Order generation method and device and electronic equipment

Abstract

The invention discloses an order generation method and device and electronic equipment. The method comprises the following steps: receiving service description information input by a user; the service description information is split through the classification model in combination with a first prompt template, classification information corresponding to the service scene type in the service description information is obtained, and the first prompt template is used for guiding the classification model to understand the service description information; service acceptance elements required by different service scenes are determined from the classification information through the extraction model in combination with a preset extraction rule and a second prompt template, and the second prompt template is used for guiding the extraction model to understand the service acceptance elements and standardizing the information extraction format of the extraction model; and determining an execution sequence of the business acceptance elements according to a preset business rule, and generating an order corresponding to the business description information according to the execution sequence. The technical problem that complex and changeable business requirements are difficult to efficiently and accurately process due to the fact that an order generation method in the related technology is low in accuracy is solved.

Description

Technical Field

[0001] The present application relates to the field of data processing, and more specifically, to an order generation method, device and electronic device. Background Art

[0002] With the rapid development of artificial intelligence technology, especially the breakthrough of large model technology, the field of natural language processing has ushered in unprecedented progress, significantly improving the ability of language understanding and generation. These technological advances are not limited to academic research, but also show excellent performance in practical applications such as text analysis, dialogue understanding, and document generation. However, although large models perform well in general language tasks, they still face challenges in specific fields, especially in the complex scenario of business acceptance.

[0003] The business acceptance process usually involves a large number of customer consultations and order processing, including diversified needs such as package changes, performance adjustments, product ordering, and disassembly. In the business acceptance process of related technologies, business personnel are usually required to manually fill in relevant order information in the business acceptance system based on understanding user needs to complete the business processing. This process is not only inefficient, but also prone to errors, especially when dealing with complex orders or non-standardized descriptions. In addition, due to the customization and flexibility of business acceptance, even if machine learning or deep learning models are used for automated order generation, it is difficult to avoid the limitations of the model's understanding of business rules and product details, especially in terms of precise matching of product names and accurate classification of business scenarios.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present application provide an order generation method, device and electronic device to at least solve the technical problem that the order generation method in the related art has low accuracy and is difficult to efficiently and accurately handle complex and changing business needs.

[0006] According to one aspect of an embodiment of the present application, an order generation method is provided, comprising: receiving business description information input by a user; splitting the business description information through a classification model in combination with a first prompt template to obtain classification information corresponding to the business description information, wherein the first prompt template is used to guide the classification model to understand the meaning and splitting logic in the business description information, and the classification information includes the business scenario type in the business description information and its corresponding classification information; determining the business acceptance element from the classification information through an extraction model in combination with preset extraction rules and a second prompt template, wherein the second prompt template is used to guide the extraction model to understand the meaning of the business acceptance element and standardize the information extraction format of the extraction model, and the business acceptance element information is used to represent the structured acceptance element information required for different business scenarios in the classification information; determining the execution order of the business acceptance elements according to the preset business rules, and generating an order corresponding to the business description information according to the execution order.

[0007] Optionally, after determining the business acceptance element from the classification information through the extraction model combined with preset extraction rules and the second prompt template, the method also includes: obtaining the name information corresponding to the sales product in the business acceptance element; vectorizing the name information through the vectorization model to obtain a first vector; performing semantic retrieval in the vector database to determine the target vector with the highest similarity to the first vector, wherein the vector database stores the vectorized representation of the full amount of standard sales product names; and determining the sales product name corresponding to the target vector as the standardized name of the sales product in the business acceptance element.

[0008] Optionally, a semantic search is performed in a vector database to determine a target vector with the highest similarity to the first vector, including: obtaining a second vector in the vector database, wherein the second vector is a vector corresponding to any standard sales product name in the vector database; determining the similarity between the first vector and each second vector in the vector database to obtain a similarity sequence; and determining the sales product vector with the highest similarity in the similarity sequence as the target vector.

[0009] Optionally, the classification model is determined by: determining an initial classification model; obtaining historical business description information under different business scenarios and historical classification information corresponding to the historical business description information; optimizing the initial classification model based on the historical business description information and the historical classification information to obtain a classification model.

[0010] Optionally, business acceptance elements are determined from classified information through an extraction model combined with preset extraction rules and a second prompt template, including: determining the complexity of the classified information, wherein the complexity is used to indicate the degree of semantic confusion of the business elements in the classified information; when the complexity is less than a preset threshold, determining the business acceptance elements from the classified information through the extraction model and the second prompt template; when the complexity is greater than or equal to the preset threshold, determining the business acceptance elements from the classified information through the extraction model, the preset extraction rules and the second prompt template.

[0011] Optionally, the execution order of business acceptance elements is determined according to preset business rules, including: splicing business acceptance elements according to preset business rules to obtain a business acceptance form corresponding to the business description information; determining the workflow logic and dependencies in the business acceptance form; and determining the execution order of the business acceptance form based on the workflow logic and dependencies.

[0012] Optionally, the execution order includes a sequential order and a parallel order, wherein when there is a dependency relationship among the business elements in the business acceptance order, the execution order of the business elements is determined to be a sequential order; when there is no dependency relationship among the business elements in the business acceptance order, the execution order of the business elements is determined to be a parallel order.

