Work order processing method, device, equipment and storage medium

Abstract

The embodiments of the present application provide a work order processing method, apparatus, device and storage medium. The method includes: obtaining a work order to be classified; inputting the work order to be classified into a preset work order classification model for prediction to obtain a candidate category; screening out a reference work order belonging to the candidate category from the classified alternative work orders in a preset set of alternative work orders; obtaining a reference semantic feature of the reference work order; inputting the work order to be classified into a preset semantic representation model to obtain the semantic feature to be classified of the work order to be classified; determining the target category of the work order to be classified based on the similarity between the semantic feature to be classified and the reference semantic feature of the reference work order, which can reduce the negative impact of the training data distribution error on the semantic representation model and the work order classification model, improve the accuracy of work order classification, and improve the accuracy of the work order being assigned to the client that matches the actual category of the work order.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a work order processing method and device, an electronic device, and a storage medium. Background Art

[0002] To facilitate problem resolution, companies have established a targeted work order system. For example, customer service personnel in various departments create work orders in response to customer inquiries. These orders are then assigned to problem solvers for processing and feedback. Furthermore, departments also require internal coordination, with individuals within the department submitting work orders to resolve issues, which are then assigned to the appropriate personnel for resolution.

[0003] As the scale of various enterprises continues to expand, the categories of work orders used to solve various problems are also increasing rapidly. The category of the work order is used to assign the work order to the client that solves the problem in the work order. If the category of the work order is selected incorrectly, the work order will be assigned to a client that does not match the actual category of the work order, resulting in the problem mentioned in the work order cannot be solved. Summary of the Invention

[0004] An embodiment of the present application provides a work order processing method to reduce the probability of a work order being incorrectly assigned.

[0005] Correspondingly, an embodiment of the present application also provides a work order processing device, an electronic device and a storage medium to ensure the implementation and application of the above method.

[0006] In order to solve the above problems, an embodiment of the present application discloses a work order processing method, which includes:

[0007] Get the work orders to be classified;

[0008] Inputting the work order to be classified into a preset work order classification model for prediction to obtain a candidate category; screening out reference work orders belonging to the candidate category from the classified candidate work orders in a preset candidate work order set;

[0009] Obtaining a reference semantic feature of the reference work order;

[0010] Inputting the work order to be classified into the preset semantic representation model to obtain the semantic features to be classified of the work order to be classified;

[0011] Determining a target category of the work order to be classified based on a similarity between the semantic feature to be classified and a reference semantic feature of the reference work order;

[0012] The work order to be classified is processed based on the target category of the work order to be classified.

[0013] The present application also discloses a work order processing and analysis device, which includes:

[0014] Receiving module, used to obtain the work orders to be classified;

[0015] A screening module, configured to screen out reference work orders belonging to the candidate category from the classified candidate work orders;

[0016] A reference semantic acquisition module is used to obtain the reference semantic features of the reference work order; a semantic extraction module inputs the work order to be classified into the preset semantic representation model to obtain the semantic features to be classified of the work order to be classified;

[0017] an output module, configured to determine a target category of the work order to be classified based on a similarity between the semantic feature to be classified and a reference semantic feature of the reference work order;

[0018] A processing module is used to process the work order to be classified based on the target category of the work order to be classified.

[0019] An embodiment of the present application further discloses an electronic device, comprising: a processor; and a memory on which executable code is stored. When the executable code is executed, the processor executes one or more methods described in the embodiments of the present application.

[0020] The embodiments of the present application also disclose one or more machine-readable media on which executable codes are stored. When the executable codes are executed, the processor executes one or more methods described in the embodiments of the present application.

[0021] Compared with the prior art, the embodiments of the present application have the following advantages:

[0022] In an embodiment of the present application, the candidate category predicted by the work order classification model is used to select a reference work order belonging to the candidate category from the classified alternative work orders. Then, the semantic feature extraction of the work order to be classified and the reference work order can be performed through the semantic representation model. The work order information is classified into the corresponding work order category based on the similarity comparison between the semantic features of the work order to be classified and the reference work order, so that the work order classification model and the semantic representation model complement each other, thereby reducing the negative impact of the training data distribution error on the semantic representation model and the work order classification model, improving the accuracy of work order classification, and further improving the accuracy of work orders being assigned to clients that match the actual category of the work order, thereby improving the processing efficiency of work orders. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a logical architecture diagram of an embodiment of a work order processing method of the present application;

[0024] Figure 2 This is a system diagram of a work order processing method according to an embodiment of the present application;

[0025] Figure 3 This is one of the interface renderings of a method for determining a work order category in an embodiment of the present application;

[0026] Figure 4 This is a schematic diagram of a work order style in an embodiment of the present application;

[0027] Figure 5 This is the second interface rendering of a work order category determination method implemented in this application;

[0028] Figure 6 This is the third interface effect diagram of a method for determining a work order category according to an embodiment of the present application;

[0029] Figure 7 This is a flowchart of the steps of an embodiment of a work order processing method of the present application;

[0030] Figure 8 This is a logic flow chart of a work order allocation method according to an embodiment of the present application;

[0031] Figure 9 This is a structural diagram of an embodiment of a work order processing device of the present application;

[0032] Figure 10 It is a structural diagram of a device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0033] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0034] In the prior art, many scenarios use a single type of machine learning model to predict work order information. For example, representation models derived through unsupervised learning based on algorithms such as TF-IDF (term frequency–inverse document frequency) and Skip-Gram (a word vector extraction algorithm) are used for classification. These representation models are affected by errors in the clustering process of the representation models during unsupervised learning and data distribution, which can affect the accuracy of subsequent classification. This is especially true when there are many work order categories but few training samples and their distribution is uneven. The accuracy of work order classification is low. Alternatively, classification models derived through supervised learning based on algorithms such as SVM (Support Vector Machine) and CNN (Convolutional Neural Network) are used for classification. However, the classification models derived through supervised learning are overly dependent on the accuracy and adequacy of sample set annotations and cannot take into account the different expressions of the same content. When encountering the same content expressed in different ways, it is difficult to accurately identify it. In addition, sample annotation relies on manual labor, which inevitably introduces human errors, resulting in limitations in the content that the classification model can recognize. It is understood that although both of the above two methods can identify the work order category, neither can avoid the negative impact of work order errors on the model classification results.

[0035] Reference Figure 1 , which shows a logical architecture flow chart of a work order processing method according to an embodiment of the present application, including:

[0036] B1, obtain sample work orders by performing stratified sampling on the full set of historical work order data;

[0037] B2, after tokenizing the extracted sample work orders, for example, filtering and / or replacing stop words / invalid words, the extracted work orders are stored in the work order corpus;

[0038] B3, perform unsupervised training on the representation model based on sample work orders in the work order corpus to obtain a semantic representation model;

[0039] B4, perform supervised training on the classification model based on sample work orders in the work order corpus to obtain a work order classification model;

[0040] B5: Deploy the trained semantic representation model and work order classification model.

[0041] B6, provides historical work order information in the work order corpus to the model prediction process.

[0042] The above steps B1 to B5 belong to the model training process of the embodiment of the present application. For details, please refer to the detailed description of model training later. The following describes the model prediction process of the model in work order classification:

[0043] C11, input the work order information to be classified into the work order classification model for prediction;

[0044] C12, the work order classification model outputs the candidate category to which the work order information to be classified belongs;

[0045] C13, the work order classification model also outputs the category probability that the work order to be classified belongs to different candidate categories;

[0046] C14, screening candidate data sets belonging to the candidate category from the historical work order information provided in step B6;

[0047] C15, submitting the candidate dataset to the prediction process of the semantic representation model;

[0048] C21, input the work order information to be classified into the semantic representation model for prediction;

[0049] C22, the semantic representation model outputs the semantic features of the work order information to be classified;

[0050] C23, calculating a similarity score between the reference semantic features of the reference work order in the candidate data set obtained in step C15 and the semantic features to be classified;

[0051] C24, combines the similarity scores of different dimensions of information in the work order to be classified and the reference work order to obtain the target semantic similarity corresponding to the reference work order;

[0052] C3 combines the target semantic similarity corresponding to the reference work order with the probability value of the candidate category to which it belongs to obtain the target score, and filters the target category of the work order information to be classified based on the target score to recommend work order classification to the user.

