Cross-platform service promotion method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of cross-platform business promotion, and discloses a cross-platform business promotion method and device, electronic equipment and a storage medium, and the method comprises the steps: carrying out the semantic and intention analysis of a natural language instruction, and extracting a structured task element; obtaining rules of each target platform, generating a unified structured task vector based on the elements and the rules, and inputting a preset multi-modal large language model to generate a content strategy adaptive to each platform; and then a cross-platform delivery strategy is formed and popularized in combination with historical platform data of each platform. The method has the beneficial technical effects that the period from demand to content and delivery scheme formation is shortened, the labor cost of platform docking and material adaptation is reduced, the consistency and compliance of cross-platform contents are improved, natural language service instructions are organically connected with multi-target platform rules and historical data, and the method is suitable for application and popularization. And automatic generation and cross-platform deployment of the popularization strategy are realized.

Description

Technical Field

[0001] This invention relates to the field of cross-platform business promotion technology, and in particular to a cross-platform business promotion method, apparatus, electronic device and storage medium. Background Technology

[0002] With the widespread adoption of mobile internet and multi-channel publishing platforms, businesses face significant technical and operational bottlenecks when promoting their products across multiple target platforms. Traditional solutions require manually writing extensive amounts of adaptation code to interface with various platform APIs and creating separate material templates for different channels, resulting in long development cycles, high costs, and poor cross-platform consistency. Furthermore, each publishing platform exhibits significant differences in material specifications, duration, tags, and policy constraints, making manual adjustments time-consuming and prone to errors. Summary of the Invention

[0003] Based on this, it is necessary to propose a cross-platform business promotion method, device, electronic device and storage medium to address the existing cross-platform business promotion problem.

[0004] A cross-platform business promotion method, the method comprising: It receives natural language business instructions from users and targets multiple business platforms for promotion. Semantic and intent parsing is performed on the natural language business instructions to extract structured task elements; Obtain the platform rules of each of the target business platforms, and generate a structured task vector based on the structured task elements and the platform rules; The structured task vector is input into a preset multimodal large language model generation engine to generate a multimodal business strategy solution; Obtain historical platform data for each of the target business platforms, and generate a cross-platform delivery strategy based on the historical platform data and the multimodal business strategy scheme; The cross-platform deployment strategy is used to deploy the multimodal business strategy solution to various target business platforms for promotion.

[0005] Furthermore, the step of deploying the multimodal business strategy solution to various target business platforms for promotion through the cross-platform delivery strategy includes: Obtain real-time feedback data after the execution of the multimodal service strategy scheme; The cross-platform delivery strategy is dynamically optimized in real time based on the real-time feedback data to obtain a real-time cross-platform delivery strategy. Based on the real-time cross-platform deployment strategy, the multimodal business strategy solution is deployed to each of the target business platforms for promotion.

[0006] Furthermore, the step of dynamically optimizing the cross-platform delivery strategy in real time based on the real-time feedback data to obtain a real-time cross-platform delivery strategy includes: Construct a strategy optimization model with click-through rate, conversion rate, and return on investment as optimization objectives; Based on the real-time feedback data, the policy optimization model is iteratively updated using a reinforcement learning algorithm. The resource allocation adjustment decisions for each target business platform are output through the iteratively updated strategy optimization model. Based on the resource allocation adjustment decision, the resource allocation ratio among the target business platforms is dynamically adjusted to obtain a real-time cross-platform deployment strategy.

[0007] Further, the step of obtaining the platform rules of each of the target business platforms and generating a structured task vector based on the structured task elements and the platform rules includes: The system queries a pre-built platform rules knowledge base to obtain the platform rules for each of the target business platforms; wherein, the platform rules include application programming interface specifications, content format requirements, and delivery strategy constraints. Based on the structured task elements, the acquired platform rules are filtered and feature extracted to obtain platform rule features; The platform rule features and the structured task elements are fused and encoded to generate a unified structured task vector.

[0008] Furthermore, after the step of inputting the structured task vector into a preset multimodal large language model generation engine to generate a multimodal business strategy solution, the method further includes: Obtain target region information from the structured task elements; Call the regional cultural feature database corresponding to the aforementioned regional information; Based on the cultural taboos and compliance rules in the regional cultural feature database, the generated multimodal business strategy scheme is adjusted for compliance.

[0009] Furthermore, the step of generating a cross-platform delivery strategy based on the historical platform data and the multimodal business strategy scheme includes: Based on the historical platform data, predict the estimated effect of achieving the business objectives on each target business platform; Based on the estimated results and the resource constraints in the structured task elements, an initial budget is allocated to each target business platform, a bidding strategy is set, and a delivery period is planned to obtain a cross-platform delivery strategy.