[0013] According to another aspect of an embodiment of the present application, an order generation device is also provided, including: a receiving module for receiving business description information input by a user; a classification module for splitting the business description information through a classification model in combination with a first prompt template to obtain classification information corresponding to the business description information, wherein the first prompt template is used to guide the classification model to understand the meaning and splitting logic in the business description information, and the classification information includes the business scenario type in the business description information and its corresponding classification information; an extraction module for determining the business acceptance element from the classification information through an extraction model in combination with preset extraction rules and a second prompt template, wherein the second prompt template is used to guide the extraction model to understand the meaning of the business acceptance element and standardize the information extraction format of the extraction model, and the business acceptance element information is used to represent the structured acceptance element information required for different business scenarios in the classification information; a generation module for determining the execution order of the business acceptance elements according to preset business rules, and generating an order corresponding to the business description information according to the execution order.

[0014] According to another aspect of an embodiment of the present application, there is also provided an electronic device, including: a memory and a processor, wherein the memory is used to store program instructions; the processor is connected to the memory and is used to execute the above-mentioned order generation method.

[0015] According to another aspect of the embodiments of the present application, a non-volatile storage medium is also provided, the non-volatile storage medium includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned order generation method by running the computer program.

[0016] According to another aspect of the embodiments of the present application, a computer program product is also provided, including computer instructions, which implement the above-mentioned order generation method when executed by a processor.

[0017] In an embodiment of the present application, business description information input by a user is received; the business description information is split through a classification model combined with a first prompt template to obtain classification information corresponding to the business description information, wherein the first prompt template is used to guide the classification model to understand the meaning and splitting logic in the business description information, and the classification information includes the business scenario type in the business description information and its corresponding classification information; the business acceptance elements are determined from the classification information through an extraction model combined with preset extraction rules and a second prompt template, wherein the second prompt template is used to guide the extraction model to understand the meaning of the business acceptance elements and standardize the information extraction format of the extraction model, and the business acceptance element information is used to represent the structured acceptance element information required for different business scenarios in the classification information; the execution order of the business acceptance elements is determined according to the preset business rules, and an order corresponding to the business description information is generated according to the execution order, thereby achieving the purpose of improving the efficiency and accuracy of business acceptance, thereby realizing the optimization of the automatic generation process of business orders, and the technical effect of intelligent understanding and efficient processing of complex business scenarios, thereby solving the technical problem that the order generation method in the related technology has low accuracy and is difficult to efficiently and accurately handle complex and changeable business needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0019] Figure 1 is a hardware structure diagram of a computer terminal for implementing an order generation method according to an embodiment of the present application;

[0020] Figure 2 is a flow chart of an order generation method according to an embodiment of the present application;

[0021] Figure 3 is a flow chart of another order generation method according to an embodiment of the present application;

[0022] Figure 4 It is a structural diagram of an order generating device according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0024] 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 sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. 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 comprising 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.

[0025] First, some nouns or terms that appear in the process of explaining the embodiments of the present application are subject to the following explanations:

[0026] LLM (Large Language Model): A deep learning model used in the field of natural language processing, with a large number of parameters, that can understand and generate complex text content. Large language models are pre-trained on massive amounts of text data to learn the deep structure and patterns of language, thus performing well in a variety of NLP tasks, such as text generation, machine translation, semantic understanding, etc.

[0027] LoRA fine-tuning: A lightweight technique for fine-tuning large models that can effectively reduce computing resource consumption while maintaining or improving the performance of the model on specific tasks by training only specific low-rank parts of the model instead of the entire model.

[0028] RAG (Retrieval-Augmented Generation): A model that combines retrieval and generation techniques, mainly used to handle open-domain question answering and text generation tasks. In the RAG model, when receiving an input query, it not only relies on the model's own knowledge to generate an answer, but also retrieves relevant information from a large external document library to enhance the generated answer or text.

[0029] Few-shot learning: refers to the method of providing a small number of examples to enhance the model's ability to solve specific tasks in the process of writing prompt templates for a large model. In this application, a small number of business scenario classification samples are provided to guide the model to learn, so that it can understand and process business descriptions similar to the examples.

[0030] Workflow engine: An automated software tool used to define, execute, and manage workflows to ensure that business scenarios are executed correctly according to the preset order and rules. In this application, it is used to automatically arrange the execution order of business scenarios to generate a complete order message.

[0031] In order to solve the problem of poor business acceptance efficiency in the related art, the present application embodiment provides an order generation method, which can be run on Figure 1 In the computer terminal shown, the computer terminal is described below.

[0032] The order generation method embodiment provided in the embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computer terminal for implementing an order generation method. Figure 1 As shown, the computer terminal 10 may include one or more (102a, 102b, ..., 102n are used to illustrate) processors (the processor may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission module 106 for communication functions connected via a wired and / or wireless network. In addition, it may also include: a display, a keyboard, a cursor control device, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, and a BUS bus. Those skilled in the art can understand that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1Different configurations are shown.

[0033] It should be noted that the one or more processors and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10. As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0034] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the order generation method in the embodiment of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the above-mentioned order generation method is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0035] The transmission module 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission module 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission module 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0036] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 .

[0037] It should be noted that, in some optional embodiments, the above Figure 1 The computer terminal shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of hardware elements and software elements. It should be noted that Figure 1This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computer terminal described above.

[0038] In the above-mentioned operating environment, an embodiment of the present application provides an order generation method embodiment. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0039] Figure 2 is a flow chart of an order generation method according to an embodiment of the present application, such as Figure 2 As shown, the method comprises the following steps:

[0040] Step S202: receiving service description information input by a user.

[0041] In the above step S202, the original business demand (description) information is collected directly from the user or the business reception staff. The business description information is usually described in a free-form natural language, which may include specific needs such as package changes, performance adjustments, product member changes, and sales product orders.

