Operation processing method and device

Abstract

The embodiment of the invention provides an operation processing method and device. The operation processing method comprises the steps that a to-be-processed operation requirement is obtained; based on the operation demand, retrieval is carried out in an operation label database to obtain at least one corresponding target operation label, the operation label database comprises a plurality of operation labels, and the plurality of operation labels are generated through a label model based on corresponding operation information; based on the target operation labels and the corresponding target operations, a step generation model is called to generate a corresponding operation execution sequence, operation steps in the operation execution sequence are executed, and the operation execution sequence comprises the target operation labels and the operation steps generated by the corresponding target operations according to the execution sequence. In the mode, large model automatic operation is used, the language instruction of the user is converted into accurate editor parameter adjustment, the learning cost of an editor with mass data is reduced, and the operation efficiency is improved.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of computer technology, and more particularly to an operation processing method and apparatus, a computing device, and a computer-readable storage medium. Background Art

[0002] As digitalization accelerates in fields like gaming and industrial design, the functional complexity of professional editors is growing exponentially. After years of iterative evolution, these features have become incredibly complex, leaving them scattered throughout the editor, often leaving them largely unused and unnoticed. For example, an editor used to support large-scale game production has over 530 parameters for adjusting ambient lighting alone, with various functional modules scattered across multiple layers of menus.

[0003] In the existing technology, user operation editors rely on two solutions: manual query and command mapping. Manual query requires users to accurately enter operation terms for search and editing, while command mapping requires binding shortcut keys to commonly used functions and only supports a limited number of pre-defined operations.

[0004] However, these solutions have inherent flaws. For example, users need to know the professional name of the target function in advance, and the editor cannot understand the user's natural language instructions, which brings extremely high learning costs to users. Therefore, a faster, more efficient and reliable solution is still needed to reduce the user's operational learning cost. Summary of the Invention

[0005] In view of this, embodiments of this specification provide an operation processing method. One or more embodiments of this specification also relate to an operation processing apparatus, a computing device, a computer-readable storage medium, and a computer program product to address technical deficiencies in the prior art.

[0006] According to a first aspect of the embodiments of this specification, an operation processing method is provided, including: Get pending operation requirements; Based on the operation requirement, searching an operation annotation database to obtain at least one corresponding target operation annotation, wherein the operation annotation database includes a plurality of operation annotations, and the plurality of operation annotations are generated by an annotation model based on corresponding operation information; Based on each target operation label and the corresponding target operation, the step generation model is called to generate a corresponding operation execution sequence, and the operation steps in the operation execution sequence are executed, wherein the operation execution sequence includes each target operation label and the corresponding target operation generated in the execution order.

[0007] According to a second aspect of the embodiments of this specification, there is provided an operation processing device, including: an acquisition module, configured to acquire an operation requirement to be processed; a determination module configured to search an operation annotation database based on the operation requirement to obtain at least one corresponding target operation annotation, wherein the operation annotation database includes a plurality of operation annotations generated by an annotation model based on corresponding operation information; The operation module is configured to call the step generation model to generate a corresponding operation execution sequence based on each target operation label and the corresponding target operation, and execute the operation steps in the operation execution sequence, wherein the operation execution sequence includes the operation steps generated by each target operation label and the corresponding target operation in an execution order.

[0008] According to a third aspect of an embodiment of this specification, a computing device is provided, including: memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the above-mentioned operation processing method are implemented.

[0009] According to a fourth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores computer-executable instructions, and when the instructions are executed by a processor, the steps of the above-mentioned operation processing method are implemented.

[0010] According to a fifth aspect of the embodiments of this specification, a computer program product is provided, comprising a computer program / instruction, which implements the steps of the above-mentioned operation processing method when executed by a processor.

[0011] One embodiment of the present specification implements the following steps: obtaining an operation requirement to be processed; based on the operation requirement, searching an operation annotation database to obtain at least one corresponding target operation annotation, wherein the operation annotation database includes multiple operation annotations, and the multiple operation annotations are generated by an annotation model based on corresponding operation information; based on each target operation annotation and the corresponding target operation, calling a step generation model to generate a corresponding operation execution sequence, and executing the operation steps in the operation execution sequence, wherein the operation execution sequence includes the target operation annotations and the operation steps generated in the execution order of the corresponding target operation. The user only needs to input the operation requirement, and the corresponding target operation annotation can be automatically retrieved from the operation annotation database, and the retrieved target operation annotation is converted into a corresponding operation execution sequence through a large model, and the operation execution sequence is automatically executed to implement the operation requirement. The operation requirement input by the user is automatically converted into a corresponding operation execution sequence and automatically executed. This intelligent operation requirement processing method does not require the user to learn a large number of complex operations, can accurately understand and execute the user's operation requirements, not only reducing the user's learning cost, but also making the entire process efficient and fast, greatly improving operational efficiency, avoiding errors caused by human operation errors, improving the accuracy and reliability of operations, and providing users with a more convenient and efficient user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a flowchart of an operation processing method provided by one embodiment of this specification; Figure 2 This is a schematic diagram of retrieving and obtaining a target operation annotation in an operation processing method provided in one embodiment of this specification; Figure 3 This is an update operation execution sequence diagram in an operation processing method provided by an embodiment of this specification; Figure 4 This is a process flow chart of an operation processing method provided by one embodiment of this specification; Figure 5 This is a schematic diagram of the structure of an operation processing device provided by one embodiment of this specification; Figure 6 This is a structural block diagram of a computing device provided by one embodiment of this specification. DETAILED DESCRIPTION

[0013] The following description sets forth many specific details to facilitate a thorough understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the scope of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0014] The terms used in one or more embodiments of this specification are for the purpose of describing specific embodiments only and are not intended to limit one or more embodiments of this specification. The singular forms "a," "an," and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0015] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining."

[0016] In addition, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0017] First, the terms involved in one or more embodiments of this specification are explained.

[0018] Faiss (Facebook AI Similarity Search): is an efficient vector similarity search library designed for fast similarity search tasks on large-scale, massive, high-dimensional vector data.

[0019] API (Application Programming Interface): is a set of predefined rules, protocols, and tools. API acts as a connector between software components, allowing different applications or services to communicate and exchange data or functions without requiring developers to understand the complex internal implementation details of each other. Its purpose is to provide applications and developers with the ability to access a set of routines based on specific software or hardware, promoting modular design, code reuse, and loosely coupled integration between systems.

[0020] JSON (JavaScript Object Notation) is a lightweight, text-based file and data exchange format. It uses a text format that is completely independent of specific programming languages ​​to store and transmit data objects structured by attribute-value pairs and ordered lists. It is easy for humans to read and write, and also easy for machines to parse and generate. Due to its language-independent nature, JSON is widely used in scenarios such as data transmission in application programming interfaces, configuration file storage, and serialization of complex data structures.

