Work order processing method, device and equipment

Abstract

The invention discloses a work order processing method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a plurality of to-be-processed work orders, and extracting key elements of each to-be-processed work order; based on the key elements, extracting a semantic vector of the corresponding to-be-processed work order; and performing clustering analysis according to the semantic vector, determining at least one co-complaint work order from the work orders to be processed, and processing each co-complaint work order. Wherein the key elements of the to-be-processed work order comprise a main body, an event, a place and the like, through semantic vector extraction and clustering analysis, the to-be-processed work orders with the same or similar key elements can be clustered to serve as a co-complaint work order, and then only the co-complaint work order needs to be processed. Therefore, the whole process of processing the work order to be processed does not depend on a predefined label system, and the method can more flexibly adapt to the actual situation that the appeal content is diversified and the enumerable type is not available.

Description

Technical Field

[0001] This application belongs to the field of data processing, and particularly relates to a work order processing method, device, equipment, and storage medium. Background Art

[0002] Natural language processing technology has shown extensive application value in the fields of text understanding and classification, and can be used in specific scenarios such as work order processing.

[0003] In the prior art, pre-training and fine-tuning are the mainstream technical paradigms of natural language processing technology. Among them, the pre-training stage refers to pre-training on a general corpus using an autoencoding or autoregressive model to learn the general laws and features of language; the fine-tuning stage refers to fine-tuning the pre-trained model using domain-related data and annotation corpora of specific tasks for a specific task or domain to adapt to the solution of specific domain problems.

[0004] However, pre-training and fine-tuning often rely on a fixed label set, that is, the model needs to process specific category labels and perform classification or prediction based on these labels. However, in the work order processing scenario, it is impossible to define all possible labels in advance, or, over time or with changes in task requirements, the originally defined category labels may need to be adjusted or new labels added, resulting in the model may not be able to effectively generalize to new labels and may require re-fine-tuning or even re-pre-training, which is time-consuming and laborious. Summary of the Invention

[0005] The purpose of the embodiments of this application is to provide a work order processing method, device, equipment, and storage medium, which can solve the problem that the current natural language processing technology cannot adapt to the actual situation where the content of the demands is diverse and not enumerable.

[0006] In a first aspect, the embodiments of this application provide a work order processing method, and the method includes:

[0007] Obtain multiple work orders to be processed, and extract the key elements of each work order to be processed;

[0008] Based on the key elements, extract the semantic vector of the corresponding work order to be processed;

[0009] Perform clustering analysis according to the semantic vector, determine at least one common complaint work order from the work orders to be processed, and process each common complaint work order.

[0010] Optionally, the extracting the key elements of each work order to be processed includes:

[0011] Input the target prompt word and each work order to be processed into the key element extraction model for processing to obtain the key elements corresponding to each work order to be processed.

[0012] Optionally, before inputting the target prompt word and each work order to be processed into the key element extraction model respectively to obtain the key elements corresponding to each work order to be processed, it includes:

[0013] Obtain multiple first sample work orders and the first sample elements of the first sample work orders;

[0014] Input the pre-obtained initial prompt word and the first sample work order into the key element extraction model for processing to obtain predicted elements;

[0015] Calculate the loss value between the predicted elements and the first sample elements;

[0016] In the case that the loss value does not meet the first preset condition, adjust the initial prompt word, and return to the step of inputting the pre-obtained initial prompt word and the first sample work order into the key element extraction model for processing until the loss value meets the first preset condition, and use the current initial prompt word as the target prompt word.

[0017] Optionally, before inputting the pre-obtained initial prompt word and the first sample work order into the key element extraction model for processing to obtain predicted elements, it further includes:

[0018] Obtain multiple second sample work orders and the second sample elements of the second sample work orders;

[0019] Initialize and generate an initial prompt word based on the second sample elements.

[0020] Optionally, the extracting the semantic vector of the corresponding work order to be processed based on the key elements includes:

[0021] For each work order to be processed, input the corresponding key elements into the work order rewriting model for processing to obtain a rewritten work order;

[0022] Input the rewritten work order into the autoencoding language model for processing to extract the semantic vector of the work order to be processed.

[0023] Optionally, the performing clustering analysis according to the semantic vector to determine at least one common complaint work order from the work orders to be processed includes:

[0024] Use the grid search method to determine the target hyperparameters;

[0025] Based on the target hyperparameters, perform clustering analysis on the semantic vectors to determine at least one common complaint work order from the work orders to be processed.

[0026] Optionally, the processing of each common complaint work order includes:

[0027] Perform semantic analysis on each complaint work order to generate a summary and a title for each complaint work order.

