An operation scheme generation method, device, equipment and medium

Abstract

The application relates to the technical field of image recognition and discloses an operation scheme generation method, device and equipment and a medium, which comprises the following steps: performing standardization preprocessing on a screenshot to obtain a normalized image, extracting text features of the normalized image, and identifying an App page type, a business scenario and a visual hotspot area of the normalized image to obtain visual features. Through the fusion of the text features and the visual features, unified multi-modal fusion features are obtained. Subsequently, a business rule engine is used to dynamically infer the features to generate a multi-dimensional structured business description containing a business line, a function module and a responsible department. Based on the description, a preliminary operation strategy is automatically matched and generated. After the strategy is implemented by an operation personnel, the system collects business index data in a preset time period and judges whether the business index data meet preset requirements. If the business index data do not meet the preset requirements, the business rule engine is optimized according to the data, and the inference step is re-executed; otherwise, the preliminary strategy is determined as a final operation strategy. The efficiency and reliability of operation scheme generation are improved.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method, apparatus, device and medium for generating operation plans. Background Technology

[0002] With the deepening development of the mobile internet and the acceleration of enterprise digital transformation, the number of internal applications (Apps) is increasing daily, with frequent version iterations. Operations staff need to process massive amounts of App page screenshots every day. These screenshots are key materials for user experience analysis, troubleshooting functional anomalies, monitoring business processes, and evaluating operational effectiveness, and are widely used in multiple departments such as product, operations, testing, and risk control.

[0003] However, current processing and analysis of app screenshots primarily rely on traditional manual methods or fragmented technical tools. Faced with the explosive growth of screenshot data, manually viewing, classifying, and summarizing each screenshot is time-consuming and labor-intensive, making it difficult to meet real-time analysis needs and becoming a bottleneck for business response and decision-making. Furthermore, screenshots themselves are unstructured image data, and the rich business semantics they contain (such as page type, core functions, key text, and control intent) cannot be directly understood and utilized by machines. The lack of efficient structured description methods makes it impossible to build a screenshot knowledge base that supports accurate and rapid retrieval, resulting in insufficient reliability. Summary of the Invention

[0004] This invention provides an operation plan generation method, apparatus, computer equipment, and medium to solve the problems of low efficiency and low reliability of existing operation plan generation methods on the market.

[0005] Firstly, a method for generating operational plans is provided, including: Obtain screenshots of preset APP services, and perform size normalization, noise reduction, and contrast enhancement processing on the screenshots to obtain standardized images; Extract text features from the standardized image, identify the App function page type, business scenario, and visual hotspot area corresponding to the standardized image, and obtain visual features; The text features and the visual features are fused to obtain multimodal fused features; The multimodal fusion features are dynamically reasoned using a preset business rule engine to generate a multi-dimensional structured description that includes business lines, functional modules and responsible departments, thus obtaining a structured business description. A preliminary operation strategy is generated based on the structured business description. Operational operations are executed based on the preliminary operation strategy. Business indicator data generated after a preset operation time is obtained. The operational operations include business adjustments made by operations personnel based on the preliminary operation strategy, or automatic execution of operational configurations based on the preliminary operation strategy. Determine whether the business indicator data meets the preset business indicator requirements; If it does not meet the requirements, the business rule engine is optimized based on the business indicator data, and then the process is returned to the previous step of using the preset business rule engine to perform dynamic reasoning on the multimodal fusion features to generate a multi-dimensional structured description containing business lines, functional modules and responsible departments, thereby obtaining a structured business description. If the conditions are met, the preliminary operational strategy will be adopted as the final operational strategy.

[0006] Secondly, an operation plan generation device is provided, including: Image processing module 101 is used to acquire screenshots of preset APP services, and perform size normalization, noise reduction and contrast enhancement processing on the screenshots to obtain standardized images. The feature extraction module 102 is used to extract text features from the standardized image, identify the App function page type, business scenario and visual hotspot area corresponding to the standardized image, and obtain visual features; The dynamic reasoning module 103 is used to perform feature fusion on the text features and the visual features to obtain multimodal fusion features, and to perform dynamic reasoning on the multimodal fusion features using a preset business rule engine to generate a multi-dimensional structured description containing business lines, functional modules and responsible departments, thereby obtaining a structured business description. The business adjustment module 104 is used to generate a preliminary operation strategy based on the structured business description, execute operation operations based on the preliminary operation strategy, and obtain business indicator data generated after a preset operation time. The operation operations include business adjustments made by operation personnel based on the preliminary operation strategy, or automatic execution of operation configuration based on the preliminary operation strategy. The indicator judgment module 105 is used to determine whether the business indicator data meets the preset business indicator requirements. If it does not meet the requirements, the business rule engine is optimized based on the business indicator data, and then the preset business rule engine is used to perform dynamic reasoning on the multimodal fusion features to generate a multi-dimensional structured description containing business lines, functional modules and responsible departments, thereby obtaining a structured business description. If it meets the requirements, the preliminary operation strategy is used as the final operation strategy.

[0007] Thirdly, a computer device is provided, 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 steps of the above-described operation plan generation method.

[0008] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the above-described operation scheme generation method.