[0042] Step S204, split the business description information through the classification model combined with the first prompt template to obtain classification information corresponding to the business description information, wherein the first prompt template is used to guide the classification model to understand the meaning and splitting logic in the business description information, and the classification information includes the business scenario type in the business description information and its corresponding classification information.

[0043] In the above step S204, after receiving the user input, the classification model fine-tuned by LoRA (the classification model is a large language model LLM) can be used, combined with the classification prompt template of the business knowledge background (that is, the above-mentioned first prompt template), to split the business description information. This process aims to convert the free text business description into structured classification information, and clearly indicate which specific business scenarios the user needs correspond to, such as "changing packages", "changing performance", "disassembling machines", etc. Among them, the role of the first prompt template is to guide the model to understand the key information (meaning or splitting logic) in the text to ensure that the model can accurately parse business needs.

[0044] Step S206, determine the business acceptance elements from the classified information through the extraction model combined with the preset extraction rules and the second prompt template, wherein the second prompt template is used to guide the extraction model to understand the meaning of the business acceptance elements and standardize the information extraction format of the extraction model, and the business acceptance element information is used to represent the structured acceptance element information required for different business scenarios in the classified information.

[0045] In the above step S206, the extraction model (or the LLM model fine-tuned in the above step S204) can be used in combination with preset rules and extraction prompt templates (i.e., the above second prompt template) to extract business acceptance elements corresponding to different business scenarios from the classified information. These elements are key data points for business processing, such as equipment number, products to be ordered, products to be canceled, construction time, etc. Among them, the second prompt template is used to help the extraction model understand the meaning of business acceptance elements and clarify the format of information extraction, and the preset extraction rules are used to ensure the accuracy and completeness of the information and avoid the problem of element confusion that may be encountered in a single extraction of a large model. For example, for the "change package" scenario, the model will extract the device number and target package name from the input text.

[0046] It should be noted that after extracting the elements, in order to ensure the standardization and accuracy of the sales product name, a vectorization model can be used to convert the sales product name into a vector representation, and a semantic search can be performed in the vector database to find the standard sales product name that best matches the user description.

[0047] Step S208, determining the execution order of the business acceptance elements according to the preset business rules, and generating an order corresponding to the business description information according to the execution order.

[0048] In the above step S208, the dependencies between different business scenarios are taken into account to ensure that the processing logic of the business scenarios is correct, for example, "changing packages" is executed before "changing product members". These business scenarios are automatically arranged through the workflow engine to ensure the correctness and efficiency of the process. Finally, the output of all business scenarios will be integrated into a complete order message and directly submitted to the business acceptance system for processing.

[0049] Through the above steps S202 to S208, the purpose of improving business acceptance efficiency and accuracy is achieved, thereby realizing the optimization of the automatic business order generation process and the technical effect of intelligent understanding and efficient processing of complex business scenarios, thereby solving the technical problem that the order generation method in the related technology has low accuracy and is difficult to efficiently and accurately handle complex and changeable business needs.

[0050] Figure 3 is a flow chart of another order generation method according to an embodiment of the present application. Figure 3 As shown, the method includes four stages: business scenario classification and decomposition, business information pre-extraction, sales name matching, and order assembly and arrangement.

[0051] Specifically, the process starts with the business description text (information) input by the user. Through the fine-tuned large language model (LLM) combined with the classification prompt template, the description text is converted into structured classification information, and each business scenario is clearly subdivided, such as changing packages, changing performance, changing product members, disassembling, etc. Subsequently, the LLM guided by the information extraction prompt template and predefined rules is used to further accurately extract the key elements in each scenario, such as device number, sales product name, time, etc., to avoid confusion between elements. In the sales product name matching link, the vectorization model is used to convert the sales product fuzzily described by the user into a vector representation, and semantic retrieval is performed in the constructed vector database to ensure that the most accurate standard sales product name is matched. Finally, the workflow engine automatically arranges the processing order of all business scenarios according to the preset business rules and scenario dependencies, integrates the extracted element information into a complete order message, and submits it to the business acceptance system for execution. This entire process cleverly combines the intelligent classification and information extraction capabilities of the big model, the semantic matching technology of the vector database, and the automated orchestration capabilities of the workflow engine, achieving automation from user free text input to standardized order generation, greatly improving the efficiency and accuracy of business processing, and providing users with a more efficient and accurate service experience.

[0052] The following combination Figure 2 and Figure 3 The above steps are explained in detail.

[0053] In the above step S204, the classification model is determined by: determining an initial classification model; obtaining historical business description information under different business scenarios and historical classification information corresponding to the historical business description information; optimizing the initial classification model based on the historical business description information and the historical classification information to obtain a classification model.

[0054] In an embodiment of the present application, the classification model is determined by combining deep learning technology with business data. First, an initial classification model is selected, which is usually a pre-trained large language model with strong natural language understanding and classification capabilities. Then, by collecting historical business description information and corresponding historical classification information under different business scenarios within the operator, the model is supervised by LoRA fine-tuning and optimization training. These historical data include a wealth of business scenario instances, such as package changes, performance adjustments, etc., as well as results that have been manually classified by business personnel. By learning these instances and classification labels, the initial classification model gradually improves the understanding and classification accuracy of business scenarios, and finally forms a classification model that can accurately identify and classify complex business scenarios, providing a solid foundation for subsequent information extraction and order generation.

