Model training method, device and apparatus, and data processing method

Abstract

The application discloses a model training method, comprising: obtaining model training original data, one piece of original data containing description information of a navigation history question and a corresponding processing script serial number; obtaining a feature vector of the navigation history question according to the description information of the navigation history question; obtaining model training input data, one piece of data input including: a feature vector of a navigation history question and a corresponding processing script serial number; training a matching model by using the model training input data to obtain a target matching model; wherein the target matching model is used for predicting a processing script serial number matched with a navigation question according to a feature vector of the navigation question. The above method is used to improve the processing and solving efficiency of the navigation question.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a model training method and device, electronic equipment and storage equipment, and relates to a data processing method and device. BACKGROUND

[0002] At present, with more and more car factories using navigation systems, more and more customers will import some development and use problems to the internal problem analysis platform, and the developers of the navigation system will be tired of dealing with customer problems.

[0003] Under the prior art, some navigation manufacturers provide a problem feedback platform for their map customers, and customers can only search some fixed known problems online through keywords, and professional navigation problems need to be handled by artificial.

[0004] In addition, the problem feedback platform of another part of navigation manufacturers can only answer the consultation problems raised by customers, and cannot automatically analyze and process other use problems encountered by customers, and the problem solving efficiency is very low.

[0005] Therefore, it is very important to have a solution that can improve the processing and solving efficiency of navigation problems under the condition of limited development manpower and reduce the cost of artificial processing. SUMMARY

[0006] The present application provides a model training method and device, electronic equipment and storage equipment to improve the processing and solving efficiency of navigation problems.

[0007] The present application provides a model training method, comprising:

[0008] Obtaining model training original data, one original data contains description information of navigation history problem and its corresponding processing script serial number;

[0009] According to the description information of the navigation history problem, the feature vector of the navigation history problem is obtained;

[0010] Obtaining model training input data, one data input includes: feature vector of navigation history problem and its corresponding processing script serial number;

[0011] Training the matching model with the model training input data to obtain the target matching model; wherein the target matching model is used to predict the processing script serial number matched with the navigation problem according to the feature vector of the navigation problem.

[0012] Optionally, it further comprises:

[0013] Pretreating the description information of the navigation history problem to obtain the pretreated problem description information;

[0014] The step of obtaining the feature vector of the navigation history problem based on the description information of the navigation history problem includes:

[0015] Based on the preprocessed problem description information, the feature vector of the navigation history problem is obtained;

[0016] The preprocessing of the description information of the navigation history problem includes at least one of the following:

[0017] Low-frequency vocabulary cleaning;

[0018] Punctuation mark cleaning;

[0019] Stop the word cleaning.

[0020] Optionally, the description information of the navigation history problem is preprocessed to obtain preprocessed problem description information, including:

[0021] Extract the main descriptive information from the description information of the navigation history problem;

[0022] The extracted subject description information is semantically segmented to obtain multiple subject description information segments;

[0023] The step of obtaining the feature vector of the navigation history problem based on the preprocessed problem description information includes:

[0024] A preprocessing model is used to perform feature transformation on each segment of the main description information to obtain the feature vector of the navigation history problem.

[0025] Optionally, obtaining the raw data for model training includes:

[0026] Retrieve descriptions of navigation history issues from the database using a script;

[0027] The scripts are used for annotation to generate training data containing descriptive information about navigation history issues and their corresponding processing script numbers.

[0028] Optionally, training the matching model using the input data of the model training method to obtain the target matching model includes:

[0029] The model training input data is input into the matching model, and the model parameters are updated through the backpropagation algorithm and the stochastic gradient descent method to output the target matching model.

[0030] This application also provides a data processing method, including:

[0031] Obtain a description of the navigation problem;

[0032] Based on the description information of the navigation problem, the feature vector of the navigation problem is obtained;

[0033] The feature vector of the navigation problem is input into the target matching model to predict the processing script number that matches the navigation problem;

[0034] Based on the processing script number, obtain the processing script corresponding to the processing script number;

[0035] The navigation problem is analyzed and processed using the processing script.

[0036] Optionally, the step of inputting the feature vector of the navigation problem into the target matching model to predict the processing script number matching the navigation problem includes:

[0037] The feature vector of the navigation problem is input into the target matching model, and the forward search algorithm is used to calculate and output the processing script number with the highest probability.

[0038] Optionally, it also includes: preprocessing the description information of the navigation problem to obtain preprocessed problem description information;

[0039] The step of obtaining the feature vector corresponding to the description information based on the description information of the navigation problem includes:

[0040] Based on the preprocessed problem description information, the feature vector of the navigation problem is obtained.