[0028] In a second aspect, an embodiment of the present application provides a work order processing device, and the device includes:

[0029] An acquisition module, configured to acquire a plurality of work orders to be processed and extract key elements of each work order to be processed;

[0030] An extraction module, configured to extract a semantic vector corresponding to the work order to be processed based on the key elements;

[0031] An analysis module, configured to perform clustering analysis according to the semantic vector, determine at least one complaint work order from the work orders to be processed, and process each complaint work order.

[0032] In a third aspect, an embodiment of the present application provides an electronic device, which includes a processor and a memory. The memory stores a program or instruction that can run on the processor. When the program or instruction is executed by the processor, the steps of the method described in the first aspect are implemented.

[0033] In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the method described in the first aspect are implemented.

[0034] In a fifth aspect, an embodiment of the present application provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor, and the processor is configured to run a program or instruction to implement the method described in the first aspect.

[0035] In a sixth aspect, an embodiment of the present application provides a computer program product, which is stored in a storage medium and is executed by at least one processor to implement the method described in the first aspect.

[0036] As can be seen from the above, after acquiring the work orders to be processed, the semantic vector of the work orders to be processed can be further obtained by extracting the key elements in the work orders to be processed. Furthermore, based on the semantic vector of the work orders to be processed, clustering analysis is performed on the work orders to be processed, and complaint work orders are determined from the work orders to be processed. Among them, the key elements of the work orders to be processed include the subject, event, location, etc. Through the extraction and clustering analysis of the semantic vector, it is possible to cluster the work orders to be processed with the same or similar key elements as a complaint work order, and then only the complaint work order needs to be processed. Therefore, the overall process of processing the work orders to be processed does not depend on a pre-defined tag system, and can more flexibly adapt to the actual situation where the appeal content is diverse and not enumerable. Description of the Drawings

[0037] Figure 1 is a flowchart of a work order processing method shown according to an exemplary embodiment;

[0038] Figure 2 is a schematic diagram of prompt optimization shown according to an exemplary embodiment;

[0039] Figure 3 is a flowchart of self - encoding language model training shown according to an exemplary embodiment;

[0040] Figure 4 is an example schematic diagram of a work order processing method shown according to an exemplary embodiment;

[0041] Figure 5 is a block diagram of a work order processing device shown according to an exemplary embodiment;

[0042] Figure 6 is a block diagram of an electronic device shown according to an exemplary embodiment;

[0043] Figure 7 is a schematic diagram of the hardware structure of an electronic device shown according to an exemplary embodiment. Detailed implementation manners

[0044] Next, the technical solutions in the embodiments of the present application will be clearly described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art belong to the scope of protection of the present application.

[0045] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, rather than to describe a specific order or sequence. It should be understood that such used data can be interchanged under appropriate circumstances so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. generally belong to the same category, and the number of objects is not limited. For example, the first object can be one or multiple. In addition, "and / or" in the specification and claims means at least one of the connected objects, and the character " / ", generally represents an "or" relationship between the associated objects before and after.

[0046] First, the nouns mentioned in the present application are explained:

[0047] Natural Language Understanding: Refers to the process of using computer technology to analyze and understand natural language. It involves multiple aspects such as semantic analysis, syntactic analysis, and context understanding, aiming to convert human language into a form that can be understood and processed by computers.

[0048] Autoregressive Model: Refers to a type of generative model used to generate subsequent parts based on the previous parts of a sequence. Its core idea is that the output at the current time can be predicted from the outputs at previous times. Typical application scenarios include language modeling, machine translation, and text generation, etc.

[0049] Autoencoder Model: Refers to an unsupervised learning model used for learning data representations. It is mainly used for feature extraction, dimensionality reduction, denoising, and generation tasks.

[0050] Large Language Model: Refers to a large-scale neural network language model trained on a vast amount of text data, usually including hundreds of millions to billions of parameters. Large models generally require a large amount of computing resources and datasets for training, but have high accuracy and generalization ability in complex tasks.

[0051] Few-shot Learning: Refers to a machine learning method aiming to learn how to classify or predict new data with only very few training samples. Few-shot learning usually includes techniques such as meta-learning, transfer learning, and template matching, which can be used to solve the problems of overfitting or underfitting that traditional machine learning methods are prone to when the amount of data is insufficient.

[0052] Prompt: Refers to a mechanism used to help the model understand the text context, usually using context information to determine the meaning of a certain word. A guiding word / prompt can be a single word, a phrase, or an entire sentence, and its role is to provide additional information for the model to better understand and process the input text.

[0053] Work Order: Refers to documents or electronic forms in the form of various work requests, complaints, consultations, etc. received by certain institutions or departments, which usually need to be processed and replied to.

[0054] Common Complaint Work Orders: Refers to a collection of work orders that involve similar locations, reflect similar content, and similar events.

