Model training corpus generation method and request information processing method

Abstract

The invention discloses a model training corpus generation method and a request information processing method. The method relates to the field of large models, and comprises the following steps: selecting a plurality of preset sentence pattern templates, obtaining characteristic variables used for filling the plurality of preset sentence pattern templates, and obtaining a plurality of characteristic variables; respectively selecting a filling variable from the plurality of characteristic variables according to the filling requirement of each preset sentence pattern template to obtain a filling variable set; generating a characteristic value of a corresponding filling variable according to the filling requirement of each preset sentence pattern template, and obtaining a target sentence set; obtaining character string information of each target statement, and forming a character string statement of each target statement by the character string information; and determining each target statement and the corresponding character string statement as a group of training data, and determining multiple groups of training data as a training corpus of the target model. Through the method and the device, the problem that the prediction accuracy of the model obtained according to corpus training is relatively low due to the fact that the corpus style is single in the related technology is solved.

Description

Technical Field

[0001] The present application relates to the field of large models, and more specifically, to a method for generating a model training corpus and a method for processing request information. Background Art

[0002] In the customer interaction system of the civil aviation industry, after receiving the user's input request, a natural language processing model is usually used to parse the user's input request and then generate structured data, such as request information in JSON format, so that the system can further process it and provide corresponding feedback.

[0003] In order to improve the processing accuracy and efficiency of the model, it is usually necessary to train the model. Traditional model training methods rely on the corpus information of users' historical input requests as training samples. These corpus information often have certain patterns and fixed styles. As a result, the process of training the model through a single sample set limits the model's learning ability, reduces training efficiency, and limits the model's generalization ability and accuracy. Especially when processing user queries with high diversity and complexity, the model cannot accurately identify input requests, which in turn affects the accuracy of the output content.

[0004] Currently, no effective solution has been proposed to address the problem in related technologies where the prediction accuracy of the model trained based on the corpus is low due to the single corpus style. Summary of the Invention

[0005] The present application provides a method for generating a model training corpus and a method for processing request information to solve the problem in related technologies that the prediction accuracy of a model obtained by training the corpus is low due to the single corpus style.

[0006] According to one aspect of the present application, a method for generating a model training corpus is provided. The method includes: selecting multiple preset sentence templates and obtaining feature variables used to fill the multiple preset sentence templates to obtain multiple feature variables; obtaining filling requirements for each preset sentence template, selecting filling variables for each preset sentence template from the multiple feature variables according to each filling requirement, and obtaining a filling variable set for each preset sentence template; generating feature values ​​of the filling variables in the corresponding filling variable set according to the filling requirements of each preset sentence template to obtain a target sentence set for each preset sentence template, wherein each target sentence set includes multiple target sentences with the same sentence template but different feature values; obtaining string information of the filling variables and string information of the feature values ​​of each target sentence, and forming a string sentence for each target sentence from the string information; determining each target sentence and the corresponding string sentence as a set of training data to obtain multiple sets of training data, and determining the multiple sets of training data as a training corpus for the target model.

[0007] Optionally, the filling requirements include precise filling, and the filling variables of each preset sentence template are selected from multiple feature variables according to each filling requirement, to obtain the filling variable set of each preset sentence template, including: for any preset sentence template, obtaining the features to be filled from the filling requirements, and selecting the feature variables corresponding to each feature to be filled from multiple feature variables to obtain the filling variable set of the preset sentence template.

[0008] Optionally, the filling requirement includes fuzzy filling, and the filling variables of each preset sentence template are selected from multiple feature variables according to each filling requirement, and the filling variable set of each preset sentence template is obtained, including: for any preset sentence template, obtaining the fixed filling feature indicated by the filling requirement, and obtaining the feature variable corresponding to the fixed filling feature to obtain the first feature variable; determining the feature variables other than the first feature variable in the multiple feature variables as second feature variables, and traversing to generate multiple second feature variable sets consisting of at least one second feature variable; combining each second feature variable set with the first feature variable in turn to obtain the filling variable set of the preset sentence template.

[0009] Optionally, determining each target sentence and the corresponding string sentence as a set of training data includes: for any target sentence, adjusting the order of the characteristic values ​​of multiple filling variables in the target sentence to obtain multiple changed sentences; determining each changed sentence and the string sentence corresponding to the target sentence as a set of training data to obtain multiple sets of training data.

[0010] Optionally, determining each target sentence and the corresponding string sentence as a set of training data includes: for any target sentence, obtaining multiple interference filling information, and adding the interference feature value of each interference filling information to the target sentence to obtain an updated target sentence; determining the updated target sentence and the string sentence corresponding to the target sentence as a set of training data.

[0011] Optionally, the filling requirements include precise filling and fuzzy filling, and the characteristic values ​​of the filling variables in the corresponding filling variable set are generated according to the filling requirements of each preset sentence template, and the target sentence set of each preset sentence template is obtained, including: for the preset sentence template whose filling requirement is precise filling, the attribute information of each filling variable is obtained, and the filling format of the filling value of the filling variable is obtained according to the attribute information, and the characteristic value of each filling variable is generated according to the filling format to obtain the target statement set; for the preset sentence template whose filling requirement is fuzzy filling, the description information of the characteristic value of each filling variable is generated, and the description information is determined as the characteristic value of the filling variable to obtain the target sentence set.

[0012] Optionally, after selecting the filling variables of each preset sentence template from multiple feature variables according to each filling requirement to obtain a filling variable set for each preset sentence template, the method also includes: for any filling variable set, obtaining the feature values ​​indicated by each filling variable, and judging whether there are repeated filling values ​​among the multiple filling values; if there are repeated filling values ​​among the multiple filling values, deleting the filling variables corresponding to the repeated filling values ​​to obtain an updated filling variable set.