[0053] It can be understood that the embodiment of the present application can first train a supervised work order classification model and an unsupervised semantic representation model, and then when classifying the work orders to be classified in the future, combine the work order classification model and the semantic representation model to determine a more accurate classification, and then recommend the work order classification to the work order to be classified.

[0054] Reference Figure 2 , which shows a system diagram of a work order processing method according to an embodiment of the present application. In the embodiment of the present application, the system includes: a client 101 and a work order system server 102.

[0055] It should be noted that the client 101 is an electronic device with a work order submission application installed. The client 101 is an electronic device such as a personal computer, mobile terminal, etc. with functions such as information display, data processing, and data transmission. The embodiments of this application do not limit it.

[0056] The work order system server 102 is a server used to provide data support to the client of the work order submission application. The system server 102 is also used to store the work order information together with the work order category to which the work order information belongs. Of course, the system server 102 also has management functions such as querying work orders, adding, deleting and modifying work orders, etc., which are not limited in the embodiments of this application.

[0057] It should be noted that, in the embodiments of this application, Figure 4 The work order information 104 is in the form of electronic text format, such as the work order information compiled by the user using the work order editing software, or the work order information automatically generated by the system according to the work task, so that the client directly extracts the text information in the work order information for subsequent model processing; of course, the work order information 104 is also first obtained by photographing or scanning the paper work order to obtain a work order image, and then using text recognition technology such as OCR (Optical Character Recognition) to recognize the work order image to obtain text information, which is not limited by the embodiment of the present application. For example Figure 4 The work order information 104 includes title information, department information, work content, position information, and of course other work order related information such as job type information and date information, which is not limited in this embodiment of the present application.

[0058] It should be noted that the work order classification model and semantic representation model in the embodiment of the present application are machine learning models pre-trained using corresponding training data. Such machine learning models are, for example, neural network models, support vector machine models, etc., and the present application does not impose any restrictions on them. The model of the embodiment of the present application is a mathematical model. A mathematical model is a scientific or engineering model constructed using mathematical logic methods and mathematical language. A mathematical model is a mathematical structure that is generally or approximately expressed using mathematical language with reference to the characteristics or quantitative dependencies of a certain system of things. This mathematical structure is a pure relational structure of a certain system portrayed by means of mathematical symbols. A mathematical model is one or a group of algebraic equations, differential equations, difference equations, integral equations or statistical equations and their combinations, which quantitatively or qualitatively describe the mutual relationship or causal relationship between the various variables of the system through these equations. In addition to mathematical models described by equations, there are also models described by other mathematical tools, such as algebra, geometry, topology, mathematical logic, etc. Among them, the mathematical model describes the behavior and characteristics of the system rather than the actual structure of the system. Among them, the simulator adopts machine learning, deep learning methods, etc. for model training. Machine learning methods may include linear regression, decision tree, random forest, xgboost, lightgbm, etc. Deep learning methods may include convolutional neural networks (CNN), long short-term memory networks (LSTM), gated recurrent units (GRU), etc.

[0059] Specifically, the work order classification model is a supervised model, which refers to a machine learning model obtained by inputting labeled training samples into the model to be trained, and setting training requirements to make the model's prediction results continuously converge to the expected standards. For example, a work order training set labeled with work order categories is input into the supervised model to be trained, and the work order category is used as the target. After each training, the effect of the trained model is tested by the work order sample test set. If the test does not meet the expectations, the model parameters are adjusted according to the model effect and the training is continued until the model prediction effect meets the expectations. The supervised model is a machine learning model based on algorithms such as Random Forest, Logistics Regression, and Xgboost (a boosting tree based on decision trees). As long as it is a model that outputs probability, the embodiments of the present application do not limit it.

[0060] The semantic representation model is an unsupervised model, which refers to an unsupervised model that inputs unlabeled training samples into the model to be trained so that the model clusters the sample features based on the similarity between the training samples. For example, a large number of unlabeled work order sample training sets are input into the unsupervised model to be trained for clustering, so that the trained unsupervised model outputs the semantic features in the model work order samples. The unsupervised model is a machine learning model based on algorithms such as Fasttext (a word vector and text classification tool), Word2vec (a group of related models used to generate word vectors), GloVe (a vectorized representation algorithm for word meaning), Transformer (machine translation), BERT (Bidirectional Encoder Representation from Transformers, bidirectional language characterizer), etc. As long as the semantic features in the text information are extracted, the embodiments of the present application are not limited to it.

[0061] like Figure 2 In an embodiment of a work order processing method of the present application, the method includes the following process:

[0062] A1: The user initiates an operation to access the work order submission interface on the client 101.

[0063] exist Figure 2 In the example, the user initiates an operation of accessing the work order submission interface in the client 101, such as triggering access to a preset website or application interface in a browser or other application.

[0064] It should be noted that in the case of a client using a mouse or keyboard input method, the user uses the mouse or keyboard to initiate the operation of accessing the work order submission interface. In the case of a client using a touch screen input method, the user uses the touch screen to access the work order submission interface.

[0065] It should be noted that the display interface in the embodiment of the present application is a work order submission interface.

[0066] A2: The client displays the work order submission interface.

[0067] After receiving the user's operation for accessing the work order submission interface, the client draws and renders the work order submission interface and displays the generated interface on the display screen. Figure 3 The work order submission interface 103 is shown, or the work order submission interface 103 is shown. Figure 5 The work order submission interface 105 is shown.

[0068] In actual applications, the client generates the work order submission interface by calling the display component of the work order submission interface to draw and render, or requests the work order system server to obtain the display data of the work order submission interface by accessing the URL link and displaying the work order submission interface in the browser.

[0069] A3: The user enters the work order to be classified in the work order submission interface.

[0070] On the work order submission interface, Figure 3 When the work order submission interface 103 is shown, the user clicks the "Select File" 1031 option on the work order submission interface 103 to select an existing work order file in text format or image format and enter the work order file as shown in FIG. Figure 4 As shown in the work order file, the work order file in image format can be obtained by on-site scanning or photographing the paper work order. For the work order file in text format, the text information can be directly extracted from it as the work order information of the work order to be classified. For the work order file in image format, the text information can be extracted through text recognition technology such as OCR as the work order information of the work order to be classified.

[0071] On the work order submission interface, Figure 5 When using the work order submission interface 105, the user enters the title information and department information of the work order in the work order information input field 1051, the title information input field 10511 and the department information input field 10512 in the work order submission interface 105, as well as information of other dimensions in the work order to be classified, so that the client 101 directly obtains the text information in the work order to be classified based on the user input.

[0072] A4, the client 101 sends the work order information to be classified to the work order system server 102.

[0073] The client 101 sends a target category acquisition request carrying the work order information to be classified to the classification system server 102 through the network connection between the client and the classification system server 102 .

[0074] A5: The client 102 processes the work order to be classified using the work order classification model and the semantic representation model to obtain a target classification.