[0041] Optional, also includes:

[0042] Based on the processing script number, obtain the consulting expert information that matches the navigation problem.

[0043] This application also provides a model training apparatus, comprising:

[0044] The model training raw data acquisition unit is used to obtain the raw data for model training. Each piece of raw data contains a description of the navigation history problem and its corresponding processing script number.

[0045] The feature vector acquisition unit is used to obtain the feature vector of the navigation history problem based on the description information of the navigation history problem;

[0046] The model training input data acquisition unit is used to obtain model training input data. One data input includes: the feature vector of the navigation history problem and its corresponding processing script number;

[0047] The model training unit is used to train a matching model using the model training input data to obtain a target matching model; wherein, the target matching model is used to predict the processing script number that matches the navigation problem based on the feature vector of the navigation problem.

[0048] This application also provides an electronic device, including:

[0049] Processor; and

[0050] The memory stores the program for the model training method. After the device is powered on and the program for the matching model training method is run by the processor, the following steps are performed:

[0051] Obtain the raw data for model training. Each raw data entry contains a description of the navigation history problem and its corresponding processing script number.

[0052] Based on the description information of the navigation history problem, the feature vector of the navigation history problem is obtained;

[0053] Obtain the model training input data. Each data input includes: the feature vector of the navigation history problem and its corresponding processing script number;

[0054] The model is used to train the matching model using the input data to obtain the target matching model; wherein, the target matching model is used to predict the processing script number that matches the navigation problem based on the feature vector of the navigation problem.

[0055] This application also provides a storage device storing a program for a model training method, which is executed by a processor to perform the following steps:

[0056] Obtain the raw data for model training. Each raw data entry contains a description of the navigation history problem and its corresponding processing script number.

[0057] Based on the description information of the navigation history problem, the feature vector of the navigation history problem is obtained;

[0058] Obtain the model training input data. Each data input includes: the feature vector of the navigation history problem and its corresponding processing script number;

[0059] The model is used to train the matching model using the input data to obtain the target matching model; wherein, the target matching model is used to predict the processing script number that matches the navigation problem based on the feature vector of the navigation problem.

[0060] This application also provides a data processing apparatus, including:

[0061] The description information acquisition unit is used to acquire description information about the navigation problem;

[0062] The feature vector acquisition unit is used to obtain the feature vector of the navigation problem based on the description information of the navigation problem;

[0063] The script sequence number prediction unit is used to input the feature vector of the navigation problem into the target matching model and predict the processing script sequence number that matches the navigation problem.

[0064] The processing script obtaining unit is used to obtain the processing script corresponding to the processing script number according to the processing script number.

[0065] The problem analysis and processing unit is used to analyze and process the navigation problem using the processing script.

[0066] Compared with the prior art, this application has the following advantages:

[0067] This application provides a model training method, comprising: obtaining raw model training data, wherein each raw data entry contains descriptive information of a navigation history problem and its corresponding processing script number; obtaining a feature vector of the historical navigation problem based on the descriptive information of the navigation history problem; obtaining model training input data, wherein each input data entry includes: the feature vector of the navigation history problem and its corresponding processing script number; training a matching model using the model training input data to obtain a target matching model; wherein the target matching model is used to predict the processing script number matching the navigation problem based on the feature vector of the navigation problem. The model training method provided in this application trains a matching model based on the feature vector and processing script number of the navigation history problem, thereby training a target matching model. The target matching model is used to obtain the processing script number matching the navigation problem based on the feature vector of the navigation problem, realizing automated matching of the descriptive information of the navigation problem with the corresponding processing script, effectively improving the efficiency of analyzing and solving problems, and reducing the cost of manual processing and problem solving. Attached Figure Description

[0068] Figure 1A This is an application scenario diagram provided in an embodiment of this application.

[0069] Figure 1 This is a flowchart of a model training method provided in the first embodiment of this application.

[0070] Figure 2 This application provides a flowchart for handling navigation problems.

[0071] Figure 3 This is a flowchart of a data processing method provided in the second embodiment of this application.

[0072] Figure 4 This is a schematic diagram of a data processing apparatus provided in the third embodiment of this application. Detailed Implementation

[0073] Numerous specific details are set forth in the following description to provide a full understanding of the invention. However, the invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0074] To better illustrate this application, we will first briefly introduce the application scenarios of the data processing method provided in the embodiments of this application.