[0010] Furthermore, the step of performing semantic and intent parsing on the natural language business instructions to extract structured task elements includes: The natural language business instructions are input into a preset cross-language pre-trained model for processing to identify the instruction information in the natural language business instructions; wherein, the instruction information includes business objectives, target regions, resource constraints, and promotion target information; The instruction information is formatted into the structured task elements of a preset structure.

[0011] A cross-platform business promotion device, the device comprising: The receiving module is used to receive natural language business instructions input by the user and to respond to multiple target business platforms applied for promotion. The extraction module is used to perform semantic and intent parsing on the natural language business instructions and extract structured task elements; The acquisition module is used to acquire the platform rules of each of the target business platforms and generate a structured task vector based on the structured task elements and the platform rules. The input module is used to input the structured task vector into a preset multimodal large language model generation engine to generate a multimodal business strategy scheme. The generation module is used to acquire historical platform data of each of the target business platforms and generate a cross-platform delivery strategy based on the historical platform data and the multimodal business strategy scheme. The deployment module is used to deploy the multimodal business strategy solution to each of the target business platforms for promotion through the cross-platform delivery strategy.

[0012] An electronic device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the following steps: It receives natural language business instructions from users and targets multiple business platforms for promotion. Semantic and intent parsing is performed on the natural language business instructions to extract structured task elements; Obtain the platform rules of each of the target business platforms, and generate a structured task vector based on the structured task elements and the platform rules; The structured task vector is input into a preset multimodal large language model generation engine to generate a multimodal business strategy solution; Obtain historical platform data for each of the target business platforms, and generate a cross-platform delivery strategy based on the historical platform data and the multimodal business strategy scheme; The cross-platform deployment strategy is used to deploy the multimodal business strategy solution to various target business platforms for promotion.

[0013] A computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the following steps: It receives natural language business instructions from users and targets multiple business platforms for promotion. Semantic and intent parsing is performed on the natural language business instructions to extract structured task elements; Obtain the platform rules of each of the target business platforms, and generate a structured task vector based on the structured task elements and the platform rules; The structured task vector is input into a preset multimodal large language model generation engine to generate a multimodal business strategy solution; Obtain historical platform data for each of the target business platforms, and generate a cross-platform delivery strategy based on the historical platform data and the multimodal business strategy scheme; The cross-platform deployment strategy is used to deploy the multimodal business strategy solution to various target business platforms for promotion.

[0014] The beneficial effects of this invention are as follows: Semantic and intent parsing of natural language instructions extracts structured task elements; rules for each target platform are then acquired, and a unified structured task vector is generated based on these elements and rules. This vector is then input into a pre-defined multimodal large language model to generate content strategies adapted to each platform; subsequently, historical platform data is combined to form a cross-platform delivery strategy and promote it. This shortens the cycle from demand to content and delivery plan formation, reduces the manpower costs of platform integration and material adaptation, improves cross-platform content consistency and compliance, and enhances resource allocation efficiency and promotional effectiveness by utilizing delivery strategies supported by historical data. It organically links natural language business instructions with multi-target platform rules and historical data to achieve automated generation and cross-platform deployment of promotion strategies. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art 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.

[0016] in: Figure 1 This is a diagram illustrating the application environment of a cross-platform business promotion method in one embodiment. Figure 2 This is a flowchart of a cross-platform business promotion method in one embodiment; Figure 3 This is a structural block diagram of a cross-platform business promotion device in one embodiment; Figure 4 This is a structural block diagram of an electronic device in one embodiment. Detailed Implementation

[0017] 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 embodiments of the present invention, and not all embodiments. 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.

[0018] Figure 1 This is a diagram illustrating a cross-platform business promotion and application environment in one embodiment. (Refer to...) Figure 1 This cross-platform business promotion method is applied to a cross-platform business promotion system. The system includes a terminal 110 and a server 120. The terminal 110 and server 120 are connected via a network. The terminal 110 can be a desktop terminal or a mobile terminal; a mobile terminal can be at least one of a mobile phone, tablet, or laptop. The server 120 can be a standalone server or a server cluster consisting of multiple servers. The terminal 110 is used to generate cross-platform advertising strategies, and the server 120 is used to promote these strategies.