[0055] At the same time, the first (classification) prompt template mentioned above can also be optimized by using a few-shot small sample learning method. Specifically, first collect a small number of representative business scenario descriptions and their correct classification annotations. These sample data constitute the cornerstone of few-shot learning. These annotated sample data are vectorized, and the business description text in the sample is converted into a vector representation using the same vectorization model. Then, these annotated samples are incorporated into the first prompt template in the form of prompts as a reference and guide for the model to process new business descriptions. When the user enters a new business description, the model not only relies on its own pre-trained knowledge, but also can refer to the few-shot samples embedded in the template to quickly understand and match the most similar business scenario type, thereby effectively improving the accuracy and efficiency of classification. This strategy provides the model with additional scene understanding clues by adding few-shot learning samples to the classification prompt template, and enhances the model on the basis of its original capabilities, enabling it to handle frequently changing business scenarios, which helps to improve the intelligence level of business automation processing. Specific small sample classification examples are as follows:

[0056] Example 1:

[0057] The user input is: 191******** The original package is changed to the Beautiful Home 129 package, and the whole house gigabit network age is ordered at half price for 25 yuan, superimposed with FTTR networking, the whole house gigabit superimposed with ancillary products one master and one slave, 4K subscription 202002-Tianyi HD Variety Membership Package (monthly package) 20 yuan / month.

[0058] The correct output is:

[0059] {

[0060] "Change package": "The original package of 191******** is changed to the package of Beautiful Home 129",

[0061] "Improve performance": "Order the whole house gigabit network age half price 25 yuan package, superimpose FTTR networking, whole house gigabit superimpose auxiliary products one master and one slave, 4K subscription 202002-Tianyi HD variety show membership package (monthly package) 20 yuan / month"

[0062] }

[0063] Example 2:

[0064] The user input is: the original package of 189******** is changed to the Meihaojia 229FTTR monthly payment package, ordering a full card with 100G, 1 yuan for 20G, 5 traffic packages at 0.1 yuan per card, a whole house gigabit package at 0 yuan for one master and one slave, specifying the Meihaojia package super gift package (1299 yuan) to pay at the front desk, and canceling all the original Stacking Le services.

[0065] The correct output is:

[0066] {

[0067] "Change package": "The original 189******** package is changed to the Meihaojia 229FTTR monthly payment package",

[0068] "Change performance": "Order a full card with 100G, 1 yuan for 20G, 0.1 yuan for each card for 5 data packages, 0 yuan for the whole house gigabit package with one master and one slave, specify the beautiful home package super enjoyment package (1299 yuan) for front desk payment, and cancel all the original stacking music services"

[0069] }

[0070] Provide several labeled business data similar to the input as few shot learning data to enhance the model's capabilities. The final first prompt template example is as follows:

[0071]

[0072]

[0073] In this step, the comprehensive application of RAG retrieval enhancement, few-shot small sample learning and classification prompt word strategy can rely on the model's inherent learning ability when processing the business description text entered by the user, and combine the support of the external knowledge base and the guidance information of specific scenarios, thereby significantly improving the accuracy and efficiency of the model's recognition and classification of business scenarios. Even in the face of complex and changing business descriptions, it can ensure that the model can quickly and accurately identify the scene, thereby providing more accurate basic information for subsequent business information pre-extraction and order generation. For example:

[0074] Enter the business description information:

[0075] The original package of device 133******** was changed to the Meihaojia 229 yuan monthly payment package. The order of primary and secondary cards remained unchanged. WIFI6 was not required. Two secondary cards were added: 898611230702143**** and 898611230702143****. Device 133******** ordered a 1 yuan mutual package, a full card 30G traffic package, and unsubscribed from the 30G directional traffic small contract; device ZH11311259 was cancelled, the terminal recycling model was ZNHE602, the serial number was SHSY2109010003, and the construction time was September 1, 2024.

[0076] The classification model fine-tuned by LoRA is combined with the classification prompt template embedded with few-shot samples to process the input business description information to obtain structured classification information. The output is:

[0077] {

[0078] "Change package": "The original package of device 133******** has been changed to the Beautiful Home 229 yuan monthly payment package",

[0079] "Product member change": "Add two additional cards: 898611230702143**** and 898611230702143****",

[0080] "Change performance": "Device 133******** subscribes to a 1 yuan package, a full card 30G traffic package, and cancels the 30G directional traffic small contract",

[0081] "Disassembly": "Equipment ZH11311259 is deregistered, terminal recycling model ZNHE602, serial number SHSY2109010003, construction time September 1, 2024"

[0082] }

[0083] In the above step S206, the business acceptance elements are determined from the classified information through the extraction model combined with the preset extraction rules and the second prompt template, including: determining the complexity of the classified information, wherein the complexity is used to indicate the degree of semantic confusion of the business elements in the classified information; when the complexity is less than a preset threshold, determining the business acceptance elements from the classified information through the extraction model and the second prompt template; when the complexity is greater than or equal to the preset threshold, determining the business acceptance elements from the classified information through the extraction model, the preset extraction rules and the second prompt template.

[0084] In the embodiment of the present application, the business information pre-extraction process determines the information extraction strategy by evaluating the semantic confusion degree of the classified information. This process combines the extraction model (the above-mentioned fine-tuned LLM can also be used), the preset extraction rules and the extraction (second) prompt template, aiming to accurately determine the business acceptance elements from the classified information. The specific analysis is as follows:

[0085] First, the system will evaluate the complexity of the classification information obtained from step S204. The complexity is evaluated based on the degree of semantic confusion between business elements. For example, when the input business description contains multiple business scenarios or element information (such as equipment number / model / serial number, etc.) is highly intertwined, the complexity will be judged as high; on the contrary, if the business description is clear and the elements are clear, the complexity is low.