[0075] The data processing method provided in this application can be applied to scenarios involving interaction between a client and a server, such as... Figure 1A When a user encounters a navigation problem while using a navigation product, the client typically establishes a connection with the server first. After the connection is established, the client sends a description of the navigation problem to the server. The server first receives the description information through the description information acquisition unit 101; then, through the feature vector acquisition unit 102, it obtains the feature vector of the navigation problem based on the description information; next, the script number prediction unit 103 inputs the feature vector of the navigation problem into the target matching model to predict the processing script number that matches the navigation problem; then, through the processing script acquisition unit 104, it obtains the processing script corresponding to the processing script number; finally, through the problem analysis and processing unit 105, it uses the script to analyze and process the navigation problem to obtain the response information. Afterwards, the server provides the response information to the client, and the client receives the response information.

[0076] The navigation product described in this application can be a smartphone navigation product or an in-vehicle navigation product. In addition, the navigation product can be a dedicated map navigation application software or other application software that integrates map navigation services, such as ride-hailing applications, information life service applications, etc.

[0077] The first embodiment of this application provides a model training method, which will be described below in conjunction with... Figure 1 , Figure 2 Let me introduce it.

[0078] like Figure 1 As shown, in step S101, the original training data of the model is obtained. Each piece of original data contains the description information of the navigation history problem and its corresponding processing script number.

[0079] The navigation history issues refer to problems encountered by customers using the navigation system during development or use. Customers can import navigation history issues into an online problem analysis platform for the navigation system.

[0080] The description of navigation history issues includes at least one of the following: defect description, expected result, actual result, cause identification, and repair suggestions. For example, the defect description for a navigation history issue is: After activating the navigation interface and returning to the system, the buttons are unclickable when re-entering navigation. Another example: The defect description for a navigation history issue is: In the lower right corner of the overview mode, the navigation icon has obvious jagged edges; the expected result is: In the lower right corner of the overview mode, the jagged edges of the navigation icon are not obvious; the actual result is: In the lower right corner of the overview mode, the jagged edges of the navigation icon are obvious.

[0081] The processing script refers to the program code that describes the navigation problem and analyzes and processes the logs.

[0082] The processing script number refers to the number of the processing script. For example, the processing script number is 26260.

[0083] In practice, the raw data for model training can be obtained through the following steps:

[0084] Retrieve descriptions of navigation history issues from the database using a script;

[0085] The scripts are used for annotation to generate training data containing descriptive information about navigation history issues and their corresponding processing script numbers.

[0086] Specifically, annotation can be performed using scripts to generate training data containing descriptive information about navigation history issues and their corresponding processing script numbers. This can include: using Python scripts to assist manual annotation in a semi-automated manner to generate training data containing descriptive information about navigation history issues and their corresponding processing script numbers.

[0087] like Figure 1 As shown, in step S102, the feature vector of the navigation history problem is obtained based on the description information of the navigation history problem.

[0088] In practical implementation, the BERT general semantic model can be used to perform feature transformation processing on the descriptive information of the navigation history problem, extracting the feature vector of the navigation history problem. BERT, a general language feature model, stands for Bidirectional Encoder Representation from Transformers.

[0089] As an implementation method, in order to improve the accuracy of model prediction, the first embodiment of this application may further include: preprocessing the description information of the navigation history problem to obtain preprocessed problem description information;

[0090] The step of obtaining the feature vector of the navigation history problem based on the description information of the navigation history problem includes:

[0091] Based on the preprocessed problem description information, the feature vector of the navigation history problem is obtained.

[0092] The preprocessing of the description information of the navigation history problem may refer to removing interfering words from the description information of the navigation history problem. Preprocessing of the description information of the navigation history problem includes at least one of the following:

[0093] Low-frequency vocabulary cleaning;

[0094] Punctuation mark cleaning;

[0095] Stop the word cleaning.

[0096] For example, "Navigation makes travel better" becomes "Navigation makes travel better" after preprocessing.

[0097] Specifically, preprocessing the description information of the navigation history problem to obtain preprocessed problem description information may include the following steps:

[0098] Extract the main descriptive information from the description information of the navigation history problem;

[0099] The extracted subject description information is semantically segmented to obtain multiple subject description information segments.

[0100] Specifically, the main descriptive information of navigation history issues is extracted, including:

[0101] Part-of-speech analysis was performed on the descriptive information of navigation history issues.

[0102] Based on the results of part-of-speech analysis, the main descriptive information is extracted from the descriptive information of navigation history questions.

[0103] The step of obtaining the feature vector of the navigation history problem based on the preprocessed problem description information includes:

[0104] A preprocessing model is used to perform feature transformation on each segment of the main description information to obtain the feature vector of the navigation history problem.

[0105] The preprocessing model is obtained by training based on the collected descriptive information of navigation history problems.

