level semantic feature representation tensor for characterizing entity types, entity descriptions, relationships between entities, and areas where entities are located; the target format includes entity descriptions, entity types, entity probabilities, entity start position indexes, and end position indexes.

10. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to perform the method of any one of claims 1 to 8 by executing the executable instructions.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images