Metafunction component calling method and device, computer equipment and readable storage medium

Abstract

The embodiment of the invention discloses a meta-function component calling method and device, computer equipment and a readable storage medium. The method comprises the following steps: acquiring a description text of a current problem, and respectively performing feature extraction and business scene recognition based on the description text to obtain a text feature vector and a business scene label; screening out a target business rule from a preset business rule base based on the business scene label; performing semantic recognition on the text feature vector to obtain an intention recognition result; and based on the intention recognition result and the target business rule, determining a meta-function component list required to be called by the current problem through the large model, and calling corresponding meta-function components in sequence according to the meta-function component list. According to the method, in combination with the business rule, the meta-function component needing to be called by the user intention can be recognized more accurately, the influence caused by large model illusion is reduced, and the calling accuracy of the meta-function component is improved.

Description

TECHNICAL FIELD

[0001] The embodiment of the application relates to the technical field of business processing, in particular to a meta-function component calling method and device, computer equipment and a readable storage medium. BACKGROUND

[0002] In a traditional large model application, the large model identifies user intent through an internal mechanism and calls meta-function components in a full connection mode. However, this mode has obvious defects. The large model may have hallucinations, that is, generate outputs that do not match the actual situation, which leads to incorrect and inaccurate calling of meta-function components, and further affects the efficiency and accuracy of business processing. For example, in an intelligent customer service scenario, the large model may misinterpret the user's question, and thus call irrelevant meta-function components, failing to provide effective solutions for the user. SUMMARY

[0003] The embodiment of the application provides a meta-function component calling method and device, computer equipment and a readable storage medium, aiming to solve the technical problem of incorrect calling of meta-function components by a large model due to hallucinations.

[0004] In a first aspect, the embodiment of the application provides a meta-function component calling method, which includes:

[0005] obtaining a description text of a current question, and respectively performing feature extraction and business scenario identification based on the description text to obtain a text feature vector and a business scenario label;

[0006] filtering out a target business rule from a preset business rule library based on the business scenario label;

[0007] performing semantic identification on the text feature vector to obtain an intent recognition result;

[0008] determining a list of meta-function components to be called for the current question by a large model based on the intent recognition result and the target business rule, and sequentially calling corresponding meta-function components according to the list of meta-function components.

[0009] In a second aspect, the embodiment of the application provides a meta-function component calling device, which includes:

[0010] an obtaining module configured to obtain a description text of a current question, and respectively perform feature extraction and business scenario identification based on the description text to obtain a text feature vector and a business scenario label;

[0011] a filtering module configured to filter out a target business rule from a preset business rule library based on the business scenario label;

[0012] an identification module configured to perform semantic identification on the text feature vector to obtain an intent recognition result;

[0013] The calling module is used to determine the list of meta-functional components to be called for the current problem based on the intent recognition result and the target business rules through the big model, and to call the corresponding meta-functional components in sequence according to the list of meta-functional components.

[0014] Thirdly, embodiments of the present invention provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the meta-functional component invocation method described in the first aspect above.

[0015] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to perform the meta-functional component invocation method described in the first aspect.

[0016] This invention provides a method, apparatus, computer device, and readable storage medium for invoking meta-functional components. The method acquires a description text of the current problem, and performs feature extraction and business scenario identification based on the description text to obtain a text feature vector and a business scenario label. Based on the business scenario label, it filters target business rules from a pre-set business rule library. It performs semantic recognition on the text feature vector to obtain an intent recognition result. Based on the intent recognition result and the target business rules, it determines a list of meta-functional components required to be invoked for the current problem using a large model, and sequentially invokes the corresponding meta-functional components according to the list. This method, combined with business rules, can more accurately identify the meta-functional components required by the user's intent, reduce the impact of the large model illusion, and improve the accuracy of meta-functional component invocation. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating an embodiment of the method for invoking meta-functional components provided in this invention.