[0106] like Figure 1 As shown, in step S103, the model training input data is obtained. One data input includes: the feature vector of the navigation history problem and its corresponding processing script number.

[0107] like Figure 1 As shown, in step S104, the matching model is trained using the model training input data to obtain the target matching model; wherein, the target matching model is used to predict the processing script number that matches the navigation problem based on the feature vector of the navigation problem.

[0108] The matching model can refer to a DNN (Deep Neural Networks) model built using PyTorch (an open-source Python machine learning library), or other neural network models, such as LSTM (Long Short-Term Memory) neural network, Bi-LSTM-CRF (Bidirectional Long Short-Term Memory Conditional Random Field), and RNN (Recurrent Neural Network).

[0109] The target matching model refers to the trained matching model obtained after training the matching model based on the model training input data.

[0110] The step of training a matching model using the input data of the model training method to obtain a target matching model includes:

[0111] The model training input data is input into the matching model, and the model parameters are updated through the backpropagation algorithm and the stochastic gradient descent method to output the target matching model.

[0112] Since the target matching model trained using only feature vectors from historical navigation problems is still inaccurate in predicting new navigation problems, it is necessary to continuously collect navigation problems that are predicted inaccurately and re-label them to obtain updated training data during the use of the target matching model. The target matching model is then iteratively trained based on the updated training data to generate an updated target matching model, making the model's predictions more accurate.

[0113] As one implementation method, it may also include:

[0114] Obtain updated training data containing a description of the navigation problem and the sequence number of the processing script;

[0115] The target matching model is iteratively trained based on the updated training data to generate an updated target matching model.

[0116] like Figure 2 When an external customer resolves a navigation problem, if the prediction is inaccurate, the problem monitoring module obtains a description of the new navigation problem. After data preprocessing, the preprocessed data and the corresponding processing script number are used as updated training data. The target matching model is iteratively trained based on the updated training data to generate an updated target matching model.

[0117] This concludes the introduction of the first embodiment of this application. The first embodiment trains a matching model based on the feature vector of the navigation history problem and the processing script number, and trains a target matching model. The target matching model is used to obtain the processing script number that matches the navigation problem based on the feature vector of the navigation history problem, thereby realizing the automatic matching of the description information of the navigation problem with the corresponding processing script, which can effectively improve the efficiency of problem analysis and solution and reduce the cost of manual processing and solution.

[0118] The second embodiment of this application provides a data processing method. The second embodiment is a method for using the target matching model trained in the first embodiment. The following is a combination of... Figure 3 Let me introduce it.

[0119] like Figure 3 As shown, in step S301, the description information of the navigation problem is obtained.

[0120] Navigation issues refer to new problems discovered by customers using the navigation system during development or use. Customers can import these new navigation issues into an online problem analysis platform for the navigation system.

[0121] In practice, the navigation problem configuration monitoring can be configured through the problem analysis platform. When a customer creates a new problem, the problem and processing script matching model prediction device is automatically triggered to obtain the description information of the navigation problem.

[0122] like Figure 3 As shown, in step S302, the feature vector of the navigation problem is obtained based on the description information of the navigation problem.

[0123] After step S301, the second embodiment of this application may further include: preprocessing the description information of the navigation problem to obtain preprocessed problem description information;

[0124] The step of obtaining the feature vector corresponding to the description information based on the description information of the navigation problem includes:

[0125] Based on the preprocessed problem description information, the feature vector of the navigation problem is obtained.

[0126] Specifically, preprocessing the description information of the navigation problem to obtain preprocessed problem description information may include the following steps:

[0127] Extract the main descriptive information from the description information of the navigation problem;

[0128] The extracted subject description information is semantically segmented to obtain multiple subject description information segments;

[0129] Specifically, the main descriptive information of the navigation problem is extracted, including:

[0130] Part-of-speech analysis is performed on the descriptive information of navigation problems;

[0131] Based on the results of part-of-speech analysis, the main descriptive information is extracted from the descriptive information of the navigation problem.

[0132] The step of obtaining the feature vector of the navigation problem based on the preprocessed problem description information includes:

[0133] A preprocessing model is used to perform feature transformation on each segment of the main description information to obtain the feature vector of the navigation problem.

[0134] like Figure 3 As shown, in step S303, the feature vector of the navigation problem is input into the target matching model to predict the processing script number that matches the navigation problem.

[0135] The step of inputting the feature vector of the navigation problem into the target matching model to predict the processing script number that matches the navigation problem includes:

[0136] The feature vector of the navigation problem is input into the target matching model, and the forward search algorithm is used to calculate and output the processing script number with the highest probability.

