A Method, Device, Equipment and Storage Medium for Optimizing Large Model Q&A

Abstract

The present application discloses a method, device, equipment and storage medium for optimizing large model question answering, which relates to the field of artificial intelligence technology. The method includes extracting elements from candidate program files of candidate web page codes obtained from a target business system to obtain at least one candidate web page element; a candidate web page element refers to a web page element containing a click trigger event; determining a target feature value tree according to the code path of the candidate web page code, at least one candidate web page element and the behavioral action characteristics of at least one candidate web page element; based on a general business large model, optimizing the initial feature value dictionary of the target business system according to the target feature value tree to obtain the target feature value dictionary of the target business system; optimizing the general business large model according to the target feature value dictionary and the private knowledge question answering library of the target business system to obtain the target business large model of the target business system. The above technical solution helps to improve the noise reduction function of the optimized large model.

Description

Technical Field

[0001] Embodiments of the present application relate to the field of computer technology, and more particularly to the field of artificial intelligence technology. Specifically, the present application relates to a method, apparatus, device, and storage medium for optimizing large model question answering. Background Art

[0002] With the advent of the AI (Artificial Intelligence) era, all walks of life have begun to embrace AI, enabling AI to empower business and helping users quickly process business;

[0003] Multi-turn conversation is considered one of the most difficult problems in the field of artificial intelligence, which involves language understanding, reasoning, and the application of common sense knowledge. Although with the development of the Internet, data information can be better shared, and multi-turn conversation has been better studied and developed, however, due to the characteristics of language itself, there are still problems with low accuracy. One of the main reasons is that when building a private large model, various private knowledge is constructed, but when users ask questions, due to the semantic interference between various private knowledge, the questions may not get the expected answers. Summary of the Invention

[0004] The present application provides a method, apparatus, device, and storage medium for optimizing large model question answering to improve the noise reduction function of the optimized large model.

[0005] According to one aspect of the present application, a method for optimizing large model question answering is provided. The method includes:

[0006] Performing element extraction on candidate program files of candidate web page codes obtained from a target business system to obtain at least one candidate web page element; the candidate web page element refers to a web page element containing a click trigger event;

[0007] Determining a target feature value tree according to the code path of the candidate web page code, the at least one candidate web page element, and the behavioral action characteristics of the at least one candidate web page element; the behavioral action characteristics are used to characterize the intention when the candidate web page element triggers a click trigger event;

[0008] Based on a general business large model, optimizing an initial feature value dictionary of the target business system according to the target feature value tree to obtain a target feature value dictionary of the target business system;

[0009] Optimizing the general business large model according to the target feature value dictionary and a private knowledge question answering library of the target business system to obtain a target business large model of the target business system.

[0010] According to another aspect of the present application, a device for optimizing large model question answering is provided. The device includes:

[0011] An element extraction module, configured to perform element extraction on candidate program files of candidate web page codes obtained from a target business system to obtain at least one candidate web page element; the candidate web page element refers to a web page element containing a click trigger event.

[0012] A feature value tree determination module, configured to determine a target feature value tree according to the code path of the candidate web page code, the at least one candidate web page element, and the behavioral action features of the at least one candidate web page element; the behavioral action features are used to characterize the intention when the candidate web page element triggers a click trigger event.

[0013] A dictionary optimization module, configured to optimize an initial feature value dictionary of the target business system based on a general business large model according to the target feature value tree to obtain a target feature value dictionary of the target business system.

[0014] A model optimization module, configured to optimize the general business large model according to the target feature value dictionary and a private knowledge Q&A library of the target business system to obtain a target business large model of the target business system.

[0015] According to another aspect of the present application, there is provided an electronic device, which includes:

[0016] One or more processors;

[0017] A memory for storing one or more programs;

[0018] When the one or more programs are executed by the one or more processors, the one or more processors implement any one of the large model Q&A optimization methods provided by the embodiments of the present application.

[0019] According to another aspect of the present application, there is provided a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements any one of the large model Q&A optimization methods provided by the embodiments of the present application.

[0020] According to another aspect of the present application, there is provided a computer program product, including a computer program, and when the computer program is executed by a processor, it implements any one of the large model Q&A optimization methods provided by the embodiments of the present application.