[0013] According to one aspect of the present application, a method for processing request information is provided. The method includes: receiving a request message sent by a target user and obtaining a request statement in the request message; inputting the request statement into a preset parsing model to obtain string information of the request statement, wherein the preset parsing model is trained using a model corpus generated using the above-mentioned model training corpus generation method; and invoking the string information to obtain reply information for the request statement.

[0014] According to another aspect of the present application, a device for generating a model training corpus is provided. The device includes: a filling unit for selecting multiple preset sentence templates and obtaining feature variables used to fill the multiple preset sentence templates, thereby obtaining multiple feature variables; a selection unit for obtaining the filling requirements of each preset sentence template and selecting the filling variables of each preset sentence template from the multiple feature variables according to each filling requirement, thereby obtaining a filling variable set for each preset sentence template; a generation unit for generating feature values ​​of the filling variables in the corresponding filling variable set according to the filling requirements of each preset sentence template, thereby obtaining a target sentence set for each preset sentence template, wherein each target sentence set includes multiple target sentences with the same sentence template but different feature values; a composition unit for obtaining string information of the filling variables and string information of the feature values ​​of each target sentence, and composing a string sentence of each target sentence from the string information; and a determination unit for determining each target sentence and the corresponding string sentence as a set of training data, thereby obtaining multiple sets of training data, and determining the multiple sets of training data as a training corpus for a target model.

[0015] According to another aspect of the present application, a device for processing request information is provided. The device includes: a receiving unit for receiving a request message sent by a target user and obtaining a request statement in the request message; a training unit for inputting the request statement into a preset parsing model to obtain string information of the request statement, wherein the preset parsing model is trained using a model corpus generated using the above-mentioned model training corpus generation method; and a calling unit for calling the string information to obtain reply information to the request statement.

[0016] According to another aspect of the present invention, a computer program product is also provided, including a computer program, which, when executed by a processor, implements a method for generating a model training corpus and a method for processing request information provided in the aforementioned embodiment of the present application.

[0017] According to another aspect of the present invention, an electronic device is also provided, comprising one or more processors and a memory; the memory stores computer-readable instructions, and the processor is used to run the computer-readable instructions, wherein when the computer-readable instructions are run, a method for generating a model training corpus and a method for processing request information provided in the aforementioned embodiment are executed.

[0018] Through the present application, the following steps are adopted: selecting multiple preset sentence templates, and obtaining feature variables for filling the multiple preset sentence templates to obtain multiple feature variables; obtaining the filling requirements of each preset sentence template, and selecting the filling variables of each preset sentence template from the multiple feature variables according to each filling requirement to obtain a filling variable set for each preset sentence template; generating feature values ​​of the filling variables in the corresponding filling variable set according to the filling requirements of each preset sentence template to obtain a target sentence set for each preset sentence template, wherein each target sentence set includes multiple target sentences with the same sentence template and different feature values; obtaining the string information of the filling variables and the string information of the feature values ​​of each target sentence, and forming a string sentence of each target sentence from the string information; determining each target sentence and the corresponding string sentence as a set of training data to obtain multiple sets of training data, and determining the multiple sets of training data as the training corpus of the target model, which solves the problem in the related art that the prediction accuracy of the model obtained by corpus training is low due to the single corpus style. By obtaining preset sentence templates of different styles and filling them with different feature values, multiple target sentences with different styles and contents are obtained, and the target sentences and corresponding string information are further used as training data in the training corpus to train the model. By generating a corpus with large differences in style and content, the training effect of the model is improved, thereby achieving the technical effect of improving the model prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0020] Figure 1 is a flowchart of a method for generating a model training corpus according to an embodiment of the present application;

[0021] Figure 2 is a flowchart of a method for processing request information provided in an embodiment of the present application;

[0022] Figure 3 is a schematic diagram of a device for generating a model training corpus according to an embodiment of the present application;

[0023] Figure 4 is a schematic diagram of a device for processing request information provided in an embodiment of the present application;

[0024] Figure 5 This is a schematic diagram of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0026] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0028] It should be noted that the method for generating the model training corpus and the method for processing request information determined in the present disclosure can be used in the large model field, and can also be used in any field other than the large model field. The application field of the method for generating the model training corpus and the method for processing request information determined in the present disclosure is not limited.

[0029] It should be noted that the collected information, user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) used in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data comply with the relevant laws, regulations and standards of the relevant regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize use or refuse use. If the user chooses to refuse, the expert decision-making process will be entered. For example, an interface is set up between this system and relevant users or institutions. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or institution through the interface, and obtain relevant information after receiving the consent information fed back by the aforementioned user or institution.

[0030] The embodiments or examples of the present disclosure are not exhaustive, but are merely illustrations of some embodiments or examples, and are not intended to be specific limitations on the scope of protection of the present disclosure. In the absence of contradiction, each step in a certain embodiment or example can be implemented as an independent example, and the steps can be arbitrarily combined. For example, a solution after removing some steps in a certain embodiment or example can also be implemented as an independent example, and the order of the steps in a certain embodiment or example can be arbitrarily exchanged. In addition, the optional methods or optional examples in a certain embodiment or example can be arbitrarily combined; in addition, the various embodiments or examples can be arbitrarily combined. For example, some or all steps of different embodiments or examples can be arbitrarily combined, and a certain embodiment or example can be arbitrarily combined with the optional methods or optional examples of other embodiments or examples.

[0031] For ease of description, some nouns or terms involved in the embodiments of the present application are explained below:

[0032] JSON: JavaScript Object Notation, JavaScript Object Notation, JSON is a lightweight data exchange format that is easy for people to read and write, and also easy for machines to parse and generate.

[0033] According to an embodiment of the present application, a method for generating a model training corpus is provided.