[0137] like Figure 3 As shown, in step S304, the processing script corresponding to the processing script number is obtained according to the processing script number.

[0138] After predicting the processing script number that matches the navigation problem, the corresponding processing script can be obtained based on the processing script number, and then the corresponding processing script can be used to analyze and process the navigation problem.

[0139] like Figure 3 As shown, in step S305, the navigation problem is analyzed and processed using the processing script.

[0140] After analyzing and processing the navigation problem according to the processing script, an analysis and processing report can be generated. The analysis and processing report includes reports in the form of text, images, etc.

[0141] As one implementation method, the second embodiment of this application further includes:

[0142] Using the processing script number, information on consulting experts matching the navigation problem is obtained.

[0143] The information about the consulting experts may be the name of the expert matching the navigation question, the ID number of the expert matching the navigation question, or other identifying information.

[0144] Based on the processing script number, information on consulting experts matching the navigation problem is obtained. By using this information, the problem can be fed back to the expert who actually handles it, thus improving the efficiency of problem handling.

[0145] like Figure 2 The last line briefly describes the process of handling external customer issues. External customers import new navigation issues (i.e., navigation problems) into the issue analysis platform. The issue analysis platform obtains the description information of the new navigation problem and preprocesses the description information (i.e.,... Figure 2 The text is preprocessed (in the text), then model prediction is performed to obtain the corresponding script and execute it. After execution, the question answer information is sent to the external customer.

[0146] This concludes the introduction of the second embodiment of this application. The second embodiment obtains the description information of the navigation problem; then, based on the description information, it obtains the feature vector of the navigation problem; next, it inputs the feature vector of the navigation problem into a target matching model to predict the processing script number matching the navigation problem; finally, based on the processing script number, it obtains the processing script corresponding to the processing script number; and then uses the processing script to analyze and process the navigation problem. This achieves automated matching of the description information of the navigation problem with the corresponding processing script, and automatically analyzes and processes the problem using the processing script, effectively improving the efficiency of analysis and problem-solving and reducing the cost of manual processing. Furthermore, predicting the corresponding script using the problem description information significantly improves both the matching hit rate and accuracy.

[0147] Corresponding to the model training method provided in the first embodiment of this application, the third embodiment of this application provides a model training device.

[0148] like Figure 4 As shown, the model training device includes:

[0149] The model training raw data acquisition unit 401 is used to obtain the model training raw data. Each piece of raw data contains a description of the navigation history problem and its corresponding processing script number.

[0150] Feature vector acquisition unit 402 is used to obtain the feature vector of the navigation history problem based on the description information of the navigation history problem;

[0151] The model training input data acquisition unit 403 is used to acquire model training input data. One data input includes: the feature vector of the navigation history problem and its corresponding processing script number.

[0152] The model training unit 404 is used to train a matching model using the model training input data to obtain a target matching model; wherein, the target matching model is used to predict the processing script number that matches the navigation problem based on the feature vector of the navigation problem.

[0153] In one embodiment, the device further includes: a preprocessing unit, used to preprocess the description information of the navigation history problem to obtain preprocessed problem description information;

[0154] The step of obtaining the feature vector of the navigation history problem based on the description information of the navigation history problem includes:

[0155] Based on the preprocessed problem description information, the feature vector of the navigation history problem is obtained;

[0156] The preprocessing of the description information of the navigation history problem includes at least one of the following:

[0157] Low-frequency vocabulary cleaning;

[0158] Punctuation mark cleaning;

[0159] Stop the word cleaning.

[0160] In one implementation, the preprocessing unit is specifically used for:

[0161] Extract the main descriptive information from the description information of the navigation history problem;

[0162] The extracted subject description information is semantically segmented to obtain multiple subject description information segments;

[0163] The step of obtaining the feature vector of the navigation problem based on the preprocessed problem description information includes:

[0164] A preprocessing model is used to perform feature transformation on each segment of the main description information to obtain the feature vector of the navigation history problem.

[0165] As one implementation method, the model training raw data acquisition unit is specifically used for:

[0166] Retrieve descriptions of navigation history issues from the database using a script;

[0167] The scripts are used for annotation to generate training data containing descriptive information about navigation history issues and their corresponding processing script numbers.

[0168] As one implementation method, the model training unit is specifically used for:

[0169] The model training input data is input into the matching model, and the model parameters are updated through the backpropagation algorithm and the stochastic gradient descent method to output the target matching model.

[0170] It should be noted that for a detailed description of the apparatus provided in the third embodiment of this application, please refer to the relevant description of the first embodiment of this application, which will not be repeated here.