[0075] After receiving the work order to be classified sent by the client 101, the work order system server 102 first inputs the work order to be classified into the pre-trained work order classification model for classification, thereby obtaining the probability value of the work order to be classified belonging to different work order categories, and screening some work order categories with larger probability values, such as the first 5 or 10 work order categories in descending order of probability values ​​as candidate categories. Then, reference work orders belonging to these candidate categories are screened out from the set of candidate work orders, and the work order to be classified is input into the pre-trained semantic representation model to obtain the semantic features to be classified of the work order to be classified. The reference semantic features of the reference work order are pre-extracted and stored in the set of candidate work orders through the semantic representation model, so there is no need to re-extract them, and the reference semantic features can be directly extracted from the set of candidate work orders. Finally, by comparing the similarity between the semantic features to be classified and the reference semantic features, the target semantic similarity between the work order to be classified and the reference work order can be obtained. Finally, the probability that the work order to be classified belongs to the candidate category of the reference work order is combined with the target semantic similarity to obtain the target score for the work order to be classified as a reference work order. The lowest-level work order category with the highest target score or the first N reference work orders in descending order of scores is used as the final output target category.

[0076] It should be noted that target semantic similarity is achieved by comparing the semantic feature to be classified with the reference semantic features of a single reference work order, thereby obtaining the target semantic similarity corresponding to the candidate category to which each reference work order belongs. For example, assuming there is one semantic feature to be classified and three reference semantic features a, b, and c, the semantic feature to be classified is compared with reference semantic feature a to obtain target semantic similarity a; the semantic feature to be classified is compared with reference semantic feature b to obtain target semantic similarity b; and the semantic feature to be classified is compared with reference semantic feature c to obtain target semantic similarity c. Target semantic similarity a corresponds to candidate category a, target semantic similarity b corresponds to candidate category b, and target semantic similarity c corresponds to candidate category c.

[0077] It should be noted that the bottom-level work order category refers to the bottom-level work order category to which the reference work order belongs, and the candidate category refers to the upper-level work order category to which the bottom-level work order category belongs.

[0078] A6: The work order system server 102 sends the target classification of the work order to be classified to the client 101.

[0079] The work order system server 102 sends the response information carrying the target category to the client 101 through the network connection between the client 101 and the work order system server 102 .

[0080] A7, the client 101 recommends displaying the target category in the work order submission interface.

[0081] After the client 101 processes the work order to be classified by the work order classification model and the semantic representation model to obtain the target category, it refers to Figure 6 , client 101 will Figure 5 In the work order submission interface 105, the option bar originally displayed as "Select work order category" 1052 is replaced with the option bar of "Select work order category" 1054. It can be seen that compared with the option bar 1052, the option bar 1054 adjusts the identified target category as recommended content to the front position of the option bar for display for user reference, so that users can easily know the work order category of the work order to be submitted. For users who submit work order information for the first time, the efficiency of submitting work orders is greatly improved, and the occurrence of errors in submitting work order categories is minimized.

[0082] It should be noted that the work order information to be classified can be sent to the server even if the user does not click the submit button in the work order submission interface. The server can use this information for classification calculations and then return the calculated target classification to the client for display and recommendation, making it convenient for users to quickly select the work order category.

[0083] A8: The user selects the work order category in the work order submission interface.

[0084] User view as Figure 6 The target category recommended in the option bar 1054 in the work order submission interface 105 shown is selected by selecting the target category. Of course, other work order categories are also selected to specify the work order category to be classified. After the selection operation is completed, the user can complete the work order submission operation process by selecting the "Submit" option 1053.

[0085] Of course, the client 101 can also directly use the identified target category as the work order category for the work to be classified, and the user only needs to select the "Submit" option 1053 to confirm.

[0086] It is understood that the target classification obtained by the method for determining the work order classification in the embodiment of the present application is not necessarily the final submitted work order classification of the work order to be classified. The identified target classification is recommended to the user, and the final decision-making power is still on the user side, thereby increasing the user's freedom of choice of work order category when submitting work order information.

[0087] A9, client 101 uploads the work order information and work order classification to the work order system server 102.

[0088] The client 101 receives the user's request Figure 6After selecting the "Submit" option 1053, the work order information and the work order category selected by the user are merged and uploaded to the work order system server 102, so that the work order system server 102 can associate the work order information and the work order category for storage for subsequent archiving and query.

[0089] Therefore, through the above method, in an embodiment of the present application, when the user submits work order information, the client classifies the work order information into the corresponding work order category through a supervised work order classification model and an unsupervised semantic representation model, and recommends it to the user for reference by comparing the similarity between the semantic features of the work order to be classified and the reference work order. This allows the work order classification model and the semantic representation model to complement each other, thereby reducing the negative impact of the training data distribution error on the semantic representation model and the work order classification model, improving the accuracy of work order classification, and thereby improving the accuracy of work orders being assigned to clients that match the actual category of the work order, thereby improving the efficiency of work order processing. It should be noted that since the application described accelerates the category recognition speed, a quick response can be achieved, and the work order can be automatically classified after the user enters the title and other content in the work order submission interface.

[0090] Reference Figure 7 , which shows a work order processing method according to an embodiment of the present application, the method includes the following steps:

[0091] Step 201: Obtain the work order to be classified.

[0092] It should be noted that, as described above, the source file of the work order to be classified can be in text or image format. When the source file is in text format, the text information is extracted from it as the work order to be classified. When the source file is in image format, the image information is first recognized using OCR recognition technology or other text recognition technology, and the text information is extracted from it as the work order to be classified.

[0093] In actual applications, considering that the work orders to be classified cannot be directly used as model input, it is necessary to pre-segment the sentences in the work orders to be classified using a word segmentation tool. For example, the "Legal Compliance and Risk Control Department" is segmented into the word forms of "Legal Compliance", "and", "Risk Control", and "Department", and the "Network and Information Security Department" is segmented into the word forms of "Network", "and", "Information Security", and "Department". The word segmentation tools use Hanlp word segmenter, Jieba word segmenter, etc., which are not limited in the embodiments of this application.

[0094] Optionally, step 201 includes receiving a work order classification request sent by a client and carrying the work order to be classified.

[0095] In an embodiment of the present application, the client, as described above, can be a client of an application with a work order submission function. The user inputs the work order to be classified in the client so that the client generates a work order classification request carrying the work order to be classified, and sends the work order classification request to the server so that the server classifies the work order to be classified through the work order processing method shown in the embodiment of the present application.

[0096] Step 202: Input the work order to be classified into a preset work order classification model for prediction to obtain candidate categories.

[0097] It should be noted that, as described above, the work order classification model is an algorithm model obtained in advance by conducting supervised training on the classification model. The machine learning model adopts a deep learning or machine learning model other than the Fasttext algorithm model, such as Random Forest, Logistics Regression, Xgboost algorithm model, etc. As long as the candidate categories of the work order to be classified are identified and the probability values ​​belonging to different candidate categories are output, the embodiment of the present application does not limit it. For example: after the work order to be classified is input into the preset work order classification model for prediction, the classification probabilities of the three candidate categories are obtained respectively, that is, the probability value of the work order to be classified being input into candidate category 1 is 50%, the probability value of belonging to candidate category 2 is 30%, and the probability value of belonging to candidate category 3 is 20%.

[0098] It should be noted that, as described above, there are multiple category levels for work orders to be classified, and the candidate categories classified by the work order classification model are to identify high-level categories such as the first-level categories or second-level categories in the work orders to be classified, thereby reducing the amount of training for the work order classification model and improving the efficiency of model training. Of course, the candidate category is any category level of the work order to be classified. The lower the category level, the more categories need to be classified, and the greater the amount of training required. Therefore, the category level to which the candidate category belongs is determined based on the affordable computing power, and this embodiment of the present application does not limit it.

[0099] Step 203 , screening out reference work orders belonging to the candidate category from the classified candidate work orders;

[0100] In the embodiment of the present application, a large number of historical work orders can be collected in advance, these work orders can be classified, and classified candidate work orders can be obtained and labeled with classifications. Then, during the online use process, reference work orders of the above candidate categories can be screened from these classified candidate work orders.