[0034] Figure 1 Flowchart of the method for generating a model training corpus according to the embodiment of the present application. Figure 1 As shown, the method includes the following steps:

[0035] Step S101 : Select multiple preset sentence templates, and obtain feature variables used to fill the multiple preset sentence templates to obtain multiple feature variables.

[0036] It should be noted that preset sentence templates are pre-designed text frameworks tailored to common civil aviation user needs, such as flight inquiries, reservation changes, and refund requests. Feature variables refer to the variable parts of a sentence template that carry information specific to the user's needs, such as departure point, destination, date, time, and flight number.

[0037] Specifically, the corpus generation system of this embodiment may be the main execution entity. When generating a training corpus, the system first selects multiple preset sentence templates from a template table. These templates cover the main types of requests that users in the civil aviation industry may make. The system then identifies and extracts feature variables from the templates. These variables represent key information in the user request, such as the "origin" and "destination" of a flight. These feature variables can be located in different sentence templates.

[0038] For example, Table 1 is a schematic table of optionally preset sentence templates. As shown in Table 1, the preset sentence templates may include:

[0039] Table 1

[0040]

[0041] Feature variables may include but are not limited to: "personnel information", "origin information", "cabin information", "time information", etc.

[0042] By obtaining preset sentence templates and feature variables, it is possible to combine the feature values ​​of the preset sentence templates and feature variables into sentences of different styles and contents, thereby obtaining a corpus with comprehensive content.

[0043] Step S102 , obtaining the filling requirements of each preset sentence template, selecting the filling variables of each preset sentence template from a plurality of characteristic variables according to each filling requirement, and obtaining a filling variable set of each preset sentence template.

[0044] Specifically, since the preset sentence templates can include precise filling templates and fuzzy filling templates, different preset sentence templates have different filling requirements. When obtaining filling variables, it is necessary to first obtain the filling requirements of each preset sentence template, and then select the filling variables required to fill each preset sentence template from the feature variables according to the filling requirements, thereby completing the selection operation of the filling variables for each preset sentence template.

[0045] For example, in the preset sentence template A, "Date" must be filled in, or "Place of Departure" is optional.

[0046] Therefore, after obtaining the filling requirements, the system will select specific instances that meet the specifications from the feature variables to form a set of filling variables for each preset sentence template, ensuring that the generated training data not only conforms to the business logic but also covers the diversity of user needs, thereby improving the training efficiency and accuracy of the model.

[0047] Step S103, generating characteristic values ​​of the filling variables in the corresponding filling variable set according to the filling requirements of each preset sentence template, and obtaining a target sentence set for each preset sentence template, wherein each target sentence set includes multiple target sentences with the same sentence template and different characteristic values.

[0048] It's important to note that the filling requirements include not only the fill-in variables but also the requirements for how to fill in the variable's characteristic values. After determining the fill-in variable sets for each preset sentence template, the characteristic values ​​of each fill-in variable in each fill-in variable set can be generated, thereby obtaining target sentences with the same style but different content.

[0049] For example, using the same template "Query flight information from [origin] to [destination] on [date]," by filling in different city names and dates, multiple specific statements can be generated, such as "Query flight information from Beijing to Shanghai on 2023-09-15" and "Query flight information from Beijing to Guangzhou on 2023-09-16." Furthermore, by adjusting the feature values ​​of variables in the sentence structure, the training dataset can be enriched, enabling the model to have stronger recognition and understanding capabilities when processing actual user requests.

[0050] Step S104 , obtaining the character string information of the filling variables and the character string information of the feature values ​​of each target sentence, and composing the character string sentence of each target sentence from the character string information.

[0051] It should be noted that string information can be JSON characters. After obtaining the target sentence, the generated target sentence can be converted to JSON, thereby converting each target sentence into the corresponding request JSON. For example, converting the filler variables and feature values ​​in the target sentence "Book economy class tickets from Beijing to Shanghai for two adults and one child, departing on September 15, 2024" into string information yields: {'adult': 2, 'child': 1, 'departure': 'Beijing', 'arrival': 'Shanghai', 'departureDate': '2024-09-15', 'cabinClass': 'ECONOMY'}, thus obtaining the string sentence of the target sentence composed of string information.

[0052] In step S105 , each target sentence and the corresponding character string sentence are determined as a set of training data to obtain multiple sets of training data, and the multiple sets of training data are determined as a training corpus of the target model.

[0053] Specifically, after obtaining each target sentence and its string sentence, the system pairs each target sentence with its corresponding string sentence to form training data, and then the training data constitutes a comprehensive training corpus for subsequent model training, thereby ensuring the richness and diversity of the corpus data in the corpus used for training the model, thereby improving the model's learning ability and enabling it to be more accurate and efficient when processing user requests in the civil aviation industry.

[0054] The method for generating a model training corpus provided in an embodiment of the present application selects multiple preset sentence templates and obtains feature variables for filling the multiple preset sentence templates to obtain multiple feature variables; obtains the filling requirements of each preset sentence template, selects the filling variables of each preset sentence template from the multiple feature variables according to each filling requirement, and obtains a filling variable set for each preset sentence template; generates feature values ​​of the filling variables in the corresponding filling variable set according to the filling requirements of each preset sentence template, and obtains a target sentence set for each preset sentence template, wherein each target sentence set includes multiple target sentences with the same sentence template and different feature values; obtains the string information of the filling variables and the string information of the feature values ​​of each target sentence, and composes a string sentence of each target sentence from the string information; determines each target sentence and the corresponding string sentence as a set of training data to obtain multiple sets of training data, and determines the multiple sets of training data as the training corpus of the target model, which solves the problem in the related art that the prediction accuracy of the model obtained by corpus training is low due to the single corpus style. By obtaining preset sentence templates of different styles and filling them with different feature values, multiple target sentences with different styles and contents are obtained, and the target sentences and corresponding string information are further used as training data in the training corpus to train the model. By generating a corpus with large differences in style and content, the training effect of the model is improved, thereby achieving the technical effect of improving the model prediction accuracy.