[0171] Corresponding to the model training method provided in the first embodiment of this application, the fourth embodiment of this application provides an electronic device. The electronic device includes:

[0172] Processor; and

[0173] The memory stores the program for the model training method. After the device is powered on and the program for the matching model training method is run by the processor, the following steps are performed:

[0174] Obtain the raw data for model training. Each raw data entry contains a description of the navigation history problem and its corresponding processing script number.

[0175] Based on the description information of the navigation history problem, the feature vector of the navigation history problem is obtained;

[0176] Obtain the model training input data. Each data input includes: the feature vector of the navigation history problem and its corresponding processing script number;

[0177] The model is used to train the matching model using the input data to obtain the target matching model; wherein, the target matching model is used to predict the processing script number that matches the navigation problem based on the feature vector of the navigation problem.

[0178] In one implementation, the electronic device further performs the following steps:

[0179] The description information of the navigation history problem is preprocessed to obtain the preprocessed problem description information;

[0180] The step of obtaining the feature vector of the navigation history problem based on the description information of the navigation history problem includes:

[0181] Based on the preprocessed problem description information, the feature vector of the navigation history problem is obtained;

[0182] The preprocessing of the description information of the navigation history problem includes at least one of the following:

[0183] Low-frequency vocabulary cleaning;

[0184] Punctuation mark cleaning;

[0185] Stop the word cleaning.

[0186] As one implementation method, the preprocessing of the description information of the navigation history problem to obtain preprocessed problem description information includes:

[0187] Extract the main descriptive information from the description information of the navigation history problem;

[0188] The extracted subject description information is semantically segmented to obtain multiple subject description information segments;

[0189] The step of obtaining the feature vector of the navigation history problem based on the preprocessed problem description information includes:

[0190] A preprocessing model is used to perform feature transformation on each segment of the main description information to obtain the feature vector of the navigation history problem.

[0191] As one implementation method, obtaining the original training data for the model includes:

[0192] Retrieve descriptions of navigation history issues from the database using a script;

[0193] The scripts are used for annotation to generate training data containing descriptive information about navigation history issues and their corresponding processing script numbers.

[0194] As one implementation method, training a matching model using the input data of the model training method to obtain a target matching model includes:

[0195] The model training input data is input into the matching model, and the model parameters are updated through the backpropagation algorithm and the stochastic gradient descent method to output the target matching model.

[0196] It should be noted that for a detailed description of the electronic device provided in the fourth embodiment of this application, please refer to the relevant description of the first embodiment of this application, which will not be repeated here.

[0197] Corresponding to the model training method provided in the first embodiment of this application, the fifth embodiment of this application provides a storage device storing a program for the model training method, which is executed by a processor to perform the following steps:

[0198] Obtain the raw data for model training. Each raw data entry contains a description of the navigation history problem and its corresponding processing script number.

[0199] Based on the description information of the navigation history problem, the feature vector of the navigation history problem is obtained;

[0200] Obtain the model training input data. Each data input includes: the feature vector of the navigation history problem and its corresponding processing script number;

[0201] The model is used to train the matching model using the input data to obtain the target matching model; wherein, the target matching model is used to predict the processing script number that matches the navigation problem based on the feature vector of the navigation problem.

[0202] It should be noted that for a detailed description of the storage device provided in the fifth embodiment of this application, please refer to the relevant description of the first embodiment of this application, which will not be repeated here.

[0203] Corresponding to the data processing method provided in the second embodiment of this application, the sixth embodiment of this application also provides a data processing apparatus, including:

[0204] The description information acquisition unit is used to acquire description information about the navigation problem;

[0205] The feature vector acquisition unit is used to obtain the feature vector of the navigation problem based on the description information of the navigation problem;

[0206] The script sequence number prediction unit is used to input the feature vector of the navigation problem into the target matching model and predict the processing script sequence number that matches the navigation problem.

[0207] The processing script obtaining unit is used to obtain the processing script corresponding to the processing script number according to the processing script number.

[0208] The problem analysis and processing unit is used to analyze and process the navigation problem using the processing script.

[0209] It should be noted that for a detailed description of the apparatus provided in the sixth embodiment of this application, please refer to the relevant description of the second embodiment of this application, which will not be repeated here.

[0210] Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of this application. Therefore, the scope of protection of this application should be determined by the scope defined in the claims of this application.

[0211] In a typical configuration, a computing device includes one or more processors (CPUs), memory-mapped input / output interfaces, network interfaces, and memory.

[0212] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0213] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include non-transitory computer-readable media, such as modulated data signals and carrier waves.