[0021] LLM (Large Language Model): A probabilistic generative model built on a deep learning architecture and pre-trained on massive unlabeled text datasets. Its core mechanism is to learn the complex statistical dependencies and linguistic patterns between tokens in text through pre-training tasks such as autoregression or masked language modeling, thereby constructing a high-dimensional implicit language knowledge representation. After pre-training, LLM can predict and generate the most likely subsequent token sequence in the form of a conditional probability distribution based on given contextual cues, leveraging its learned parameterized knowledge. It demonstrates strong contextual learning, zero-shot and few-shot learning capabilities, and versatility in performing a variety of downstream natural language processing tasks.

[0022] Embedding is a core technology that maps discrete, high-dimensional symbolic objects into a continuous, low-dimensional, dense vector space. Its core goal is to capture and encode the intrinsic semantic, grammatical, or relational properties of objects through learned vector representations. In this vector space, vectors corresponding to semantically or relationally similar objects are geometrically closer, effectively supporting feature representation and pattern recognition in downstream machine learning tasks.

[0023] In this specification, an operation processing method is provided. This specification also relates to an operation processing device, a computing device, a computer-readable storage medium, and a computer program product, which are described in detail one by one in the following embodiments.

[0024] See also Figure 1 , Figure 1 A flowchart of an operation processing method provided according to an embodiment of the present specification is shown, which specifically includes the following steps 102-106.

[0025] Step 102: Obtain the operation requirements to be processed.

[0026] Among them, operational requirements refer to requirements related to specific operations, which may be requirements generated in various operational scenarios such as software operations and business operations. The operational requirements can be in the form of text, images, instructions, etc.

[0027] In actual implementation, the user can input the current operation requirements in the displayed interactive interface. For example, in a game editing scenario, the operation requirements input by the user are usually natural language instructions for adjusting game parameters, such as: inputting "the fog is too thick"; in a video editing scenario, the operation requirements input by the user are usually operation instructions for adjusting video parameters, such as "the video resolution is too low."

[0028] Step 104: Based on the operation requirement, search the operation annotation database to obtain at least one corresponding target operation annotation, wherein the operation annotation database includes multiple operation annotations, and the multiple operation annotations are generated through the annotation model based on the corresponding operation information.

[0029] The operation annotation database is a structured knowledge base built using semantic annotation technology. Its core function is to convert low-level operation information into searchable semantic units with natural language descriptions. Operation annotations are semantic descriptions generated from data using LLM, explaining the data's function and purpose. Target operation annotations are semantic descriptions retrieved from the operation annotation database based on user operation requirements and used to generate executable operation sequences.

[0030] Specifically, the operation annotation database can quickly process large-scale data, build a structured retrieval space through multi-LLM annotations and Faiss index, and efficiently perform similarity search. The data annotation model generates corresponding multiple operation annotations from multiple specific operation information and stores them in the operation annotation database. The operation annotation database contains multiple operation annotations for multiple different operations. After obtaining the user's operation request instruction, the database is quickly searched for a target operation annotation corresponding to at least one target operation information based on the user's operation instruction.

[0031] For example, in a game editing scenario, the operation annotation database serves as a semantic vector knowledge base for engine parameters, storing original parameter paths and their vectorized descriptions, such as "global fog effect concentration coefficient." When a user requests an operation like "the fog is too thick," the generated target operation annotation is associated with that parameter and outputs a scenario-based instruction, such as "reduce the fog concentration in the scene." In a video editing scenario, the operation annotation database becomes a semantic index library for video special effects parameters, storing special effects parameters and their descriptions, such as "adjust the saturation of the highlight area." When a user requests "make the highlights more vivid," the generated target operation annotation is dynamically associated with that parameter and outputs a scenario-based instruction, such as "increase the saturation of the highlights."

[0032] In an optional implementation of this embodiment, based on the operation requirement, searching in the operation annotation database to obtain at least one corresponding target operation annotation includes: Call the expansion model to expand the operation requirements and obtain at least two corresponding expansion requirements; Based on each expansion requirement, a search is performed among multiple operation annotations in the operation annotation database to obtain at least one corresponding target operation annotation.

[0033] Among them, the expansion model is a large model that can be based on natural language processing technology, such as a neural network model based on deep learning. It can expand the input content by learning a large amount of text data, so that the user's original operation requirements can be deconstructed into multi-dimensional expressions. The expanded requirements are the result of the expansion model expanding the user requirements, which represents the multi-dimensional expression of the user's operation intention and is used to retrieve the target operation annotation in the operation annotation database.

[0034] In actual implementation, the existing operation requirements can be deeply expanded by calling a special expansion model to ensure that no less than two rich and effective expansion requirements can be obtained, making the requirements more detailed and comprehensive, and generating expansion requirements in multiple dimensions. Based on these expansion requirements, a systematic search is carried out in multiple operation annotations in the operation annotation database to obtain at least one target operation annotation that meets the requirements.

[0035] It should be noted that the operation annotation database is a vector database, and the operation annotations in the operation annotation database are operation annotation vectors. Therefore, in the specific implementation, at least two expansion requirements can be encoded separately to obtain the requirement encoding query vector corresponding to each expansion requirement. Based on the requirement encoding query vector, search and match are performed in multiple operation annotation vectors in the operation annotation database to obtain the corresponding target operation annotation vector, decode the target operation annotation vector, and obtain the target operation annotation.

[0036] In the embodiments of this specification, for a user's single operation requirement, multiple extended requirements are generated by calling the expansion model, thereby preventing the problem of low retrieval accuracy and recall rate in the operation annotation database due to vague and single requirements. The retrieval accuracy and recall rate are greatly improved through requirement expansion, thereby ensuring the accurate realization of user needs.

[0037] In an optional implementation of this embodiment, the operation annotations included in the operation annotation database are operation annotation vectors, and the operation annotation vectors include interface annotation vectors and parameter annotation vectors. The operation annotation database is constructed in the following manner: Get the built-in code interface file and parameter file supported by the target engine; Extracting an interface list and a parameter list based on the code interface file and the parameter file, wherein the interface list includes the extracted interface information of at least one interface, and the parameter list includes the extracted parameter information of at least one parameter; The interface information and parameter information are annotated through the annotation model to obtain the target interface annotation and target parameter annotation, the target interface annotation and target parameter annotation are encoded to obtain the interface annotation vector and parameter annotation vector, and the operation annotation database is constructed.

[0038] The code interface file is a list of executable API names exposed to the outside world by the code, which can be the names of multiple APIs. The parameter file is a list of readable parameters that the engine specifically exposes to the editor. By modifying the parameters, the parameters can be passed to the engine to affect the final execution results.

[0039] Specifically, two data sources are obtained, namely the code interface file and the parameter file supported by the engine. The interface file defines the function API list that can be parsed by the program. The parameter file is organized in a JSON file structure. The nodes have the parameter names and the adjustable parameter ranges and parameter types. After obtaining the data source, the file needs to be anonymized to obtain the interface API list and parameter list of the file, where the interface list includes the interface information of at least one interface, and the parameter list includes the parameter information of at least one parameter. The interface information represents the function of the code, including the function signature and its scope constraints in the system, which is used to define the corresponding actions that the system can perform; the parameter information is a structured description of the engine state configuration, including the control unit of the system state, which is used to represent the corresponding adjustable state of the system.