[0055] In order to ensure the accuracy of obtaining filling variables, optionally, in the method for generating the model training corpus provided in the embodiment of the present application, the filling requirements include precise filling, and the filling variables of each preset sentence template are selected from multiple feature variables according to each filling requirement, and the filling variable set of each preset sentence template is obtained, including: for any preset sentence template, the features to be filled are obtained from the filling requirements, and the feature variables corresponding to each feature to be filled are selected from multiple feature variables to obtain the filling variable set of the preset sentence template.

[0056] It should be noted that accurate filling means that each feature to be filled in the preset sentence template needs to be accurately filled, and the filled feature value also needs to ensure accuracy.

[0057] Specifically, for any pre-set sentence template, first identify all feature variables that need to be filled with data from the filling requirements, i.e., the features to be filled. For example, in the sentence template "Query flight information from [departure city] to [destination] on [date]", the departure city, destination, and date are the features to be filled.

[0058] Next, the system needs to select the variables that meet the requirements from multiple feature variables as the filling items based on the filling requirements of each feature. For example, if the departure and destination cities must be valid city names, the system will select from a list of all valid city names; if there are specific requirements for the date format, the system will generate a date that conforms to that format.

[0059] The result of this process is that for each preset sentence template, the system is able to generate a set of filling variables, each variable in the set meets the filling requirements, thereby ensuring that the training data sentences generated after filling are grammatically, semantically and logically accurate.

[0060] For example, when the preset sentence template is "Book a flight from {origin information} to {destination information} {class information} for {personnel information}, with the departure date being {time information}", it is necessary to determine which features in the preset sentence template are features to be filled in based on the filling requirements. When the following contents in the preset sentence template are determined: {personnel information}, {origin information}, {destination information}, {class information}, {time information} are features to be filled in, the features representing the above features to be filled all need to be filled in using feature values, and then the feature variables corresponding to each feature to be filled in are obtained from multiple feature variables, thereby obtaining a set of filling variables for the preset sentence template.

[0061] This embodiment ensures the accuracy of subsequent acquisition of characteristic values ​​of each filling variable by accurately determining the filling variable set of the preset sentence template under the precise filling requirement, thereby avoiding the occurrence of missing characteristic values.

[0062] In order to ensure the accuracy of obtaining filling variables, optionally, in the method for generating a model training corpus provided in an embodiment of the present application, the filling requirements include fuzzy filling, and the filling variables of each preset sentence template are selected from multiple feature variables according to each filling requirement, and the filling variable set of each preset sentence template is obtained, including: for any preset sentence template, obtaining the fixed filling feature indicated by the filling requirement, and obtaining the feature variable corresponding to the fixed filling feature to obtain a first feature variable; determining the feature variables other than the first feature variable in the multiple feature variables as second feature variables, and traversing to generate multiple second feature variable sets consisting of at least one second feature variable; and combining each second feature variable set with the first feature variable in turn to obtain the filling variable set of the preset sentence template.

[0063] It should be noted that fuzzy filling represents that the preset sentence template under this filling requirement only needs to fill in the characteristic values ​​of some filling features, and the characteristic values ​​of other filling features can be left blank, and the characteristic values ​​of the filling features can be filled in by fuzzy filling. For example, the filling variable of the preset sentence template can include "travel time", and the filling value of the filling variable can be a fuzzy time, such as "7 days", without the need to fill in specific time information.

[0064] Specifically, for a preset sentence template whose filling requirement is fuzzy filling, it is first necessary to obtain a fixed filling feature, that is, a filling feature that must be filled in when filling the preset sentence template, and determine the feature variable corresponding to the fixed filling feature as the first feature variable.

[0065] For example, if "place of origin" is a fixed filling feature, then the filling feature "place of departure" corresponding to this feature can be the first feature variable.

[0066] Furthermore, after determining the first feature variable, other feature variables among the multiple feature variables can be selected as second feature variables, and a traversal operation can be performed to generate multiple sets consisting of at least one second feature variable, each set will contain different instances of the second feature variable to reflect the diversity and uncertainty of user input.

[0067] For example, the second feature variable may include: "personnel information", "cabin information", "number of people", then the second feature variable set may include: "personnel information", "cabin information", "number of people", {"personnel information", "cabin information"}, {"cabin information", "number of people"}, {"personnel information", "number of people"}, {"personnel information", "cabin information", "number of people"}.

[0068] After obtaining the first feature variable and multiple sets of second feature variables, the system will perform the final combination step. For each generated set of second feature variables, the system will combine it with the first feature variable to form a complete set of filled variables. Each data in this set represents an instance of the preset sentence template, but through the changes in the second feature variable, the instances of the preset sentence template show rich diversity in semantics and structure.

[0069] For example, the preset sentence template is: "Depart from {origin information}, the destination is {destination information}, and the remaining features are optional." At this time, it can be determined that in the preset sentence template, {origin information} and {destination information} are the first feature variables, and the other feature variables among the multiple feature variables are also the second feature variables. If the second feature variable is {"personnel information", "cabin information"}, the preset sentence template obtained is: Depart from {origin information}, the destination is {destination information}, "personnel information", "cabin information".

[0070] This embodiment enriches the learning materials of the model by introducing the fuzzy filling and traversal combination strategy, enhances the adaptability and generalization ability of the model in actual application scenarios, and thus improves the accuracy of the model's output results.

[0071] In order to improve the diversity of training data, optionally, in the method for generating a model training corpus provided in an embodiment of the present application, each target sentence and the corresponding string sentence are determined as a set of training data, including: for any target sentence, the order of the characteristic values ​​of multiple filling variables in the target sentence is adjusted to obtain multiple changed sentences; each changed sentence and the string sentence corresponding to the target sentence are determined as a set of training data to obtain multiple groups of training data.