[0086] If the complexity of the evaluation is less than the preset threshold, that is, the business description is relatively simple and the elements are clearly classified, the system will use the extraction model combined with the second prompt template to directly extract information. The second prompt template is used here as a guide for the model to process the input text, helping the model focus on key business elements and improving the accuracy and efficiency of extraction.

[0087] When the complexity of the classified information reaches or exceeds the preset threshold, it means that there are semantic confusions between multiple scenarios or elements in the business description information. At this time, the system will introduce preset extraction rules to cooperate with the extraction model and the second prompt template to perform more detailed information analysis. Preset extraction rules usually include the processing logic of business scenarios, the correlation judgment between elements, etc. They are combined with models and templates to more effectively separate and identify various business elements, and ensure the accuracy of information extraction even under complex conditions.

[0088] The second prompt template example is as follows:

[0089]

[0090]

[0091] In this strategy, the system can flexibly adjust the information extraction method according to the complexity of the classified information, which not only ensures that information extraction is completed quickly and efficiently in simple scenarios, but also ensures that in complex scenarios, through the deep combination of rule assistance, models and templates, the business acceptance elements can be accurately captured. For example:

[0092] Example 1: For "improving performance", the degree of semantic confusion is low, and the extraction model and the second prompt template can be directly used to determine the business acceptance elements from the classification information.

[0093] Input: Device 133******** orders a 1 yuan package, a full card 30G traffic package, and cancels the 30G directional traffic small contract.

[0094] Output:

[0095] {

[0096] "Device Number": ["133********"],

[0097] "Products to be ordered": ["1 yuan mutual package", "Full card 30G data package"],

[0098] "Products to be cancelled": ["30G directional traffic small contract"]

[0099] }

[0100] Example 2: For "disassembly", the semantic confusion is high, and it is necessary to determine the business acceptance elements from the classified information through the extraction model, the second prompt template and the preset extraction rules. In order to avoid the large model confusing the equipment number / model / serial number and other elements in a single extraction, the information preset extraction rules for the disassembly scene can be as follows:

[0101] General rules: extract the dismantling equipment number, required construction time, and dismantling reason.

[0102] Rule 1: If there are keywords such as "model" / "serial code" / "serial number", extract the terminal recycling model and terminal recycling serial code.

[0103] Rule 2: If there is a keyword "Flexible Charge" / "Flexible Refund", extract the associated order number.

[0104] Input: Equipment ZH11311259 is cancelled, terminal recycling model ZNHE602, serial number SHSY2109010003, construction time September 1, 2024.

[0105] According to the general rules, the output is:

[0106] {

[0107] "Disassembly equipment number": ["ZH11311259"],

[0108] "Required construction time": "2024-09-01",

[0109] "Reason for disassembly": "Terminal recycling"

[0110] }

[0111] Since the input contains the keywords "model" / "serial code" / "serial number", according to rule 1, the output is:

[0112] {

[0113] "Terminal recycling model": ["ZNHE602"],

[0114] "Terminal recovery code": ["SHSY2109010003"]

[0115] }

[0116] Optionally, after determining the business acceptance element corresponding to the business description information, the above method also includes: obtaining the name information corresponding to the sales product in the business acceptance element; vectorizing the name information through a vectorization model to obtain a first vector; performing semantic retrieval in a vector database to determine a target vector with the highest similarity to the first vector, wherein the vector database stores vectorized representations of all standard sales product names; and determining the sales product name corresponding to the target vector as the standardized name of the sales product in the business acceptance element.

[0117] Among them, semantic retrieval is performed in the vector database to determine the target vector with the highest similarity to the first vector, including: obtaining a second vector in the vector database, wherein the second vector is a vector corresponding to any standard sales product name in the vector database; determining the similarity between the first vector and each second vector in the vector database to obtain a similarity sequence; and determining the sales product vector with the highest similarity in the similarity sequence as the target vector.

[0118] In the embodiment of the present application, the process of determining the business acceptance element corresponding to the business description information and standardizing the sales product name is intended to ensure that the sales product name involved in the business acceptance element is consistent with the standard sales product name in the operator's order acceptance system. The specific process may be as follows:

[0119] After the information pre-extraction phase is completed, the system will obtain a series of business acceptance elements, including the sales product name information entered by the user. These name information may be vague, non-standardized, or slightly different from the standard sales product name. For example, the user may say "order a 20G traffic package", but the specific traffic package ordered corresponds to which sales product in the acceptance system, which needs to be further clarified.

[0120] Secondly, the vectorization model is used to convert these non-standardized name information into vector representation, that is, the first vector is obtained. This vectorization process is based on natural language processing technology, which converts text information into mathematical vectors for subsequent similarity calculation and retrieval.

[0121] Next, a semantic search is performed in the vector database. It should be noted that the system maintains a vector database in advance, which stores the vectorized representation of all standard sales product names, namely the second vector. When it is necessary to determine the standard correspondence of a non-standard sales product name, the system will calculate the similarity between the first vector and each second vector in the vector database. Among them, the similarity calculation can use methods such as cosine similarity or L2 distance to measure the semantic proximity between two vectors.

[0122] By calculating the similarity between the first vector and all the second vectors, the system generates a similarity sequence containing similarity values ​​with each standardized sales product name in the database. The system identifies the highest value in the similarity sequence, and the corresponding sales product vector is the target vector, which means that the sales product name is closest to the non-standardized description entered by the user.