[0101] In an embodiment of the present application, an alternative work order set can be preset, and the alternative work order set includes classified reference work orders. Then, in step 203, the reference work orders belonging to the candidate category can be screened out from the alternative work orders in the preset alternative work order set. Of course, the classified alternative work orders may not be stored in the form of an alternative work order set, but may be directly stored in a list or other manner. It should be noted that the preset alternative work order set is a data set obtained by collecting historical work orders in advance and classifying the historical work orders, so the alternative work orders therein are all marked with work order categories. For example: there is an alternative work order set {{A}, {B}, {C}, {D}, {E}...}, and the candidate categories predicted for the work orders to be classified are A, C, and E, then part or all of the alternative work orders are extracted from the sets {A}, {C}, and {E} in the alternative work order set as reference work orders.

[0102] Step 204: obtaining a reference semantic feature of the reference work order;

[0103] The reference semantic features are obtained by processing the reference work order through a preset semantic representation model. In practical applications, for reference work orders, the reference work order can be processed in advance using a semantic representation model to obtain its reference semantic features, and then recorded. For example, the reference semantic features are recorded in correspondence with the reference work order in the aforementioned alternative work order set, or the reference semantic features are recorded in correspondence with the reference work order in the list. This method does not require real-time extraction of reference semantic features, which can save classification calculation time.

[0104] In an embodiment of the present application, the alternative work orders in the alternative work order set can be pre-extracted with semantic features by inputting a semantic representation model, and the semantic features are associated and stored in the alternative work order set. Thus, after the reference work order is screened out during actual use, the reference semantic features corresponding to the reference work order can be extracted from the alternative work order set, eliminating the need to extract semantic features again from the reference work order after each screening, thereby effectively improving the efficiency of work order classification. Of course, if the reference work order is a new work order obtained in real time, semantic features can also be extracted by inputting the reference work order into the semantic representation model after the reference work order is screened out. This embodiment of the present application does not impose any restrictions on this.

[0105] It should be noted that, after obtaining a reference work order, it is also possible to input the reference work order into the semantic representation model to extract the corresponding reference semantic features. This method is equivalent to calculating the reference semantic features of the reference work order in real time, which requires additional time for calculation. In step 205, the work order to be classified is input into the preset semantic representation model to obtain the semantic features to be classified of the work order to be classified.

[0106] It should be noted that, as described above, the semantic representation model is a semantic representation model obtained in advance through unsupervised learning. The semantic representation model is obtained using different natural language processing pre-training methods, such as: Fasttext, Word2vec model, GloVe model, Transformer model, BERT model, etc.

[0107] In practical applications, the semantic representation model at least includes a sub-model for outputting a semantic representation vector, such as a BERT model, etc. Of course, the model for outputting a bidirectional semantic representation vector also includes other models, which are not limited in the embodiments of the present application. Then, each word segment of the work order to be classified is input into the semantic representation layer, and the word vector corresponding to each word segment can be obtained, that is, the semantic feature to be classified. The above-mentioned reference semantic features are similar and will not be repeated here.

[0108] Step 206 : Determine the target category of the work order to be classified based on the similarity between the semantic feature to be classified and the reference semantic feature of the reference work order.

[0109] It should be noted that, when there are multiple levels of work order categories for the work orders to be classified, the target category refers to the work order category at the bottom level, so the category level to which the candidate category belongs is greater than or equal to the category level of the target category. It is understood that the determination of the work order category usually requires determination to the bottom level of the work order category in order to reflect the adequacy of the work order classification. If it is only divided into high-level or mid-level work order categories, the work orders to be classified cannot be accurately classified. Of course, the highest level or mid-level work order categories of the target work order category are not limited in the embodiments of the present application.

[0110] In an embodiment of the present application, the similarity between the semantic feature to be classified and the reference semantic feature is obtained by comparing the cosine similarity method, the sine similarity comparison method, syms (a vector comparison function) of the semantic feature to be classified and the reference semantic feature, or by calculating the Euclidean distance, Manhattan distance, or Chebyshev distance between the two. The embodiment of the present application does not limit it.

[0111] Furthermore, after calculating the similarity between the reference semantic features of each reference work order and the semantic features to be classified, the bottom-level work order category of the reference work order with the highest similarity is taken as the target category of the work order to be classified by similarity sorting. For example: the similarity of reference work order 1 is calculated to be 90%, the similarity of reference work order 2 is 70%, and the similarity of reference work order 3 is 50%, then the bottom-level work order category of reference work order 1 can be selected as the target category of the work order to be classified.

[0112] Alternatively, two similarities of the reference work orders are calculated using two similarity calculation methods, namely similarity 1 and similarity 2. The comprehensive similarities obtained by combining the two similarities 2 are sorted, and the bottom-level work order category of the reference work order with the largest comprehensive similarity value is used as the target category of the work order to be classified.

[0113] This is just an exemplary description. The calculation method and screening method of similarity are set according to actual needs and are not limited in the embodiments of the present application.

[0114] Step 207: Process the work order to be classified based on the target category of the work order to be classified.

[0115] In the embodiment of the present application, after the target category of the work order to be classified is determined, the work order to be classified can be processed based on the target category of the work order to be classified according to actual needs.

[0116] Optionally, step 207 includes: allocating the work order to be classified to a processing client corresponding to the target category.

[0117] It can be understood that in the embodiment of the present application, for one or more work orders to be classified, after determining the target category thereof, they can be assigned to the processing client that handles the target category, so that the corresponding processing personnel can process the work order, such as solving the technical problems raised in the work order, contacting the customers mentioned in the work order to solve the problems raised by the customers, etc. After processing the work order, the processing client can also click to complete the work order and then send a completion notification to the first client. Figure 8 , showing a logic flow chart of a work order assignment method of the present application, wherein user 1 inputs a work order and a work order category to the first client 106, and the first client 106 uploads the work order and work order category input by user 1 to the work order system server 107, and then the work order system server selects user 2 who is responsible for processing the work order according to its corresponding work order category, and then assigns the work order to the second client 108 where user 2 is located, and user 2 processes the work order on the second client 108. In actual applications, users whose service capability range includes the work order category and who are in an idle state can be selected as assignment objects, or the work order can be assigned to the second client of user 2 specified by user 1, and this embodiment of the present application does not limit it. It should be noted that, Figure 8 The second client 108 may be a processing client for processing work orders.

[0118] Optionally, step 201 includes: after receiving input on the work order submission interface of the first client, obtaining the work order to be classified in the work order submission interface.

[0119] It can be understood that in the embodiments of the present application, Figure 6 In the work order submission interface, the work order information entered by the user, such as title information, department information, etc., can be directly sent to the server for processing.

[0120] Furthermore, step 206 includes returning the target category to the first client and displaying it on the work order submission interface for selection. As described in A6 and A7 above, the server can return the confirmed target category to the client, which then displays and recommends the target category on the work order submission interface to facilitate user selection. Alternatively, the first client can directly set the work order category on the work order submission interface to the target category, eliminating the need for the user to select the target category again, thus reducing user complexity.

[0121] In actual applications, the server can send a display instruction carrying the target type to the client, and the display instruction is used to instruct the client to display the target type on the work order submission interface. In an embodiment of the present application, after the server obtains the target category of the work order to be classified according to the work order classification determination method provided in this application, it generates a display instruction carrying the target category and sends the display instruction to the client. After the client parses the display instruction to extract the target category, it displays the target category by updating the work order submission interface so that the user can view the target type. It is worth noting that the target type is only for the user's reference. The user can directly select the target type to submit the work order to be classified to the server, or can independently select other work order types to submit the work order, so as to improve the flexibility of work order category selection when submitting the work order.