[0072] Specifically, when generating target sentences, in order to improve the diversity of training data, after obtaining each target sentence, the order of the characteristic values ​​of the filling variables in the target sentence can be adjusted, so as to use different adjustment methods to obtain different changed sentences.

[0073] Furthermore, although the order of the fill-in variables in the target sentence is changed, the content of the sentence does not change. Therefore, the multiple change sentences corresponding to the target sentence still correspond to the same string sentence. Each change sentence and the corresponding string sentence can be combined into a set of training data, thereby expanding the training data consisting of a set of target sentences and string sentences to multiple training data consisting of multiple change sentences and string sentences. By adjusting the order of the feature values, more diverse training instances are created, so that the training instances present different grammatical orders while maintaining the sentence structure, which helps the model learn the possible input diversity of users.

[0074] For example, if the target sentence is: "Departing tomorrow, from Beijing to Shanghai, flight", it can be changed to: "From Beijing to Shanghai, flight, departing tomorrow" and "From Beijing to Shanghai, departing tomorrow, flight", thereby obtaining multiple sets of training data with the same semantics but different grammatical order.

[0075] This embodiment creates more varied sentences as training examples by adjusting the order of feature values ​​in the target sentence, which significantly improves the adaptability and processing efficiency of the model.

[0076] In order to improve the diversity of training data, optionally, in the method for generating a model training corpus provided in an embodiment of the present application, each target sentence and the corresponding string sentence are determined as a set of training data, including: for any target sentence, obtaining multiple interference filling information, and adding the interference feature value of each interference filling information to the target sentence to obtain an updated target sentence; and determining the updated target sentence and the string sentence corresponding to the target sentence as a set of training data.

[0077] It should be noted that noise filler refers to information that can be added or modified at will without affecting the core meaning of the target sentence. This information may include additional descriptions, irrelevant details, or randomly inserted words, thereby simulating the redundancy and randomness of user input requests when the model is used. A noise feature value is a specific instance of noise filler information and can be any value that conforms to preset rules to reflect the diversity of information.

[0078] Specifically, the system can retrieve multiple pieces of interference filling information from a preset interference information library. The type and quantity of such information can be adjusted based on actual needs. Subsequently, the system can determine the interference characteristic value of each piece of interference filling information and add the interference characteristic value to the target sentence, thereby obtaining multiple updated target sentences. Different updated target sentences have different interference characteristic values, or the interference characteristic values ​​belong to different pieces of interference filling information.

[0079] By introducing interference filling information, the complexity and diversity of training data can be enhanced. By simulating the user's possible redundant expressions or the addition of irrelevant details, the model can learn how to ignore or process these interferences to accurately capture and understand the user's true intentions.

[0080] For example, the target sentence can be: "Query flight tickets from Beijing to Shanghai." The updated target sentence can be: "Query flight tickets from Shenzhen to Hangzhou, preferably blue." Since it is impossible to objectively select the color of the plane, "preferably blue" is interference information.

[0081] This embodiment adds interference information to the target sentence, thereby expanding the sentences in the training data and providing the model with training materials that are closer to the real user interaction environment. This not only helps to improve the model's anti-interference ability, but also enhances its performance in actual customer service, providing customers with a more personalized and accurate service experience.

[0082] Optionally, in the method for generating a model training corpus provided in an embodiment of the present application, the filling requirements include precise filling and fuzzy filling, and the characteristic values ​​of the filling variables in the corresponding filling variable set are generated according to the filling requirements of each preset sentence template, and the target sentence set of each preset sentence template is obtained, including: for the preset sentence template whose filling requirement is precise filling, the attribute information of each filling variable is obtained, and the filling format of the filling value of the filling variable is obtained according to the attribute information, and the characteristic value of each filling variable is generated according to the filling format to obtain the target sentence set; for the preset sentence template whose filling requirement is fuzzy filling, the description information of the characteristic value of each filling variable is generated, and the description information is determined as the characteristic value of the filling variable to obtain the target sentence set.

[0083] Specifically, for preset sentence templates that require precise filling, since the content in this sentence needs to be accurate, it is necessary to obtain the attribute information of each filling variable, and then obtain the filling format of the filling value of the filling variable based on the attribute information, so that when generating the characteristic value, it is ensured that the characteristic value meets the requirements of the filling format, and thus ensure that after the preset sentence template is filled, the sentence obtained is a precisely filled sentence.

[0084] For example, when the preset sentence template is "Book a flight from {origin information} to {destination information} {class information} for {personnel information}, with the departure date being {time information}", the attribute information of the origin information requires the accurate city name to be filled in, and the filling format is: accurate city name. At this time, when generating the feature value, it is necessary to identify whether the generated feature value is an accurate city name, so as to ensure the accuracy of the information after the sentence is filled in.

[0085] Furthermore, for preset sentence templates with fuzzy filling requirements, since there is no need to generate precise feature values, descriptive information of the feature values ​​of each filling variable can be generated, and the descriptive information can be used to describe the filling value, thereby ensuring the fuzziness of the filling value and the comprehensiveness of the model training.

[0086] For example, when the preset sentence template is "Book a flight for {personnel information} from {origin information} to {destination information} {class information}, with the departure date being {time information}", since it is a fuzzy sentence, the origin information does not need to be filled in with the exact name, but with descriptive information such as "coastal city" or "provincial capital city", so as to ensure the comprehensiveness of the generated training data.

[0087] This embodiment achieves the technical effect of increasing the complexity and comprehensiveness of model training data by using different feature value generation methods to generate fill-in feature values ​​in different preset sentence templates.