[0019] like Figure 2 As shown, in one embodiment, a cross-platform business promotion method is provided. This method can be applied to both terminals and servers; this embodiment uses terminal application as an example. The cross-platform business promotion method specifically includes the following steps: S1: Receives natural language business instructions input by the user and multiple target business platforms applied for promotion; S2: Perform semantic and intent parsing on the natural language business instructions to extract structured task elements; S3: Obtain the platform rules of each of the target business platforms, and generate a structured task vector based on the structured task elements and the platform rules; S4: Input the structured task vector into the preset multimodal large language model generation engine to generate a multimodal business strategy scheme; S5: Obtain historical platform data for each of the target business platforms, and generate a cross-platform delivery strategy based on the historical platform data and the multimodal business strategy scheme; S6: Deploy the multimodal business strategy solution to each of the target business platforms for promotion through the cross-platform delivery strategy.

[0020] As described in step S1 above, the system receives natural language business instructions input by the user and multiple target business platforms to be promoted. These instructions can be received through various access methods, such as text box input, file upload, speech-to-text results, or structured requests via API. User instructions include not only promotional goals (e.g., target audience, budget limit, time period, KPIs) but also a list of target business platforms to be deployed (e.g., platform identifiers, platform account credentials, or authorization tokens). Upon receiving the instructions, the system performs format verification and security checks (verifying token validity, filtering sensitive words, input length, and encoding), and identifies and standardizes the platform list (mapping different expressions to a unified internal channel ID). If the user does not explicitly specify a platform, candidate platform recommendations can be triggered, for example, intelligently selecting multiple target business platforms based on the instruction information.

[0021] As described in step S2 above, the natural language business instructions are semantically and intently parsed to extract structured task elements. The received natural language text undergoes multi-level parsing: first, the text is cleaned (denoising, error correction, and standardization) by the word segmentation, language recognition, and preprocessing modules; then, a finely tuned cross-language pre-trained model (e.g., XLM-RoBERTa) is called to perform intent classification and entity extraction, identifying elements such as business objectives, target regions, delivery time windows, budget / resource constraints, promotional object attributes, and compliance / prohibition items. The parsing results are then processed by a rule and confidence verification module; low-confidence items trigger supplementary interactions or secondary parsing. Finally, the extracted information is mapped to a predefined data structure (structured task elements), including field definitions, priorities, optional value sets, and uncertainty annotations, so that downstream modules can make decisions and perform constraint satisfaction checks. Specifically, semantic and intent parsing can use a finely tuned BERT or XLM-RoBERTa model for named entity recognition and intent classification, outputting a structured task element JSON object.

[0022] As described in step S3 above, the platform rules of each target business platform are obtained, and a structured task vector is generated based on the structured task elements and the platform rules. Technical and compliance rules (such as material format, duration threshold, tag / topic restrictions, API rate, and placement qualification requirements) of each target platform are retrieved from the platform rule knowledge base or the platform's real-time interface. The rules are then filtered and feature extracted, and rule items related to the current task are abstracted into standardized features (such as maximum video duration, supported languages, sensitive word list, bidding model type, etc.). Next, the structured task elements and these platform rule features are fused and encoded: feature engineering or a neural encoder is used to convert heterogeneous information such as text / numerical / Boolean data into a unified vector representation, forming a structured task vector. This vector contains both business intent information and embeds platform constraints and executable parameter space, providing the generation engine with machine-readable constraints and optimization dimensions. Conflict points and priorities are also marked in the vector for subsequent decision-making.

[0023] As described in step S4 above, the structured task vector is input into a preset multimodal large language model generation engine to generate a multimodal business strategy solution. The structured task vector, along with necessary context (platform rule summaries, cultural compliance points, historical template examples), is used as input to the multimodal large language model. Through prompt engineering or conditional decoding, differentiated content strategies and creative templates (text copy, image / video scripts, audio prompts, interaction designs, etc.) for each platform are generated. The generation engine supports multimodal input / output and can embed metadata (such as language version, compliance tags, size / duration specifications) in the output. The generation process can undergo multiple iterations. After the initial model output is completed, it is automatically reviewed by a compliance detector and rule validator. Non-compliant items trigger model correction, and manual review can be triggered based on confidence thresholds. Finally, a structured multimodal business strategy solution is produced, serving as an executable content blueprint for use by the delivery strategy module.