[0019] Figure 2 for Figure 1 A schematic diagram of a sub-process of step S110 in one embodiment;

[0020] Figure 3 This is a schematic block diagram of a meta-functional component invocation device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0022] It should be understood that, when used in the specification and the appended claims, the terms "comprise" and "include" indicate the presence of described features, integers, steps, operations, elements, and / or components, but do not exclude one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0023] It should also be understood that the terms used in the present application specification are only for the purpose of describing particular embodiments and are not intended to limit the present application. As used in the present application specification and the appended claims, the singular forms "a", "an" and "the" are intended to include the plural forms, unless the context clearly indicates otherwise.

[0024] It should be further understood that the term "and / or" used in the present application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

[0025] Please refer to Figure 1 A flowchart of a method for invoking a meta-function component is provided for the embodiments of the present application, and the method comprises steps S110-S140.

[0026] In step S110, a description text of a current problem is obtained, and feature extraction and business scenario recognition are respectively performed based on the description text to obtain a text feature vector and a business scenario label.

[0027] In the present embodiment, the description text of the current problem is obtained through an interactive platform, and the description text at least contains a natural language description of a requirement and context information of the current problem, such as historical dialogue records, usage habits, etc. The description text is cleaned to remove irrelevant characters, stop words and other noise information. Among them, regular expressions are used to remove punctuation marks and special characters, for example, the re.sub(r'[^\w\s]',”,text) statement (Python example) is used to remove punctuation marks in the text. A predefined stop word list is used to filter common meaningless words, such as using the stop word list in the NLTK library for filtering.

[0028] Further, the semantic recognition function of the large model is used to perform semantic recognition on the description text, to identify the business scenario of the current problem and obtain a business scenario label. The business scenario label can be determined according to an actual application workflow scenario, and can be an order processing workflow, a customer service work order processing workflow, or a purchase application approval workflow.

[0029] As shown in Figure 2 In an embodiment, step S110 includes:

[0030] Step S210, performing word segmentation on the description text to obtain a word segmentation set;

[0031] Step S220, calculating the frequency of each word segmentation in the word segmentation set in the description text and performing normalization processing to obtain a first weight coefficient of each word segmentation relative to the description text;

[0032] Step S230, calculating the inverse corpus distribution coefficient of each word segmentation in the word segmentation set relative to a preset corpus to obtain a second weight coefficient of each word segmentation relative to the corpus;

[0033] Step S240, multiplying the first weight coefficient and the second weight coefficient to obtain a semantic weight of each word segmentation in the word segmentation set;

[0034] Step S250, sorting the word segmentations in the word segmentation set according to the semantic weights, taking the top several as keywords, and vectorizing the keywords to obtain the text feature vector.

[0035] In this embodiment, vectorization conversion is achieved by calculating the importance of each word segmentation in the description text. By combining the appearance frequency of the word segmentation and the global distribution characteristics of the corpus, a differentiated weight evaluation system is constructed, which specifically includes:

[0036] First weight coefficient calculation: the appearance frequency f(t, d) of the word segmentation t in the description document d is counted, and the first weight coefficient LW(t, d) is obtained through normalization processing:

[0037]

[0038] Where max(f(d)) represents the frequency of the most frequently appearing word segmentation in the description text d, which reflects the relative importance of the word segmentation t in the description text;

[0039] Second weight coefficient calculation: for the distribution characteristics of the word segmentation t in the preset corpus D, the word segmentation inhibition coefficient is defined as the second weight coefficient IGD(t, D):

[0040]

[0041] wherein N represents the total number of documents in the preset corpus, and the corpus stores description documents of different business scenarios, which describe the processing flow, task, function and attribute required for the business scenario, n t represents the number of documents containing the word t. The word suppression coefficient suppresses the weight of high-frequency words through logarithmic transformation, and highlights low-frequency but certain distinguishing words.

[0042] Semantic weight calculation:

[0043] The first weight coefficient and the second weight coefficient are multiplied to obtain the semantic weight SW(t, d, D) of the word:

[0044] SW(t, d, D) = LW(t, d) x IGD(t, d, D),

[0045] wherein the higher the semantic weight, the stronger the semantic contribution of the word t in the document d relative to the corpus D. Through the above method, the text can be converted into a numerical feature vector with the semantic weight of the word as the component, effectively capturing the core semantic features of the text. Finally, the words in the word set are sorted according to the semantic weight, and the top several words are selected as the keywords, and the keywords are vectorized to obtain the text feature vector describing the text.