[0123] Finally, the system maps the target vector back to its original sales product name, which is the standard sales product name corresponding to the non-standard description. Through this process, the system ensures that even if the user enters a non-standard description, it can accurately match the standard sales product name in the operator's order acceptance system, providing accurate sales product information for subsequent order assembly and arrangement.

[0124] In the above process, by calculating the similarity between the user description vector and each standard sales product vector in the database, the sales product with the highest similarity is identified as the standardized name, ensuring that the sales product information in the order is consistent with the operator's internal standards, and improving the accuracy and efficiency of order generation. Among them, the use of vectorization and semantic retrieval technology effectively solves the matching problem between the user input description and the standard sales product name, providing strong support for automated order generation, promoting the intelligence of business processes, and improving user experience and the reliability of business execution.

[0125] For example, when the output is "1 yuan mutual package", the query result is as follows (in this application, a vectorization model with a dimension of 1768 is selected to vectorize the sales product name, and the L2 distance is used to calculate the vector distance). The result should select the sales product name with the smallest vector distance, so the result is "201603 Family Enjoy New Fusion Local 1 Yuan Mutual Discount Package". The specific correspondence between the sales product name and the L2 vector distance is shown in Table 1:

[0126] Table 1 Correspondence between sales product names and L2 vector distances

[0127]

[0128]

[0129] In addition, in the embodiment of the present application, a fast semantic search based on a mapping table can also be used to ensure that the common sales product names in the business acceptance elements can be quickly and accurately converted into standard sales product names in the operator's order acceptance system. The specific process can be as follows:

[0130] First, collect and organize the common high-frequency sales product names. Use the pre-trained vectorization model to vectorize each sales product name and generate the corresponding vector representation. Store each sales product name and its corresponding vector in a hash map, where the key of the hash map is the sales product name and the value is the corresponding vector representation.

[0131] Secondly, the fast search feature of the hash map is used to achieve instant access to the sales product name. For example, an approximate nearest neighbor search algorithm (such as Annoy, Faiss, etc.) is used to build a vector-based fast retrieval system, which will allow efficient retrieval of the vectors in the map to find the sales product name that is most similar to the sales product description entered by the user. At the same time, a query vector threshold is specified to distinguish the similarity level of the query results to ensure that the most relevant standard sales product names are retrieved.

[0132] When the user input contains non-standardized sales product names, the same vectorization model is used to vectorize these names to generate user input vectors. The user input vector is compared with the vectors in the hash mapping table, and the vector closest to the user input vector (i.e., the smallest cosine distance or L2 distance) is found through the fast retrieval system. Based on the retrieved vector, the corresponding standardized sales product name is obtained from the mapping table.

[0133] Finally, the retrieved standardized sales product name is output as the official name of the sales product in the business acceptance element. Based on the standardized name, the system can automatically identify its business attributes (such as package type, traffic package, optional package, etc.) and perform subsequent order assembly and business execution according to established business rules and processes.

[0134] In the above process, the mapping table not only provides a direct association between the sales product name and the vector representation, but also realizes efficient semantic-based matching through a fast retrieval mechanism, enabling operators to quickly convert the sales products in the user description into standard names recognizable by the system, thereby speeding up the automatic generation of orders and ensuring the accuracy and consistency of business processes.

[0135] In the above step S208, the execution order of the business acceptance elements is determined according to the preset business rules, including: splicing the business acceptance elements according to the preset business rules to obtain a business acceptance form corresponding to the business description information; determining the workflow logic and dependency relationship in the business acceptance form; and determining the execution order of the business acceptance form according to the workflow logic and dependency relationship.

[0136] Optionally, the execution order includes a sequential order and a parallel order, wherein when there is a dependency relationship among the business elements in the business acceptance order, the execution order of the business elements is determined to be a sequential order; when there is no dependency relationship among the business elements in the business acceptance order, the execution order of the business elements is determined to be a parallel order.

[0137] In the embodiment of the present application, the execution order of the business acceptance elements is determined based on preset business rules to ensure the correct execution process of each task in the business acceptance form. The specific process can be as follows:

[0138] First, the business acceptance elements obtained from step S206 are integrated and spliced ​​according to the preset business rules to form a complete business acceptance form. This form contains all the business operations that need to be performed, such as changing packages, changing performance, changing product members, and disassembling, and each scenario contains specific business element information.

[0139] Secondly, analyze the workflow logic of each task in the business acceptance form to identify the dependencies between different business elements. For example, you may need to change the package before you can order a data package, or you cannot perform a disassembly operation before completing the product member change. By analyzing these dependencies, the system can build a reasonable task execution flowchart.

[0140] Determine the execution order of each task in the business acceptance form based on the workflow logic and dependencies. If there are dependencies between business elements, determine the execution order of these elements as a sequence to ensure that the next task can be carried out only after the previous task is completed; if there are no dependencies between business elements or they can be executed independently, then their execution order can be a parallel sequence, that is, multiple tasks can be carried out at the same time to speed up the business processing flow. For example, "changing the package" may need to be completed before "changing product members" and "changing performance", while "disassembling the machine" may be arranged after all change operations. In this way, it can ensure that even in the face of complex multi-business scenarios, each business operation can be executed in the correct logic and order, avoiding errors and omissions in the execution process, and improving the efficiency and accuracy of business processing.