[0040] In practice, desensitization involves manually checking code interface and parameter files for sensitive information such as company information and text comments unrelated to the code. If so, the relevant sensitive information is removed from the code interface and parameter files. Furthermore, since parameter files are generally named using standard proprietary names, desensitization can be omitted and only the code interface files can be desensitized.

[0041] Among them, the target interface annotation and target parameter annotation are the language descriptions of the technical means generated after multi-model collaborative annotation. The interface information in the interface list and the parameter information in the parameter list are annotated through the annotation model, and the obtained target interface annotation and target parameter annotation are encoded to obtain the interface annotation vector and parameter annotation vector. The interface annotation vector and parameter annotation vector are the conversion of the target interface annotation and target parameter annotation into machine-understandable mathematical representations.

[0042] For example, in a game editing scenario, if a user issues an operation request "the fog is too thick", the annotation model can map the parameter "reduce the fog concentration" to the corresponding annotation vector; in a video editing scenario, if a user issues an operation request "make the picture more cinematic", the annotation model can map the parameter "enhance the cinematic color tone" to the corresponding annotation vector.

[0043] It should be noted that after obtaining the annotation vectors, they are stored separately and the operation annotation database is constructed based on the Faiss vector data index, and the vectorization standard of the editor parameters is established for subsequent fast semantic matching and retrieval according to user needs.

[0044] In an optional implementation of this embodiment, the interface information and parameter information are annotated using an annotation model to obtain a target interface annotation and a target parameter annotation, including: Generate a first annotation prompt word based on the interface information, call the interface annotation model based on the first annotation prompt word, and generate a target interface annotation corresponding to the interface information; A second annotation prompt word is generated based on the parameter information, and a parameter annotation model is called based on the second annotation prompt word to generate a target parameter annotation corresponding to the parameter information.

[0045] Among them, for each interface information in the interface list, a corresponding first annotation prompt word can be generated. The first annotation prompt word refers to specific prompt information used to guide the interface annotation model for annotation. It can contain some rules, examples or requirements for annotation content, etc. Through the interface annotation model, the interface information in the first annotation prompt word is annotated to generate a target interface annotation corresponding to the interface information. The interface information may include the function name, parameter name, return value, etc. of the corresponding interface. The target interface annotation generated by the interface annotation model is the annotation content of each information in the interface information, and marks and explains the functions, parameters, return values, etc. involved in the interface, including the parameter name, type, value range, function, etc.

[0046] For each parameter information in the parameter list, a corresponding second annotation prompt word can be generated. The second annotation prompt word refers to specific prompt information used to guide the parameter annotation model for annotation. It can contain some rules, examples or requirements for annotation content, etc. Through the parameter annotation model, the parameter information in the second annotation prompt word is annotated to generate the target parameter annotation corresponding to the parameter information. The parameter information can include the actual parameter name in the engine and the data type and default value corresponding to this name. The target parameter annotation generated by the parameter annotation model is the annotation content of each information in the parameter information.

[0047] In actual implementation, the interface annotation model can be called, and the first annotation prompt word can be input into the interface annotation model. Under the guidance of the first annotation prompt word, the interface annotation model annotates the interface information in the first annotation prompt word and generates and inputs the corresponding target interface annotation. The parameter annotation model can also be called, and the second annotation prompt word can be input into the parameter annotation model. Under the guidance of the second annotation prompt word, the parameter annotation model annotates the parameter information in the second annotation prompt word and generates and inputs the corresponding target parameter annotation.

[0048] It should be noted that after obtaining the target interface annotation and target parameter annotation, the target interface annotation and target parameter annotation can be encoded to obtain the interface annotation vector and parameter annotation vector, and to construct an operation annotation database. Specifically, a specific encoding algorithm (such as a word embedding algorithm) can be used to convert the textual interface annotation vector and parameter annotation vector into a numerical vector. After encoding, the "interface annotation vector" and "parameter annotation vector" are obtained. Each "interface annotation vector" or "parameter annotation vector" corresponds to a number index, and the corresponding operation annotation database is constructed. In other words, the operation annotation database is a key-value pair (K-Value) database, where the key (K) refers to the number index of each row, and the value (Value) refers to the "interface annotation vector" or "parameter annotation vector" obtained by encoding.

[0049] As an example, in a game editing scenario, for a code interface file, a regular expression can be used to extract a list of exposed interfaces, typically named in the form of return_type function_name(params...). Each interface in this list can include information such as the function name, parameter names, and return value. For example, function_name (function name), which indicates the general function function; params (params), which indicates the parameter names and their meanings; and return_type (return value), which indicates the current return type.

[0050] For the interface information of any interface, the first annotation prompt word can be generated, such as "f''. Please analyze the C++ code in the current <>. This is the first pure virtual function <{function}>. You need to summarize the <function name, parameter name, return value> and add function comments, parameter comments, and return value comments. The function comment can be up to 20 words, the parameter comments can be up to 20 words each, and the return value comments can be up to 20 words each. If the function or parameter name does not have clear behavior information, <unknown> is directly output. The output result is in the following format:\n' \.

[0051] F'<Function name: Function comment; Return value: Return value comment; Parameter name 1: Parameter comment; Parameter name 2: Parameter comment>'. ” By inputting the first annotation prompt word into the interface annotation model, a target interface annotation in the format of <function name: function annotation; return value: return value annotation; parameter name 1: parameter annotation; parameter name 2: parameter annotation> can be obtained.

[0052] For the parameter file, similarly, the parameter list in the engine can be parsed. For the parameter information of any parameter in the parameter list, a second annotation prompt word can be generated, such as "f' This is a definition of editable parameters in the game in a game engine editor. Please analyze the meaning of the key-value pairs in the current <>'s json. Key name: <{function_name}>, ' \ f' Key type <{function_type}>, ' \ f' You need to add a description according to the key name and key type, and the description is at most 20 characters, ' \ f' Use dots in the key name to mark the hierarchical meaning of the current parameter. So there are two types of descriptions for the key that need to be generated. \n' \ f''1. Key intuitive description: Keep the dot format to generate the intuitive meaning of the key, that is, directly translate the meaning of each word within each level. \n' \ f' 2. Key hierarchical description: Remove the dots, and in the order from left to right according to the hierarchical level, translate what this key as a whole expresses as adjusting a sub-parameter under a certain major function. ' \ f' If the key name has no clear behavior information, directly output <unknown>, ' \ f' This is an editor for a game engine, so when translating, the professional expressions of words in the game industry need to be fully considered. ' \ f' Output the results in the following format: \n' \ f' <Key name: Key intuitive description: Key hierarchical description: Key type> '"。

[0053] Input the second annotation prompt word into the parameter annotation model to obtain the target parameter annotation in the format of <key name: key intuitive description: key hierarchical description: key type>.