[0055] Text Classification Label System: Refers to a set of predefined categories or label collections used in text classification tasks. This system provides a structured framework for text classification, helping to classify text content into specific categories.

[0056] Hyperparameter: Refers to a special type of parameter in machine learning or deep learning algorithms. Different from the ordinary parameters that the model can learn during training, hyperparameters are manually set before training. They are used to control the learning process itself and can be understood as parameters for configuring the corresponding learning algorithm.

[0057] In the field of text understanding and classification, natural language processing technology has demonstrated extensive application value. Pre-training and fine-tuning are the mainstream technical paradigms of natural language processing technology. Among them, the pre-training stage refers to pre-training on a general corpus using an autoencoder or autoregressive model to learn the general laws and features of language; the fine-tuning stage refers to, for a specific task or domain, using domain-related data and annotated corpus of specific tasks to fine-tune the pre-trained model to adapt to the solution of specific domain problems.

[0058] In the prior art, the fine-tuning stage often relies on a fixed set of labels. The model needs to process specific category labels and make classifications or predictions based on these labels. However, in some cases, it is not possible to define all possible labels in advance, or, over time or with changes in task requirements, the originally defined category labels may need to be adjusted or new labels added.

[0059] In this case, the model may not be able to effectively generalize to new labels, and it is necessary to re-fine-tune, and even possibly re-pre-train, which is time-consuming and laborious, and there is room for further improvement in the flexibility of the category label system. Based on this, the present application proposes a work order processing method to solve the above problems.

[0060] The following combines the accompanying drawings to detail the work order processing method provided by the embodiments of the present application through specific embodiments and their application scenarios.

[0061] Figure 1 It is a flowchart of a work order processing method shown according to an exemplary embodiment. The work order processing method includes the following steps.

[0062] In step S11, obtain multiple work orders to be processed and extract the key elements of each work order to be processed.

[0063] In practical applications, the work orders to be processed may come from multiple channels, such as the complaint hotline of the customer service center, the online customer service system, emails, social media, etc. These work orders usually contain key information such as the customer's problem description, contact information, time and location of the problem occurrence.

[0064] After obtaining the work orders to be processed, the key elements of each work order to be processed can be extracted. These key elements are the basis for subsequent processing and analysis. For example, the key elements may include the subject, event, location, etc. To extract these key elements, some technical means can be adopted, such as natural language processing technology, machine learning technology, etc., to automatically extract key information from the text of the work order to be processed and organize it into a structured data format for subsequent processing and analysis.

[0065] For example, a set of work orders O that meet specific conditions can be screened out from a large amount of work order data. The set of work orders includes multiple work orders to be processed. Among them, the specific conditions may be set based on multiple dimensions such as the content, time, location, and priority of the work order data. The purpose of setting specific conditions is to narrow the processing scope and improve the efficiency and accuracy of subsequent steps.

[0066] In one implementation, the key elements of each work order to be processed are extracted, including:

[0067] The target prompt word and each work order to be processed are respectively input into the key element extraction model for processing to obtain the key elements corresponding to each work order to be processed.

[0068] That is to say, based on the target prompt word P and the key element extraction large model LLM E , the key elements E of the work order to be processed are extracted. Among them, the key element extraction model LLM E is a model specifically designed to extract key elements from work orders to be processed. The target prompt word P plays a crucial role in the key element extraction process and can provide additional context information for the key element extraction model LLM E to help the key element extraction model LLM E more accurately identify and understand the content of the work order to be processed, and then accurately extract the key elements.

[0069] Specifically, first, the target prompt word P and each work order to be processed can be respectively input into the key element extraction model LLM E , and the key element extraction model LLM E will use its own algorithms and training data to parse and understand the input target prompt word P and the work order to be processed, so as to extract the key elements related to the target prompt word P included in the work order to be processed.

[0070] This process can be expressed as:

[0071] E = LLM E (P, S)

[0072] where P is the target prompt word, S is the work order to be processed, and E is the key element of the work order to be processed.

[0073] In one implementation, before the target prompt word and each work order to be processed are respectively input into the key element extraction model for processing to obtain the key elements corresponding to each work order to be processed, it includes:

[0074] Obtain multiple first sample work orders and the first sample elements of the first sample work orders;

[0075] Input the pre - obtained initial prompt and the first sample work order into the key element extraction model for processing to obtain predicted elements;

[0076] Calculate the loss value between the predicted elements and the first sample elements;

[0077] In the case where the loss value does not meet the first preset condition, adjust the initial prompt and return to the step of inputting the pre - obtained initial prompt and the first sample work order into the key element extraction model for processing until the loss value meets the first preset condition, and use the current initial prompt as the target prompt.