[0009] The above-mentioned operational plan generation method, apparatus, computer equipment, and storage medium can obtain a business screenshot of a preset APP business, perform size normalization, noise reduction, and contrast enhancement on the screenshot to obtain a standardized image, eliminate differences caused by factors such as device, resolution, and lighting, provide stable and high-quality input for subsequent feature extraction, and improve the robustness and parsing accuracy of the overall system. Text features are extracted from the standardized image, and the App function page type, business scenario, and visual hotspot area corresponding to the standardized image are identified to obtain visual features, providing a semantic basis for understanding interface functions, business intentions, and key information. The text features and visual features are then fused to obtain multimodal fusion features, achieving deep complementarity and semantic alignment of text and visual information, forming a unified and rich joint feature representation. This overcomes the limitations of separating text and image information in traditional solutions. A preset business rule engine is used to dynamically process the multimodal fusion features. Dynamic reasoning is used to generate a multi-dimensional structured description including business lines, functional modules, and responsible departments, resulting in a structured business description. A preliminary operational strategy is generated based on this structured business description, improving operational response speed and the intelligence level of decision support. Operational operations are executed according to the preliminary operational strategy, and business indicator data generated after a preset operational time is obtained. These operational operations include business adjustments by operational personnel based on the preliminary operational strategy, or automatic execution of operational configurations based on the preliminary operational strategy. It is determined whether the business indicator data meets preset business indicator requirements. If not, the business rule engine is optimized based on the business indicator data, and then the process returns to the previous stage. Dynamic reasoning is performed on the multi-modal fusion features using the preset business rule engine to generate a multi-dimensional structured description including business lines, functional modules, and responsible departments, resulting in a structured business description. If the description meets the requirements, the preliminary operational strategy is adopted as the final operational strategy, improving the efficiency and reliability of operational plan generation. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of an application environment for an operation plan generation method according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating an operation plan generation method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of an operation plan generation device in one embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention; Figure 5 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] The operation plan generation method provided in this embodiment of the invention can be applied to, for example, Figure 1 In this application environment, the client communicates with the server via a network. The server can obtain the user's bank card status information from the client, and determine whether there is a card on the preset card reader based on the status information. If there is no card on the preset card reader, the server returns to the steps of obtaining the user's bank card status information. If there is a card on the preset card reader, the server obtains the bank card information, reads the user's ID card to obtain ID card data, generates front and back images of the ID card based on the ID card data, obtains the key information of the preset symmetric key, obtains the plaintext key input by the user, and determines whether the key information and the plaintext key are the same. If the key information and the plaintext key are different, the user authentication is deemed to have failed. If the key information and the plaintext key are the same, the user authentication is deemed to have passed, and service access is performed based on the bank card information and the front and back images of the ID card. This ensures the reliability of user ID card verification, and finally, service access is performed based on the bank card information and the front and back images of the ID card, improving the security of financial transactions. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0014] Please see Figure 2 As shown, Figure 2 A flowchart illustrating the operation plan generation method provided in this embodiment of the invention includes the following steps: S1. Obtain screenshots of preset APP services, and perform size normalization, noise reduction, and contrast enhancement processing on the screenshots to obtain standardized images.

[0015] In this embodiment of the invention, the APP business screenshot refers to image data that reflects the interface state and business content of a specific APP at a certain moment, collected and stored in advance during the internal operation or testing process of an enterprise through manual capture, automated test script triggering, or system log recording. These screenshots usually include real user interfaces, control layouts, text information, and business elements (such as product displays, payment buttons, error pop-ups, etc.).

[0016] In this embodiment of the invention, the step of performing size normalization, noise reduction, and contrast enhancement processing on the business screenshot to obtain a standardized image includes: The screenshot of the APP service is binarized to obtain a binarized image; The binarized image is scaled to a preset target size to obtain a size-normalized image; The noise type of the size-normalized image is analyzed to obtain the noise type analysis results; Based on the noise type analysis results, a denoising algorithm is selected to denoise the size-normalized image to obtain a denoised image. The denoised image is subjected to contrast enhancement processing to obtain the standardized image.

[0017] In detail, the binarization of the APP business screenshot to obtain a binarized image involves converting the screenshot into a grayscale image of the APP business. Then, by setting a global threshold or using an adaptive thresholding algorithm, each pixel of the grayscale image is traversed, and pixels with grayscale values ​​higher than the threshold are set to white (value 255), while those with grayscale values ​​lower than the threshold are set to black (value 0).

[0018] In detail, the binarized image is scaled to a preset target size to obtain a size-normalized image. The aspect ratio of the current binarized image size to the target size (e.g., 800x600 pixels) is calculated. A scaling ratio that maintains the original aspect ratio is selected, and image scaling is performed using algorithms such as bilinear interpolation or region interpolation.

[0019] In detail, the analysis of noise type in the size-normalized image, and the resulting noise type analysis, is determined by analyzing the statistical characteristics of pixel values ​​and frequency domain features. For example, by calculating the gray-level histogram of a local region of the image, if a large number of isolated extremely black or extremely white pixels appear, it can be inferred to be salt-and-pepper noise; if the pixel value distribution exhibits a smooth Gaussian shape, it may be Gaussian noise.

[0020] In detail, the step of selecting a denoising algorithm based on the noise type analysis results to denoise the size-normalized image and obtain a denoised image involves dynamically selecting the denoising algorithm based on the noise type analysis results. If it is salt-and-pepper noise, median filtering is used, traversing the image with a sliding window and replacing the center pixel value with the median pixel value within the window. If it is Gaussian noise, Gaussian filtering is used, convolving the image with a Gaussian kernel to smooth the noise.

[0021] In detail, the contrast enhancement processing of the denoised image to obtain the standardized image is achieved by calculating the gray-level histogram of the image and mapping the original gray-level values ​​to a new distribution through the cumulative distribution function, thereby expanding the dynamic range of the image.

[0022] S2. Extract text features from the standardized image, identify the App function page type, business scenario, and visual hotspot area corresponding to the standardized image, and obtain visual features.

[0023] In this embodiment of the invention, the extraction of text features from the standardized image involves inputting the standardized image into a pre-trained OCR engine (such as a model based on the Qwen2.5-VL-32B text branch). This engine detects text regions in the image using a deep convolutional neural network and uses a recurrent neural network combined with an attention mechanism to identify character sequences within each region. It has been specifically trained to enhance the handling of unconventional fonts, artistic fonts, and low-contrast text in App interfaces. Subsequently, the system performs structured processing on the identified text, recording the content of each text block, its precise coordinate bounding box in the image, the recognition confidence, and performing basic semantic grouping (such as associating adjacent titles with the body text), ultimately outputting structured text features.