[0122] In an embodiment of the present application, the candidate category predicted by the work order classification model is used to screen out reference work orders belonging to the candidate category from the alternative data set, and then the semantic feature extraction of the work order to be classified and the reference work order is performed through the semantic representation model. The work order information is classified into the corresponding work order category based on the similarity comparison between the semantic features of the work order to be classified and the reference work order, so that the work order classification model and the semantic representation model complement each other, thereby reducing the negative impact of the training data distribution error on the semantic representation model and the work order classification model, improving the accuracy of work order classification, and further improving the accuracy of work orders being assigned to clients that match the actual category of the work order, thereby improving the processing efficiency of work orders.

[0123] Optionally, in one embodiment of the present application, step 206 includes:

[0124] Step S11 : determining a target semantic similarity between the semantic feature to be classified and the reference semantic feature.

[0125] In the embodiment of the present application, as described above, the target semantic similarity is obtained using one or more similarity calculation methods. When there are multiple similarity calculation methods, the final target semantic similarity is calculated by weighted summation, maximum value or minimum value, which is not limited in the embodiment of the present application.

[0126] Step S11 : determining a target score for the reference work order based on the target semantic similarity and the probability value of the work order to be classified belonging to the candidate category.

[0127] It should be noted that, as described above, the probability value that the work order to be classified belongs to the candidate category is the probability value output by the work order classification model when outputting the candidate category.

[0128] In an embodiment of the present application, since the work order classification model includes a probability value when outputting a candidate category, the work order classification model indicates the possibility that the work order to be classified belongs to different candidate categories. Therefore, the target semantic similarity and the probability value are comprehensively used to evaluate the target score of each reference work order, so that the work order classification model and the semantic representation model compensate for each other's errors, thereby avoiding the situation where the final target category is significantly different from the actual work order due to the error of a certain model.

[0129] In practical applications, the target score is the weighted sum of the target semantic similarity and the probability value that the work order to be classified belongs to the candidate category, or the positive value of the difference between the probability values, or the product of the probability values. For example, for a reference work order, the target semantic similarity between it and the work order to be classified is 80%, and the probability value that the work order to be classified belongs to the candidate category to which the reference work order belongs is 70%. Then, the target semantic similarity and the probability value are added to obtain a target score of 1.5, or the positive value of the difference between the probability values ​​is taken to obtain a target score of 0.1. This is only an exemplary description and is not limited to this embodiment of the present application.

[0130] Step S12: The bottom-level work order category to which the reference work order with the largest target score belongs is used as the target category of the work order to be classified. The bottom-level work order category is a lower-level category corresponding to the candidate category.

[0131] It should be noted that, considering that the work order categories are divided into multiple levels, for example, the first-level category has several second-level categories, the second-level category has several third-level categories, and so on, the lower the level of the category, the more categories it contains, and the amount of training for the work order classification model also increases dramatically. Therefore, for the work order classification model, by classifying high-level work order categories, such as the first-level or second-level work order categories, the required amount of training is greatly reduced compared to classifying lower-level work order categories. As for the subsequent classification of work order categories at the lower level, the embodiment of the present application is to screen out the reference work order that is most similar to the work order to be classified from the reference work order by comparing the reference semantic features of the reference work order extracted by the semantic representation model and the semantic features to be classified for similarity, thereby taking the lower-level work order category of the reference work order as the final target category of the work order to be classified. The work order category mapping table is used to query the work order categories of all higher levels through the bottom-level work order category, thereby also determining the work order categories of other levels. Of course, in actual applications, categories of other levels can be selected as target categories and trained with these categories during training.

[0132] Referring to Table 1, a category reference table of a reference work order is shown as an example:

[0133]

[0134] Table 1

[0135] Although Reference Work Order 1 and Reference Work Order 2 both belong to Candidate Category 1, their corresponding underlying Categories 1 and 2 are different. Therefore, further screening is performed based on their corresponding target scores. Since 0.81 is greater than 0.72, Reference Work Order 1's underlying Category 1 is selected as the target category for the work order to be classified. This shows that even though the reference work orders belong to the same candidate category and have the same probability of belonging to Candidate Category 1, the specific content of the reference work orders leads to differences in reference semantic features. Ultimately, the target semantic similarity between Reference Work Order 1 and the work order to be classified is greater than that between Reference Work Order 2, resulting in a higher target score for Reference Work Order 1.

[0136] This embodiment of the application uses a combination of supervised and unsupervised learning to classify only the upper-level work order categories of the lower-level work order categories using a work order classification model. For the lower-level work order categories, similarity is measured to measure semantic similarity with reference work orders to classify the work orders into their corresponding specific categories. This effectively reduces the training workload of the work order classification model and improves its training efficiency.

[0137] Optionally, in an embodiment of the present application, step S11 includes: multiplying the target semantic similarity by the probability value of the work order to be classified belonging to the candidate category to obtain the target category of the work order to be classified.

[0138] In an embodiment of the present application, since the target semantic similarity and the probability value are both percentage probabilities less than or equal to 1, by multiplying the target semantic similarity and the probability value, the impact of a single model error on the final classification result is reduced, thereby improving the accuracy of work order classification.

[0139] For example, assume that there are: Reference Work Order 1, whose corresponding target semantic similarity is 0.9 and whose probability value is 0.9, and whose target score is 0.81; Reference Work Order 2, whose corresponding target semantic similarity is 0.8 and whose probability value is 0.9, and whose target score is 0.72; Reference Work Order 3, whose corresponding target semantic similarity is 0.9 and whose probability value is 0.7, and whose target score is 0.63; Reference Work Order 1, whose corresponding target semantic similarity is 0.5 and whose probability value is 0.5, and whose target score is 0.25. It can be seen that although the target semantic similarity of Reference Work Order 3 is higher than that of Reference Work Order 2, its probability value of 0.7 is much lower than the probability value of 0.9 of Reference Work Order 2, so the target score of Reference Work Order 3 is lower than that of Reference Work Order 2. This is obviously because the error in the work order classification model leads to a large gap between the target semantic similarity calculated by the semantic features output by the semantic representation model and the target semantic similarity. By combining the probability value with the target semantic similarity, the negative impact of this error on the final classification result is effectively reduced.

[0140] Optionally, in one embodiment of the present application, step S11 includes:

[0141] Step S111 : when the semantic feature to be classified includes semantic features of at least two dimensions, obtaining first semantic similarities between the semantic feature to be classified and the reference semantic feature in the same dimension respectively.

[0142] It should be noted that the dimensions of semantic features are used to represent different types of information in work order information, such as the title information dimension, the job information dimension, and the time information dimension. It should be understood that since different types of information in work order information are not comparable, for example, a person's name has no correlation with his or her position, when performing semantic similarity comparison, the semantic features of information under the same dimension need to be compared to ensure the accuracy of the similarity comparison.

[0143] like Figure 4There are four dimensions of information in the work order information shown, namely: title, department, position and work content. Therefore, feature extraction is performed on the information of these four dimensions through the semantic representation model to obtain title semantic features, department semantic features, position semantic features and work content semantic features. Therefore, when comparing feature similarity, the title semantic features between the semantic features to be classified and the reference semantic features are compared to obtain title semantic similarity; the department semantic features between the semantic features to be classified and the reference semantic features are compared to obtain department semantic similarity; the position semantic features between the semantic features to be classified and the reference semantic features are compared to obtain position semantic similarity; the work content semantic features between the semantic features to be classified and the reference semantic features are compared to obtain work content semantic similarity. This is just an exemplary description, and the embodiments of the present application are not limited to it.

[0144] Step S112: determining the target semantic similarity according to the first semantic similarity in at least some dimensions.

[0145] In an embodiment of the present application, when there are multiple first semantic similarities in different dimensions, some or all of the first semantic similarities are selected and combined to serve as the target semantic similarity of the reference work order.