[0021] This application extracts elements from the candidate program files of the candidate web page code obtained from the target business system to obtain at least one candidate web page element; a candidate web page element refers to a web page element that contains a click trigger event; according to the code path of the candidate web page code, at least one candidate web page element, and the behavioral action characteristics of at least one candidate web page element, a target feature value tree is determined; the behavioral action characteristics are used to characterize the intention when the candidate web page element triggers a click trigger event; based on the general business large model, according to the target feature value tree, the initial feature value dictionary of the target business system is optimized to obtain the target feature value dictionary of the target business system; according to the target feature value dictionary and the private knowledge Q&A library of the target business system, the general business large model is optimized to obtain the target business large model of the target business system. The above technical solution helps to improve the noise reduction function of the optimized large model by constructing a target feature value tree, optimizing the feature value dictionary according to the target feature value tree, and optimizing the general large model in combination with the private knowledge Q&A library. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flowchart of a large model Q&A optimization method provided in Embodiment 1 of the present application;

[0023] Figure 2 is a flowchart of a large model Q&A optimization method provided in Embodiment 2 of the present application;

[0024] Figure 3 is a schematic structural diagram of a large model Q&A optimization device provided in Embodiment 3 of the present application;

[0025] Figure 4 is a schematic structural diagram of an electronic device for implementing the large model Q&A optimization method of Embodiment 4 of the present application. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the solution of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0027] It should be noted that in the description and claims of this application and the above-mentioned drawings, terms such as "first" and "second" are used to distinguish similar objects and do not necessarily describe a specific order or sequence. It should be understood that such data used can be interchanged under appropriate circumstances so that the embodiments of this application described herein can be implemented in an order different from those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0028] In addition, it should also be noted that in the technical solution of this application, the collection, storage, use, processing, transmission, provision, and disclosure of relevant data such as candidate program files and candidate web page elements comply with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0029] Embodiment 1

[0030] Figure 1 is a flowchart of a large model question-answering optimization method provided according to Embodiment 1 of this application. This embodiment is applicable to the situation of optimizing the noise reduction function of a business question-answering large model and can be executed by a large model question-answering optimization device. The large model question-answering optimization device can be implemented in the form of hardware and / or software and can be configured in a computer device, such as a server. As Figure 1 shown, the method includes:

[0031] S110. Extract elements from candidate program files of candidate web page codes obtained from a target business system to obtain at least one candidate web page element.

[0032] In this embodiment, the target business system refers to a customized system in a specific field or industry. In the application of artificial intelligence or machine learning, the target business system can improve the intelligence level of the system and more efficiently process business requirements and data analysis by introducing technologies such as large models, private knowledge bases, and eigenvalue dictionaries. Candidate web page codes refer to web page source codes obtained from the target business system that may contain information on web page elements to be extracted. Candidate program files refer to files containing candidate web page codes and are used to perform specific tasks on a computer. Candidate web page elements refer to web page elements identified in a web page that contain click trigger events, such as button, link, etc. elements; click trigger events refer to specific operations or behaviors triggered by a user clicking on a certain web page element (such as a button, link, etc.), usually page jump, form submission, etc.

[0033] Exemplarily, extract the HTML (HyperText Markup Language) elements containing click trigger events from the candidate web page code obtained from the target business system.

[0034] Optionally, before obtaining the candidate web page code from the target business system, the candidate web page code can be subjected to a consistency check with the inference program history container. If the consistency check fails, then obtain the candidate web page code.

[0035] In this embodiment, the inference program history container refers to a concept used to store and manage the history of the inference process; it can record the events, decisions, inference paths, and corresponding inputs and outputs that occur during the inference process.

[0036] It can be understood that if the consistency check passes, it proves that the candidate web page code has been processed according to the method of this application and no secondary processing is required.

[0037] S120. Determine the target feature value tree according to the code path of the candidate web page code, at least one candidate web page element, and the behavioral action characteristics of at least one candidate web page element.

[0038] In this embodiment, the code path refers to the path or position in the web page code used to identify a certain element. The behavioral action characteristic refers to the description of a specific action or behavior related to the web page element, especially the action or event that the system should execute when the user clicks on the web page element. This characteristic is used to characterize the intention when the candidate web page element triggers a click trigger event; this can include page jumps, data submissions, etc. The target feature value tree refers to a hierarchical structure constructed by extracting the behavioral action characteristics of the web page element and its code path, which describes the characteristics of the web page element and its relationship with other web page elements.

[0039] Optionally, use the code path of the candidate web page code as the root node of the target feature value tree, and use the element names of at least one candidate web page element as the secondary nodes of the target feature value tree to obtain a candidate feature value tree; determine the target feature value tree according to the candidate feature value tree and the behavioral action characteristics of at least one candidate web page element.

[0040] In this embodiment, the root node refers to the starting point or origin of the entire tree in the target feature value tree; the root node represents the initial state of the entire web page or the core element of the page structure. The secondary node refers to the direct child node under the root node, which represents a direct extension or more specific element of the root node. The candidate feature value tree refers to a hierarchical structure constructed with the root node and secondary nodes.