[0024] In this embodiment of the invention, the step of identifying the App function page type, business scenario, and visual hotspot area corresponding to the standardized image to obtain visual features includes: The global features of the standardized image are extracted by the convolutional layer based on the pre-trained visual model to obtain the global feature vector. The visual model is used to identify regions in the standardized image that contain preset visual elements as candidate regions, thereby obtaining a list of candidate regions; The probability of the standardized image belonging to each predefined App function page type is calculated using the fully connected layer of the visual model based on the global feature vector, and the probability calculation result is obtained. Based on the probability calculation results, the App function page type with the highest probability is confirmed as the final App function page type; Identify the business scenario label of the standardized image based on the global feature vector and the candidate region list; The visual model is used to determine whether each candidate region in the candidate region list belongs to a predefined visual hotspot, and the determination result is obtained. The visual hotspot regions of the standardized image are identified based on the judgment results. The final App function page type, the business scenario tags, and the visual hotspot areas are structurally encapsulated to obtain visual features.

[0025] In detail, the convolutional layer based on the pre-trained visual model extracts global features from the standardized image to obtain a global feature vector. This is achieved by performing sliding window calculations on the image using multiple convolutional kernels, extracting hierarchical features from low-level edges and textures to high-level semantics, and finally outputting a fixed-dimensional feature tensor. This tensor is the global feature vector that represents the overall content of the image. It compresses the visual information of the image and provides a foundation for subsequent classification and recognition.

[0026] In detail, the step of using the visual model to identify regions containing preset visual elements in the standardized image as candidate regions, and obtaining a candidate region list, involves using a Region Proposal Network (RPN) algorithm to generate multiple bounding box proposals on the image that may contain meaningful visual elements (such as buttons, icons, text blocks, etc.). Each proposal consists of a rectangle coordinate and a confidence score indicating that it contains preset visual elements (such as specific shapes, colors, textures). The system filters and sorts these proposals based on a confidence threshold, outputting a list containing multiple candidate regions.

[0027] In detail, the process of using the fully connected layers of the visual model to calculate the probability that the standardized image belongs to each predefined App function page type based on the global feature vector, and obtaining the probability calculation result, involves a non-linear transformation through one or more fully connected layers. The number of neurons in the final fully connected layer is equal to the number of predefined App function page types (e.g., homepage, product page, payment page, etc.), and the Softmax activation function is used to convert the output into a probability distribution for each category.

[0028] In detail, the step of identifying the business scenario label of the standardized image based on the global feature vector and the candidate region list involves fusing the global feature vector with the features of the candidate regions (extracted through pooling or convolution) to form a comprehensive feature representation, which is then input into a preset classifier to predict one or more business scenario labels (such as "shopping", "social", "finance") to which the image belongs.

[0029] In detail, the step of using the visual model to determine whether each candidate region in the candidate region list belongs to a predefined visual hotspot and obtaining the determination result involves first cropping each candidate region from the original image and adjusting it to the input size required by the model. Then, it is input into a binary or multi-class sub-network (which may share some convolutional layers with the main model) to determine whether the region belongs to a predefined visual hotspot category (such as an advertisement space, button area, promotional label, etc.).

[0030] S3. Perform feature fusion on the text features and the visual features to obtain multimodal fusion features.

[0031] In this embodiment of the invention, the feature fusion of the text features and the visual features to obtain multimodal fusion features includes: The text features and the visual features are aligned to obtain aligned text features and aligned visual features. Calculate the attention weights of the aligned text features and the aligned visual features, and perform feature enhancement on the aligned text features and the aligned visual features based on the attention weights to obtain enhanced text features and enhanced visual features. Construct a bilinear interaction matrix based on the enhanced text features and the enhanced visual features; The enhanced text features and the enhanced visual features are extracted based on the bilinear interaction matrix to obtain the multimodal fusion features.

[0032] In detail, the feature alignment of the text features and the visual features to obtain aligned text features and aligned visual features is achieved by processing the text features and visual features separately through two independent linear transformation layers (fully connected layers), and performing linear projection on the two types of features through independent weight matrices and bias terms, so that the output feature vector has a uniform feature dimension length.

[0033] In detail, the process involves calculating the attention weights of the aligned text features and the aligned visual features, and then performing feature enhancement on the aligned text features and the aligned visual features based on these attention weights to obtain enhanced text features and enhanced visual features. This is done by using the aligned text features as the query and the aligned visual features as the key and value, respectively. The attention weights from text to vision are calculated by: calculating the similarity between the query and the key using a dot product, dividing by a scaling factor, and then applying the Softmax function to obtain the weight distribution. These weights are then used to perform a weighted sum of the values ​​to obtain the text-side enhanced features. Similarly, using the aligned visual features as the query and the aligned text features as the key and value, the attention from vision to text is calculated to obtain the visual-side enhanced features.

[0034] In detail, the step of constructing a bilinear interaction matrix based on the enhanced text features and the enhanced visual features involves setting the enhanced text features as a d-dimensional column vector t and the enhanced visual features as a d-dimensional column vector v, and then performing an outer product operation. Specifically, this involves calculating the transpose of t multiplied by v: M = t v^T yields a d×d matrix, which is the bilinear interaction matrix.

[0035] In detail, the step of extracting the fusion features of the enhanced text features and the enhanced visual features based on the bilinear interaction matrix to obtain the multimodal fusion features involves flattening the bilinear interaction matrix M into a long vector of d^2 dimensions, mapping the bilinear interaction matrix to the target output dimension (e.g., 256 dimensions) using a weight matrix, adding a bias term and an activation function, and finally outputting a fusion feature, i.e., the multimodal fusion feature.

[0036] S4. Utilize a preset business rule engine to perform dynamic reasoning on the multimodal fusion features to generate a multi-dimensional structured description that includes business lines, functional modules, and responsible departments, thereby obtaining a structured business description.