[0141] Finally, the workflow engine is used to automatically edit according to the execution order to ensure that the business scenarios can be executed in the correct order. Taking the four business scenarios of changing packages, changing product members, changing performance, and disassembling as examples, the result of the arrangement is that changing packages is executed first, changing product members and changing performance are combined and executed, and finally the disassembly scenario is executed. Finally, the output of all scenarios is integrated into a complete order message and submitted to the business acceptance system to complete the entire business processing flow. The order message results are shown below:

[0142]

[0143]

[0144]

[0145]

[0146]

[0147] To sum up, determining the execution order of business acceptance elements through preset business rules is a key link in the entire order automatic generation process. It ensures the reasonable arrangement of various tasks in the business acceptance form, provides logical support for automated business processing, and is an important guarantee for achieving efficient and accurate order generation.

[0148] In the embodiments of the present application, the automation and intelligence level of the traditional business acceptance process is significantly improved through intelligent classification and disassembly of business scenarios, precise pre-extraction of business information, semantic matching of sales product names, and rule-based order assembly and arrangement. Specifically, by using LoRA fine-tuning technology to enhance the classification ability of the model, combined with few-shot learning and classification prompt templates, it can more accurately identify and segment complex business scenarios; the information pre-extraction stage adopts a multi-round large model + rule-driven method to effectively avoid information confusion and improve the accuracy of information extraction; in the matching of sales product names, the vector database is innovatively used for semantic approximate retrieval, which solves the problem of inconsistency between product descriptions and standard names, and ensures the accuracy of order elements; finally, the scene dependencies and execution order determined by the workflow engine according to business rules are used to achieve efficient assembly and arrangement of order information. On the whole, this application provides an efficient, accurate and flexible order automatic generation solution by integrating advanced natural language processing technology and database technology, which greatly improves business processing efficiency and user experience, and is an important technological innovation in the field of operator automated business acceptance.

[0149] According to an embodiment of the present application, an order generation device is provided. It should be noted that the order generation device of the embodiment of the present application can be used to execute the order generation method provided in the embodiment of the present application. The order generation device provided in the embodiment of the present application is introduced below.

[0150] Figure 4 is a structural diagram of an order generation device provided according to an embodiment of the present application. Figure 4 As shown, the device comprises:

[0151] A receiving module 40, configured to receive service description information input by a user;

[0152] A classification module 42, configured to split the business description information by combining the classification model with the first prompt template to obtain classification information corresponding to the business description information, wherein the first prompt template is used to guide the classification model to understand the meaning and splitting logic in the business description information, and the classification information includes the business scenario type in the business description information and its corresponding classification information;

[0153] An extraction module 44 is used to determine the business acceptance element from the classified information by combining the extraction model with the preset extraction rules and the second prompt template, wherein the second prompt template is used to guide the extraction model to understand the meaning of the business acceptance element and standardize the information extraction format of the extraction model, and the business acceptance element information is used to represent the structured acceptance element information required for different business scenarios in the classified information;

[0154] The generating module 46 is used to determine the execution order of the business acceptance elements according to the preset business rules, and to generate an order corresponding to the business description information according to the execution order.

[0155] Through the receiving module 40, classification module 42, extraction module 44 and generation module 46 in the above-mentioned order generation device, the purpose of improving business acceptance efficiency and accuracy is achieved, thereby realizing the optimization of the automatic generation process of business orders, and the technical effect of intelligent understanding and efficient processing of complex business scenarios, thereby solving the technical problem that the order generation method in the relevant technology has low accuracy and is difficult to efficiently and accurately handle complex and changeable business needs.

[0156] In the order generation device provided in the embodiment of the present application, the classification module is also used to determine the initial classification model; obtain historical business description information under different business scenarios and historical classification information corresponding to the historical business description information; optimize the initial classification model based on the historical business description information and the historical classification information to obtain a classification model.

[0157] In the order generation device provided in the embodiment of the present application, the extraction module is also used to determine the complexity of the classification information, wherein the complexity is used to indicate the degree of semantic confusion of the business elements in the classification information; when the complexity is less than a preset threshold, the business acceptance elements are determined from the classification information through the extraction model and the second prompt template; when the complexity is greater than or equal to the preset threshold, the business acceptance elements are determined from the classification information through the extraction model, the preset extraction rules and the second prompt template.

[0158] In the order generation device provided in the embodiment of the present application, the extraction module is also used to obtain the name information corresponding to the sales product in the business acceptance element; vectorize the name information through a vectorization model to obtain a first vector; perform semantic retrieval in the vector database to determine the target vector with the highest similarity to the first vector, wherein the vector database stores the vectorized representation of the full amount of standard sales product names; and determine the sales product name corresponding to the target vector as the standardized name of the sales product in the business acceptance element.

[0159] In the order generation device provided in the embodiment of the present application, the extraction module is also used to obtain a second vector in the vector database, wherein the second vector is a vector corresponding to any standard sales product name in the vector database; determine the similarity between the first vector and each second vector in the vector database to obtain a similarity sequence; and determine the sales product vector with the highest similarity in the similarity sequence as the target vector.

[0160] In the order generation device provided in the embodiment of the present application, the generation module is also used to splice business acceptance elements according to preset business rules to obtain a business acceptance form corresponding to the business description information; determine the workflow logic and dependencies in the business acceptance form; and determine the execution order of the business acceptance form based on the workflow logic and dependencies.

[0161] In the order generation device provided in the embodiment of the present application, the generation module is also used to determine the execution order of the business elements as a sequential order when there is a dependency relationship between the business elements in the business acceptance order; when there is no dependency relationship between the business elements in the business acceptance order, determine the execution order of the business elements as a parallel order.