[0041] Further, for each secondary node of the candidate eigenvalue tree, if the behavioral action feature of the candidate web element corresponding to the secondary node is to open a new page, determine whether the new page contains a web element with a click trigger event; if so, use the element name of the web element on the new page as the subordinate child node of the secondary node, and continue to create child nodes according to the behavioral action feature of the web element corresponding to the subordinate child node until the behavioral action feature of the web element corresponding to the last node of the branch of the secondary node is to open a form or the new page does not contain a web element with a click trigger event.

[0042] Furthermore, for each secondary node of the candidate eigenvalue tree, if the behavioral action feature of the candidate web element corresponding to the secondary node is to submit a form, mark it as a submission operation and end the creation operation of the child nodes of the secondary node.

[0043] Exemplarily, extract the html element containing the click trigger event in the code. After scanning the element, add a root node to the eigenvalue tree. The node name of the root node is the path of the code, and create a secondary node under this node. The node name is the Chinese name of the element. At the same time, determine the intention of the element when triggering the click event, whether it is to submit a form or open a new page. If it is to open a new page, determine whether the new page contains an html element with a click trigger event. If it contains, repeat the data extraction and insert the created child nodes into the eigenvalue tree to complete the collection of metadata.

[0044] S130. Based on the general business large model, optimize the initial eigenvalue dictionary of the target business system according to the target eigenvalue tree to obtain the target eigenvalue dictionary of the target business system.

[0045] In this embodiment, the general business large model refers to a trained artificial intelligence model for question answering, which can be applied to different business scenarios and can process and analyze data, features, and knowledge in various business systems. The initial eigenvalue dictionary refers to the basic features and descriptions of different web elements in the target business system, which are usually preset at the beginning of the system and can be used for subsequent model optimization. The target eigenvalue dictionary refers to the dictionary obtained after optimization according to the requirements of the target business system. It contains feature descriptions that more conform to the actual requirements of the target business system and are used to improve the system performance and accuracy.

[0046] Optionally, using the target eigenvalue tree as the input, generate at least one action vocabulary through the general business large model; save the at least one action vocabulary to the initial eigenvalue dictionary to obtain the target eigenvalue dictionary of the target business system, so as to realize the optimization of the initial eigenvalue dictionary.

[0047] S140. Optimize the general business large model according to the target feature value dictionary and the private knowledge Q&A library of the target business system to obtain the target business large model of the target business system.

[0048] In this embodiment, the private knowledge Q&A library refers to a database specifically constructed for a specific business system, which contains the domain knowledge and Q&A data of the business system and can be used to provide specialized knowledge support for the model. The target business large model refers to the model obtained by optimizing the general business large model in combination with the private knowledge Q&A library and the feature value dictionary of the target business system, and is optimized and adjusted specifically for the requirements of the specific business system.

[0049] Optionally, when it is recognized that the private knowledge Q&A library of the target business system is increased, the Q&A knowledge of the private knowledge Q&A library is subjected to intention reasoning through the target feature value dictionary, and the target feature value dictionary and the reasoning result are saved to the reasoning program history container to optimize the general business large model and obtain the target business large model of the target business system.

[0050] Specifically, the intention reasoning for the Q&A knowledge of the private knowledge Q&A library can be to perform vector matching between the private knowledge Q&A library of the target business system and the target feature value dictionary to determine the question type of at least one question in the private knowledge Q&A library that matches the target feature value dictionary; for each question, according to the question type of the question and the corresponding answer type, determine the Q&A knowledge type corresponding to the question in the private knowledge Q&A library.

[0051] In this embodiment, the question type refers to the classification and nature of the questions in the Q&A library; for example, the question may involve querying a certain feature, performing calculations, obtaining suggestions, etc.; the question type is usually associated with the theme or goal of the Q&A, and determining the question type helps the system accurately understand the nature of the question. The answer type refers to the form or structure of the answer to the question; for example, the answer may be a specific numerical value, text description, recommended measure, etc.; the answer type is closely related to the question type and determines how to generate an appropriate answer according to the requirements of the question. The Q&A knowledge type is a way of classifying Q&A pairs according to the nature of the question and the structure of the answer; each question and its answer combination will be assigned to a specific knowledge type to ensure accurate classification and retrieval of Q&A pairs in the knowledge base.

[0052] Exemplarily, when adding to the private knowledge Q&A library, first perform vector matching on the questions of the knowledge to the eigenvalue dictionary to determine the feature type of the question matching the eigenvalue dictionary. If the feature type includes menu or behavior inference, determine again according to the answer of the knowledge. If the type of the knowledge answer is dynamic knowledge, set the type of this Q&A knowledge as behavior inference. If it is static knowledge, further compare the knowledge content with the eigenvalue dictionary to determine the eigenvalue type. If the eigenvalue type is still multiple, judge the vector score of the knowledge matching, and mark the one with the largest score as the eigenvalue type. If no match is found in the eigenvalue dictionary during the entire knowledge intention reasoning, the knowledge can be reasoned again to check if it includes actions on the menu such as opening or accessing. If a match can be found, it is defined as a menu. If no match can be found, it is set as such and that's all.