[0054] Encode the target interface annotations and target parameter annotations to obtain the corresponding interface annotation vectors and parameter annotation vectors. Determine the number of interface annotation vectors and parameter annotation vectors, and generate corresponding number indexes respectively. One number index corresponds to identifying one vector (the number of interface annotation vectors or parameter annotation vectors), and construct an operation annotation database. The operation annotation database includes multiple number index - vector (the number of interface annotation vectors or parameter annotation vectors) data pairs, such as number index "1" corresponding to vector X1, number index "2" corresponding to vector X2,..., number index "n" corresponding to vector Xn.

[0055] In the embodiments of this specification, each interface information can be labeled through the interface labeling model to obtain the corresponding target interface labeling; each parameter information can be labeled through the parameter labeling model to obtain the corresponding target parameter labeling; the target interface labeling and the target parameter labeling are encoded to obtain the interface labeling vector and the parameter labeling vector, and an operation labeling database is constructed so that after the user inputs the operation requirements to be processed, the constructed operation labeling database can be directly searched to obtain the relevant interfaces and parameters, and adjustments can be automatically made to meet the user's needs.

[0056] In an optional implementation of this embodiment, calling the interface annotation model based on the first annotation prompt word to generate the target interface annotation corresponding to the interface information includes: Based on the first annotation prompt word, a first interface annotation model is called to generate a first interface annotation corresponding to the interface information; and based on the first annotation prompt word, a second interface annotation model is called to generate a second interface annotation corresponding to the interface information; Extracting a first annotation text from the first interface annotation, and extracting a second annotation text from the second interface annotation; A tag similarity between the first tag text and the second tag text is determined, and a target interface tag is determined based on the tag similarity.

[0057] The first annotation text is the annotation result generated by the first annotation model, and the second annotation text is the annotation result generated by the second annotation model.

[0058] In actual implementation, the first interface annotation model and the second interface annotation model are different large models that have been trained and are capable of performing interface annotation. The first annotation prompt word is input into the first interface annotation model, and the first interface annotation model can be used to perform interface annotation to obtain the first interface annotation; furthermore, the first annotation prompt word can be input into the second interface annotation model, and the second interface annotation model can be used to perform interface annotation to obtain the second interface annotation. In other words, the same first annotation prompt word is annotated using two different large models, and two interface annotations corresponding to the first annotation prompt word are obtained.

[0059] Next, extract the first annotation text from the first interface annotation and the second annotation text from the second interface annotation. Use an appropriate algorithm (such as edit distance or cosine similarity) to calculate the similarity between the first and second annotation texts. This similarity can measure the closeness of the two annotation texts in terms of content and semantics. Based on the calculated annotation similarity, select an appropriate rule to determine the final target interface annotation.

[0060] As an example, for the output of two annotated texts, extract the text description output about the function name or parameter name, perform Chinese word vector encoding, calculate the vector cosine similarity of the final embedding data, and determine the target interface annotation based on the similarity.

[0061] For example, in a game editing scenario, for the first annotation prompt word corresponding to a certain interface information, the first interface annotation model generates the description "Set the global fog effect concentration coefficient", and the second interface annotation model outputs "Dynamically adjust the scene fog density", and the similarity between the two is calculated; in a video editing scenario, for the first annotation prompt word corresponding to a certain interface information, the first model generates "Load the color lookup table and set the intensity" for the color adjustment function, and the second model outputs "Apply a movie-level LUT filter", and the similarity between the two is calculated.

[0062] In the embodiments of this specification, by using different interface annotation models to annotate the interface information separately, and then comparing the annotation results to determine a more appropriate target interface annotation, the risk of cognitive bias of the large model can be eliminated, the knowledge blind spots of a single model can be avoided, and the accuracy and reliability of the interface information annotation can be improved.

[0063] It should be noted that, in addition to the above-mentioned interface annotation through dual models, parameter annotation can also be performed through dual models. The specific implementation method is similar to that of interface annotation through dual models, and the embodiments of this specification will not be repeated here.

[0064] Of course, in actual implementation, in addition to the above-mentioned dual-model interface annotation / parameter annotation, more models can also be used to perform interface annotation / parameter annotation, and more accurate annotation results can be selected. The embodiments of this specification do not limit this.

[0065] In an optional implementation of this embodiment, determining the target interface annotation based on annotation similarity includes: When the annotation similarity is greater than the similarity threshold, the interface annotation with fewer characters in the first annotation text and the second annotation text is used as the target interface annotation; When the annotation similarity is not greater than or equal to the similarity threshold, the evaluation model is called to evaluate the first interface annotation and the second interface annotation, and the target interface annotation is selected based on the evaluation result.

[0066] The similarity threshold is a value that can be set according to specific circumstances and is used to determine the similarity between the annotation results of the first annotation prompt word of the same interface information by two interface annotation models.

[0067] In actual implementation, after obtaining the annotation similarity, if the similarity is greater than the similarity threshold, it means that the annotation results of the two interface annotation models for the same interface information are highly consistent. In this case, the one with a shorter text description can be selected as the target interface annotation to save storage space. In another case, if the annotation similarity of the two annotation texts is not greater than or equal to the similarity threshold, it means that the annotation results of the two interface annotation models for the same interface information are quite different. In this case, the evaluation model can be further called to evaluate the annotation results output by the two interface annotation models, output the scores of the two annotation results respectively, and select the one with the higher score as the final target interface annotation.

[0068] The evaluation model is a pre-trained model capable of evaluating the quality of input information. It is used to evaluate the quality of the annotation results output by the two interface annotation models. Furthermore, the evaluation model can assess the annotation results based on factors such as language fluency, consistency of function / parameter names, terminology standardization, and behavioral consistency.

[0069] Continuing with the above example, we set the similarity threshold to 0.95. In the game editing scenario, for the first annotation prompt word corresponding to a certain interface information, the first interface annotation model generates the first interface annotation "Set global fog effect concentration coefficient", and the second interface annotation model generates the second interface annotation "Dynamically adjust scene fog density". Calculate the similarity between the two. Assuming the similarity is 0.82, call the evaluation model at this time and score according to dimensions such as "behavioral consistency" and "game terminology standardization". Assuming that the score of the first interface annotation "Set global fog effect concentration coefficient" is 8.5, and the score of the second interface annotation "Dynamically adjust scene fog density" is 7.2, it means that the first interface annotation generated by the first interface annotation model is more accurate. At this time, the first interface annotation "Set global fog effect concentration coefficient" can be used as the final target interface annotation.

[0070] It should be noted that if the similarity of the annotated text results is high, it means that the ambiguity of the function naming is low and the credibility of the arguments of different LLMs on the results is also high. If the similarity is low, introducing a third evaluation model as a referee model for scoring can ensure the reliability of key annotations.

[0071] Step 106: Based on each target operation label and the corresponding target operation, call the step generation model to generate a corresponding operation execution sequence, and execute the operation steps in the operation execution sequence, wherein the operation execution sequence includes each target operation label and the corresponding target operation generated in the execution order.