[0046] Step S120, screening the target business rule from the preset business rule library based on the business scenario label;

[0047] In this embodiment, the specific business scenario label is defined according to the actual application scenario. The business rules are specified according to the rules that may be involved in the application scenario. For example, when the business scenario is an order processing workflow, the state nodes include order receiving, order verification, inventory query, and delivery processing in turn; wherein the business trigger rule of order receiving can be defined as "the system receives an external order API request", and the associated meta-function component is "order parsing engine"; the business trigger rule of order acceptance can be defined as "① order integrity check passes ② amount ≥ n yuan", and the associated meta-function component is "order compliance check tool"; the business trigger rule of inventory query can be defined as "the commodity xxxx code is valid and the inventory > 0", and the associated meta-function component is "warehouse management system"; the business trigger rule of delivery processing can be defined as "the logistics order number is generated successfully", and the associated meta-function component is "logistics interface platform". When the business scenario is a customer service work order processing workflow, the state nodes include work order creation, work order distribution, work order processing, and work order confirmation in turn; wherein the business trigger rule of work order creation can be defined as "the user submits a work order through the customer service system", and the associated meta-function component is "work order entry module"; the business trigger rule of work order distribution can be defined as "① match processing group according to business type ② priority ≥ n automatically upgrade", and the associated meta-function component is "intelligent order allocation engine"; the business trigger rule of work order processing can be defined as "the handler submits a solution", and the associated meta-function component is "knowledge base retrieval tool"; the business trigger rule of work order confirmation can be defined as "the customer feedback confirms the solution", and the associated meta-function component is "satisfaction survey component".

[0048] It should be noted that the user can consider the actual business scenario and define the business rules according to the possible business. In addition, new business rules can be generated based on historical business rules through machine learning algorithms, and all business rules can be stored to obtain a business rule library.

[0049] Further, for rule learning based on machine learning, a decision tree algorithm can be used. Specifically, it includes: obtaining historical business data; constructing a decision tree algorithm, training the decision tree algorithm based on the historical business data, and adjusting using a preset weight to obtain a business decision tree; generating rules based on the business decision tree to obtain at least one business rule; and collecting all business rules generated by the decision tree to obtain a business rule library. Wherein, the decision tree model outputs different decisions according to different feature values. For example, the rule can be "if the user order amount exceeds a certain threshold, then the senior logistics meta-function component is called first", and the decision tree can learn and generate rules according to order amount, user historical consumption records, etc.

[0050] Further, in order to perform business rule screening, the business scenario label is vectorized to obtain a scenario feature vector; a text vector corresponding to each business rule in the business rule library is obtained, the similarity between the scenario feature vector and each text vector is calculated according to a preset similarity algorithm, and the business rule corresponding to the text vector with the highest similarity is taken as a target business rule. The calculation formula is as follows:

[0051] cos (θ) = ||A|| ||B|| A·B,

[0052] wherein A and B represent the scenario feature vector and the text vector of the business rule respectively, A·B represents the dot product of the vectors, and ||A|| and ||B|| represent the modulus of the vectors.

[0053] Step S130, performing semantic recognition on the text feature vector to obtain an intent recognition result;

[0054] Step S140, determining a list of meta-function component to be called for the current problem based on the intent recognition result and the target business rule, and sequentially calling corresponding meta-function components according to the list of meta-function components.

[0055] In the embodiment, the existing large model is fine-tuned by using a preset corpus, the text feature vector is subjected to semantic recognition based on the fine-tuned large model to obtain an intent recognition result of the current problem, the list of meta-function component to be called for the current problem is determined based on the intent recognition result and the target business rule by the large model (the list of meta-function component contains at least one meta-function component), a corresponding communication channel is established according to the meta-function component, and the corresponding meta-function component is sequentially called according to the list of meta-function component. For example, the intent recognition result of the current problem is determined as “business trip travel”, the target business rule matched is “1. query weather; 2. order travel tool ticket; 3. submit leave approval”, the large model determines that the meta-function component for querying the weather is “weather forecast tool”, the meta-function component for ordering the travel tool ticket is “XX ticket purchase tool”, and the meta-function component for submitting the leave approval is “XX company office affairs tool” according to the knowledge learned by training.