[0162] An embodiment of the present application also provides an electronic device, including: a memory and a processor, wherein the memory is used to store program instructions; the processor is connected to the memory and is used to execute the above-mentioned order generation method.

[0163] It should be noted that the above electronic equipment is used to execute Figure 2 The order generation method shown, therefore the relevant explanations and instructions in the above order generation method are also applicable to the electronic device and will not be repeated here.

[0164] An embodiment of the present application also provides a non-volatile storage medium, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned order generation method by running the computer program.

[0165] It should be noted that the above non-volatile storage medium is used to execute Figure 2 The order generation method shown, therefore the relevant explanations in the above order generation method are also applicable to the non-volatile storage medium, and will not be repeated here.

[0166] An embodiment of the present application also provides a computer program product, including computer instructions, which implement the above-mentioned order generation method when executed by a processor.

[0167] It should be noted that the above-mentioned computer program product is used to execute Figure 2 The order generating method shown, therefore the relevant explanations and instructions in the above order generating method are also applicable to the computer program product and will not be repeated here.

[0168] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0169] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0170] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0171] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0172] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0173] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0174] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. An order generation method, characterized in that: include: Receive business description information input by the user; Splitting the business description information by combining a classification model with a first prompt template to obtain classification information corresponding to the business description information, wherein the first prompt template is used to guide the classification model to understand the meaning and splitting logic in the business description information, and the classification information includes the business scenario type in the business description information and its corresponding classification information; Determine the business acceptance elements from the classified information by combining the extraction model with the preset extraction rules and the second prompt template, wherein the second prompt template is used to guide the extraction model to understand the meaning of the business acceptance elements and standardize the information extraction format of the extraction model, and the business acceptance element information is used to represent the structured acceptance element information required for different business scenarios in the classified information; The execution order of the business acceptance elements is determined according to preset business rules, and an order corresponding to the business description information is generated according to the execution order.

2. The method according to claim 1, characterized in that After determining the business acceptance element from the classification information by combining the extraction model with the preset extraction rule and the second prompt template, the method further includes: Obtaining name information corresponding to the sales product in the business acceptance element; Performing vectorization processing on the name information through a vectorization model to obtain a first vector; Performing semantic search in a vector database to determine a target vector having the highest similarity to the first vector, wherein the vector database stores vectorized representations of all standard sales product names; The sales product name corresponding to the target vector is determined as the standardized name of the sales product in the business acceptance element.

3. The method according to claim 2, characterized in that Performing semantic search in the vector database to determine the target vector with the highest similarity to the first vector includes: Acquire a second vector in the vector database, wherein the second vector is a vector corresponding to any standard sales product name in the vector database; Determine the similarity between the first vector and each second vector in the vector database to obtain a similarity sequence; The sales product vector with the highest similarity in the similarity sequence is determined as the target vector.

4. The method according to claim 1, characterized in that: The classification model is determined by: Determine the initial classification model; Acquire historical business description information under different business scenarios and historical classification information corresponding to the historical business description information; The initial classification model is optimized according to the historical business description information and the historical classification information to obtain the classification model.

5. The method according to claim 1, characterized in that Determining the business acceptance element from the classification information by combining the extraction model with the preset extraction rule and the second prompt template includes: Determining the complexity of the classification information, wherein the complexity is used to represent the degree of semantic confusion of business elements in the classification information; When the complexity is less than a preset threshold, determining the business acceptance element from the classification information by using the extraction model and the second prompt template; When the complexity is greater than or equal to the preset threshold, the business acceptance element is determined from the classification information by using the extraction model, the preset extraction rule and the second prompt template.

6. The method according to claim 1, characterized in that Determine the execution order of the business acceptance elements according to the preset business rules, including: splicing the business acceptance elements according to the preset business rules to obtain a business acceptance form corresponding to the business description information; Determine the workflow logic and dependencies in the business acceptance form; The execution order of the business acceptance form is determined according to the workflow logic and dependency relationship.

7. The method according to claim 6, characterized in that The execution order includes sequential order and parallel order, wherein: When there is a dependency relationship between the business elements in the business acceptance order, determining the execution order of the business elements as a sequential order; When there is no dependency relationship between the business elements in the business acceptance order, the execution order of the business elements is determined to be a parallel order.

8. An order generating device, characterized in that: include: A receiving module, used for receiving business description information input by a user; a classification module, configured to split the business description information by combining a classification model with a first prompt template to obtain classification information corresponding to the business description information, wherein the first prompt template is used to guide the classification model to understand the meaning and splitting logic in the business description information, and the classification information includes the business scenario type in the business description information and its corresponding classification information; An extraction module, used to determine the business acceptance element from the classified information by combining an extraction model with a preset extraction rule and a second prompt template, wherein the second prompt template is used to guide the extraction model to understand the meaning of the business acceptance element and standardize the information extraction format of the extraction model, and the business acceptance element information is used to represent the structured acceptance element information required for different business scenarios in the classified information; A generation module is used to determine the execution order of the business acceptance elements according to preset business rules, and generate an order corresponding to the business description information according to the execution order.

9. An electronic device, characterized in that: include: A memory and a processor, wherein the memory is used to store program instructions; The processor, connected to the memory, is used to execute the order generation method described in any one of claims 1 to 7.

10. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the order generation method described in any one of claims 1 to 7 by running the computer program.

11. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by a processor, the order generation method described in any one of claims 1 to 7 is implemented.

Images