[0024] As described in step S5 above, historical platform data for each of the target business platforms is obtained, and a cross-platform delivery strategy is generated based on the historical platform data and the multimodal business strategy scheme. First, relevant historical data, including historical click-through rates, conversion rates, cost per click, audience response characteristics, and time-period distribution, is obtained from the internal historical database or the data APIs of each platform. The data is then cleaned, normalized, and feature-engineered (e.g., time-series aggregation, regression features, audience segmentation). Subsequently, based on these historical performance indicators and the content / channel matching prediction model in the multimodal business strategy scheme, the expected performance of different content on each platform is evaluated. Combining the budget and resource constraints in the structured task elements, an optimization algorithm (e.g., linear programming, heuristic allocation, or reinforcement learning initialization strategy) is used to generate a cross-platform delivery strategy. The initial budget allocation, bidding type, delivery rhythm, and target audience targeting parameters for each platform are clarified, and quantifiable expected KPIs and uncertainty estimates are output for deployment and subsequent optimization.

[0025] As described in step S6 above, the multimodal business strategy is deployed to various target business platforms for promotion through the cross-platform delivery strategy. During the deployment phase, the cross-platform delivery strategy and multimodal content strategy are converted into executable API calls and resource packages for each platform: this includes formatting text and materials to target specifications, generating platform-specific delivery plans, and encapsulating them into execution instruction sequences. Lightweight containers or cloud function mechanisms are used to manage execution components during deployment, handling authentication, rate limiting, and retry logic. The plans are then distributed to each platform. After deployment, monitoring and closed-loop operation are implemented: real-time delivery feedback metrics are collected and reported to the optimization module to enable strategy adjustments or rollbacks. Simultaneously, compliance and security checks (such as regional qualifications and sensitive content blocking) are performed during deployment, and audit logs are retained to ensure traceability and incident recovery.

[0026] In one embodiment, step S6, which involves deploying the multimodal business strategy to various target business platforms for promotion using the cross-platform delivery strategy, includes: S601: Obtain real-time feedback data after the execution of the multimodal service strategy scheme; S602: Based on the real-time feedback data, the cross-platform delivery strategy is dynamically optimized in real time to obtain a real-time cross-platform delivery strategy; S603: Based on the real-time cross-platform deployment strategy, the multimodal business strategy solution is deployed to each of the target business platforms for promotion.

[0027] As described in step S601 above, real-time feedback data after the execution of the multimodal business strategy is obtained. A continuous connection is established with the runtime interfaces of each target business platform to collect feedback data during the campaign process in real time, including but not limited to: impressions, clicks, click-through rate (CTR), conversions / conversion rate (CVR), display time distribution, user interaction behavior (such as likes, comments, and shares), cost, reach rate, and platform-returned error / rejection reasons. To ensure data availability, a multi-channel collection mechanism (such as platform webhooks, API polling, log push, and third-party data aggregation services) needs to be implemented. Preprocessing (timestamp unification, ID mapping, deduplication, and packet loss retries) is performed at the collection end. The collected raw data enters a stream processing pipeline for cleaning (outlier filtering, missing value imputation), aggregation (grouping and summarizing by time window, creative version, and audience), and normalization (unit conversion, distribution standardization). The data is then written to a low-latency time-series database and a short-term cache to support online optimization. Simultaneously, platform privacy and compliance requirements must be followed (such as user data anonymization, geographical restrictions, and access permission verification), and audit logs must be recorded for traceability and retraining.

[0028] As described in step S602 above, the cross-platform advertising strategy is dynamically optimized in real time based on the real-time feedback data to obtain a real-time cross-platform advertising strategy. Using real-time or near-real-time data streams as input, the current advertising strategy is evaluated and updated using predefined target metrics and reward functions. First, a configured reward function is constructed or loaded. This function can integrate multiple metrics such as CTR, CVR, cost per conversion (CPA), return on investment (ROI), and brand exposure quality, and assign weights to different targets. Online or incremental learning algorithms (such as online reinforcement learning, constrained strategy gradients, actor-critic structures, or Bayesian optimization-based rapid testers) are used to iteratively update the strategy model, generating optimization suggestions for parameters such as inter-platform resource allocation, bid adjustments, audience targeting correction, and display frequency control. Safety constraints and cold start protection are incorporated into the optimization process (such as budget caps, minimum exposure guarantees, and thresholds for sharp fluctuations). The effects of the changes are automatically verified through A / B or multi-armed slot machine experiments. To reduce risks, updates are divided into suggestion and execution layers: first, they are verified in a sandbox or low-traffic canary environment. If the effect is improved and there are no compliance issues, they are promoted to the full traffic. All optimization actions, causal basis, and performance changes are recorded for rollback and model retraining.