[0037] In this embodiment of the invention, the step of using a preset business rule engine to dynamically reason about the multimodal fusion features to generate a multi-dimensional structured description including business lines, functional modules, and responsible departments, resulting in a structured business description, includes: The multimodal fusion features are parsed into a set of semantic labels containing confidence levels, resulting in a semantic label set; Search for business line nodes associated with the semantic tag set in a preset knowledge graph to obtain the business line set; In the knowledge graph, functional sub-modules associated with the business line set and the semantic tag set are identified to obtain functional module identification results; Based on the identification results of the functional modules, the responsible departments are associated in the preset enterprise organizational structure mapping table to obtain the responsible department association results; The semantic tag set, the business line set, the functional module identification results, and the responsible department association results are structurally encapsulated to obtain a structured business description.

[0038] In detail, parsing the multimodal fusion features into a set of semantic labels with confidence scores, resulting in a semantic label set, involves inputting the multimodal fusion feature vector into a pre-trained lightweight parsing network. This parsing network typically consists of fully connected layers and a Softmax classifier, and is trained on labeled data to learn how to map the fusion features to basic semantic concepts in a knowledge graph (such as UI component types "button / input box," business keywords "annualized interest rate," visual elements "ad placement," etc.). For each possible semantic label, the network outputs a confidence score between 0 and 1. The system then filters out high-confidence labels based on a preset threshold (e.g., 0.6), ultimately outputting a structured list, i.e., the semantic label set.

[0039] In detail, the step of searching for business line nodes associated with the semantic tag set in the preset knowledge graph to obtain the business line set involves performing a traversal query in a pre-built business knowledge graph (usually stored in the form of a graph database). A graph query language (such as Cypher or Gremlin) is used to find "business line" type nodes (such as "consumer finance" and "e-commerce") that are directly connected to the semantic tags in the semantic tag set through relationships such as "belongs to" or "associates with".

[0040] In detail, the process of identifying functional sub-modules associated with the business line set and the semantic tag set in the knowledge graph, and obtaining functional module identification results, involves limiting the search scope based on the business line set, and then matching the semantic tag set with the definition pattern of each functional module based on the limited search scope. Each functional module is defined in the knowledge graph as a set of required semantic tags (e.g., the "loan application" module may require the simultaneous appearance of "amount input box," "term selector," and "interest rate text"). The system uses a rule engine or graph pattern matching algorithm to find all functional modules that are fully satisfied by the current semantic tag set.

[0041] In detail, the structured encapsulation of the semantic tag set, the business line set, the functional module identification results, and the responsible department association results to obtain a structured business description involves assembling these elements according to a predefined, standardized data pattern. This data pattern defines fixed field names, data types, and nesting relationships. By creating a new data object and filling the corresponding fields with the semantic tag set, the business line set, the functional module identification results, and the responsible department association results, a structured business description is obtained.

[0042] S5. Generate a preliminary operation strategy based on the structured business description, execute operation operations based on the preliminary operation strategy, and obtain business indicator data generated after a preset operation time. The operation operations include business adjustments made by operation personnel based on the preliminary operation strategy, or automatic execution of operation configuration based on the preliminary operation strategy.

[0043] In this embodiment of the invention, generating a preliminary operation strategy based on the structured business description includes: Extract key field features from the structured business description; Based on the key field features, a set of matching strategies is obtained by performing strategy matching in a preset strategy template library. Extract strategy parameters from the structured business description, and fill the matching strategy set with parameters according to the strategy parameters to obtain a filled strategy set; Obtain historical operational data of the APP business, and prioritize the filling strategy set according to the historical operational data to obtain a sorting strategy set; The sorting strategy set is formatted and encapsulated to obtain a preliminary operation strategy.

[0044] In detail, the extraction of key field features from the structured business description is based on a predefined list of "key fields" configurations, from which specified field values ​​are extracted.

[0045] In detail, the step of performing strategy matching in a preset strategy template library based on the key field features to obtain a set of matching strategies involves using a preset matching engine (such as the Drools rule engine or a code-based logic judge) to compare the key field features one by one with the triggering conditions of all strategy templates.

[0046] In detail, the step of extracting strategy parameters from the structured business description and filling the matching strategy set with these parameters to obtain a filled strategy set involves extracting detailed parameters from the structured business description used to specify the strategy, such as specific app_name (application name), responsible_dept (responsible department), error code, user ID fragment, or interface element coordinates. These parameters correspond to variable placeholders reserved in the strategy template (e.g., "optimize loading speed for the {module} module of {app_name}"). By traversing the matching strategy set, the placeholders in each template are replaced with the extracted actual parameter values, thereby instantiating the general strategy framework into operable strategy statements specific to the current scenario.

[0047] In detail, the process of acquiring historical operational data of the APP business and prioritizing the set of population strategies based on this historical operational data to obtain a ranked set of strategies involves querying a historical operational database to obtain execution data of historical strategies similar to the current population strategy, such as average adoption rate, problem resolution rate, implementation time, and return on investment. For each strategy, a comprehensive priority score is calculated based on its type and content. This score is typically determined by multiple weighted factors, including historical performance data, the urgency of the current abnormal state (e.g., "crash" level is higher than "notification" level), the estimated scope of affected users, and the resource cost of strategy implementation. After calculation, the system sorts the set of population strategies in descending order based on the scores.

[0048] In detail, the process of formatting and encapsulating the sorting strategy set to obtain a preliminary operation strategy involves organizing each strategy in the sorting strategy set and its additional information (such as priority scores, key parameters, and historical performance references) according to a preset, unified output template.