[0146] For example, if there is a title semantic similarity of 0.5, a department semantic similarity of 0.7, a position semantic similarity of 0.6, and a work content semantic similarity of 0.6, then only the title semantic similarity of 0.5 and the department semantic similarity of 0.7 are weighted and summed to obtain a target semantic similarity of 0.5*0.5+0.7*0.5=0.6, or the four semantic similarities of 0.5, 0.7, 0.6, and 0.6 are weighted and summed to obtain a target semantic similarity of 0.5*0.25+0.7*0.25+0.6*0.25+0.6*0.25=0.6. This embodiment of the present application is not limited thereto.

[0147] Optionally, in an embodiment of the present application, step S11 includes: calculating the cosine similarity between the semantic feature to be classified and the reference semantic feature to obtain the target semantic similarity.

[0148] In the embodiment of the present application, the target semantic similarity can be calculated by the following formula (1):

[0149]

[0150] Among them, A represents the semantic feature to be classified, B represents the reference semantic feature, similarity A,Brepresents the target semantic similarity, n represents the dimension of the semantic feature to be classified and the reference semantic feature, Ai represents the value of the i-th dimension in the semantic feature to be classified A, and Bi represents the value of the i-th dimension in the reference semantic feature B.

[0151] Optionally, in an embodiment of the present application, the semantic features to be classified include: semantic features of titles to be classified and semantic features of departments to be classified.

[0152] It should be noted that, as described above, the semantic features of the title to be classified are obtained by extracting semantic features from the title information in the work order to be classified using a semantic representation model, and the semantic features of the department to be classified are obtained by extracting semantic features from the department information in the work order to be classified using a semantic representation model. Statistically, it was found that the title information and department information in the work order information have a high correlation between the work order and the work order category at the upper level of the work order. Therefore, the embodiment of the present application uses information of these two dimensions as semantic features to be extracted, which effectively reduces the amount of training for the semantic representation model, and improves the correlation between the semantic features extracted by the semantic representation model and the work order category, thereby improving the accuracy of work order classification.

[0153] Optionally, in one embodiment of the present application, step 202 includes:

[0154] In step S21 , the work order to be classified is input into a preset work order classification model for prediction to obtain probability values ​​of multiple work order categories.

[0155] It should be noted that, as described above, the output of the work order classification model is the probability value of the unclassified work order belonging to different work order categories.

[0156] Step S22: taking N work order categories whose probability values ​​meet preset conditions as candidate categories.

[0157] In the embodiment of the present application, in order to reduce the computational complexity of subsequent similarity comparison, some work order categories whose probability values ​​meet preset requirements are selected as candidate categories participating in subsequent similarity comparison, so N is a positive integer less than or equal to the total number of work order categories.

[0158] In practical applications, the preset condition is that the work order categories are ranked at the front, middle, or back in descending order of probability values. For example, assuming that N is 5, the work order categories ranked at the front 5 in descending order of probability values ​​are selected as candidate categories, or the work order categories in the middle position and the two positions before and after the middle position in the probability value ranking are selected as candidate categories. Of course, this is only an exemplary description, and the embodiments of the present application do not limit it. The embodiments of the present application reduce the workload required for subsequent similarity comparison and improve the efficiency of work order classification by screening some work order categories as candidate work orders based on the probability values ​​output by the work order classification model.

[0159] Optionally, in an embodiment of the present application, the method further includes:

[0160] Step S31: Obtain a first training set, where the first training set includes a first sample work order and a category label corresponding to the first sample work order.

[0161] It should be noted that the first training set is collected through classified historical work orders, and stratified sampling is performed from historical work orders of different work order categories according to a preset ratio. The sampling ratio is determined according to the number of work order categories and the amount of data required for the first training set. After sampling, the first sample work order is added with a category label based on the work order category to which it belongs, completing the sample labeling. Historical work orders are obtained by clustering and grouping historical data according to work order categories, so that historical work orders of different work order categories are divided into one group. When sampling, the group data with the largest number of historical work orders is sampled, and the group data with a smaller number is retained to avoid the situation where the number of work orders of a certain work order category is small and the sampling quantity is insufficient. For example: the amount of data required for the first training set is 10,000, and the number of work order categories is 20. Then, 500 samples are extracted from the historical work orders corresponding to each work order category and input into the first training set. Of course, random sampling is also used to randomly extract from historical work orders, which is not limited in the embodiments of the present application.

[0162] Step S32: Perform supervised training on the classification model based on the first training set to obtain the work order classification model.

[0163] In an embodiment of the present application, after the first training set is divided into a training set and a validation set in a ratio of, for example, 7:3 or 8:2, the category label is used as the training target, the training set is input into the classification model for training, and after each training, the validation set is input into the trained classification model for prediction. The prediction effect of the model is characterized by the loss value calculated by the loss function or the accuracy obtained by comparison between the obtained predicted category and the category label in the validation set. When the prediction effect meets expectations, the supervised training of the model is confirmed to be completed, and the trained classification model is used as the work order classification model. At least the department dimension information in the first sample work order is spliced ​​into a sentence as a prediction factor and input into the classification model for single-label classification. For example, the department information and title information in the first sample work order are spliced ​​into a sentence and input into the classification model for training, or the title information and job type information in the first sample work order are spliced ​​into a sentence and input into the classification model for training. The embodiment of the present application does not limit this.

[0164] In actual applications, the training process of the work order classification model is completed on the work order system server side and then deployed to the client.

[0165] The embodiment of the present application improves the efficiency of work order classification by pre-training a work order classification model for direct use during actual work order classification.

[0166] Optionally, in an embodiment of the present application, the method further includes:

[0167] Step S41: Obtain a second training set, where the second training set includes a second sample work order.

[0168] It should be noted that the second training set is also obtained by referring to the method of obtaining the first training set in step S31, which will not be repeated here. Unlike the first training set, since the semantic model adopts an unsupervised training method, the second sample work order in the second training set does not need to be accompanied by a category label.

[0169] Step S42: performing unsupervised training on the semantic model based on the second training set to obtain the semantic representation model.

[0170] In an embodiment of the present application, as described above, the semantic model is an unsupervised model based on algorithms such as Fasttext, Word2vec, GloVe, Transformer, and BERT. The unlabeled second training set is input into the initialized semantic model for training, so that the semantic model extracts semantic features from the second sample work order based on the correlation between the various dimensions of information in the second sample work order. For example, the title information and department information from the second sample work order are input into the semantic model for training, resulting in a semantic representation model that extracts high-dimensional semantic features from the title information and department information.

[0171] In actual applications, the training process of the semantic representation model is completed on the work order system server side and then deployed to the client.

[0172] The embodiment of the present application improves the efficiency of work order classification by pre-training a semantic representation model for direct use in actual work order classification.

[0173] Optionally, in one embodiment of the present application, step 101 includes: deleting stop words in the work order to be classified; matching invalid information in the work order to be classified through a preset regular expression, and deleting the invalid information and / or replacing the invalid information with preset information, wherein the invalid information includes: at least one of: URL information, digital information, symbol information, repeated information, and error information.

[0174] It should be noted that stop words refer to words that are pre-set and do not participate in model prediction. Specifically, specific stop words can be searched through keyword queries, such as: ah, oh, ha, and other interjections that are not related to the semantics of the work order, etc., and the embodiments of this application do not limit them.

[0175] Invalid information refers to specific types of words and / or symbols that are irrelevant to the ticket's semantics. Examples include URLs in the URL format, numbers such as Arabic numerals and Roman numerals, symbols such as exclamation marks, question marks, and periods, erroneous information such as Martian text and garbled characters, and repeated information. Preset information refers to information that is pre-set to replace specific invalid information.

[0176] In practical applications, you can first write a regular expression for the target field where the invalid information to be processed is located, such as the numeric expression "\d+", which matches pure numbers. Then, delete the parts that are not relevant to the expression semantics, or replace the parts that are relevant to the expression semantics but are too long and messy, such as replacing the URL with a preset information URL, and finally replace them. This embodiment of the application is not limited to this.