[0053] It can be understood that the focus of the optimization of the business large model in this application lies in the addition of the eigenvalue dictionary and the private knowledge Q&A library to solve the problem that various private knowledge may have mutual semantic interference, resulting in the question not being answered as expected, and improving the noise reduction ability of the large model.

[0054] In the embodiment of this application, element extraction is performed on the candidate program files of the candidate web page code obtained from the target business system to obtain at least one candidate web page element; a candidate web page element refers to a web page element that contains a click trigger event; according to the code path of the candidate web page code, at least one candidate web page element, and the behavior action features of at least one candidate web page element, a target eigenvalue tree is determined; the behavior action features are used to characterize the intention when the candidate web page element triggers the click trigger event; based on the general business large model, according to the target eigenvalue tree, the initial eigenvalue dictionary of the target business system is optimized to obtain the target eigenvalue dictionary of the target business system; according to the target eigenvalue dictionary and the private knowledge Q&A library of the target business system, the general business large model is optimized to obtain the target business large model of the target business system. The above technical solution, by constructing the target eigenvalue tree, optimizing the eigenvalue dictionary according to the target eigenvalue tree, and combining the private knowledge Q&A library to optimize the general large model, helps to improve the noise reduction function of the optimized large model.

[0055] Embodiment 2

[0056] Figure 2It is a flowchart of a large model question and answer optimization method provided by Embodiment 2 of the present application. Based on the technical solutions of the above embodiments, this embodiment refines "optimizing the initial eigenvalue dictionary of the target business system according to the target eigenvalue tree based on the general business large model to obtain the target eigenvalue dictionary of the target business system" into "traversing and converting the eigenvalue format from the bottom-level nodes of the target eigenvalue tree upwards to convert the target eigenvalue tree into a string feature list; saving the string feature list into the open menu feature of the initial eigenvalue dictionary to obtain the candidate eigenvalue dictionary of the target business system; optimizing the candidate eigenvalue dictionary of the target business system according to the target eigenvalue tree based on the general business large model to obtain the target eigenvalue dictionary of the target business system". It should be noted that for the parts not detailed in the embodiments of the present application, reference can be made to the relevant descriptions of other embodiments. As Figure 2 shown, the method includes:

[0057] S210. Extract elements from the candidate program files of the candidate web page code obtained from the target business system to obtain at least one candidate web page element.

[0058] S220. Determine the target eigenvalue tree according to the code path of the candidate web page code, at least one candidate web page element, and the behavioral action characteristics of at least one candidate web page element.

[0059] S230. Traverse and convert the eigenvalue format from the bottom-level nodes of the target eigenvalue tree upwards to convert the target eigenvalue tree into a string feature list.

[0060] In this embodiment, traversing from the bottom up means starting from the bottom-level nodes (leaf nodes) of the tree and traversing layer by layer upwards to the root node; each layer may contain combinations of features, which are finally merged into higher-level features or feature sets. The eigenvalue format conversion refers to converting the eigenvalues (which may be original numerical values, texts, etc.) extracted from the target eigenvalue tree into a standard format (such as a string) that meets the requirements of the business system; this can make the features easier to use and understand in different business systems. The string feature list refers to organizing all the converted eigenvalues (such as the eigenvalues obtained from the bottom-level nodes of the tree) into a list represented in string format; this list contains all the formatted features for subsequent operations or analyses.

[0061] S240. Save the string feature list into the open menu feature of the initial eigenvalue dictionary to obtain the candidate eigenvalue dictionary of the target business system.

[0062] In this embodiment, the open menu feature refers to an interface interaction method used to display and select available features; in a business system, features may be presented to users or the system in the form of a menu for selection; these features can help users or the system optimize their tasks or decisions.

[0063] Exemplarily, according to the feature value tree, the last-level nodes of the feature value tree are processed upwards to form a string feature list of the entire feature value tree, and the type of this list is defined as the open menu feature and saved to the feature value dictionary.

[0064] S250. Based on the general business large model, according to the target feature value tree, optimize the candidate feature value dictionary of the target business system to obtain the target feature value dictionary of the target business system.

[0065] Optionally, extract nouns from each node of the target feature value tree to obtain at least one node noun; for each node noun, input the node noun into the general business large model to obtain at least one action vocabulary corresponding to the node noun; save the at least one action vocabulary corresponding to the at least one node noun to the candidate feature value dictionary to obtain the target feature value dictionary of the target business system.