[0088] Optionally, in the method for generating a model training corpus provided in an embodiment of the present application, after selecting the filling variables of each preset sentence template from multiple feature variables according to each filling requirement to obtain a filling variable set for each preset sentence template, the method also includes: for any filling variable set, obtaining the feature values ​​indicated by each filling variable, and judging whether there are repeated filling values ​​among the multiple filling values; if there are repeated filling values ​​among the multiple filling values, deleting the filling variables corresponding to the repeated filling values ​​to obtain an updated filling variable set.

[0089] Specifically, for any filling variable set, since the filling variable set contains multiple filling variables, it is possible to obtain the characteristic values ​​indicated by each filling variable, and determine whether there are repeated filling values ​​among the multiple filling values. If there are repeated filling values, the filling variables with repeated filling values ​​are deleted, thereby reducing the number of repeated characteristic values ​​in the filling variable set, and ensuring the authenticity and accuracy of the statement obtained based on the filling variable set.

[0090] For example, for elements with a decision relationship, if one of them already has a restriction, the other component is simply assumed to have a mask. If both the departure date and return date are determined to be generated through the masker, the stay period element is considered not to be generated at this time, thus avoiding a large number of duplicate feature values.

[0091] According to an embodiment of the present application, a method for processing request information is also provided.

[0092] Figure 2 This is a flowchart of a method for processing request information provided in accordance with an embodiment of the present application. Figure 2 As shown, the method includes the following steps:

[0093] Step S201: Receive a request message sent by a target user and obtain a request statement in the request message.

[0094] It should be noted that the target user refers to any user who sends a request message to the system. The request message refers to the request instruction sent by the user to the server through the client. The request statement is the part of the request message that directly expresses the user's needs.

[0095] Specifically, the execution entity of this embodiment is an information processing system. After receiving a message from a target user, the system first needs to identify and extract the request statement. This process usually involves speech recognition (if the request information is in voice form), text extraction, and semantic analysis. Its purpose is to simplify the complex information sent by the user into a request statement that can be directly processed for subsequent parsing and reply.

[0096] For example, suppose a customer expresses the need to "query the flight information from Beijing to Shanghai tomorrow" to the customer service over the phone. The customer service needs to convert this voice information into text and identify the request sentence "query the flight information from Beijing to Shanghai tomorrow" as input for subsequent processing.

[0097] In step S202, the request sentence is input into a preset parsing model to obtain character string information of the request sentence, wherein the preset parsing model is obtained by training a model corpus, and the model corpus is generated by the above-mentioned method for generating a model training corpus.

[0098] It should be noted that the preset parsing model is a trained natural language processing model, and its training corpus is generated by the above steps S101 to S105, which contains a large number of structured and unstructured user request and response instances.

[0099] Specifically, after obtaining the text information, the system inputs the extracted request statement into the preset parsing model, and uses the model to convert the request statement expressed in natural language into string information that the computer can understand and process. Through the parsing of the model, the system can identify key elements in the request, such as departure city, destination, date, etc., and prepare for the subsequent generation of reply information.

[0100] For example, using the flight query request above as an example, the preset parsing model parses "Query for tomorrow's flights from Beijing to Shanghai" and converts it into a structured string, perhaps similar to {"action":"query","departure_city":"Beijing","destination":"Shanghai","date":"tomorrow"}. This conversion process ensures that the system accurately understands the user's specific needs.

[0101] Step S203: call the character string information to obtain the reply information of the request statement.

[0102] Specifically, after obtaining the string information of the request statement, the system will call the corresponding business logic or database interface to retrieve matching flight information from the database, construct a reply statement, or perform specific business operations, thereby generating reply information based on the key parameters in the string information.

[0103] For example, assuming that the system parsing module has converted the user's needs into structured string information, such as {"action":"query","departure_city":"Beijing","destination":"Shanghai","date":"2023-09-15"}, the system will retrieve the flight information from Beijing to Shanghai on September 15, 2023 from the database based on this information, and then construct a reply statement, for example, "Tomorrow there is XX flight from Beijing to Shanghai, with a departure time of YY and an arrival time of ZZ", thereby completing the reply process of the requested information.

[0104] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0105] The present application also provides a device for generating a model training corpus. It should be noted that the device for generating a model training corpus in the present application can be used to execute the method for generating a model training corpus provided in the present application. The following describes the device for generating a model training corpus provided in the present application.

[0106] Figure 3 Schematic diagram of a device for generating a model training corpus according to an embodiment of the present application. Figure 3 As shown, the device includes: a filling unit 31, a selection unit 32, a generation unit 33, a composition unit 34, and a determination unit 35.

[0107] The filling unit 31 is used to select a plurality of preset sentence templates and obtain characteristic variables used to fill the plurality of preset sentence templates to obtain a plurality of characteristic variables.

[0108] The selection unit 32 is used to obtain the filling requirements of each preset sentence template, and select the filling variables of each preset sentence template from multiple feature variables according to each filling requirement to obtain a filling variable set of each preset sentence template.

[0109] The generation unit 33 is used to generate characteristic values ​​of the filling variables in the corresponding filling variable set according to the filling requirements of each preset sentence template, and obtain a target sentence set for each preset sentence template, wherein each target sentence set includes multiple target sentences with the same sentence template but different characteristic values.

[0110] The composition unit 34 is used to obtain the character string information of the filling variables and the character string information of the feature values ​​of each target sentence, and compose the character string sentence of each target sentence from the character string information.

[0111] The determining unit 35 is configured to determine each target sentence and the corresponding string sentence as a set of training data, obtain multiple sets of training data, and determine the multiple sets of training data as a training corpus of the target model.