[0049] In this embodiment of the invention, the step of executing operational operations according to the preliminary operational strategy and obtaining business indicator data generated after a preset operational time includes operational personnel making business adjustments based on the preliminary operational strategy, or automatically executing operational configurations based on the preliminary operational strategy. When the preset operational personnel adopt and implement the system-generated preliminary operational strategy (e.g., modifying APP interface elements, adjusting business processes, or launching new functions), the system will automatically collect and summarize the quantitative effect data generated by the specific APP business modules affected by the strategy within a preset observation period (e.g., 24 hours, 7 days, or a complete business cycle) after the strategy takes effect. This data is the business indicator data, which typically includes user behavior indicators (e.g., click-through rate, conversion rate, page dwell time), business performance indicators (e.g., order volume, transaction amount, credit approval rate), and system status indicators (e.g., error rate, loading time), used to objectively evaluate the actual business impact after the strategy is implemented.

[0050] S6. Determine whether the business indicator data meets the preset business indicator requirements.

[0051] In this embodiment of the invention, determining whether the business indicator data meets the preset business indicator requirements involves automatically comparing the real-time collected business indicator data (such as click-through rate and conversion rate) with the thresholds or target ranges (business indicator requirements) set in advance for different business scenarios. This comparison is based on explicit logical rules or statistical tests (such as determining whether the current value is greater than the target lower limit, whether it is within the range, or whether it has a statistically significant improvement), and outputs a binary judgment result ("meets" or "does not meet").

[0052] If it does not meet the requirements, proceed to S7, optimize the business rule engine based on the business indicator data, and then return to S4, use the preset business rule engine to perform business attribution analysis on the business screenshot based on the multimodal fusion features to obtain a structured business description.

[0053] In this embodiment of the invention, optimizing the business rule engine based on the business indicator data includes: Calculate the difference between the business indicator data and the business indicator requirements to obtain the business indicator difference, and generate an effect score based on the business indicator difference; The traceability data is obtained by generating a complete data chain during the process of tracing the business indicator data, which includes a structured business description, multimodal fusion features, and a globally unique number of the business screenshot. Analyze the contribution of the traceability data chain to the nodes in the preset knowledge graph; The direction is adjusted based on the effect score and the node contribution. The business rule engine is adjusted based on the parameter adjustment direction and the preset adjustment step size to obtain the business rule engine after preliminary adjustment. Based on the node contribution, the rule base and knowledge graph in the initially adjusted business rule engine are structurally optimized to obtain the final business rule engine.

[0054] In detail, the step of generating an effectiveness score based on the difference in business metrics involves using a preset scoring function (such as weighted summation, normalized mapping, or piecewise function) to comprehensively transform the difference in business metrics into a quantified effectiveness score. For example, if the conversion rate increases beyond the target, a positive score is generated; if the target is not met or a negative effect occurs, a negative or low score is generated.

[0055] In detail, the process of tracing the source of the business indicator data generates a complete data chain that includes a structured business description, multimodal fusion features, and a globally unique ID for the business screenshot. This results in a source data chain, which is obtained by querying and linking all relevant intermediate data in logs or databases based on a unique transaction ID that runs through the entire processing flow: from the initial original screenshot ID and upload record, to the multimodal fusion feature vector generated after parsing, to the structured business description inferred by the rule engine, and finally linked to the business indicator data for this evaluation.

[0056] In detail, the analysis of the contribution of the traceability data chain to nodes in the preset knowledge graph involves locating the specific knowledge graph nodes and paths triggered by the rule engine during this inference process (e.g., which business line nodes and functional module nodes were activated). Subsequently, attribution analysis is used to evaluate the contribution of these nodes to the final generated structured business description (and subsequent strategies). Specifically, this can be achieved by calculating the confidence of node activation, the indispensability of nodes in the inference path, or by using more complex algorithms (such as SHAP values) to quantify the impact of each node on the final output, thereby identifying key and secondary nodes.

[0057] In detail, the adjustment direction based on the effect score and the node contribution confirmation parameter is as follows: for nodes with high contribution and positive effect score (good effect) and their associated rules, the adjustment direction is "strengthening", such as increasing their weight or lowering the trigger threshold; for nodes and rules with high contribution but negative effect score (poor effect), the adjustment direction is "weakening" or "correction", such as reducing their weight or raising the trigger threshold.

[0058] In detail, the structural optimization of the rule base and knowledge graph in the initially adjusted business rule engine based on the node contribution to obtain the final business rule engine involves disabling or deleting rules and knowledge graph nodes with consistently low contribution or that have never been triggered, after evaluation, in order to simplify the system and merge redundant rules with similar logic.

[0059] If the conditions are met, then execute S8 and adopt the preliminary operation strategy as the final operation strategy.

[0060] In this embodiment of the invention, when the system determines that the business indicator data meets the preset requirements, the preliminary operation strategy will be automatically marked as "verified effective". The system will move it from the suggestion pool to the formal strategy library and complete the format standardization to adapt to the downstream execution system (such as the work order system, A / B testing platform). At the same time, the system will automatically push the strategy details, implementation instructions and correlation analysis report to the designated responsible personnel or departments through the preset interface, and start the tracking process of the strategy implementation status, thereby completing the transformation from machine-generated suggestions to deployable and traceable formal operation decisions.

[0061] As can be seen, in the above scheme, the original screenshot is binarized, its size normalized, noise analyzed and removed, and contrast enhanced to obtain a standardized image. Subsequently, a pre-trained OCR engine extracts the text content and location from the image as text features, while a visual model is used to identify the App's functional page types, business scenarios, and visual hotspots to form visual features. The key innovation lies in the deep fusion of text and visual features into a unified multimodal feature through feature alignment, cross-modal attention mechanisms, and a bilinear interaction matrix. Next, based on this feature, a business rule engine and knowledge graph are used for dynamic reasoning to automatically generate a structured business description containing business lines, functional modules, and responsible departments. Based on this description, a preliminary operational strategy is generated by matching strategy templates, filling in parameters, and sorting. After the strategy is implemented by operations personnel, the system collects business indicator data within a preset time period and determines whether it meets the requirements. If it does not meet the requirements, the system traces back to the source based on the indicator data, analyzes the contribution of knowledge graph nodes, and optimizes the parameters and structure of the rule engine to achieve self-evolution and re-analysis; if it meets the requirements, the strategy is solidified into the final solution, improving the reliability of the generated operational plan.