[0072] Among them, the step generation model is a model that converts each target operation into an operation execution sequence. The operation execution sequence refers to an operation sequence that the engine can recognize and conforms to the engine constraints, including the target operation label and the corresponding target operation steps generated in the execution order.

[0073] It should be noted that the step generation model is based on the corresponding operation execution sequence generated by each target operation label and the target operation, which refers to the operation parameters that the game engine can recognize, and realizes the conversion of the user's operation requirements input in natural language into machine parameters that the engine can understand.

[0074] As an example, Figure 2 This is a schematic diagram of retrieving and obtaining a target operation annotation in an operation processing method provided in one embodiment of this specification, such as Figure 2 As shown, in a game editing scenario, the user inputs an operation requirement to be processed: "The fog is too thick." Based on this operation requirement, a search is performed in the operation annotation database, and the target operation annotations obtained include: "Reduce fog density," "Increase fog height," "Increase fog starting distance," "Reduce fog opacity," and "Adjust fog color to a lighter hue." The target operation annotation "Reduce fog density" corresponds to the target operation "HeightFog.FogDensity," the target operation annotation "HeightFog.FogHeight" corresponds to the target operation "Increase fog height," the target operation annotation "HeightFog.FogstartDistance" corresponds to the operation annotation "Increase fog starting distance," the target operation annotation "HeightFog.Fogopacity" corresponds to the operation annotation "Reduce fog opacity," and the target operation annotation "HeightFog.FogColor" corresponds to the operation annotation "Adjust fog color to a lighter hue."

[0075] By inputting each target operation label and the corresponding target operation into the step generation model, the corresponding operation execution sequence can be obtained: "This operation process first reduces the density of the fog, then adjusts the height and starting distance of the fog to make the fog look less dense, and finally reduces the opacity of the fog and adjusts the color of the fog to a lighter tone to improve the visual experience. The retrieved target operations and corresponding target operation labels are: ["HeightFog.FogDensity", "Reduce fog density"], ["HeightFog.FogHeight", "Increase fog height"], ["HeightFog.FogstartDistance", "Increase fog starting distance"], ["HeightFog.Fogopacity", "Reduce fog opacity"], ["HeightFog.FogColor", "Adjust fog color to a lighter tone"]. The recommended operation execution order is: reduce fog density, increase fog height, increase fog starting distance, reduce fog opacity, and adjust fog color to a lighter tone."

[0076] In an optional implementation of this embodiment, executing the operation steps in the operation execution sequence includes: Convert each operation step in the operation execution sequence into executable operation parameters of the target engine to obtain an updated operation execution sequence, wherein the target engine is the engine of each operation step in the operation execution sequence to be executed; In the target engine, each operation step is executed sequentially based on the executable operation parameters in the updated operation execution sequence.

[0077] The executable operation parameters of the target engine refer to converting the generated operation steps into standardized instructions that can be directly executed by the engine. These instructions are in a format that the engine can understand and process, just like clear commands given to the engine.

[0078] Specifically, the target engine is the program or system responsible for executing each step in the operation execution sequence. It has its own specific functions and operating rules, and the operation steps need to be converted into a form that it can recognize and execute. Each operation step in the operation execution sequence is converted into executable operation parameters for the target engine. Executable operation parameters refer to several supported adjustment directions predefined in the engine to enable linkage with its built-in functions.

[0079] In actual implementation, each operation step in the operation execution sequence can be converted into executable operation parameters for the target engine to obtain an updated operation execution sequence. The updated operation execution sequence includes executable operation parameters that the target engine can recognize and execute. This allows the target engine to execute each operation step in sequence based on the executable operation parameters in the updated operation execution sequence, thus achieving automatic execution of each operation step by the target engine. In other words, this process is like translating one language into another language that the target engine can understand. As a result, the target engine can execute each operation step in sequence according to the order of the executable operation parameters in the updated operation execution sequence, ultimately completing the entire task.

[0080] In the embodiments of this specification, after automatically generating the corresponding operation execution sequence based on the target operation annotations and corresponding target operations obtained through retrieval, the operation execution sequence can be converted into executable operation parameters that can be automatically recognized and processed by the target engine, so that the target engine can automatically execute each operation step without manual execution, and can better link to the built-in functions of the engine.

[0081] In an optional implementation of this embodiment, converting each operation step in the operation execution sequence into executable operation parameters of the target engine to obtain an updated operation execution sequence includes: Determine the operations to be performed for each operation step in the operation execution sequence; Determine the executable operation parameters corresponding to the to-be-executed operation of each operation step based on the operation parameters corresponding to the preset operation in the target engine; Based on the executable operation parameters corresponding to each operation step, an update operation execution sequence is generated.

[0082] Among them, the operation parameters corresponding to the preset operations in the target engine are to pre-define several adjustable directions supported by the engine and the corresponding adjustment parameters in the engine in order to be able to call the built-in functions of the engine. These parameters can be used to control the engine to perform operations in an expected manner. For example, the operation parameters corresponding to the preset operations in the target engine include: the operation parameter corresponding to the preset operation "increase" is "executable operation parameter inc" - the parameter is gradually increased with a step size of 1; the operation parameter corresponding to the preset operation "decrease" is "executable operation parameter dec" - the parameter is gradually decreased with a step size of 1; the operation parameter corresponding to the preset operation "enable XX function" is "executable operation parameter..." - the corresponding function item is enabled; the operation parameter corresponding to the preset operation "disable" is "executable operation parameter..." - the corresponding function item is disabled.

[0083] In actual implementation, each operation step in the operation execution sequence can be converted into executable operation parameters according to the operation parameters corresponding to the preset operation pre-defined by the target engine, and the executable operation parameters can be added to the corresponding steps to generate the corresponding update operation execution sequence.

[0084] Using the above example, for Figure 2 In the operation execution sequence shown, step 1 is: "HeightFog.FogDensity", which is used to decrease the fog density parameter. The target engine pre-configured preset operation parameters are queried to determine that the executable operation parameter for step 1 is "dec", i.e., decrease. This identifies the "decrease" operation in the target engine. In this case, the target engine executable operation parameter "dec" can be added after step 1. Step 2 is: "HeightFog.FogHeight", which is used to increase the fog height. The target engine pre-configured preset operation parameters are queried to determine that the executable operation parameter for step 2 is "inc", i.e., increase. This identifies the "increase" operation in the target engine. The target engine executable operation parameter "inc" can be added after step 2. Step 3 is: "HeightFog.FogstartDistance", which is used to increase the fog start distance. The target engine pre-configured preset operation parameters are queried to determine that the executable operation parameter for step 3 is "inc". The target engine executable operation parameter "inc" can be added after step 3. Step 4 is: "HeightFog.Fogopacity", "Reduce fog opacity", determine that the executable operation parameter of step 4 in the target engine is "dec", and add the executable operation parameter of the target engine "dec" after step 4. Step 5 is: "HeightFog.FogColor", "Adjust fog color to a lighter tone", determine that the executable operation parameter of step 5 in the target engine is "dec", and add the executable operation parameter of the target engine "dec" after step 5. Figure 3 The update operation execution sequence shown is Figure 3 This is a schematic diagram of an update operation execution sequence in an operation processing method provided in an embodiment of this specification.