[0056] Further, in order to facilitate subsequent analysis and optimization, after sequentially calling the corresponding meta-function component according to the list of meta-function component, the execution result returned by the called tool is received and recorded in the log.

[0057] The method obtains a description text of a current question, and respectively performs feature extraction and business scenario recognition based on the description text to obtain a text feature vector and a business scenario label; a target business rule is screened out from a preset business rule library based on the business scenario label; semantic recognition is performed on the text feature vector to obtain an intention recognition result; based on the intention recognition result and the target business rule, a list of meta-function component to be called for the current question is determined through a large model, and corresponding meta-function components are sequentially called according to the list of meta-function components. The method can more accurately recognize the meta-function component to be called by the user intention in combination with the business rule, reduce the influence of large model hallucination, and improve the accuracy of the meta-function component calling.

[0058] The embodiment of the application further provides a meta-function component calling device, which is used for executing any one of the foregoing embodiments of the meta-function component calling method. Specifically, refer to Figure 3 , Figure 3 is a schematic block diagram of the meta-function component calling device provided by the embodiment of the application. The meta-function component calling device 100 can be configured in a server.

[0059] As shown in Figure 3 , the meta-function component calling device 100 comprises an acquisition module 110, a screening module 120, an identification module 130 and a calling module 140.

[0060] The acquisition module 110 is used for acquiring a description text of a current question, and respectively performing feature extraction and business scenario recognition based on the description text to obtain a text feature vector and a business scenario label;

[0061] The screening module 120 is used for screening a target business rule from a preset business rule library based on the business scenario label;

[0062] The identification module 130 is used for performing semantic recognition on the text feature vector to obtain an intention recognition result;

[0063] The calling module 140 is used for determining a list of meta-function component to be called for the current question through a large model based on the intention recognition result and the target business rule, and sequentially calling corresponding meta-function components according to the list of meta-function components.

[0064] In an embodiment, the acquisition module 110 comprises:

[0065] A word segmentation unit is used for performing word segmentation on the description text to obtain a word segmentation set;

[0066] A first calculation unit is used for calculating the frequency of each word segmentation in the word segmentation set in the description text and performing normalization processing to obtain a first weight coefficient of each word segmentation relative to the description text;

[0067] a second computing unit configured to calculate a distribution coefficient of each word in the word set relative to a preset corpus, to obtain a second weight coefficient of each word relative to the corpus;

[0068] a third computing unit configured to multiply the first weight coefficient and the second weight coefficient, to obtain a semantic weight of each word in the word set;

[0069] a sorting unit configured to sort the words in the word set according to the semantic weight, to take a plurality of words at the top of the sorting as keywords, and to vectorize the keywords to obtain the text feature vector.

[0070] In an embodiment, the screening module 120 comprises:

[0071] a vectorization processing unit configured to vectorize the business scenario label to obtain a scenario feature vector;

[0072] a similarity calculation unit configured to obtain a text vector corresponding to each business rule in the business rule library, to calculate a similarity between the scenario feature vector and each text vector, and to take a business rule corresponding to a text vector with the highest similarity as the target business rule.

[0073] In an embodiment, the meta-function component calling apparatus 100 further comprises:

[0074] a recording module configured to receive an execution result returned by a called tool, and to record the execution result in a log.

[0075] In an embodiment, the meta-function component calling apparatus 100 further comprises:

[0076] a denoising module configured to perform data cleaning on the description text to remove noise information.

[0077] The embodiment of the present application further provides a computer device comprising a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein the processor implements the meta-function component calling method as described above when executing the computer program.

[0078] In another embodiment of the present application, a computer readable storage medium is provided. The computer readable storage medium can be a non-volatile computer readable storage medium. The computer readable storage medium stores a computer program, which, when executed by a processor, causes the processor to execute the meta-function component calling method as described above.

[0079] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the devices, apparatuses and units described above can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here. Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, each example has been described in the foregoing description in terms of its general functionality, without regard to its specific combination of hardware and software. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0080] In several embodiments provided by the present application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely schematic, for example, the division of the units is merely logical functional division, and actual implementation can have another division, or units with the same function can be combined into one unit, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, or can be electrical, mechanical or other form of connection.