[0062] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0063] In one embodiment, an operation plan generation device is provided, which corresponds one-to-one with the operation plan generation method in the above embodiments. For example... Figure 3 As shown, the operation plan generation device includes an image processing module 101, a feature extraction module 102, a dynamic reasoning module 103, a business adjustment module 104, and an indicator judgment module 105. Detailed descriptions of each functional module are as follows: Image processing module 101 is used to acquire screenshots of preset APP services, and perform size normalization, noise reduction and contrast enhancement processing on the screenshots to obtain standardized images. The feature extraction module 102 is used to extract text features from the standardized image, identify the App function page type, business scenario and visual hotspot area corresponding to the standardized image, and obtain visual features; The dynamic reasoning module 103 is used to perform feature fusion on the text features and the visual features to obtain multimodal fusion features, and to perform dynamic reasoning on the multimodal fusion features using a preset business rule engine to generate a multi-dimensional structured description containing business lines, functional modules and responsible departments, thereby obtaining a structured business description. The business adjustment module 104 is used to generate a preliminary operation strategy based on the structured business description, execute operation operations based on the preliminary operation strategy, and obtain business indicator data generated after a preset operation time. The operation operations include business adjustments made by operation personnel based on the preliminary operation strategy, or automatic execution of operation configuration based on the preliminary operation strategy. The indicator judgment module 105 is used to determine whether the business indicator data meets the preset business indicator requirements. If it does not meet the requirements, the business rule engine is optimized based on the business indicator data, and then the preset business rule engine is used to perform dynamic reasoning on the multimodal fusion features to generate a multi-dimensional structured description containing business lines, functional modules and responsible departments, thereby obtaining a structured business description. If it meets the requirements, the preliminary operation strategy is used as the final operation strategy.

[0064] In one embodiment, when performing the size normalization, noise reduction, and contrast enhancement processing on the business screenshot to obtain a standardized image, the image processing module 101 is specifically used for: The screenshot of the APP service is binarized to obtain a binarized image; The binarized image is scaled to a preset target size to obtain a size-normalized image; The noise type of the size-normalized image is analyzed to obtain the noise type analysis results; Based on the noise type analysis results, a denoising algorithm is selected to denoise the size-normalized image to obtain a denoised image. The denoised image is subjected to contrast enhancement processing to obtain the standardized image.

[0065] In one embodiment, when the feature extraction module 102 performs the step of identifying the App function page type, business scenario, and visual hotspot region corresponding to the standardized image to obtain visual features, it is specifically used for: The global features of the standardized image are extracted by the convolutional layer based on the pre-trained visual model to obtain the global feature vector. The visual model is used to identify regions in the standardized image that contain preset visual elements as candidate regions, thereby obtaining a list of candidate regions; The probability of the standardized image belonging to each predefined App function page type is calculated using the fully connected layer of the visual model based on the global feature vector, and the probability calculation result is obtained. Based on the probability calculation results, the App function page type with the highest probability is confirmed as the final App function page type; Identify the business scenario label of the standardized image based on the global feature vector and the candidate region list; The visual model is used to determine whether each candidate region in the candidate region list belongs to a predefined visual hotspot, and the determination result is obtained. The visual hotspot regions of the standardized image are identified based on the judgment results. The final App function page type, the business scenario tags, and the visual hotspot areas are structurally encapsulated to obtain visual features.

[0066] In one embodiment, the dynamic reasoning module 103, when performing feature fusion on the text features and the visual features to obtain multimodal fused features, is specifically used for: The text features and the visual features are aligned to obtain aligned text features and aligned visual features. Calculate the attention weights of the aligned text features and the aligned visual features, and perform feature enhancement on the aligned text features and the aligned visual features based on the attention weights to obtain enhanced text features and enhanced visual features. Construct a bilinear interaction matrix based on the enhanced text features and the enhanced visual features; The enhanced text features and the enhanced visual features are extracted based on the bilinear interaction matrix to obtain the multimodal fusion features.

[0067] In one embodiment, the dynamic reasoning module 103, when performing dynamic reasoning on the multimodal fusion features using a preset business rule engine to generate a multi-dimensional structured description including business lines, functional modules, and responsible departments, and obtaining a structured business description, is specifically used for: The multimodal fusion features are parsed into a set of semantic labels containing confidence levels, resulting in a semantic label set; Search for business line nodes associated with the semantic tag set in a preset knowledge graph to obtain the business line set; In the knowledge graph, functional sub-modules associated with the business line set and the semantic tag set are identified to obtain functional module identification results; Based on the identification results of the functional modules, the responsible departments are associated in the preset enterprise organizational structure mapping table to obtain the responsible department association results; The semantic tag set, the business line set, the functional module identification results, and the responsible department association results are structurally encapsulated to obtain a structured business description.

[0068] In one embodiment, the business adjustment module 104, when executing the generation of a preliminary operation strategy based on the structured business description, is specifically used for: Extract key field features from the structured business description; Based on the key field features, a set of matching strategies is obtained by performing strategy matching in a preset strategy template library. Extract strategy parameters from the structured business description, and fill the matching strategy set with parameters according to the strategy parameters to obtain a filled strategy set; Obtain historical operational data of the APP business, and prioritize the filling strategy set according to the historical operational data to obtain a sorting strategy set; The sorting strategy set is formatted and encapsulated to obtain a preliminary operation strategy.