[0085] In the embodiments of this specification, the target engine may predefine operation parameters corresponding to preset operations, thereby converting the generated operation steps into a form that the target engine can recognize and execute, so that the target engine automatically executes each operation step, thereby improving execution efficiency and accuracy.

[0086] In an optional implementation of this embodiment, the target engine sequentially executes each operation step based on the executable operation parameters in the update operation execution sequence, including: Determining a target operation type corresponding to a target operation step in the update operation execution sequence, wherein the target operation step is any operation step in the update operation execution sequence; If the target operation type of the target operation step is an interface type, then based on the function mapping table, find the function pointer corresponding to the target operation step, inject the executable operation parameters in the target operation step into the corresponding target function to execute the target operation step; If the target operation type of the target operation step is a parameter type, the executable operation parameters of the target operation step are passed to the target engine through the built-in parameter passing method of the target engine to execute the target operation step.

[0087] The target operation step refers to one of the operation steps in the update operation execution sequence, and it is necessary to determine whether the operation type of the target operation step is an interface type or a parameter type.

[0088] Specifically, for the target operation type of the target operation step, its operation type field is first parsed, and the interface annotation data and parameter annotation data are classified and stored in the operation annotation database. Based on the source of the operation step, it can be determined whether the corresponding operation type is an interface type or a parameter type.

[0089] It should be noted that the interface annotation vector obtained by annotating the interface information in the interface list can be stored in code_embbedding.faiss; the parameter annotation vector obtained by annotating the parameter information in the parameter list can be stored in data_embbedding.faiss. When retrieving the target operation annotation, it is necessary to calculate the similarity between the user-entered requirement encoding (query embbeding) and the vectors stored in code_embbedding.faiss and data_embbedding.faiss, and the similarity results of code_embbedding.faiss and data_embbedding.faiss can be recorded in two different dicts respectively. "Dict" is the abbreviation of "dictionary", which refers to the dictionary data type in computer programming. It is a mutable container model that can store objects of any type, stores data in the form of key-value pairs, and quickly accesses the corresponding value through the key. Specifically, Code_emd_dict {stores the code-side embbeding data closest to the query embbeding}; Data_emb_dict {stores the data-side embbeding data closest to the query embbeding}. Therefore, based on the source of the operation step, it can be determined whether the current operation step comes from the code interface file or the parameter file, and the corresponding operation step can be executed accordingly.

[0090] In actual implementation, for any operation step in the update operation execution sequence (i.e., the target operation step), the corresponding operation type must first be determined. Operation types may vary, such as interface type and parameter type. If the target operation step's operation type is an interface type, the corresponding function pointer is found using a function mapping table. A function mapping table is a data structure that establishes a correspondence between operation steps and function pointers. After finding the corresponding function pointer, the executable operation parameters from the target operation step are injected into the corresponding target function. Executable operation parameters are specific data that the target engine can recognize and execute. After injecting these parameters, the target function can perform the corresponding operation based on these parameters, thereby completing the target operation step in the target engine. If the target operation step's operation type is a parameter type, the target engine's built-in parameter passing method is used to pass the executable operation parameters of the target operation step to the target engine. The target engine is a program module with specific functions. After receiving these parameters, it executes the target operation step according to its own logic.

[0091] Using the above example, Figure 3 As shown, for step 1 in the update operation execution sequence, its operation type is determined to be an interface type. At this time, the function pointer corresponding to "HeightFog.FogDensity" in the target operation step can be found based on the function mapping table, and the executable operation parameter "dec" in the target operation step can be injected into the corresponding target function to execute the target operation step according to the operation parameters set by the executable operation parameter "dec" (such as reducing the step size), so that the target engine can automatically execute step 1.

[0092] It should be noted that when processing the update operation execution sequence, the operation type corresponding to each operation step in the update operation execution sequence is first determined, and then based on the different operation types, the corresponding method is used to automatically execute the operation step in the target engine. There is no need for the user to execute the corresponding operation steps one by one, which improves operation efficiency and reduces errors that may be caused by human operations.

[0093] One embodiment of this specification realizes that the user only needs to input the operation requirements, and the corresponding target operation annotation can be automatically retrieved from the operation annotation database. The retrieved target operation annotation is converted into the corresponding operation execution sequence through the large model, and the operation execution sequence is automatically executed to realize the operation requirement. It realizes the automatic conversion of the operation requirements input by the user into the corresponding operation execution sequence and automatic execution. This intelligent operation requirement processing method does not require the user to learn a large number of complex operations, and can accurately understand and execute the user's operation requirements. It not only reduces the user's learning cost, but also makes the entire process efficient and fast, greatly improving the operation efficiency, avoiding errors caused by human operation errors, improving the accuracy and reliability of operations, and bringing a more convenient and efficient user experience to users.

[0094] The following combined Figure 4 , taking the application of the operation processing method provided in this specification in the AI-assisted game editor scenario as an example, the operation processing method is further explained. Figure 4 A processing flow chart of an operation processing method provided in one embodiment of this specification is shown, which specifically includes the following steps.

[0095] Step 402: The user inputs an operation requirement in a natural language instruction.

[0096] Among them, natural language instructions are instructions made by users to the game editor, such as "the fog is too thick."

[0097] Step 404: The large model generates a plurality of different operation annotations based on different operation information types, and retrieves a target operation annotation from the operation annotation database.

[0098] Two different LLMs are used to mark the results simultaneously, and a third LLM is used to evaluate the similarity of the two results and select the best one to be stored in the database. Specifically: Call two annotation LLMs: gpt4 and claude3.5. The annotation LLM is responsible for outputting the input information according to the prompt and generating the corresponding text description: gpt4-output and claude3.5-output.

[0099] For each output, the text description of the function or parameter name is extracted and encoded into a Chinese word vector. The resulting embeddings are then subjected to vector cosine similarity calculations. If the similarity is above 0.95, indicating that the two contents are very similar, the one with the shorter text description is selected for storage. If the similarity is below 0.95, the two outputs are combined into a unified prompt, and another referee LLM model, claude3.7, is called to generate a score for both. The scoring criteria include language fluency and function / parameter name consistency. The AI ​​model gives a comprehensive score to both and selects the one with the higher score for storage.

[0100] The key is the exact function name or parameter name within the engine, and the value is an interpretable Chinese annotation and calling method description. After annotating and obtaining a key-value file of several codes or parameters, use the open source embedding model to encode the value line by line to obtain the underlying embedding data set.

[0101] Step 406: Based on the operation annotation results, the large model generates multiple continuous steps of adjustment feedback to the user. The user can directly use the game editor to make adjustments according to the specific operation, or directly feedback to the game engine to adjust the engine parameters according to the defined adjustment type.