[0029] As described in step S603 above, the multimodal business strategy is deployed to each of the target business platforms for promotion based on the real-time cross-platform delivery strategy. The optimized real-time delivery strategy parameters are combined with the multimodal content strategy and transformed into deployment instructions executable by each target platform. First, the materials are re-rendered and adapted according to the platform rules (size, encoding, duration, copy localization and compliance labeling), and a platform-specific API call sequence and delivery plan (including budget allocation table, bidding strategy, audience targeting, delivery time period and frequency limit) are generated. The deployment adopts a progressive release mechanism: the update is first sent to a lightweight container or cloud function, and these execution units call the target platform interface to place orders or update ad groups according to the plan; at the same time, the delivery results (order success rate, response latency, platform return errors) are monitored and idempotent retry or rollback strategies are executed when an anomaly occurs. Deployment also includes version management (release version number, change description), Canary testing and tiered deployment (small traffic → scaling up → full deployment), and linked alarms (instant notification of manual intervention for KPI drops or compliance alarms). The operational data generated during the deployment process flows back to the monitoring and optimization module, forming a continuous closed loop, enabling the deployment to respond quickly to real-time optimization decisions and operate stably within the boundaries of security and compliance.

[0030] In one embodiment, step S602, which involves dynamically optimizing the cross-platform delivery strategy in real time based on the real-time feedback data to obtain a real-time cross-platform delivery strategy, includes: S6021: Construct a strategy optimization model with click-through rate, conversion rate, and return on investment as optimization objectives; S6022: Based on the real-time feedback data, the policy optimization model is iteratively updated using a reinforcement learning algorithm; S6023: Output resource allocation adjustment decisions for each of the target business platforms through the iteratively updated strategy optimization model; S6024: Based on the resource allocation adjustment decision, dynamically adjust the resource allocation ratio among the target business platforms to obtain a real-time cross-platform deployment strategy.

[0031] As described in step S6021 above, a strategy optimization model is constructed with click-through rate (CTR), conversion rate (CVR), and return on investment (ROI) as optimization objectives. First, the measurement system and summary criteria for the optimization objectives are determined. Multi-dimensional indicators such as CTR, CVR, and ROI are standardized into a unified scoring scale. Weights or priorities for each indicator are set based on business objectives. Then, a reward function (or multi-objective reward vector) is defined, which can take the form of a weighted sum, hierarchical constraint optimization, or a vectorized multi-objective function. Delayed feedback and attribution issues are considered (e.g., conversion values ​​are discounted according to attribution windows). Based on this, a suitable strategy optimization model framework is selected (e.g., a strategy network based on policy gradients, a Q-network based on value functions, or an actor-critic structure), and the model parameters, action space, and state representation are initialized. The state vector contains the real-time characteristics of the current campaign (platform metrics, creative identifiers, time period information, audience distribution, etc.), and the action space is defined as adjustable resource dimensions (budget ratio, bid, display frequency, audience weight, etc.). At the same time, to ensure stability, constraints (budget upper and lower limits, daily fluctuation thresholds, platform compliance restrictions) and a cold start mechanism are designed to keep the online learning process safe and controllable.

[0032] As described in step S6022 above, the policy optimization model is iteratively updated using a reinforcement learning algorithm based on the real-time feedback data. Real-time or near-real-time collected feedback data is used as environmental signals, and reinforcement learning (RL) methods are employed to train the policy model online or incrementally. The specific process includes: converting real-time feedback (such as CTR, CVR, actual spending, and conversion events per minute or hour) into state transition and immediate reward samples, constructing an experience replay pool or streaming sample window; performing gradient calculation and parameter updates based on the selected RL algorithm (such as deep Q-learning, Proximal Policy Optimization (PPO), or DDPG / TD3, etc.), and updating the policy network or value network. The training process introduces exploratory policies (such as ε-greedy or random policy noise), and prevents policy oscillations through soft updates, target networks, and regularization. Furthermore, online updates employ phased verification: updates are first verified in a sandbox or low-traffic canary cluster, allowing the model to undergo A / B testing under real-world distributions. Only after meeting predetermined safety constraints is the application scaled up. The entire update process records training loss, changes in expected rewards, and actual business metrics for tracking and backtracking.

[0033] As described in step S6023 above, the resource allocation adjustment decision for each of the target business platforms is output through the iteratively updated strategy optimization model. The trained or updated strategy model calculates and outputs specific action suggestions based on currently observed state inputs (including real-time feedback, historical performance, audience changes, etc.), which are resource allocation adjustment decisions for each target business platform. The output is usually presented in the form of numerical vectors, including budget increase / decrease ratios, bid upper and lower limits adjustments, audience targeting parameter modifications, suggestions for optimizing ad placement time periods, and creative display priorities for each platform. The output results are post-processed: the model actions are mapped to executable business units (e.g., converting the ratio values ​​into specific monetary budgets, mapping the bidding strategy to the maximum click limit), and compliance and security checks are performed in conjunction with a constraint checker (ensuring that the daily budget does not exceed the limit, the bid does not exceed the threshold, etc.). At the same time, to improve operability, confidence indicators and impact estimates (expected CTR / ROI changes) are generated, and the decisions are provided in two modes: suggestions or automatic execution. Decisions with high confidence and verified through sandbox can be automatically issued, while decisions with low confidence or high risk are verified first through manual confirmation or small-scale trials.