[0069] In one embodiment, the indicator judgment module 105, when executing the optimization of the business rule engine based on the business indicator data, is specifically used for: Calculate the difference between the business indicator data and the business indicator requirements to obtain the business indicator difference, and generate an effect score based on the business indicator difference; The traceability data is obtained by generating a complete data chain during the process of tracing the business indicator data, which includes a structured business description, multimodal fusion features, and a globally unique number of the business screenshot. Analyze the contribution of the traceability data chain to the nodes in the preset knowledge graph; The direction is adjusted based on the effect score and the node contribution. The business rule engine is adjusted based on the parameter adjustment direction and the preset adjustment step size to obtain the business rule engine after preliminary adjustment. Based on the node contribution, the rule base and knowledge graph in the initially adjusted business rule engine are structurally optimized to obtain the final business rule engine.

[0070] This invention provides an operational plan generation device that binarizes, normalizes, analyzes and removes noise, and enhances contrast of original screenshots to obtain standardized images. Subsequently, a pre-trained OCR engine extracts text content and location from the image as text features, while a visual model identifies App function page types, business scenarios, and visual hotspots to form visual features. The key innovation lies in the deep fusion of text and visual features into unified multimodal features through feature alignment, cross-modal attention mechanisms, and a bilinear interaction matrix. Next, based on these features, a business rule engine and knowledge graph are used for dynamic reasoning to automatically generate a structured business description containing business lines, functional modules, and responsible departments. Based on this description, a preliminary operational strategy is generated by matching strategy templates, filling parameters, and sorting. After the strategy is implemented by operations personnel, the system collects business indicator data within a preset time period and determines whether it meets the requirements. If it does not meet the requirements, the system traces back to the source based on the indicator data, analyzes the contribution of knowledge graph nodes, and optimizes the parameters and structure of the rule engine to achieve self-evolution and re-analysis; if it meets the requirements, the strategy is solidified into the final plan, improving the reliability of operational plan generation.

[0071] Specific limitations regarding the operation plan generation device can be found in the limitations of the operation plan generation method described above, and will not be repeated here. Each module in the aforementioned operation plan generation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0072] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown. The computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of an operation plan generation method on the server side.

[0073] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 5As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps on the client side of an operation plan generation method.

[0074] In one embodiment, a computer device is provided, 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 perform the following steps: Obtain screenshots of preset APP services, and perform size normalization, noise reduction, and contrast enhancement processing on the screenshots to obtain standardized images; Extract text features from the standardized image, identify the App function page type, business scenario, and visual hotspot area corresponding to the standardized image, and obtain visual features; The text features and the visual features are fused to obtain multimodal fused features; The multimodal fusion features are dynamically reasoned using a preset business rule engine to generate a multi-dimensional structured description that includes business lines, functional modules and responsible departments, thus obtaining a structured business description. A preliminary operation strategy is generated based on the structured business description. Operational operations are executed based on the preliminary operation strategy. Business indicator data generated after a preset operation time is obtained. The operational operations include business adjustments made by operations personnel based on the preliminary operation strategy, or automatic execution of operational configurations based on the preliminary operation strategy. Determine whether the business indicator data meets the preset business indicator requirements; If it does not meet the requirements, the business rule engine is optimized based on the business indicator data, and then the process is returned to the previous step of using the preset business rule engine to perform dynamic reasoning on the multimodal fusion features to generate a multi-dimensional structured description containing business lines, functional modules and responsible departments, thereby obtaining a structured business description. If the conditions are met, the preliminary operational strategy will be adopted as the final operational strategy.

[0075] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Obtain screenshots of preset APP services, and perform size normalization, noise reduction, and contrast enhancement processing on the screenshots to obtain standardized images; Extract text features from the standardized image, identify the App function page type, business scenario, and visual hotspot area corresponding to the standardized image, and obtain visual features; The text features and the visual features are fused to obtain multimodal fused features; The multimodal fusion features are dynamically reasoned using a preset business rule engine to generate a multi-dimensional structured description that includes business lines, functional modules and responsible departments, thus obtaining a structured business description. A preliminary operation strategy is generated based on the structured business description. Operational operations are executed based on the preliminary operation strategy. Business indicator data generated after a preset operation time is obtained. The operational operations include business adjustments made by operations personnel based on the preliminary operation strategy, or automatic execution of operational configurations based on the preliminary operation strategy. Determine whether the business indicator data meets the preset business indicator requirements; If it does not meet the requirements, the business rule engine is optimized based on the business indicator data, and then the process is returned to the previous step of using the preset business rule engine to perform dynamic reasoning on the multimodal fusion features to generate a multi-dimensional structured description containing business lines, functional modules and responsible departments, thereby obtaining a structured business description. If the conditions are met, the preliminary operational strategy will be adopted as the final operational strategy.

[0076] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0077] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0078] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0079] Finally, it should be noted that if any software tools or components not belonging to this company appear in the embodiments of the application, they are merely illustrative examples and do not represent actual use. The embodiments described above are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for generating an operation plan, characterized in that, include: Obtain screenshots of preset APP services, and perform size normalization, noise reduction, and contrast enhancement processing on the screenshots to obtain standardized images; Extract text features from the standardized image, identify the App function page type, business scenario, and visual hotspot area corresponding to the standardized image, and obtain visual features; The text features and the visual features are fused to obtain multimodal fused features; The multimodal fusion features are dynamically reasoned using a preset business rule engine to generate a multi-dimensional structured description that includes business lines, functional modules and responsible departments, thus obtaining a structured business description. A preliminary operation strategy is generated based on the structured business description. Operational operations are executed based on the preliminary operation strategy. Business indicator data generated after a preset operation time is obtained. The operational operations include business adjustments made by operations personnel based on the preliminary operation strategy, or automatic execution of operational configurations based on the preliminary operation strategy. Determine whether the business indicator data meets the preset business indicator requirements; If it does not meet the requirements, the business rule engine is optimized based on the business indicator data, and then the process is returned to the previous step of using the preset business rule engine to perform dynamic reasoning on the multimodal fusion features to generate a multi-dimensional structured description containing business lines, functional modules and responsible departments, thereby obtaining a structured business description. If the conditions are met, the preliminary operational strategy will be adopted as the final operational strategy.