[0081] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, i.e. they can be located in one place or distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiments of the present application.

[0082] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be realized in the form of hardware or software functional unit.

[0083] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a storage medium. Based on such understanding, the technical solutions of the present application essentially or the part of the prior art that contributes to the present application, or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in various embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0084] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present application, and these modifications or replacements should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method of invoking a metafunction component, the method comprising: The method comprises the following steps: obtaining a description text of a current problem, and performing feature extraction and business scenario recognition based on the description text respectively to obtain a text feature vector and a business scenario label; filtering a target business rule from a preset business rule library based on the business scenario label; performing semantic recognition on the text feature vector to obtain an intention recognition result; determining a list of meta-function components to be called for the current problem by a large model based on the intention recognition result and the target business rule, and sequentially calling corresponding meta-function components according to the list of meta-function components.

2. The meta-function component invocation method of claim 1, wherein, The feature extraction and business scenario recognition based on the description text comprise the following steps: performing word segmentation on the description text to obtain a word segmentation set; normalizing the frequency of each word in the word segmentation set in the description text to obtain a first weight coefficient of each word relative to the description text; calculating a word suppression coefficient of each word in the word segmentation set relative to a preset corpus to obtain a second weight coefficient of each word relative to the corpus, and the word suppression coefficient is obtained by logarithmic transformation to suppress the weight of high-frequency words; multiplying the first weight coefficient and the second weight coefficient to obtain a semantic weight of each word in the word segmentation set; sorting the words in the word segmentation set according to the semantic weight, taking the top several as keywords, and vectorizing the keywords to obtain the text feature vector.

3. The meta-function component invocation method of claim 2, wherein, The semantic weight is calculated according to the following formula: SW(t,d,D)=LW(t,d)×IGD(t,d,D), wherein, LW(t,d) represents the first weight coefficient, IGD(t,D) represents the second weight coefficient, d represents the description document, t represents the word segmentation, f(t,d) represents the statistical frequency of the word segmentation t appearing in the description document d, max(f(d)) represents the frequency of the most frequently appearing word segmentation in the description text d, N represents the total number of documents in the preset corpus, n t represents the number of documents containing the word segmentation t, and SW(t,d,D) represents the semantic weight.

4. The meta-function component invocation method of claim 1, wherein, The construction of the business rule library comprises the following steps: obtaining historical business data; constructing a decision tree algorithm, training the decision tree algorithm based on the historical business data, and adjusting the preset weight to obtain a business decision tree; generating rules based on the business decision tree to obtain at least one business rule; collecting all business rules generated by the decision tree to obtain the business rule library.

5. The meta-function component invocation method of claim 1, wherein, The filtering of the target business rule from the preset business rule library based on the business scenario label comprises the following steps: vectorizing the business scenario label to obtain a scenario feature vector; obtaining a text vector corresponding to each business rule in the business rule library, calculating the similarity between the scenario feature vector and each text vector, and taking the business rule corresponding to the text vector with the highest similarity as the target business rule.

6. The meta-function component invocation method of claim 1, wherein, After sequentially calling the corresponding meta-function components according to the list of meta-function components, the method comprises the following steps: receiving an execution result returned by the called tool and recording the execution result in a log.

7. The metacomponent invocation method of claim 1, wherein, After obtaining the description text of the current problem, the method comprises the following steps: performing data cleaning on the description text to remove noise information.

8. A metafunction component invocation method, comprising: The method comprises the following steps: an obtaining module, configured to obtain a description text of a current problem, and perform feature extraction and business scenario recognition based on the description text to obtain a text feature vector and a business scenario label; a filtering module, configured to filter a target business rule from a preset business rule library based on the business scenario label; an identification module, configured to perform semantic recognition on the text feature vector to obtain an intention recognition result; The calling module is configured to determine a list of meta-function component to be called for the current problem based on the intention recognition result and target business rules, and sequentially call corresponding meta-function components according to the list of meta-function components. 9.A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the meta-function component calling method according to any one of claims 1 to 5 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, which, when executed by a processor, causes the processor to perform the meta-function component calling method according to any one of claims 1 to 5.

Images