[0066] In this embodiment, a node refers to each point in a tree, which can be a single feature or a combination of multiple features. Noun extraction refers to extracting nouns (i.e., feature names) from each node of the target feature value tree; each node represents a specific feature, and extracting these feature names is for subsequent processing and analysis. A node noun refers to the feature name or related noun represented by each node. For example, "number of rooms", "area", "price", etc. Action vocabulary refers to the actions or operations related to the node noun analyzed and output by the general business large model; for example, if the node noun is "number of rooms", the general business large model may return related action vocabulary such as "calculate" or "evaluate", etc.

[0067] Exemplarily, extract the nouns of each node in the entire feature value tree and generate action words in the general business large model. The question of the general business large model is "Please generate possible action vocabulary according to this xxx noun", and save all the verbs returned by the large model to the feature value dictionary.

[0068] It can be understood that the optimization of the feature value dictionary in this application is to optimize the feature set of the general business large model, so as to achieve the optimization of the general business large model.

[0069] S260. According to the target feature value dictionary and the private knowledge Q&A library of the target business system, optimize the general business large model to obtain the target business large model of the target business system.

[0070] In an alternative embodiment, for the convenience of understanding the optimization of the business large model in this application, the use of the optimized model is taken as an example for illustration as follows:

[0071] When a user asks a question to the large model, the user's question and the request address of the current browser are used as parameters for access at the same time. After receiving the question, the large model service retrieves in the model knowledge. During the entire knowledge retrieval process, multiple knowledge contents may be retrieved. If one piece of knowledge is retrieved, it is directly returned to the user; if there are multiple pieces of knowledge, they are sorted according to the type and matching scores of the retrieved knowledge. If the score of the knowledge is the highest but it includes knowledge types of menus and behaviors, reasoning is performed again based on the request address of the current browser to determine whether the current request address is the home page or a business menu. If it is the home page, the knowledge of the menu type is preferentially used as the main classification. If it is not the home page, the knowledge of behaviors is preferentially used as the main classification knowledge for return; after the main classification return knowledge is determined, the retrieved questions need to be sorted for recommended knowledge again, and this sorting is just the opposite of the main classification knowledge sorting. If it is the home page, the behavior type is at the front. If it is a function, the menu classification is preferred to complete the encapsulation of the entire Q&A knowledge.

[0072] In another alternative embodiment, it is also possible to obtain the browser address type and question type during the actual use of the model user; match the browser address type with the question type. If they do not match, a personal Q&A behavior library of the model user is established in the target business large model, and the user data of the model user is saved in the personal Q&A behavior library; in the case of identifying that the model user uses it again, the personal Q&A behavior library is preferentially used for noise reduction. The personal Q&A behavior library is a database established specifically for each user to record all questions, browsing behaviors, and interaction data of the user during the use of the system.

[0073] In this embodiment, the browser address type refers to the category of different URL (Uniform Resource Locator) addresses or paths generated when a user uses an application or accesses a web page. The question type refers to the nature or category of a question when a user asks a question.

[0074] Exemplarily, after the Q&A knowledge is returned to the user, if the user uses the recommended knowledge and brings the browser address and the type of the question during the use process, when the large model receives the document again, it determines whether the type of the current browser address matches the type of the question. If it is found that the types do not match, the personal Q&A behavior library of the user is started to be established, and the user, address, and type are saved in the Q&A library. If the user asks questions again, the question noise reduction is preferentially performed in the behavior library. First, the browser address of the Q&A is matched. If a match can be found, the most matching Q&A type is used as the answer to the main knowledge.

[0075] It can be understood that by matching the browser address type with the question type and establishing and using the personal Q&A behavior library, more accurate personalized services can be provided for the model users, optimizing the user experience, reducing noise, and improving the response efficiency of the target business large model.

[0076] In the embodiment of the present application, element extraction is performed on the candidate program files of the candidate web page code obtained from the target business system to obtain at least one candidate web page element; a candidate web page element refers to a web page element that contains a click trigger event; according to the code path of the candidate web page code, at least one candidate web page element, and the behavioral action characteristics of at least one candidate web page element, a target feature value tree is determined; the behavioral action characteristics are used to characterize the intention when the candidate web page element triggers the click trigger event; traversing and feature value format conversion are performed from the bottom-level nodes of the target feature value tree upwards to convert the target feature value tree into a string feature list; the string feature list is saved into the open menu feature of the initial feature value dictionary to obtain the candidate feature value dictionary of the target business system; based on the general business large model, the candidate feature value dictionary of the target business system is optimized according to the target feature value tree to obtain the target feature value dictionary of the target business system; according to the target feature value dictionary and the private knowledge Q&A library of the target business system, the general business large model is optimized to obtain the target business large model of the target business system. The above technical solution helps to improve the noise reduction function of the optimized large model by constructing a target feature value tree, optimizing the feature value dictionary according to the target feature value tree, and optimizing the general large model in combination with the private knowledge Q&A library.