[0177] The embodiment of the present application deletes stop words in the work order information to be classified, and deletes and / or replaces invalid information therein, thereby reducing the negative impact of information in the work order information to be classified that is irrelevant to the work order semantics on model prediction and improving the accuracy of work order classification.

[0178] It should be noted that for the method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present application are not limited by the order of the actions described, because according to the embodiments of the present application, certain steps are performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present application.

[0179] Based on the above embodiment, this embodiment further provides a work order processing device, which is applied to electronic devices such as clients and servers.

[0180] Reference Figure 9 , which shows a schematic structural diagram of a work order processing device according to an embodiment of the present application, specifically comprising:

[0181] Receiving module 301, used to obtain the work order to be classified;

[0182] Prediction module 302, used to input the work order to be classified into a preset work order classification model for prediction to obtain candidate categories;

[0183] A screening module 303 is used to screen out reference work orders belonging to the candidate category from the classified candidate work orders;

[0184] A reference semantics acquisition module 304 is configured to acquire reference semantic features of the reference work order;

[0185] The semantic extraction module 305 is used to input the work order to be classified into the preset semantic representation model to obtain the semantic features of the work order to be classified;

[0186] An output module 306 is configured to determine a target category of the work order to be classified based on a similarity between the semantic feature to be classified and a reference semantic feature of the reference work order;

[0187] The processing module 307 is configured to process the work order to be classified based on the target category of the work order to be classified.

[0188] Optionally, the output module 306 is further configured to:

[0189] Determining a target semantic similarity between the semantic feature to be classified and the reference semantic feature;

[0190] Determining a target score for the reference work order based on the target semantic similarity and a probability value of the work order to be classified belonging to the candidate category;

[0191] The bottom-level work order category to which the reference work order with the largest target score belongs is used as the target category of the work order to be classified, where the bottom-level work order category is the lower-level category corresponding to the candidate category.

[0192] Optionally, the output module 306 is further configured to:

[0193] The target semantic similarity is multiplied by the probability value that the work order to be classified belongs to the candidate category to obtain the target category of the work order to be classified.

[0194] Optionally, the output module 306 is further configured to:

[0195] In a case where the semantic feature to be classified includes semantic features of at least two dimensions, respectively obtaining a first semantic similarity between the semantic feature to be classified and the reference semantic feature in the same dimension;

[0196] The target semantic similarity is determined according to the first semantic similarity in at least some dimensions.

[0197] Optionally, the output module 306 is further configured to:

[0198] The cosine similarity between the semantic feature to be classified and the reference semantic feature is calculated to obtain the target semantic similarity.

[0199] Optionally, the semantic features to be classified include: semantic features of titles to be classified and semantic features of departments to be classified.

[0200] Optionally, the prediction module 302 is further configured to:

[0201] Inputting the work order to be classified into a preset work order classification model for prediction to obtain probability values ​​of multiple work order categories;

[0202] The N work order categories whose probability values ​​meet the preset conditions are taken as candidate categories.

[0203] Optionally, the device further comprises: a first training module for

[0204] Obtain a first training set, where the first training set includes a first sample work order and a category label corresponding to the first sample work order;

[0205] Based on the first training set, supervised training is performed on the classification model to obtain the work order classification model.

[0206] Optionally, the device further comprises: a second training module for

[0207] Obtaining a second training set, where the second training set includes a second sample work order;

[0208] According to the second training set, unsupervised training is performed on the semantic model to obtain the semantic representation model.

[0209] Optionally, the receiving module 301 is further configured to:

[0210] Deleting stop words in the work order to be classified;

[0211] And / or, invalid information in the work order to be classified is matched by a preset regular expression, and the invalid information is deleted and / or replaced with preset information, wherein the invalid information includes: at least one of: URL information, digital information, symbol information, repeated information, and error information.

[0212] The optional processing module 307 is further used to: assign the work order to be classified to the processing client corresponding to the target category

[0213] Optionally, the receiving module 301 is further configured to:

[0214] After receiving input on the work order submission interface of the first client, obtaining the work order to be classified in the work order submission interface;

[0215] The processing module 307 is further configured to return the target category to the first client and display the target category on the work order submission interface for selection.

[0216] In an embodiment of the present application, the candidate category predicted by the work order classification model is used to screen out reference work orders belonging to the candidate category from the alternative data set, and then the semantic feature extraction of the work order to be classified and the reference work order is performed through the semantic representation model. The work order information is classified into the corresponding work order category based on the similarity comparison between the semantic features of the work order to be classified and the reference work order, so that the work order classification model and the semantic representation model complement each other, thereby reducing the negative impact of the training data distribution error on the semantic representation model and the work order classification model, improving the accuracy of work order classification, and further improving the accuracy of work orders being assigned to clients that match the actual category of the work order, thereby improving the processing efficiency of work orders.

[0217] An embodiment of the present application further provides a non-volatile readable storage medium, which stores one or more modules (programs). When the one or more modules are applied to a device, the device executes instructions (instructions) of each method step in the embodiment of the present application.

[0218] The present application provides one or more machine-readable media having instructions stored thereon, which, when executed by one or more processors, cause an electronic device to perform one or more of the methods described in the above embodiments. In the present application, the electronic device includes various types of devices such as a client, a server (cluster), etc.

[0219] The embodiments of the present disclosure may be implemented as a device configured as desired using any appropriate hardware, firmware, software, or any combination thereof, and the device may include electronic devices such as a client, a server (cluster), etc. Figure 10 An exemplary apparatus 400 that may be used to implement various embodiments described herein is schematically illustrated.

[0220] For one embodiment, Figure 10 An exemplary apparatus 400 is shown having one or more processors 402, a control module (chip set) 404 coupled to at least one of the processor(s) 402, a memory 406 coupled to the control module 404, a non-volatile memory (NVM) / storage device 408 coupled to the control module 404, one or more input / output devices 410 coupled to the control module 404, and a network interface 412 coupled to the control module 404.

[0221] Processor 402 may include one or more single-core or multi-core processors, and processor 402 may include any combination of general-purpose processors or dedicated processors (e.g., graphics processors, application processors, baseband processors, etc.). In some embodiments, apparatus 400 can function as a client, server (cluster), or other device described in the embodiments of the present application.

[0222] In some embodiments, the apparatus 400 may include one or more computer-readable media (e.g., memory 406 or NVM / storage 408) having instructions 414 and one or more processors 402 configured in conjunction with the one or more computer-readable media to execute the instructions 414 to implement a module to perform the actions described in the present disclosure.

[0223] For one embodiment, control module 404 may include any suitable interface controller to provide any suitable interface to at least one of processor(s) 402 and / or any suitable device or component in communication with control module 404 .

[0224] The control module 404 may include a memory controller module to provide an interface to the memory 406. The memory controller module is a hardware module, a software module, and / or a firmware module.

[0225] The memory 406 can be used, for example, to load and store data and / or instructions 414 for the device 400. For one embodiment, the memory 406 can include any suitable volatile memory, such as a suitable DRAM. In some embodiments, the memory 406 can include double data rate type four synchronous dynamic random access memory (DDR4 SDRAM).

[0226] For one embodiment, the control module 404 may include one or more input / output controllers to provide interfaces to the NVM / storage device 408 and the input / output device(s) 410 .

[0227] For example, NVM / storage 408 may be used to store data and / or instructions 414. NVM / storage 408 may include any suitable non-volatile memory (e.g., flash memory) and / or may include any suitable non-volatile storage device(s) (e.g., one or more hard disk drives (HDDs), one or more compact disk (CD) drives, and / or one or more digital versatile disk (DVD) drives).