[0034] As described in step S6024 above, the resource allocation ratio among the target business platforms is dynamically adjusted based on the resource allocation adjustment decision to obtain a real-time cross-platform delivery strategy. The adjustment decision is first converted into API-level execution instructions for each platform, including updating ad group budgets, modifying bidding strategies, adjusting targeting conditions and delivery time windows, and prioritizing or replacing creatives. A gradual deployment strategy is adopted: changes are first issued in canary traffic or small sample partitions, and key indicators are monitored in real time. After confirming no abnormalities, the changes are gradually expanded to full delivery. If a KPI decline or compliance alarm is detected, a rollback or automatic reduction of the adjustment range is triggered. Furthermore, the change history, execution results, and related causal data are continuously recorded during implementation to provide samples for subsequent model training. Linkage with the monitoring and alarm system is maintained to ensure timely retry or adjustment of the allocation strategy if an interface error, rate limit, or order rejection occurs on any platform. Ultimately, an executable cross-platform delivery strategy based on real-time feedback and closed-loop updates is formed, achieving a dynamic balance between effectiveness and safety constraints.

[0035] In one embodiment, step S3, which involves obtaining the platform rules of each of the target business platforms and generating a structured task vector based on the structured task elements and the platform rules, includes: S301: Query the pre-set platform rule knowledge base to obtain the platform rules of each of the target business platforms; wherein, the platform rules include application programming interface specifications, content format requirements and delivery strategy constraints; S302: Based on the structured task elements, the acquired platform rules are filtered and feature extracted to obtain platform rule features; S303: The platform rule features and the structured task elements are fused and encoded to generate a unified structured task vector.

[0036] As described in step S301 above, the pre-set platform rule knowledge base is queried to obtain the platform rules of each of the target business platforms. These platform rules include application programming interface (API) specifications, content format requirements, and delivery strategy constraints. First, corresponding records are retrieved from the local or remote "Platform Rule Knowledge Base" based on the list of target business platforms. This knowledge base stores the terms and technical specifications of each platform in structured entries, including API endpoints and parameter descriptions, authentication and call frequency limits, material size and encoding formats, copy length and tag usage specifications, applicable delivery regions / qualification requirements, and platform-specific delivery strategy constraints (such as bidding models, bidding time windows, prohibited content lists, etc.). During the retrieval process, platform identification standardization is performed (mapping different user input expressions to internal platform IDs), and knowledge base version control and effective timestamps are considered to obtain the currently effective rules. For rules not explicitly stated in the knowledge base, real-time API calls or crawling logic can be triggered to obtain the latest platform descriptions, which are then parsed and stored in a structured manner. It can also include rule integrity verification and caching strategies, that is, to cache frequently accessed platform rules locally to reduce latency, and to set up automatic change detection and notification for rule updates to ensure the timeliness and consistency of the rules used.

[0037] As described in step S301 above, the acquired platform rules are filtered and feature extracted based on the structured task elements to obtain platform rule features. The retrieved rule entries are refined and semantically extracted according to the needs of the current task. First, based on the key fields in the structured task elements (such as target region, content format, budget range, audience attributes, etc.), a subset of rules closely related to this promotion is filtered out. For example, if the task requires short video format, entries such as video duration threshold, resolution, frame rate, subtitles, and audio constraints are retained first. Then, the selected rules are characterized: numerical rules (such as duration, size, and call limits) are extracted into quantifiable features; boolean or enumeration rules (such as whether links are allowed or whether specific tags are allowed) are converted into binary or one-hot vectors; textual rules (such as sensitive word lists and descriptive compliance clauses) are subject to keyword extraction and semantic vectorization, and compliance priority is marked. Rule parsers, regular expression / template matching, and pre-trained language models can be used to help extract the meaning of complex clauses. At the same time, conflicts or priorities between rules are identified (such as regional restrictions covering global rules), and these conflict points are recorded in the form of metadata to generate a structured and semantically rich set of "platform rule features" for subsequent coding.