[0077] Embodiment III

[0078] Figure 3 FIG. is a structural schematic diagram of a large model Q&A optimization device provided according to Embodiment III of the present application, which is applicable to the situation of optimizing the noise reduction function of the business Q&A large model. The large model Q&A optimization device can be implemented in the form of hardware and / or software, and the large model Q&A optimization device can be configured in a computer device, such as a server. As Figure 3 shown, the device includes:

[0079] An element extraction module 310, configured to perform element extraction on the candidate program files of the candidate web page code obtained from the target business system to obtain at least one candidate web page element; a candidate web page element refers to a web page element that contains a click trigger event;

[0080] An eigenvalue tree determination module 320, configured to determine a target eigenvalue tree according to the code path of a candidate web page code, at least one candidate web page element, and the behavioral action features of at least one candidate web page element; the behavioral action features are used to characterize the intention when the candidate web page element triggers a click trigger event;

[0081] A dictionary optimization module 330, configured to optimize the initial eigenvalue dictionary of the target business system based on a general business large model according to the target eigenvalue tree, so as to obtain the target eigenvalue dictionary of the target business system;

[0082] A model optimization module 340, configured to optimize the general business large model according to the target eigenvalue dictionary and the private knowledge Q&A library of the target business system, so as to obtain the target business large model of the target business system.

[0083] In the embodiment of the present application, by performing element extraction on the candidate program files of the candidate web page code obtained from the target business system, at least one candidate web page element is obtained; the candidate web page element refers to a web page element that contains a click trigger event; according to the code path of the candidate web page code, at least one candidate web page element, and the behavioral action features of at least one candidate web page element, a target eigenvalue tree is determined; the behavioral action features are used to characterize the intention when the candidate web page element triggers a click trigger event; based on the general business large model, according to the target eigenvalue tree, the initial eigenvalue dictionary of the target business system is optimized to obtain the target eigenvalue dictionary of the target business system; according to the target eigenvalue dictionary and the private knowledge Q&A library of the target business system, the general business large model is optimized to obtain the target business large model of the target business system. The above technical solution helps to improve the noise reduction function of the optimized large model by constructing a target eigenvalue tree, optimizing the eigenvalue dictionary according to the target eigenvalue tree, and optimizing the general large model in combination with the private knowledge Q&A library.

[0084] Optionally, the dictionary optimization module 330 includes:

[0085] A format conversion unit, configured to traverse from the leaf nodes of the target eigenvalue tree from bottom to top and perform eigenvalue format conversion, so as to convert the target eigenvalue tree into a string feature list;

[0086] A list saving unit, configured to save the string feature list to the open menu feature of the initial eigenvalue dictionary to obtain a candidate eigenvalue dictionary of the target business system;

[0087] A dictionary optimization unit, configured to optimize the candidate eigenvalue dictionary of the target business system based on the general business large model according to the target eigenvalue tree, so as to obtain the target eigenvalue dictionary of the target business system.

[0088] Optionally, the dictionary optimization unit is specifically configured to:

[0089] Extract nouns from each node of the target feature value tree to obtain at least one node noun;

[0090] For each node noun, input the node noun into the general business large model to obtain at least one action vocabulary corresponding to the node noun;

[0091] Save at least one action vocabulary corresponding to at least one node noun to the candidate feature value dictionary to obtain the target feature value dictionary of the target business system.

[0092] Optionally, the feature value tree determination module 320 includes:

[0093] A node creation unit, configured to use the code path of the candidate web page code as the root node of the target feature value tree, and use the element names of at least one candidate web page element as the secondary nodes of the target feature value tree to obtain a candidate feature value tree;

[0094] A feature value tree construction unit, configured to determine the target feature value tree according to the candidate feature value tree and the behavioral action features of at least one candidate web page element.

[0095] Optionally, the feature value tree construction unit is specifically configured to:

[0096] For each secondary node of the candidate feature value tree, if the behavioral action feature of the candidate web page element corresponding to the secondary node is to open a new page, determine whether the new page contains a web page element with a click trigger event;

[0097] If so, use the element name of the web page element of the new page as the subordinate child node of the secondary node, and continue to create child nodes according to the behavioral action features of the web page element corresponding to the subordinate child node until the behavioral action feature of the web page element corresponding to the last node of the branch of the secondary node is to open a form or the new page does not contain a web page element with a click trigger event.