[0102] After annotating several key-value files like these or parameters, the key is the exact function name or parameter name within the engine, and the value is an interpretable Chinese annotation and call method description. Specifically, using an open source embedding model, the value is encoded line by line to obtain the underlying embedding data set: Based on the open-source Faiss vector data indexing algorithm, a large index is constructed for all embeddings and cached locally. When processing user input, the LLM open-source model is first used to appropriately expand the user's demand query to obtain up to three possible potential user demand queries 123. The same embedding model is then used to encode query 123 to obtain the corresponding query_emb123. Then, Faiss's built-in algorithm is used to select the top K most relevant APIs or parameters.

[0103] Furthermore, in order to realize intelligent one-key operation, it is possible to support the continuous adjustment of relevant parameters to directly let the scene affect user needs. For example, if the user inputs "the fog is too thick", after guiding the AI ​​to support continuous step adjustment, the corresponding operation suggestion sequence is obtained, such as Figure 2 As shown in the figure, in order to meet the needs, AI recommends gradually adjusting the fog effect through steps 1-5. The user adjusts the game editor according to the operation steps given by AI to achieve the user's goal. In order to better link it with the built-in functions of the engine, several adjustment directions supported by the engine are pre-defined, such as "inc, dec, enable, disable" and other operations. When executing user needs, the engine adjustment direction is also used as the output requirement, and the latest output is as follows: Figure 3 As shown, after the editor gets the AI ​​output, it can directly adjust the engine parameters according to this defined adjustment type to directly achieve the user's purpose.

[0104] Corresponding to the above method embodiment, this specification also provides an operation processing device embodiment, Figure 5 FIG. 1 shows a schematic diagram of the structure of an operation processing device provided by an embodiment of this specification. Figure 5 As shown, the device includes: An acquisition module 502 is configured to acquire an operation requirement to be processed; The determining module 504 is configured to search an operation annotation database based on the operation requirement to obtain at least one corresponding target operation annotation, wherein the operation annotation database includes a plurality of operation annotations, and the plurality of operation annotations are generated based on corresponding operation information using an annotation model; The operation module 506 is configured to call the step generation model to generate a corresponding operation execution sequence based on each target operation label and the corresponding target operation, and execute the operation steps in the operation execution sequence, wherein the operation execution sequence includes each target operation label and the operation steps generated in the execution order of the corresponding target operation.

[0105] Optionally, the determining module 504 is further configured to: Call the expansion model to expand the operation requirements and obtain at least two corresponding expansion requirements; Based on each expansion requirement, a search is performed among multiple operation annotations in the operation annotation database to obtain at least one corresponding target operation annotation.

[0106] Optionally, the determining module 504 is further configured to: Get the built-in code interface file and parameter file supported by the target engine; Extracting an interface list and a parameter list based on the code interface file and the parameter file, wherein the interface list includes the extracted interface information of at least one interface, and the parameter list includes the extracted parameter information of at least one parameter; The interface information and parameter information are annotated through the annotation model to obtain the target interface annotation and target parameter annotation, the target interface annotation and target parameter annotation are encoded to obtain the interface annotation vector and parameter annotation vector, and the operation annotation database is constructed.

[0107] Optionally, the determining module 504 is further configured to: Generate a first annotation prompt word based on the interface information, call the interface annotation model based on the first annotation prompt word, and generate a target interface annotation corresponding to the interface information; A second annotation prompt word is generated based on the parameter information, and a parameter annotation model is called based on the second annotation prompt word to generate a target parameter annotation corresponding to the parameter information.

[0108] Optionally, the determining module 504 is further configured to: Based on the first annotation prompt word, a first interface annotation model is called to generate a first interface annotation corresponding to the interface information; and based on the first annotation prompt word, a second interface annotation model is called to generate a second interface annotation corresponding to the interface information; Extracting a first annotation text from the first interface annotation, and extracting a second annotation text from the second interface annotation; A tag similarity between the first tag text and the second tag text is determined, and a target interface tag is determined based on the tag similarity.

[0109] Optionally, the determining module 504 is further configured to: When the annotation similarity is greater than the similarity threshold, the interface annotation with fewer characters in the first annotation text and the second annotation text is used as the target interface annotation; When the annotation similarity is not greater than or equal to the similarity threshold, the evaluation model is called to evaluate the first interface annotation and the second interface annotation, and the target interface annotation is selected based on the evaluation result.

[0110] Optionally, the operation module 506 is further configured to: Convert each operation step in the operation execution sequence into executable operation parameters of the target engine to obtain an updated operation execution sequence, wherein the target engine is the engine of each operation step in the operation execution sequence to be executed; In the target engine, each operation step is executed sequentially based on the executable operation parameters in the updated operation execution sequence.

[0111] Optionally, the operation module 506 is further configured to: Determine the operations to be performed for each operation step in the operation execution sequence; Determine the executable operation parameters corresponding to the to-be-executed operation of each operation step based on the operation parameters corresponding to the preset operation in the target engine; Based on the executable operation parameters corresponding to each operation step, an update operation execution sequence is generated.

[0112] Optionally, the operation module 506 is further configured to: Determining a target operation type corresponding to a target operation step in the update operation execution sequence, wherein the target operation step is any operation step in the update operation execution sequence; If the target operation type of the target operation step is an interface type, then based on the function mapping table, find the function pointer corresponding to the target operation step, inject the executable operation parameters in the target operation step into the corresponding target function to execute the target operation step; If the target operation type of the target operation step is a parameter type, the executable operation parameters of the target operation step are passed to the target engine through the built-in parameter passing method of the target engine to execute the target operation step.

[0113] The above is a schematic diagram of an operation processing device according to this embodiment. It should be noted that the technical solution of the operation processing device and the technical solution of the above-mentioned operation processing method are based on the same concept. For details not described in detail in the technical solution of the operation processing device, please refer to the description of the technical solution of the above-mentioned operation processing method.

[0114] Figure 66 shows a block diagram of a computing device 600 according to one embodiment of the present disclosure. Components of the computing device 600 include, but are not limited to, a memory 610 and a processor 620. The processor 620 is connected to the memory 610 via a bus 630, and a database 650 is used to store data.

[0115] Computing device 600 also includes an access device 640 that enables computing device 600 to communicate via one or more networks 660. Examples of such networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. Access device 640 may include one or more of any type of network interface (e.g., a network interface card (NIC)) whether wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, or a near field communication (NFC) interface.

[0116] In one embodiment of the present specification, the above components of the computing device 600 and Figure 6 Other components not shown in the figure may also be connected to each other, for example, via a bus. Figure 6 The computing device structure block diagram shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art may add or replace other components as needed.

[0117] Computing device 600 can be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, personal digital assistant, laptop computer, notebook computer, netbook computer, etc.), a mobile phone (e.g., a smartphone), a wearable computing device (e.g., a smartwatch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or personal computer (PC). Computing device 600 can also be a mobile or stationary server.

[0118] The processor 620 is configured to execute the following computer-executable instructions, which implement the steps of the above-mentioned operation processing method when executed by the processor.

[0119] The above is a schematic diagram of a computing device according to this embodiment. It should be noted that the technical solution of the computing device and the technical solution of the above-mentioned operation processing method are of the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the above-mentioned operation processing method.