[0214] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

Claims

1. A model training method, wherein, include: Obtain the original data for model training. Each piece of original data contains a description of navigation history problems and its corresponding processing script number. The navigation history problems refer to problems discovered by customers using the navigation system during development or use. The processing script number is the number of the processing script. The processing script is the program code that analyzes and processes the description information of the navigation problems. The process of obtaining the feature vector of the navigation history problem based on its description information includes: performing part-of-speech analysis on the description information of the navigation history problem; extracting main description information from the description information of the navigation history problem based on the results of the part-of-speech analysis; performing semantic segmentation on the extracted main description information to obtain multiple segments of main description information; and using a preprocessing model to perform feature transformation on each segment of the multiple segments of main description information to obtain the feature vector of the navigation history problem. The preprocessing model is based on a general semantic model and is trained using the collected description information of the navigation history problem. Obtain the model training input data. Each data input includes: the feature vector of the navigation history problem and its corresponding processing script number; The model is trained using the input data to train a matching model, thereby obtaining a target matching model. The target matching model is used to predict the processing script number matching the navigation problem based on the feature vector of the navigation problem, to obtain the processing script corresponding to the processing script number, and to use the processing script to analyze and process the navigation problem, thereby obtaining a response to the navigation problem. The response is then sent to the customer.

2. The method according to claim 1, wherein, It also includes: preprocessing the description information of the navigation history problem to obtain preprocessed problem description information; The step of obtaining the feature vector of the navigation history problem based on the description information of the navigation history problem includes: Based on the preprocessed problem description information, the feature vector of the navigation history problem is obtained; The preprocessing of the description information of the navigation history problem includes at least one of the following: Low-frequency vocabulary cleaning; Punctuation mark cleaning; Stop the word cleaning.

3. The method according to claim 1, wherein, The obtained raw data for model training includes: Retrieve descriptions of navigation history issues from the database using a script; The scripts are used for annotation to generate training data containing descriptive information about navigation history issues and their corresponding processing script numbers.

4. The method according to claim 1, wherein, The step of training a matching model using the input data of the model training method to obtain a target matching model includes: The model training input data is input into the matching model, and the model parameters are updated through the backpropagation algorithm and the stochastic gradient descent method to output the target matching model.

5. A data processing method, wherein, include: Obtain descriptive information about navigation problems, which refer to new problems discovered by customers using the navigation system during development or use; The process of obtaining the feature vector of the navigation problem based on its description information includes: performing part-of-speech analysis on the description information of the navigation problem; extracting main description information from the description information of the navigation problem based on the results of the part-of-speech analysis; performing semantic segmentation on the extracted main description information to obtain multiple segments of main description information; and using a preprocessing model to perform feature transformation on each segment of the multiple segments of main description information to obtain the feature vector of the navigation problem. The preprocessing model is based on a general semantic model and is trained using the collected description information of historical navigation problems. The feature vector of the navigation problem is input into the target matching model to predict the processing script number that matches the navigation problem. The processing script number is the number of the processing script. Based on the processing script number, the processing script corresponding to the processing script number is obtained. The processing script is program code that analyzes and processes the descriptive information of the navigation problem. The processing script is used to analyze and process the navigation problem to obtain answer information for the navigation problem; The response to the question will be sent to the customer.

6. The method according to claim 5, wherein, The step of inputting the feature vector of the navigation problem into the target matching model to predict the processing script number that matches the navigation problem includes: The feature vector of the navigation problem is input into the target matching model, and the forward search algorithm is used to calculate and output the processing script number with the highest probability.

7. The method according to claim 5, wherein, It also includes: preprocessing the description information of the navigation problem to obtain preprocessed problem description information; The step of obtaining the feature vector corresponding to the description information based on the description information of the navigation problem includes: Based on the preprocessed problem description information, the feature vector of the navigation problem is obtained.

8. A model training device, wherein, include: The model training raw data acquisition unit is used to acquire the model training raw data. Each piece of raw data contains a description of navigation history problems and its corresponding processing script number. The navigation history problems refer to problems discovered by customers using the navigation system during development or use. The processing script number is the number of the processing script. The processing script is the program code that analyzes and processes the description information of the navigation problems. The feature vector acquisition unit is used to obtain the feature vector of the navigation history problem based on the description information of the navigation history problem. This includes: performing part-of-speech analysis on the description information of the navigation history problem; extracting main description information from the description information of the navigation history problem based on the results of the part-of-speech analysis; performing semantic segmentation on the extracted main description information to obtain multiple segments of main description information; and using a preprocessing model to perform feature transformation on each segment of the multiple segments of main description information to obtain the feature vector of the navigation history problem. The preprocessing model is based on a general semantic model and is trained using the collected description information of the navigation history problem. The model training input data acquisition unit is used to obtain model training input data. One data input includes: the feature vector of the navigation history problem and its corresponding processing script number; The model training unit is used to train a matching model using the model training input data to obtain a target matching model; wherein, the target matching model is used to predict the processing script number matching the navigation question based on the feature vector of the navigation question, obtain the processing script corresponding to the processing script number based on the processing script number, use the processing script to analyze and process the navigation question, obtain the question answer information for the navigation question, and send the question answer information to the customer.