[0038] As described in step S303 above, the platform rule features and the structured task elements are fused and encoded to generate a unified structured task vector. To fuse the task elements and platform rule features into a machine-readable unified representation, heterogeneous features are first preprocessed and standardized: numerical features are normalized, categorical features are embedded or one-hot encoded, and text features are represented using semantic embedding (such as sentence vectors), while retaining meta-information such as rule priority, effective time, and uncertainty confidence. Then, a fusion strategy is used to aggregate these encodings into a high-dimensional vector. Concatenation-projection, attention-weighted fusion, or neural network-based encoders (such as multi-layer feedforward networks or Transformer encoding layers) can be used to learn the interaction relationships between features. During the fusion encoding process, encoding constraint information and conflict markers are displayed (e.g., through additional dimensions or mask mechanisms), as well as marking which dimensions are hard constraints that must be followed and which are soft constraints that can be optimized. This makes it easier for the subsequent generation engine to follow during inference. The final output structured task vector contains both business intent information and embedded platform executable constraints and priorities, forming a unified input for conditional decoding by the multimodal large language model generation engine. At the same time, it retains reversible mappings to support interpretable retrieval and backtracking of a certain dimension in the vector.

[0039] In one embodiment, after step S4 of inputting the structured task vector into a preset multimodal large language model generation engine to generate a multimodal business strategy scheme, the method further includes: S501: Obtain target region information from the structured task elements; S502: Call the regional cultural feature database corresponding to the regional information; S503: Adjust the generated multimodal business strategy scheme for compliance based on the cultural taboos and compliance rules in the regional cultural feature database.

[0040] As described in step S501 above, target region information is obtained from the structured task elements. The target region information is read and standardized from the previously extracted and formatted structured task elements. This information may include multi-dimensional attributes such as country, administrative region, city, language, time zone, and specific market identifier. First, the standardization module maps the region description, which the user may provide in natural language or abbreviations, to an internally unified geographic code (such as ISO 3166 country / region code, administrative division ID, or custom market ID), and supplements necessary contextual data (such as legal jurisdiction, main language, main currency, and internet regulatory category). Simultaneously, multi-level parsing is performed based on detailed instructions in the task elements (such as "Southeast Asian market," "Middle Eastern Arabic user," or a regional site of an e-commerce platform) to determine the precise granularity of the applied cultural characteristics (national, regional, or city level). To ensure the reliability of subsequent processing, the confidence level of the region information is scored. If the confidence level is low or ambiguous, supplementary interaction is triggered or a candidate region set is used for parallel evaluation. The final output is a structured, searchable target region identifier with version and confidence metadata, which serves as input for calling the regional cultural feature library.

[0041] As described in step S502 above, the regional cultural feature database corresponding to the regional information is invoked. Using the standardized regional identifier generated in step S501 as the search key, a query is initiated to the regional cultural feature database. This database stores various cultural and compliance entries in a machine-readable structured format (such as JSON / OWL ontology), including but not limited to color symbolism, religious and customary taboos, key legal regulatory points (advertising law, data protection, protection of minors), holidays and prohibited times, music / sound usage restrictions, sensitive lists of symbols and images, language tone preferences, and localized wording suggestions. The search process supports hierarchical rollback, prioritizing the retrieval of fine-grained rules corresponding to specific cities or sub-regions; if these rules are not found, rollback to national or regional rules is performed. Entries from third-party compliance databases, platform public policies, and internal review experience libraries are also merged. The feature database returns not only the rule text but also structured attributes (rule type, hard constraint / soft suggestion identifier, effective date, source, and confidence level). Version control and timeliness checks are also performed; if a rule has expired or conflicts with the platform's current rules, a rule refresh process is triggered, or a high-risk entry requiring manual review is marked. The final output can be directly used for automated detection and adjustment of regional cultural features.

[0042] As described in step S503 above, the generated multimodal business strategy scheme is adjusted for compliance based on the cultural taboos and compliance rules in the regional cultural feature database. The multimodal business strategy scheme is compared item by item with the regional cultural features obtained in step S502, and an automated compliance correction process is executed. First, detection is performed: for text, prohibited expressions are identified through sensitive word matching, semantic similarity judgment, and entity recognition; for visual materials, potential violation elements are identified through image recognition, color analysis, and symbol detection; for audio / video content, whether the music type, speech wording, and image combination violate taboos are detected. For detected violations, the system adopts different strategies based on the "hard constraints / soft suggestions" identifiers in the feature library. For hard constraints (such as legal prohibitions or platform-mandated bans), it enforces corrections or directly blocks the violation, using rule replacement, rewriting copy, replacing materials, or regenerating creative content. For soft suggestions (such as cultural style or wording), it prompts the engineering team to adjust the multimodal model to generate a version that better suits local preferences, or provides multiple alternatives for manual selection. Compliance adjustments are iteratively verified, meaning that automatic detection is performed again after each modification, and manual review is initiated when necessary to handle high-risk or uncertain cases. All adjustments are recorded in a change log, including the reason for the change and the responsible person (or the automation module identifier), and the final compliant strategy is returned to the campaign module to ensure that regional cultural compliance requirements are met while preserving creative effects and campaign objectives as much as possible.