[0098] Optionally, the feature value tree construction unit is further specifically configured to:

[0099] For each secondary node of the candidate feature value tree, if the behavioral action feature of the candidate web page element corresponding to the secondary node is to submit a form, mark it as a submission operation and end the sub-node creation operation of the secondary node.

[0100] The large model question answering optimization device provided by the embodiments of the present application can execute the large model question answering optimization method provided by any embodiment of the present application, and has the corresponding functional modules and beneficial effects for executing various large model question answering optimization methods.

[0101] According to the embodiments of the present application, the present application also provides an electronic device, a readable storage medium, and a computer program product.

[0102] Example 4

[0103] Figure 4 It is a schematic structural diagram of an electronic device 410 that implements the large model question and answer optimization method of the embodiments of the present application. The electronic device is intended to represent various forms of digital computers, such as, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, personal digital processors, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are only examples and are not intended to limit the implementation of the present application described herein and / or claimed.

[0104] As Figure 4 shown, the electronic device 410 includes at least one processor 411, and a memory communicatively connected to the at least one processor 411, such as a read-only memory (ROM) 412, a random access memory (RAM) 413, etc. Among them, the memory stores a computer program executable by the at least one processor. The processor 411 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 412 or the computer program loaded from the storage unit 418 into the random access memory (RAM) 413. In the RAM 413, various programs and data required for the operation of the electronic device 410 can also be stored. The processor 411, the ROM 412, and the RAM 413 are connected to each other through a bus 414. The input / output (I / O) interface 415 is also connected to the bus 414.

[0105] Multiple components in the electronic device 410 are connected to the I / O interface 415, including: an input unit 416, such as a keyboard, a mouse, etc.; an output unit 417, such as various types of displays, speakers, etc.; a storage unit 418, such as a magnetic disk, an optical disk, etc.; and a communication unit 419, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 419 allows the electronic device 410 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0106] The processor 411 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 411 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The processor 411 executes the various methods and processes described above, such as the large model question answering optimization method.

[0107] In some embodiments, the large model question answering optimization method can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as the storage unit 418. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 410 via the ROM 412 and / or the communication unit 419. When the computer program is loaded into the RAM 413 and executed by the processor 411, one or more steps of the large model question answering optimization method described above can be executed. Alternatively, in other embodiments, the processor 411 can be configured for the large model question answering optimization method by any other suitable means (e.g., by means of firmware).

[0108] The various embodiments of the systems and technologies described above in this document can be implemented in digital electronic circuit systems, integrated circuit systems, application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-chip systems (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs, which can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special or general programmable processor, and can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0109] The computer program for implementing the method of the present application can be written in any combination of one or more programming languages. These computer programs can be provided to the processor of a general-purpose computer, a special-purpose computer, or other programmable large model question answering optimization devices, such that when the computer program is executed by the processor, the functions / operations specified in the flowchart and / or block diagram are implemented. The computer program can be executed entirely on the machine, partially on the machine, as an independent software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0110] In the context of the present application, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, the computer-readable storage medium can be a machine-readable signal medium. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0111] To provide for interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the electronic device. Other kinds of devices can also be used to provide for interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic, speech, or tactile input).

[0112] The systems and techniques described herein can be implemented in a computing system that includes backend components (such as, for example, a data server), or a computing system that includes middleware components (such as, for example, an application server), or a computing system that includes frontend components (such as, for example, a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (such as, for example, a communication network). Examples of a communication network include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0113] A computing system may include a client and a server. The client and the server are generally far from each other and usually interact via a communication network. The relationship between the client and the server is created by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system, and solves the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services.

[0114] It should be understood that various forms of the processes shown above can be used, steps can be reordered, added or deleted. For example, the steps recited in this application can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of this application can be achieved, and no limitation is made herein.

[0115] The above specific embodiments do not constitute a limitation on the protection scope of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application shall be included within the protection scope of this application.