[0078] It can be understood that in the work order processing flow, extracting the key elements of each work order to be processed is a crucial step. To ensure the accuracy of key element extraction, it is necessary to pre - train and optimize the key element extraction model. Among them, the first sample work order serves as the training data for the key element extraction model, and its quantity should be large enough, and it should cover various types and problems to ensure the generalization ability of the key element extraction model. The first sample work order is manually annotated, so the real first sample elements corresponding to the first sample work order can be obtained.

[0079] That is to say, the initial prompt P 0 and the first sample work order S 1 can be input into the key element extraction model LLM E to obtain predicted elements and calculate the loss value D between the real first sample elements E and the predicted elements through the evaluation model, and then optimize the initial prompt through the prompt optimization model LLM until D meets the first preset condition, and denote the optimized initial prompt as P, which is the target prompt. P

[0080] The loss value is an index that measures the difference between the predicted value output by the key element extraction model and the pre - annotated real value, that is, the difference between the predicted elements and the first sample elements. A specific loss function can be used to calculate the loss value, including but not limited to mean square error, cross - entropy, etc. The smaller the loss value, the closer the predicted value of the key element extraction model is to the real value.

[0081] Among them, the first preset condition can be that the loss value is less than the first threshold, or it can also be that the number of iterations of the loss value reaches the second threshold, etc., which is not specifically limited.

[0082] In one implementation, the definition of the loss value D is as follows:

[0083]

[0084] Among them, Represents the distance metric between the predicted element and the first sample element, E i Represents the i-th first sample element, Represents the i-th predicted element, ‖E i ‖ is the number of elements in the set E i The set E i Is the set of first sample elements, and m is the number of first sample work orders S 1 Of.

[0085] In one implementation, before inputting the pre-obtained initial prompt word and the first sample work order into the key element extraction model for processing to obtain the predicted element, it further includes:

[0086] Obtain multiple second sample work orders and the second sample elements of the second sample work orders;

[0087] Based on the second sample elements, initialize and generate the initial prompt word.

[0088] It can be understood that in the work order processing flow, in order to improve the accuracy and efficiency of the key element extraction model, a series of preprocessing work needs to be carried out before model training, obtain multiple second sample work orders and second sample elements, and initialize and generate the initial prompt word based on the second sample elements.

[0089] Specifically, first, a small sample set S including multiple work order data can be obtained; then, the small sample set S is divided into the set S 0 And the set S 1 , where the set S 1 Is the first sample work order, and the set S 0 Is the second sample work order, used to initialize the initial prompt word P 0 .

[0090] The initial prompt word is designed to guide the key element extraction model to focus on the key elements in the work order to be processed and improve the accuracy of key element extraction. For example, the second sample elements can be statistically analyzed to find the elements with higher occurrence frequencies, and then a set of initial prompt words can be generated based on these elements.

[0091] Figure 2 Is a schematic diagram of prompt word optimization shown according to an exemplary embodiment.

[0092] Among them, first, based on the second sample work order S 0 , initialize and generate the initial prompt word P 0 ; The key element extraction model LLM E Based on the initial prompt word P 0 Perform key element extraction on the first sample work order S 1 To obtain the predicted element The evaluation model calculates the prediction elements The loss value between the first sample element E prompts the prompt optimization model LLM P Based on the loss value, the initial prompt P 0 is adjusted until the loss value meets the first preset condition, and the current initial prompt is used as the target prompt P.

[0093] In step S12, based on the key elements, the semantic vector of the corresponding work order to be processed is extracted.

[0094] After the key elements are extracted, the semantic vector of the corresponding work order to be processed can be generated. The semantic vector converts text information into a numerical vector representation, which can capture the semantic information in the text information, so that similar texts have a close distance in the vector space.

[0095] The key elements are the core information of the work order to be processed and are crucial for understanding the content and intention of the work order to be processed. Therefore, semantic analysis of the work order to be processed can be achieved by extracting the semantic vector.

[0096] Among them, when generating the semantic vector, the following methods can be adopted:

[0097] Weighted processing: Weight the key elements so that they play a more important role in the vector representation. This can be achieved by adjusting the weight parameters of the word embedding or sentence embedding model;

[0098] Feature fusion: Fuse the features of the key elements with other text features to generate a more comprehensive semantic vector. This can be achieved by means such as concatenation, summation, or weighted summation;

[0099] Attention mechanism: Introduce the attention mechanism to focus on the importance of the key elements in the text. The attention mechanism can dynamically adjust the corresponding vector representation according to the importance of the key elements.

[0100] In one implementation, based on the key elements, the semantic vector of the corresponding work order to be processed is extracted, including:

[0101] For each work order to be processed, the corresponding key elements are input into the work order rewriting model for processing to obtain the rewritten work order;

[0102] The rewritten work order is input into the auto-encoding language model for processing to extract the semantic vector of the work order to be processed.