[0228] NVM / storage device 408 may include storage resources that are physically part of the device on which apparatus 400 is installed, or it may be accessible to the device without being part of the device. For example, NVM / storage device 408 may be accessible over a network via input / output device(s) 410.

[0229] (One or more) input / output devices 410 may provide an interface for apparatus 400 to communicate with any other appropriate devices, including communication components, audio components, sensor components, etc. Network interface 412 may provide an interface for apparatus 400 to communicate via one or more networks. Apparatus 400 may wirelessly communicate with one or more components of a wireless network according to any of one or more wireless network standards and / or protocols, for example, accessing a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G, 5G, etc., or a combination thereof for wireless communication.

[0230] For one embodiment, at least one of the processor(s) 402 may be packaged together with the logic of one or more controllers (e.g., a memory controller module) of the control module 404. For one embodiment, at least one of the processor(s) 402 may be packaged together with the logic of one or more controllers of the control module 404 to form a system-in-package (SiP). For one embodiment, at least one of the processor(s) 402 may be integrated on the same die with the logic of one or more controllers of the control module 404. For one embodiment, at least one of the processor(s) 402 may be integrated on the same die with the logic of one or more controllers of the control module 404 to form a system-on-chip (SoC).

[0231] In various embodiments, device 400 is, but is not limited to, a server, a desktop computing device, or a client such as a mobile computing device (e.g., a laptop computing device, a handheld computing device, a tablet computer, a netbook, etc.). In various embodiments, device 400 may have more or fewer components and / or a different architecture. For example, in some embodiments, device 400 includes one or more cameras, a keyboard, a liquid crystal display (LCD) screen (including a touchscreen display), a non-volatile memory port, multiple antennas, a graphics chip, an application-specific integrated circuit (ASIC), and a speaker.

[0232] Among them, the main control chip can be used in the device as a processor or control module, sensor data, location information, etc. are stored in the memory or NVM / storage device, the sensor group can be used as an input / output device, and the communication interface may include a network interface.

[0233] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0234] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0235] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, clients (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data analysis client to produce a machine, so that the instructions executed by the processor of the computer or other programmable data analysis client generate instructions for implementing the process in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0236] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data analysis client to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, the instruction device being implemented in the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0237] These computer program instructions may also be loaded onto a computer or other programmable data analysis client so that a series of operational steps are executed on the computer or other programmable client to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable client provide for implementing the process described in the flow. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0238] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they become aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0239] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include," "comprise," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or client that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or client. In the absence of further limitations, an element defined by the phrase "comprises a..." does not exclude the presence of additional identical elements in the process, method, article, or client that includes the element.

[0240] The above is a detailed introduction to a data analysis method and device, an electronic device and a storage medium provided by the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for general technical personnel in this field, based on the ideas of the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A work order processing method, characterized in that: The method comprises: Get the work orders to be classified; Input the work order to be classified into a preset work order classification model for prediction to obtain a candidate category; wherein the work order category has multiple levels, and the category level to which the candidate category belongs is greater than the bottom level; Selecting reference work orders belonging to the candidate category from pre-collected classified candidate work orders; Obtaining a reference semantic feature of the reference work order; Inputting the work order to be classified into the preset semantic representation model to obtain the semantic features to be classified of the work order to be classified; Based on the similarity between the semantic features to be classified and the reference semantic features of the reference work order, the bottom-level work order category of the reference work order with the highest similarity is selected as the target category of the work order to be classified, where the bottom-level work order category is the lower-level category corresponding to the candidate category; The work order to be classified is processed based on the target category of the work order to be classified.

2. The method according to claim 1, characterized in that The selecting, based on the similarity between the semantic feature to be classified and the reference semantic feature of the reference work order, the bottom-level work order category of the reference work order with the highest similarity as the target category of the work order to be classified includes: Determining a target semantic similarity between the semantic feature to be classified and the reference semantic feature; Determining a target score for the reference work order based on the target semantic similarity and a probability value of the work order to be classified belonging to the candidate category; The bottom-level work order category to which the reference work order with the largest target score belongs is used as the target category of the work order to be classified.

3. The method according to claim 2, characterized in that The step of determining the target score of the reference work order based on the target semantic similarity and the probability value of the work order to be classified belonging to the candidate category includes: The target semantic similarity is multiplied by the probability value that the work order to be classified belongs to the candidate category to obtain the target category of the work order to be classified.

4. The method according to claim 2, characterized in that The determining of the target semantic similarity between the semantic feature to be classified and the reference semantic feature includes: In a case where the semantic feature to be classified includes semantic features of at least two dimensions, respectively obtaining a first semantic similarity between the semantic feature to be classified and the reference semantic feature in the same dimension; The target semantic similarity is determined according to the first semantic similarity in at least some dimensions.

5. The method according to claim 2, characterized in that The determining of the target semantic similarity between the semantic feature to be classified and the reference semantic feature includes: The cosine similarity between the semantic feature to be classified and the reference semantic feature is calculated to obtain the target semantic similarity.

6. The method according to any one of claims 1 to 5, characterized in that: The step of inputting the work order to be classified into a preset work order classification model for prediction to obtain candidate categories includes: Inputting the work order to be classified into a preset work order classification model for prediction to obtain probability values ​​of multiple work order categories; The N work order categories whose probability values ​​meet the preset conditions are taken as candidate categories.

7. The method according to claim 1, characterized in that The method further comprises: Obtain a first training set, where the first training set includes a first sample work order and a category label corresponding to the first sample work order; Based on the first training set, supervised training is performed on the classification model to obtain the work order classification model.

8. The method according to claim 1, characterized in that The method further comprises: Obtaining a second training set, where the second training set includes a second sample work order; According to the second training set, unsupervised training is performed on the semantic model to obtain the semantic representation model.

9. The method according to claim 1, characterized in that The step of obtaining the work order to be classified includes: Deleting stop words in the work order to be classified; And / or, invalid information in the work order to be classified is matched by a preset regular expression, and the invalid information is deleted and / or replaced with preset information, wherein the invalid information includes: at least one of: URL information, digital information, symbol information, repeated information, and error information.

10. The method according to claim 1, characterized in that The obtaining of the work order to be classified includes: After receiving input on the work order submission interface of the first client, obtaining the work order to be classified in the work order submission interface; The processing of the work order to be classified based on the target category of the work order to be classified includes: The target category is returned to the first client, and the target category is displayed on the work order submission interface for selection.

11. The method according to claim 1, wherein The processing of the work order to be classified based on the target category of the work order to be classified includes: Allocate the work order to be classified to a processing client corresponding to the target category.

12. A work order processing device, characterized in that: The device comprises: Receiving module, used to obtain work orders to be classified; A prediction module is used to input the work order to be classified into a preset work order classification model for prediction to obtain a candidate category; wherein the work order category has multiple levels, and the category level to which the candidate category belongs is greater than the bottom level; A screening module, configured to screen out reference work orders belonging to the candidate category from pre-collected and classified candidate work orders; A reference semantic acquisition module is used to obtain the reference semantic features of the reference work order; a semantic extraction module is used to input the work order to be classified into the preset semantic representation model to obtain the semantic features to be classified of the work order to be classified; An output module is configured to select, based on the similarity between the semantic features to be classified and the reference semantic features of the reference work orders, the bottom-level work order category of the reference work order with the highest similarity as the target category of the work order to be classified, where the bottom-level work order category is a lower-level category corresponding to the candidate category; A processing module is used to process the work order to be classified based on the target category of the work order to be classified.

13. An electronic device, characterized in that: include: processor; and A memory having executable code stored thereon, which, when the executable code is executed, causes the processor to execute the work order processing method according to any one of claims 1 to 11.

14. One or more machine-readable media having executable codes stored thereon, which, when executed, enable a processor to execute the work order processing method according to any one of claims 1 to 11.

Images