Claims

1. A method for optimizing large model question answering, characterized in that, Including: Performing element extraction on candidate program files of candidate web page codes obtained from a target business system to obtain at least one candidate web page element; The candidate web page element refers to a web page element containing a click trigger event; Determining a target feature value tree according to the code path of the candidate web page code, the at least one candidate web page element, and the behavioral action characteristics of the at least one candidate web page element; The behavioral action characteristics are used to characterize the intention when the candidate web page element triggers a click trigger event; Based on a general business large model, optimizing the initial feature value dictionary of the target business system according to the target feature value tree to obtain the target feature value dictionary of the target business system; Optimizing the general business large model according to the target feature value dictionary and the private knowledge Q&A library of the target business system to obtain the target business large model of the target business system; Among them, the optimizing the initial feature value dictionary of the target business system according to the target feature value tree based on the general business large model to obtain the target feature value dictionary of the target business system includes: Traversing and performing feature value format conversion from the leaf nodes of the target feature value tree from bottom to top to convert the target feature value tree into a string feature list; Saving the string feature list to the open menu feature of the initial feature value dictionary to obtain the candidate feature value dictionary of the target business system; Optimizing the candidate feature value dictionary of the target business system according to the target feature value tree based on the general business large model to obtain the target feature value dictionary of the target business system; Among them, the optimizing the candidate feature value dictionary of the target business system according to the target feature value tree based on the general business large model to obtain the target feature value dictionary of the target business system includes: Performing noun extraction on each node of the target feature value tree to obtain at least one node noun; For each node noun, inputting the node noun into the general business large model to obtain at least one action vocabulary corresponding to the node noun; Saving the at least one action vocabulary corresponding to the at least one node noun to the candidate feature value dictionary to obtain the target feature value dictionary of the target business system.

2. The method according to claim 1, wherein Determining the target feature value tree according to the code path of the candidate web page code, the at least one candidate web page element, and the behavioral action characteristics of the at least one candidate web page element includes: Taking the code path of the candidate web page code as the root node of the target feature value tree and taking the element names of the at least one candidate web page element as the secondary nodes of the target feature value tree to obtain a candidate feature value tree; Determining the target feature value tree according to the candidate feature value tree and the behavioral action characteristics of the at least one candidate web page element.

3. The method according to claim 2, characterized in that, Determining the target feature value tree according to the candidate feature value tree and the behavioral action characteristics of the at least one candidate web page element includes: For each second-level node of the candidate eigenvalue tree, if the behavioral action feature of the candidate web page element corresponding to the second-level node is to open a new page, determine whether the new page contains a web page element with a click trigger event; If so, use the element name of the web page element of the new page as the lower-level child node of the second-level node, and continue to create child nodes according to the behavioral action feature of the web page element corresponding to the lower-level child node until the behavioral action feature of the web page element corresponding to the last node of the branch of the second-level node is to open a form or the new page does not contain a web page element with a click trigger event.

4. The method according to claim 3, characterized in that, The method further includes: For each second-level node of the candidate eigenvalue tree, if the behavioral action feature of the candidate web page element corresponding to the second-level node is to submit a form, mark it as a submission operation and end the creation operation of the child nodes of the second-level node.

5. An optimization device for large model question answering, characterized in that, It includes: An element extraction module, configured to perform element extraction on the candidate program files of the candidate web page code obtained from the target business system to obtain at least one candidate web page element; The candidate web page element refers to a web page element that contains a click trigger event; An eigenvalue tree determination module, configured to determine a target eigenvalue tree according to the code path of the candidate web page code, the at least one candidate web page element, and the behavioral action feature of the at least one candidate web page element; The behavioral action feature is used to characterize the intention when the candidate web page element triggers a click trigger event; A dictionary optimization module, configured to optimize the initial eigenvalue dictionary of the target business system based on a general business large model according to the target eigenvalue tree to obtain the target eigenvalue dictionary of the target business system; A model optimization module, configured to optimize the general business large model according to the target eigenvalue dictionary and the private knowledge Q&A library of the target business system to obtain the target business large model of the target business system; Optionally, the dictionary optimization module includes: A format conversion unit, configured to traverse from the last-level node of the target eigenvalue tree from bottom to top and perform eigenvalue format conversion to convert the target eigenvalue tree into a string feature list; A list storage unit, configured to store the string feature list in the open menu feature of the initial eigenvalue dictionary to obtain a candidate eigenvalue dictionary of the target business system; A dictionary optimization unit, configured to optimize the candidate eigenvalue dictionary of the target business system based on a general business large model according to the target eigenvalue tree to obtain the target eigenvalue dictionary of the target business system; Optionally, the dictionary optimization unit is specifically configured to: Extract nouns from each node of the target eigenvalue tree to obtain at least one node noun; For each node noun, input the node noun into the general business large model to obtain at least one action vocabulary corresponding to the node noun; Save the at least one action vocabulary corresponding to the at least one node noun to the candidate eigenvalue dictionary to obtain the target eigenvalue dictionary of the target business system.

6. An electronic device, characterized in that, It includes: One or more processors; A memory, configured to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the large model question and answer optimization method according to any one of claims 1-4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the large model question and answer optimization method according to any one of claims 1-4.

8. A computer program product, comprising a computer program, characterized in that, The computer program implements the large model question and answer optimization method according to any one of claims 1-4 when executed by the processor.

Images