[0103] Among them, the main purpose of the work order rewriting model is to reorganize and optimize the key elements in the work order to be processed, and generate a more standardized and normalized rewritten work order. This step helps to eliminate problems such as redundant information, typos, and colloquial expressions in the original work order, and improve the accuracy and efficiency of subsequent processing. The main purpose of the autoencoder language model is to convert the text information in the rewritten work order into low-dimensional semantic vectors. This step helps to capture the semantic information in the text, so that similar work orders have similar distances in the vector space, providing convenience for subsequent classification, matching and other tasks.

[0104] That is to say, based on the obtained key element E, through the work order rewriting model M m Rewrite the original work order to be processed, and then use the autoencoder language model to extract the semantic vector V of the work order to be processed. Among them, the autoencoder language model can be the pre-trained RetroMAE (Pre-Training Retrieval-oriented Language Models Via MaskedAuto-Encoder) model. This process can be expressed as:

[0105] V = RetroMAE(M m (E))

[0106] Figure 3 It is a flowchart showing the training of an autoencoder language model according to an exemplary embodiment.

[0107] Among them, take "The citizen called to report: There is a problem that a high-voltage wire is 1 meter above the ground across the feedback information road." as an example of the original work order information. This is the input content that the model needs to understand and rewrite.

[0108] Semantic vector extraction: After the original work order information is input into the system, first, in order to train the autoencoder language model, some words in the original work order information will be randomly masked (that is, replaced with placeholders, such as [Mask]). This is shown in the flowchart through example texts such as "The citizen called to report: There is a [Mask] across the [Mask]. There is a [Mask] 1 meter above the ground.", "[Mask] called to [Mask]: There is a problem that a [Mask] is 1 meter above the [Mask] across the [Mask] information road." The purpose of masking is to let the RetroMAE model learn to predict and restore the masked words according to the context information.

[0109] Then, the RetroMAE model is used to extract the semantic vectors of the original work orders. These semantic vectors can capture the key information and context relationships in the work order text. Data exchange can occur between RetroMAE and the Decoder. This is an essential part of the model training process, ensuring that the semantic vectors can be effectively utilized to guide the decoding process of the Decoder.

[0110] Next, the masked work order information is then passed to the Decoder component for processing. The task of the Decoder is to predict and recover the masked words in the original work order information based on the input masked information and the previously extracted semantic vectors.

[0111] After the Decoder decodes, the differences between the decoding results and the original work orders can be compared, and the model parameters of RetroMAE and the Decoder can be adjusted according to the differences. This process will be iterated continuously until the model can accurately recover the masked words based on the semantic vectors and context information.

[0112] In summary, this flowchart describes the training process of the work order semantic rewriting model, which includes key steps such as the input of the original work order information, the extraction of semantic vectors, masking processing, Decoder decoding, and the adjustment of model parameters. Through this process, the model can learn to perform accurate semantic rewriting based on the semantic information and context relationships of the work order text.

[0113] In step S13, clustering analysis is performed based on the semantic vectors to determine at least one common complaint work order from the work orders to be processed, and each common complaint work order is processed.

[0114] Clustering analysis is an unsupervised learning method that can divide the samples in a dataset into several clusters, such that the samples within the same cluster have a high degree of similarity, while the samples between different clusters have a low degree of similarity. In work order processing, through the clustering analysis of semantic vectors, work orders with similar problems or requests can be divided into the same cluster, that is, common complaint work orders. This helps to quickly identify work orders with common problems and facilitates subsequent unified processing.

[0115] In one implementation, performing clustering analysis based on the semantic vectors to determine at least one common complaint work order from the work orders to be processed includes:

[0116] Using the grid search method to determine the target hyperparameters;

[0117] Based on the target hyperparameters, performing clustering analysis on the semantic vectors to determine at least one common complaint work order from the work orders to be processed.

[0118] Among them, Grid Search is an exhaustive search method that optimizes model performance by traversing given parameter combinations. In clustering analysis, Grid Search can help find the optimal clustering algorithm parameters, such as the number of clusters, distance metric methods, initialization methods, etc., thereby improving the clustering effect.

[0119] After determining the target hyperparameters, it is necessary to select a suitable clustering algorithm to perform clustering analysis on the semantic vectors. Commonly used clustering algorithms include K-means, DBSCAN (Density-Based Spatial Clustering of Applications with Noise), hierarchical clustering, HDBScan (Hierarchical Density-Based Spatial Clustering of Applications with Noise), etc. According to the distribution characteristics of the data and business requirements, the most suitable clustering algorithm can be selected.