2. The operation plan generation method as described in claim 1, characterized in that, The process of normalizing the size, denoising, and enhancing the contrast of the screenshot to obtain a standardized image includes: The screenshot of the APP service is binarized to obtain a binarized image; The binarized image is scaled to a preset target size to obtain a size-normalized image; The noise type of the size-normalized image is analyzed to obtain the noise type analysis results; Based on the noise type analysis results, a denoising algorithm is selected to denoise the size-normalized image to obtain a denoised image. The denoised image is subjected to contrast enhancement processing to obtain the standardized image.

3. The method for generating an operation plan as described in claim 1, characterized in that, The process of identifying the App function page type, business scenario, and visual hotspot area corresponding to the standardized image to obtain visual features includes: The global features of the standardized image are extracted by the convolutional layer based on the pre-trained visual model to obtain the global feature vector. The visual model is used to identify regions in the standardized image that contain preset visual elements as candidate regions, thereby obtaining a list of candidate regions; The probability of the standardized image belonging to each predefined App function page type is calculated using the fully connected layer of the visual model based on the global feature vector, and the probability calculation result is obtained. Based on the probability calculation results, the App function page type with the highest probability is confirmed as the final App function page type; Identify the business scenario label of the standardized image based on the global feature vector and the candidate region list; The visual model is used to determine whether each candidate region in the candidate region list belongs to a predefined visual hotspot, and the determination result is obtained. The visual hotspot regions of the standardized image are identified based on the judgment results. The final App function page type, the business scenario tags, and the visual hotspot areas are structurally encapsulated to obtain visual features.

4. The operation plan generation method as described in claim 1, characterized in that, The feature fusion of the text features and the visual features to obtain multimodal fusion features includes: The text features and the visual features are aligned to obtain aligned text features and aligned visual features. Calculate the attention weights of the aligned text features and the aligned visual features, and perform feature enhancement on the aligned text features and the aligned visual features based on the attention weights to obtain enhanced text features and enhanced visual features. Construct a bilinear interaction matrix based on the enhanced text features and the enhanced visual features; The enhanced text features and the enhanced visual features are extracted based on the bilinear interaction matrix to obtain the multimodal fusion features.

5. The method for generating an operation plan as described in claim 1, characterized in that, The process involves using a pre-defined business rule engine to dynamically reason about the multimodal fusion features to generate a multi-dimensional structured description that includes business lines, functional modules, and responsible departments, resulting in a structured business description, including: The multimodal fusion features are parsed into a set of semantic labels containing confidence levels, resulting in a semantic label set; Search for business line nodes associated with the semantic tag set in a preset knowledge graph to obtain the business line set; In the knowledge graph, functional sub-modules associated with the business line set and the semantic tag set are identified to obtain functional module identification results; Based on the identification results of the functional modules, the responsible departments are associated in the preset enterprise organizational structure mapping table to obtain the responsible department association results; The semantic tag set, the business line set, the functional module identification results, and the responsible department association results are structurally encapsulated to obtain a structured business description.

6. The method for generating an operation plan as described in claim 1, characterized in that, The step of generating a preliminary operational strategy based on the structured business description includes: Extract key field features from the structured business description; Based on the key field features, a set of matching strategies is obtained by performing strategy matching in a preset strategy template library. Extract strategy parameters from the structured business description, and fill the matching strategy set with parameters according to the strategy parameters to obtain a filled strategy set; Obtain historical operational data of the APP business, and prioritize the filling strategy set according to the historical operational data to obtain a sorting strategy set; The sorting strategy set is formatted and encapsulated to obtain a preliminary operation strategy.

7. The method for generating an operation plan as described in claim 1, characterized in that, The step of optimizing the business rule engine based on the business metric data includes: Calculate the difference between the business indicator data and the business indicator requirements to obtain the business indicator difference, and generate an effect score based on the business indicator difference; The traceability data is obtained by generating a complete data chain during the process of tracing the business indicator data, which includes a structured business description, multimodal fusion features, and a globally unique number of the business screenshot. Analyze the contribution of the traceability data chain to the nodes in the preset knowledge graph; The direction is adjusted based on the effect score and the node contribution. The business rule engine is adjusted based on the parameter adjustment direction and the preset adjustment step size to obtain the business rule engine after preliminary adjustment. Based on the node contribution, the rule base and knowledge graph in the initially adjusted business rule engine are structurally optimized to obtain the final business rule engine.

8. An operation plan generation device, characterized in that, include: The image processing module is used to acquire screenshots of preset APP services, and to perform size normalization, noise reduction and contrast enhancement processing on the screenshots to obtain standardized images. The feature extraction module is used to extract text features from the standardized image, identify the App function page type, business scenario and visual hotspot area corresponding to the standardized image, and obtain visual features; The dynamic reasoning module is used to perform feature fusion on the text features and the visual features to obtain multimodal fusion features. The module then uses a preset business rule engine to perform dynamic reasoning on the multimodal fusion features to generate a multi-dimensional structured description that includes business lines, functional modules and responsible departments, thus obtaining a structured business description. The business adjustment module is used to generate a preliminary operation strategy based on the structured business description, execute operation operations based on the preliminary operation strategy, and obtain business indicator data generated after a preset operation time. The operation operations include business adjustments made by operation personnel based on the preliminary operation strategy, or automatic execution of operation configuration based on the preliminary operation strategy. The indicator judgment module is used to determine whether the business indicator data meets the preset business indicator requirements. If it does not meet the requirements, the business rule engine is optimized based on the business indicator data, and then the preset business rule engine is used to perform dynamic reasoning on the multimodal fusion features to generate a multi-dimensional structured description containing business lines, functional modules and responsible departments, thus obtaining a structured business description. If it meets the requirements, the preliminary operation strategy is used as the final operation strategy.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the operation plan generation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, The computer program is processed The steps of the operation plan generation method as described in any one of claims 1 to 7 are implemented during execution.

Images