[0112] The model training corpus generation device provided by the embodiment of the present application includes a filling unit 31 selecting multiple preset sentence templates and obtaining feature variables for filling the multiple preset sentence templates to obtain multiple feature variables; a selection unit 32 obtaining the filling requirements of each preset sentence template, selecting the filling variables of each preset sentence template from the multiple feature variables according to each filling requirement, and obtaining a filling variable set for each preset sentence template; a generation unit 33 generating feature values ​​of the filling variables in the corresponding filling variable set according to the filling requirements of each preset sentence template, and obtaining a target sentence set for each preset sentence template, wherein each target sentence set includes multiple target sentences with the same sentence template but different feature values; a composition unit 34 obtaining string information of the filling variables and string information of the feature values ​​of each target sentence, and composing a string sentence of each target sentence from the string information; a determination unit 35 determining each target sentence and the corresponding string sentence as a set of training data to obtain multiple sets of training data, and determining the multiple sets of training data as the training corpus of the target model. The invention solves the problem in the related art that the prediction accuracy of the model obtained by corpus training is low due to the single corpus style. By obtaining preset sentence templates of different styles and filling them with different feature values, multiple target sentences with different styles and contents are obtained, and the target sentences and corresponding string information are further used as training data in the training corpus to train the model. By generating a corpus with large differences in style and content, the training effect of the model is improved, thereby achieving the technical effect of improving the model prediction accuracy.

[0113] Optionally, in the device for generating the model training corpus provided in the embodiment of the present application, the filling requirement includes precise filling, and the selection unit 32 includes: a first acquisition module, which is used to obtain the features to be filled from the filling requirements for any preset sentence template, and select the feature variables corresponding to each feature to be filled from multiple feature variables to obtain a set of filling variables for the preset sentence template.

[0114] Optionally, in the device for generating the model training corpus provided in an embodiment of the present application, the filling requirement includes fuzzy filling, and the selection unit 32 includes: a second acquisition module, used to obtain the fixed filling feature indicated by the filling requirement for any preset sentence template, and obtain the feature variable corresponding to the fixed filling feature to obtain the first feature variable; a first determination module, used to determine the feature variables other than the first feature variable among multiple feature variables as second feature variables, and traverse to generate multiple second feature variable sets consisting of at least one second feature variable; a combination module, used to combine each second feature variable set with the first feature variable in turn to obtain the filling variable set of the preset sentence template.

[0115] Optionally, in the device for generating the model training corpus provided in the embodiment of the present application, the determination unit 35 includes: an adjustment module, which is used to adjust the order of the characteristic values ​​of multiple filling variables in the target sentence for any target sentence, so as to obtain multiple changed sentences; and a second determination module, which is used to determine the string sentence corresponding to each changed sentence and the target sentence as a set of training data, so as to obtain multiple sets of training data.

[0116] Optionally, in the device for generating the model training corpus provided in the embodiment of the present application, the determination unit 35 includes: a third acquisition module, used to obtain multiple interference filling information for any target sentence, and add the interference feature value of each interference filling information to the target sentence to obtain an updated target sentence; a third determination module, used to determine the updated target sentence and the string sentence corresponding to the target sentence as a set of training data.

[0117] Optionally, in the generation device of the model training corpus provided in the embodiment of the present application, the filling requirements include precise filling and fuzzy filling, and the generation unit 33 includes: a fourth acquisition module, which is used to obtain the attribute information of each filling variable for a preset sentence template with a filling requirement of precise filling, and obtain the filling format of the filling value of the filling variable according to the attribute information, and generate the characteristic value of each filling variable according to the filling format to obtain a target sentence set; a generation module, which is used to generate description information of the characteristic value of each filling variable for a preset sentence template with a filling requirement of fuzzy filling, and determine the description information as the characteristic value of the filling variable to obtain a target sentence set.

[0118] Optionally, in the device for generating the model training corpus provided in the embodiment of the present application, after selecting the filling variables of each preset sentence template from multiple feature variables according to each filling requirement to obtain a filling variable set for each preset sentence template, the device also includes: a judgment unit, which is used to obtain the feature values ​​indicated by each filling variable for any filling variable set, and to judge whether there are repeated filling values ​​among the multiple filling values; and a deletion unit, which is used to delete the filling variables corresponding to the repeated filling values ​​when there are repeated filling values ​​among the multiple filling values, so as to obtain an updated filling variable set.

[0119] The generation device of the above-mentioned model training corpus includes a processor and a memory. The above-mentioned filling unit 31, selection unit 32, generation unit 33, composition unit 34, determination unit 35, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize the corresponding functions.

[0120] Figure 4 Schematic diagram of a device for processing request information according to an embodiment of the present application. Figure 4 As shown, the device includes:

[0121] The receiving unit 41 is configured to receive a request message sent by a target user and obtain a request statement in the request message.

[0122] The training unit 42 is used to input the request sentence into a preset parsing model to obtain the character string information of the request sentence, wherein the preset parsing model is obtained by training the model corpus, and the model corpus is generated by the above-mentioned model training corpus generation method.

[0123] The calling unit 43 is used to call the character string information to obtain the reply information of the request statement.

[0124] The device for generating the above-mentioned model training corpus includes a processor and a memory. The above-mentioned receiving unit 41, training unit 42, calling unit 43, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize corresponding functions.

[0125] The processor contains a kernel, which retrieves the corresponding program unit from memory. One or more kernels can be configured. By adjusting kernel parameters, this solves the problem in related technologies where the prediction accuracy of models trained on corpora is low due to the single corpus style.

[0126] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0127] An embodiment of the present invention provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, the method for generating a model training corpus and the method for processing request information are implemented.

[0128] An embodiment of the present invention provides a processor, which is used to run a program, wherein when the program is run, the method for generating the model training corpus and the method for processing request information are executed.

[0129] Figure 5 is a schematic diagram of an electronic device provided according to an embodiment of the present application, such as Figure 5 As shown, an embodiment of the present invention provides an electronic device 50 including a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps of the above-described method for generating a model training corpus and the method for processing request information are implemented. The device herein may be a server, a PC, a PAD, a mobile phone, or the like.