[0120] That is to say, the most suitable hyperparameters θ can be obtained through the Grid Search method using clustering models C such as HDBScab. c , and automatically determine the number n of co-occurring cases;

[0121]

[0122] Among them, O 1 ,..., O n represents the determined co-occurring work orders, and V o represents the semantic vectors corresponding to the set O of work orders to be processed.

[0123] In one implementation, processing each co-occurring work order includes:

[0124] Performing semantic analysis on each co-occurring work order to generate a summary and title for each co-occurring work order.

[0125] The main purpose of summary generation is to simplify the long text content in the co-occurring work order into a short and refined text summary. Through the summary, the main content and problems of the work order can be quickly understood, providing convenience for subsequent processing. The main purpose of title generation is to generate a concise, clear, and attractive title for the co-occurring work order. Through the title, the theme and type of the work order can be quickly understood, providing convenience for subsequent classification, retrieval, and other tasks.

[0126] That is to say, for the co-occurring work order O in the co-occurring work order set [O 1 ,..., O n ​i , a corresponding abstract generation prompt can be constructed Title generation prompt Using the large language model LLM B Generate a common complaint work order O i Abstract s of i And title t i .

[0127] This process can be expressed as:

[0128]

[0129] Figure 4 is an example schematic diagram of a work order processing method shown according to an exemplary embodiment. It includes the following processes: constructing a small sample set for extracting key elements of the work order; optimizing the element extraction prompt based on the small sample set for extracting work order elements and the multi-task processing ability of the large model; extracting the key elements of the work order set based on the optimized element extraction prompt; combining the extracted key elements of the work order, and using the work order semantic extraction model to obtain the semantic representation of the work order; automatically selecting the hyperparameters of the HDBScan clustering algorithm by grid search, and performing work order clustering to obtain common complaint cases; realizing the generation of the abstract and title of the common complaint cases through the semantic understanding and semantic generation capabilities of the large model.

[0130] In this way, by constructing a small number of work order element extraction samples, the common complaint cases in the work order set are adaptively identified, adapting to the situation where the appeal content involved in the work order processing process is diverse and not enumerable. At the same time, by adopting the zero-shot and few-shot learning methods, the dependence on labeled data is greatly reduced, and the generalization performance of the method is improved.

[0131] As can be seen from the above, for the technical solution provided by the embodiment of the present application, after obtaining the work order to be processed, the semantic vector of the work order to be processed can be further obtained by extracting the key elements in the work order to be processed. Furthermore, based on the semantic vector of the work order to be processed, clustering analysis is performed on the work order to be processed, and common complaint work orders are determined from the work orders to be processed. Among them, the key elements of the work order to be processed include the subject, event, location, etc. Through the extraction and clustering analysis of the semantic vector, work orders to be processed with the same or similar key elements can be clustered as a common complaint work order, and then only the common complaint work order needs to be processed. Therefore, the overall process of processing the work order to be processed does not depend on a predefined label system, and can more flexibly adapt to the actual situation where the appeal content is diverse and not enumerable.

[0132] For the work order processing method provided by the embodiment of the present application, the execution subject can be a work order processing device. In the embodiment of the present application, the method of the terminal access executed by the work order processing device is taken as an example to illustrate the device of the work order processing method provided by the embodiment of the present application.

[0133] Figure 5 It is a block diagram of a work order processing device shown according to an exemplary embodiment, including:

[0134] An acquisition module 201, configured to acquire a plurality of work orders to be processed and extract key elements of each work order to be processed;

[0135] An extraction module 202, configured to extract a semantic vector corresponding to the work order to be processed based on the key elements;

[0136] An analysis module 203, configured to perform clustering analysis according to the semantic vector, determine at least one common complaint work order from the work orders to be processed, and process each common complaint work order.

[0137] As can be seen from the above, for the technical solution provided by the embodiments of the present application, after acquiring the work order to be processed, the semantic vector of the work order to be processed can be further obtained by extracting the key elements in the work order to be processed. Furthermore, based on the semantic vector of the work order to be processed, clustering analysis is performed on the work order to be processed to determine the common complaint work order from the work orders to be processed. Among them, the key elements of the work order to be processed include the subject, event, location, etc. Through the extraction and clustering analysis of the semantic vector, it is possible to cluster the work orders to be processed with the same or similar key elements as a common complaint work order, and then only the common complaint work order needs to be processed. Therefore, the overall process of processing the work order to be processed does not depend on a predefined tag system and can more flexibly adapt to the actual situation where the appeal content is diverse and not enumerable.

[0138] For the work order processing method provided by the embodiments of the present application, the execution entity may be a terminal access terminal. In the embodiments of the present application, taking the terminal access terminal executing the terminal access method as an example, the device for the work order processing method provided by the embodiments of the present application is described.