[0120] An embodiment of the present specification further provides a computer-readable storage medium storing computer-executable instructions, which implement the steps of the above-mentioned operation processing method when executed by a processor.

[0121] The above is a schematic scheme of a computer-readable storage medium of this embodiment. It should be noted that the technical scheme of the storage medium and the technical scheme of the above-mentioned operation processing method are based on the same concept. For details not described in detail in the technical scheme of the storage medium, please refer to the description of the technical scheme of the above-mentioned operation processing method.

[0122] An embodiment of the present specification further provides a computer program, wherein when the computer program is executed in a computer, the computer is caused to execute the steps of the above-mentioned operation processing method.

[0123] The above is an illustrative solution of a computer program of this embodiment. It should be noted that the technical solution of the computer program and the technical solution of the above-mentioned operation processing method are of the same concept. For details not described in detail in the technical solution of the computer program, please refer to the description of the technical solution of the above-mentioned operation processing method.

[0124] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0125] The computer instructions include computer program code, which may be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium may include any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium. It should be noted that the content of the computer-readable medium may be appropriately increased or decreased based on the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media does not include electric carrier signals and telecommunication signals.

[0126] It should be noted that for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited by the order of the actions described, because according to the embodiments of this specification, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of this specification.

[0127] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0128] The preferred embodiments disclosed above are intended only to help illustrate this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations can be made based on the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that those skilled in the art can better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. An operation processing method, characterized in that: include: Get pending operation requirements; Based on the operation requirement, searching an operation annotation database to obtain at least one corresponding target operation annotation, wherein the operation annotation database includes a plurality of operation annotations, and the plurality of operation annotations are generated by an annotation model based on corresponding operation information; Based on each target operation label and the corresponding target operation, the step generation model is called to generate a corresponding operation execution sequence, and the operation steps in the operation execution sequence are executed, wherein the operation execution sequence includes each target operation label and the corresponding target operation generated in the execution order.

2. The operation processing method according to claim 1, characterized in that: The step of executing the operation steps in the operation execution sequence includes: Convert each operation step in the operation execution sequence into executable operation parameters of a target engine to obtain an updated operation execution sequence, wherein the target engine is an engine to execute each operation step in the operation execution sequence; In the target engine, each operation step is executed in sequence based on the executable operation parameters in the update operation execution sequence.

3. The operation processing method according to claim 2, characterized in that: The step of converting each operation step in the operation execution sequence into executable operation parameters of the target engine to obtain an updated operation execution sequence includes: Determining an operation to be performed for each operation step in the operation execution sequence; Determining executable operation parameters corresponding to the to-be-executed operations of each operation step based on the operation parameters corresponding to the preset operations in the target engine; The update operation execution sequence is generated based on the executable operation parameters corresponding to each operation step.

4. The operation processing method according to claim 2, characterized in that: The target engine sequentially executes each operation step based on the executable operation parameters in the update operation execution sequence, including: Determining a target operation type corresponding to a target operation step in the update operation execution sequence, wherein the target operation step is any operation step in the update operation execution sequence; If the target operation type of the target operation step is an interface type, searching for a function pointer corresponding to the target operation step based on a function mapping table, and injecting the executable operation parameters in the target operation step into the corresponding target function to execute the target operation step; If the target operation type of the target operation step is a parameter type, the executable operation parameters of the target operation step are passed to the target engine through the parameter transfer method built into the target engine to execute the target operation step.

5. The operation processing method according to claim 1, characterized in that: The step of searching an operation annotation database based on the operation requirement to obtain at least one corresponding target operation annotation includes: Calling the expansion model to expand the operation requirement and obtain at least two corresponding expansion requirements; Based on each expansion requirement, a search is performed among multiple operation annotations in the operation annotation database to obtain at least one corresponding target operation annotation.

6. The operation processing method according to any one of claims 1 to 5, characterized in that: The operation annotations included in the operation annotation database are operation annotation vectors, and the operation annotation vectors include interface annotation vectors and parameter annotation vectors. The operation annotation database is constructed in the following manner: Get the built-in code interface file and parameter file supported by the target engine; Extracting an interface list and a parameter list based on the code interface file and the parameter file, wherein the interface list includes the extracted interface information of at least one interface, and the parameter list includes the extracted parameter information of at least one parameter; The interface information and the parameter information are annotated by the annotation model to obtain a target interface annotation and a target parameter annotation, the target interface annotation and the target parameter annotation are encoded to obtain an interface annotation vector and a parameter annotation vector, and the operation annotation database is constructed.

7. The operation processing method according to claim 6, characterized in that: The step of labeling the interface information and the parameter information by using the labeling model to obtain a target interface labeling and a target parameter labeling includes: generating a first annotation prompt word based on the interface information, calling an interface annotation model based on the first annotation prompt word, and generating a target interface annotation corresponding to the interface information; A second annotation prompt word is generated based on the parameter information, and a parameter annotation model is called based on the second annotation prompt word to generate a target parameter annotation corresponding to the parameter information.

8. The operation processing method according to claim 7, characterized in that: The calling of the interface annotation model based on the first annotation prompt word to generate a target interface annotation corresponding to the interface information includes: Invoking a first interface annotation model based on the first annotation prompt word to generate a first interface annotation corresponding to the interface information, and invoking a second interface annotation model based on the first annotation prompt word to generate a second interface annotation corresponding to the interface information; Extracting a first annotation text from the first interface annotation, and extracting a second annotation text from the second interface annotation; Determine the annotation similarity between the first annotation text and the second annotation text, and determine the target interface annotation based on the annotation similarity.

9. The operation processing method according to claim 8, characterized in that: The determining the target interface annotation based on the annotation similarity includes: When the annotation similarity is greater than a similarity threshold, the interface annotation with fewer characters in the first annotation text and the second annotation text is used as the target interface annotation; When the annotation similarity is not greater than or equal to the similarity threshold, an evaluation model is called to evaluate the first interface annotation and the second interface annotation, and the target interface annotation is selected based on the evaluation result.

10. An operation processing device, characterized in that: include: an acquisition module, configured to acquire an operation requirement to be processed; a determination module configured to search an operation annotation database based on the operation requirement to obtain at least one corresponding target operation annotation, wherein the operation annotation database includes a plurality of operation annotations generated by an annotation model based on corresponding operation information; The operation module is configured to call the step generation model to generate a corresponding operation execution sequence based on each target operation label and the corresponding target operation, and execute the operation steps in the operation execution sequence, wherein the operation execution sequence includes the operation steps generated by each target operation label and the corresponding target operation in an execution order.

11. A computing device, characterized in that include: memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the operation processing method according to any one of claims 1 to 9 are implemented.

12. A computer-readable storage medium, characterized in that It stores computer-executable instructions, which, when executed by a processor, implement the steps of the operation processing method according to any one of claims 1 to 9.

13. A computer program product, characterized in that The method comprises a computer program / instruction, which, when executed by a processor, implements the steps of the operation processing method according to any one of claims 1 to 9.

Images