[0130] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program that initializes the steps of the above-mentioned method for generating a model training corpus and the method for processing request information.

[0131] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0132] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (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 processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. 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.

[0133] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0134] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0135] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0136] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0137] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules 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 technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0138] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0139] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A method for generating a model training corpus, characterized in that: include: Selecting a plurality of preset sentence templates and obtaining feature variables for filling the plurality of preset sentence templates to obtain a plurality of feature variables; Obtaining a filling requirement for each preset sentence template, and selecting a filling variable for each preset sentence template from the plurality of characteristic variables according to each filling requirement, to obtain a filling variable set for each preset sentence template; Generate characteristic values ​​of filling variables in a corresponding filling variable set according to the filling requirements of each preset sentence template, and obtain a target sentence set for each preset sentence template, wherein each target sentence set includes multiple target sentences with the same sentence template but different characteristic values; Obtaining character string information of filling variables and character string information of feature values ​​of each target sentence, and composing a character string sentence of each target sentence from the character string information; Each target sentence and the corresponding string sentence are determined as a set of training data to obtain multiple sets of training data, and the multiple sets of training data are determined as the training corpus of the target model.

2. The method according to claim 1, characterized in that The filling requirement includes precise filling, and the filling variables of each preset sentence template are selected from the multiple feature variables according to each filling requirement, and the filling variable set of each preset sentence template is obtained, including: For any preset sentence template, the features to be filled are obtained from the filling requirements, and the feature variables corresponding to each feature to be filled are selected from the multiple feature variables to obtain a filling variable set of the preset sentence template.

3. The method according to claim 1, characterized in that The filling requirement includes fuzzy filling, and the filling variables of each preset sentence template are selected from the multiple feature variables according to each filling requirement, and the filling variable set of each preset sentence template includes: For any preset sentence template, obtain the fixed filling feature indicated by the filling requirement, and obtain the feature variable corresponding to the fixed filling feature to obtain a first feature variable; Determine the characteristic variables other than the first characteristic variable among the plurality of characteristic variables as second characteristic variables, and traverse to generate a plurality of second characteristic variable sets consisting of at least one second characteristic variable; Each second feature variable set is combined with the first feature variable in sequence to obtain a filling variable set of the preset sentence template.

4. The method according to claim 1, wherein Determining each target sentence and the corresponding string sentence as a set of training data includes: For any target sentence, the order of the characteristic values ​​of multiple filling variables in the target sentence is adjusted to obtain multiple changed sentences; The character string sentences corresponding to each changed sentence and the target sentence are determined as a set of training data to obtain multiple sets of training data.

5. The method according to claim 1, wherein Determining each target sentence and the corresponding string sentence as a set of training data includes: For any target sentence, obtain multiple interference filling information, and add the interference feature value of each interference filling information to the target sentence to obtain an updated target sentence; The updated target sentence and the character string sentence corresponding to the target sentence are determined as a set of training data.

6. The method according to claim 1, characterized in that The filling requirements include precise filling and fuzzy filling. The characteristic values ​​of the filling variables in the corresponding filling variable set are generated according to the filling requirements of each preset sentence template. The target sentence set of each preset sentence template includes: For the preset sentence template whose filling requirement is the precise filling, obtaining attribute information of each filling variable, obtaining a filling format of a filling value of the filling variable according to the attribute information, and generating a feature value of each filling variable according to the filling format to obtain the target sentence set; For the preset sentence template whose filling requirement is the fuzzy filling, description information of the characteristic value of each filling variable is generated, and the description information is determined as the characteristic value of the filling variable to obtain the target sentence set.

7. The method according to claim 1, characterized in that After selecting a filling variable for each preset sentence template from the plurality of characteristic variables according to each filling requirement to obtain a filling variable set for each preset sentence template, the method further includes: For any set of filling variables, obtain the characteristic value indicated by each filling variable, and determine whether there is a repeated filling value among the multiple filling values; In the case where the repeated filling value exists in the multiple filling values, the filling variables corresponding to the repeated filling value are deleted to obtain an updated filling variable set.

8. A method for processing request information, characterized in that: include: Receive a request message from a target user and obtain a request statement in the request message; Inputting the request sentence into a preset parsing model to obtain character string information of the request sentence, wherein the preset parsing model is obtained by training a model corpus, and the model corpus is generated by the method for generating a model training corpus according to any one of claims 1 to 7; The character string information is called to obtain reply information of the request statement.

9. A device for generating a model training corpus, characterized in that: include: A filling unit, configured to select a plurality of preset sentence templates and obtain feature variables for filling the plurality of preset sentence templates to obtain a plurality of feature variables; a selection unit, configured to obtain a filling requirement for each preset sentence template, and select a filling variable for each preset sentence template from the plurality of characteristic variables according to each filling requirement, to obtain a filling variable set for each preset sentence template; a generating unit, configured to generate characteristic values ​​of filling variables in a corresponding filling variable set according to the filling requirements of each preset sentence template, and obtain a target sentence set for each preset sentence template, wherein each target sentence set includes a plurality of target sentences having the same sentence template but different characteristic values; a composition unit, configured to obtain character string information of filling variables and character string information of feature values ​​of each target sentence, and compose a character string sentence of each target sentence from the character string information; The determination unit is used to determine each target sentence and the corresponding string sentence as a set of training data, obtain multiple sets of training data, and determine the multiple sets of training data as the training corpus of the target model.

10. A device for processing request information, characterized in that: include: A receiving unit, configured to receive a request message sent by a target user and obtain a request statement in the request message; a training unit, configured to input the request statement into a preset parsing model to obtain character string information of the request statement, wherein the preset parsing model is obtained by training a model corpus, and the model corpus is generated by the method for generating a model training corpus according to any one of claims 1 to 7; The calling unit is used to call the character string information to obtain reply information of the request statement.

Images