[0139] The work order processing device in the embodiments of the present application may be an electronic device or a component in an electronic device, such as an integrated circuit or a chip. The electronic device may be a terminal or other devices other than terminals. Exemplarily, the electronic device may be a mobile phone, a tablet computer, a laptop computer, a handheld computer, a vehicle-mounted electronic device, a Mobile Internet Device (MID), an augmented reality (AR) / virtual reality (VR) device, a robot, a wearable device, an ultra-mobile personal computer (UMPC), a netbook, or a personal digital assistant (PDA), etc. It may also be a server, a Network Attached Storage (NAS), a personal computer (PC), a television (TV), a teller machine, or a self-service machine, etc. The embodiments of the present application do not make specific limitations.

[0140] The work order processing device provided in the embodiments of the present application can implement Figures 1 to 4 each process implemented by the method embodiments. To avoid repetition, it will not be elaborated here.

[0141] Optionally, as Figure 6 shown, the embodiments of the present application further provide an electronic device 500, including a processor 501 and a memory 502. A program or instruction that can run on the processor 501 is stored on the memory 502. When the program or instruction is executed by the processor 501, it implements each step of the above work order processing method embodiment and can achieve the same technical effect. To avoid repetition, it will not be elaborated here.

[0142] It should be noted that the electronic devices in the embodiments of the present application include the above-mentioned mobile electronic devices and non-mobile electronic devices.

[0143] Figure 7 Schematic diagram of the hardware structure of an electronic device for implementing the embodiments of the present application.

[0144] The electronic device 1000 includes, but is not limited to: a radio frequency unit 1001, a network module 1002, an audio output unit 1003, an input unit 1004, a sensor 1005, a display unit 1006, a user input unit 1007, an interface unit 1008, a memory 1009, and a processor 1010 and other components.

[0145] Those skilled in the art can understand that the electronic device 1000 may further include a power source (such as a battery) for powering each component. The power source can be logically connected to the processor 1010 through a power management system, so as to realize functions such as management of charging, discharging, and power consumption management through the power management system. Figure 7 The structure of the electronic device shown in Figure 7 does not limit the electronic device. The electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements, which will not be elaborated here.

[0146] As can be seen from the above, in the technical solution provided by the embodiments of the present application, after obtaining the work order to be processed, the semantic vector of the work order to be processed can be further obtained by extracting the key elements in the work order to be processed. Furthermore, based on the semantic vector of the work order to be processed, clustering analysis is performed on the work order to be processed to determine the co-complaint work order from the work orders to be processed. Among them, the key elements of the work order to be processed include the subject, event, location, etc. Through the extraction and clustering analysis of the semantic vector, it is possible to cluster the work orders to be processed with the same or similar key elements as a co-complaint work order, and then only the co-complaint work order needs to be processed. Therefore, the overall process of processing the work order to be processed does not depend on a predefined tag system and can more flexibly adapt to the actual situation where the content of the demands is diverse and not enumerable.

[0147] It should be understood that in the embodiments of the present application, the input unit 1004 may include a Graphics Processing Unit (GPU) 10041 and a microphone 10042. The graphics processor 10041 processes the image data of static pictures or videos obtained by an image capture device (such as a camera) in a video capture mode or an image capture mode. The display unit 1006 may include a display panel 10061, and the display panel 10061 may be configured in the form of a liquid crystal display, an organic light emitting diode, etc. The user input unit 1007 includes at least one of a touch panel 10071 and other input devices 10072. The touch panel 10071 is also called a touch screen. The touch panel 10071 may include two parts: a touch detection device and a touch controller. The other input devices 10072 may include, but are not limited to, a physical keyboard, function keys (such as volume control keys, power on / off keys, etc.), a trackball, a mouse, a joystick, which will not be elaborated here.

[0148] The memory 1009 can be used to store software programs and various data. The memory 1009 may mainly include a first storage area for storing programs or instructions and a second storage area for storing data. Among them, the first storage area can store an operating system, application programs or instructions required for at least one function (such as a sound playback function, an image playback function, etc.). In addition, the memory 1009 can include volatile memory or non-volatile memory, or the memory 1009 can include both volatile and non-volatile memory. Among them, the non-volatile memory can be a Read-Only Memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an Electrically Erasable PROM (EEPROM), or a flash memory. The volatile memory can be a Random Access Memory (RAM), a Static RAM (SRAM), a Dynamic RAM (DRAM), a Synchronous DRAM (SDRAM), a Double Data Rate SDRAM (DDR SDRAM), an Enhanced SDRAM (ESDRAM), a Synch link DRAM (SLDRAM), and a Direct Rambus RAM (DRRAM). The memory 109 in the embodiments of the present application includes, but is not limited to, these and any other suitable types of memory.