9. An electronic device, wherein, include: processor; as well as The memory stores the program for the model training method. After the device is powered on and the processor runs the program for the matching model training method, it performs the following steps: Obtain the original data for model training. Each piece of original data contains a description of navigation history problems and its corresponding processing script number. The navigation history problems refer to problems discovered by customers using the navigation system during development or use. The processing script number is the number of the processing script. The processing script is the program code that analyzes and processes the description information of the navigation problems. The process of obtaining the feature vector of the navigation history problem based on its description information includes: performing part-of-speech analysis on the description information of the navigation history problem; extracting main description information from the description information of the navigation history problem based on the results of the part-of-speech analysis; performing semantic segmentation on the extracted main description information to obtain multiple segments of main description information; and using a preprocessing model to perform feature transformation on each segment of the multiple segments of main description information to obtain the feature vector of the navigation history problem. The preprocessing model is based on a general semantic model and is trained using the collected description information of the navigation history problem. Obtain the model training input data. Each data input includes: the feature vector of the navigation history problem and its corresponding processing script number; The model is trained using the input data to train a matching model, thereby obtaining a target matching model. The target matching model is used to predict the processing script number matching the navigation problem based on the feature vector of the navigation problem, to obtain the processing script corresponding to the processing script number, and to use the processing script to analyze and process the navigation problem, thereby obtaining a response to the navigation problem. The response is then sent to the customer.

10. A storage device, wherein, A program containing model training methods is executed by the processor, performing the following steps: Obtain the original data for model training. Each piece of original data contains a description of navigation history problems and its corresponding processing script number. The navigation history problems refer to problems discovered by customers using the navigation system during development or use. The processing script number is the number of the processing script. The processing script is the program code that analyzes and processes the description information of the navigation problems. The process of obtaining the feature vector of the navigation history problem based on its description information includes: performing part-of-speech analysis on the description information of the navigation history problem; extracting main description information from the description information of the navigation history problem based on the results of the part-of-speech analysis; performing semantic segmentation on the extracted main description information to obtain multiple segments of main description information; and using a preprocessing model to perform feature transformation on each segment of the multiple segments of main description information to obtain the feature vector of the navigation history problem. The preprocessing model is based on a general semantic model and is trained using the collected description information of the navigation history problem. Obtain the model training input data. Each data input includes: the feature vector of the navigation history problem and its corresponding processing script number; The model is trained using the input data to train a matching model, thereby obtaining a target matching model. The target matching model is used to predict the processing script number matching the navigation problem based on the feature vector of the navigation problem, to obtain the processing script corresponding to the processing script number, and to use the processing script to analyze and process the navigation problem, thereby obtaining a response to the navigation problem. The response is then sent to the customer.

11. A data processing apparatus, wherein, include: The description information acquisition unit is used to acquire description information of navigation problems, wherein the navigation problems refer to new problems discovered by customers using the navigation system during development or use. The feature vector acquisition unit is used to obtain the feature vector of the navigation problem based on the description information of the navigation problem. This includes: performing part-of-speech analysis on the description information of the navigation problem; extracting main description information from the description information of the navigation problem based on the results of the part-of-speech analysis; performing semantic segmentation on the extracted main description information to obtain multiple segments of main description information; and using a preprocessing model to perform feature transformation on each segment of the multiple segments of main description information to obtain the feature vector of the navigation problem. The preprocessing model is based on a general semantic model and is trained using collected description information of historical navigation problems. The script number prediction unit is used to input the feature vector of the navigation problem into the target matching model and predict the processing script number that matches the navigation problem. The processing script number is the number of the processing script. The processing script obtaining unit is used to obtain the processing script corresponding to the processing script number according to the processing script number, wherein the processing script is program code that analyzes and processes the description information of the navigation problem. The problem analysis and processing unit is used to analyze and process the navigation problem using the processing script to obtain the problem response information for the navigation problem; The response to the question will be sent to the customer.

Images