[0149] The processor 1010 may include one or more processing units; optionally, the processor 1010 integrates an application processor and a modem processor. Among them, the application processor mainly processes operations related to the operating system, user interface, and application programs, etc., and the modem processor mainly processes wireless communication signals, such as a baseband processor. It can be understood that the above modem processor may not be integrated into the processor 1010.

[0150] The embodiments of the present application also provide a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, it implements each process of the above embodiment of the work order processing method and can achieve the same technical effect. To avoid repetition, it will not be elaborated here.

[0151] Among them, the processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory ROM, random access memory RAM, magnetic disks, or optical discs, etc.

[0152] Another embodiment of the present application provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement each process of the above embodiment of the work order processing method, and can achieve the same technical effects. To avoid repetition, it will not be elaborated here.

[0153] It should be understood that the chip mentioned in the embodiments of the present application may also be referred to as a system-on-chip, system chip, chip system, or system-on-chip, etc.

[0154] The embodiments of the present application provide a computer program product. The program product is stored in a storage medium and is executed by at least one processor to implement each process of the above embodiment of the work order processing method, and can achieve the same technical effects. To avoid repetition, it will not be elaborated here.

[0155] It should be noted that in this article, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including that element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in a reverse order according to the functions involved. For example, the methods described may be performed in an order different from that described, and various steps may be added, omitted, or combined. Additionally, the features described with reference to certain examples may be combined in other examples.

[0156] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-described example methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. The computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions for causing a terminal (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present application.

[0157] The embodiments of the present application have been described above in conjunction with the accompanying drawings. However, the present application is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present application, those of ordinary skill in the art can also make many forms without departing from the purpose of the present application and the scope protected by the claims, and all of them belong to the protection scope of the present application.

Claims

1. A work order processing method, characterized in that: The method comprises: Get multiple pending work orders and extract the key elements of each pending work order; Based on the key elements, extracting the semantic vector of the corresponding work order to be processed; Cluster analysis is performed based on the semantic vectors, at least one common work order is determined from the work orders to be processed, and each common work order is processed.

2. The work order processing method according to claim 1, characterized in that: The key elements of each work order to be processed are extracted, including: The target prompt word and each pending work order are respectively input into the key element extraction model for processing to obtain the key elements corresponding to each pending work order.

3. The work order processing method according to claim 2, characterized in that: The target prompt word and each work order to be processed are respectively input into the key element extraction model for processing to obtain the key element corresponding to each work order to be processed, including: Acquire a plurality of first sample work orders and first sample elements of the first sample work orders; Inputting the pre-acquired initial prompt words and the first sample work order into a key factor extraction model for processing to obtain prediction factors; Calculating a loss value between the prediction element and the first sample element; When the loss value does not meet the first preset condition, the initial prompt word is adjusted, and the process returns to the step of inputting the pre-acquired initial prompt word and the first sample work order into the key element extraction model for processing until the loss value meets the first preset condition, and the current initial prompt word is used as the target prompt word.

4. The work order processing method according to claim 3, characterized in that: Before inputting the pre-acquired initial prompt word and the first sample work order into the key factor extraction model for processing to obtain the prediction factor, the method further includes: Acquire a plurality of second sample work orders and second sample elements of the second sample work orders; Initial prompt words are generated based on the second sample elements.

5. The work order processing method according to claim 1, characterized in that: The extracting the corresponding semantic vector of the work order to be processed based on the key elements includes: For each work order to be processed, the corresponding key elements are input into the work order rewriting model for processing to obtain a rewritten work order; The rewritten work order is input into an autoencoding language model for processing, and a semantic vector of the work order to be processed is extracted.

6. The work order processing method according to claim 1, characterized in that: The performing cluster analysis according to the semantic vector to determine at least one common complaint work order from the to-be-processed work orders includes: Use grid search method to determine target hyperparameters; Based on the target hyperparameter, cluster analysis is performed on the semantic vector to determine at least one common work order from the work orders to be processed.

7. The work order processing method according to claim 1, characterized in that: The processing of each joint complaint work order includes: Perform semantic analysis on each co-appeal ticket and generate a summary and title for each co-appeal ticket.

8. A work order processing device, characterized in that: The device comprises: The acquisition module is used to acquire multiple pending work orders and extract key elements of each pending work order; An extraction module, used for extracting the semantic vector of the corresponding work order to be processed based on the key elements; The analysis module is used to perform cluster analysis based on the semantic vector, determine at least one common work order from the work orders to be processed, and process each common work order.

9. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the work order processing method as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by the processor, the steps of the work order processing method according to any one of claims 1 to 7 are implemented.

Images