[0043] In one embodiment, step S5, which generates a cross-platform delivery strategy based on the historical platform data and the multimodal business strategy scheme, includes: S511: Based on the historical platform data, predict the estimated effect of achieving the business objectives on each target business platform; S512: Based on the estimated results and the resource constraints in the structured task elements, allocate initial budgets, set bidding strategies, and plan delivery periods for each target business platform to obtain a cross-platform delivery strategy.

[0044] As described in step S511 above, based on the historical platform data, the estimated effect of achieving the business objectives on each target business platform is predicted. First, the acquired historical platform data is cleaned and feature-engineered, including time series smoothing, missing value imputation, outlier removal, and aggregated statistics by activity, creative, and audience dimensions (such as hourly / daily / weekly CTR, CVR, CPA, and display distribution). Then, a predictive model is constructed to estimate the expected performance of different platforms, creatives, and audience segments under given deployment conditions. Various modeling methods can be used: time series forecasting (such as ARIMA and Prophet) is used for periodic trends, and regression models or Gradient Boosting Tree (GBDT) are used for causal effect estimation. Alternatively, a multi-touchpoint attribution model or a backward model based on causal attribution can be used to estimate the contribution of different channels to conversion. In the case of sparse data or cold start, cross-platform transfer learning, Bayesian priors or industry benchmarks can be introduced to smooth the estimation, and confidence intervals and uncertainty quantification can be given for the prediction results. The prediction output includes the expected CTR, CVR, cost per conversion (CPA), estimated conversion volume and corresponding confidence interval under given budget / bid assumptions, and may generate multi-scenario (conservative, baseline, aggressive) predictions for decision-making reference. All prediction results are used for subsequent resource allocation and risk control.

[0045] As described in step S512 above, based on the estimated results and the resource constraints in the structured task elements, initial budgets are allocated to each target business platform, bidding strategies are set, and campaign periods are planned to obtain a cross-platform campaign strategy. Using the prediction results from S511 and the resource constraints explicitly defined in the structured task elements (such as total budget cap, minimum exposure requirements, KPI targets, compliance restrictions, and time period preferences) as input, a resource allocation optimization problem is constructed and the initial campaign strategy is solved. The optimization model can employ linear / integer programming, constraint optimization, or a multi-objective optimization framework. The objective function is to maximize the expected conversion rate or ROI, or, under multi-objective conditions, a weighted sum or Pareto front method can be used to balance CTR, CVR, and cost. The bidding strategy is set according to the platform's bidding model (CPM / CPC / CPA) and mapped to a specific bid cap or bid range, taking into account the bidding density during the campaign period. Regarding cost fluctuations, dayparting is used to suggest efficient time slots or segmented budgets based on historical performance and audience activity to improve marginal benefits. It also incorporates safety constraints and buffer strategies (such as daily budget caps, maximum adjustment ranges, and legal / compliance exclusion periods). Simulations or Monte Carlo simulations are performed to evaluate the robustness of the strategy under different uncertainty scenarios. The final cross-platform delivery strategy includes initial budget allocation tables for each platform, bidding and auction strategies for each platform, delivery time windows and frequency controls, audience targeting suggestions, and expected KPIs and uncertainty assessments, serving as the initial configuration for deployment and subsequent online optimization.

[0046] In one embodiment, step S2, which involves semantic and intent parsing of the natural language business instructions to extract structured task elements, includes: S201: Input the natural language business instruction into a preset cross-language pre-trained model for processing to identify the instruction information in the natural language business instruction; wherein, the instruction information includes business objectives, target regions, resource constraints, and promotion target information; S202: Format the instruction information into the structured task elements of a preset structure.

[0047] As described in step S201 above, the natural language business instructions are input into a preset cross-language pre-trained model for processing to identify the instruction information in the natural language business instructions; wherein, the instruction information includes business objectives, target regions, resource constraints, and promotion target information. First, the original input is preprocessed, including character encoding standardization, noise removal, language detection, and sentence / segmentation processing, to support the recognition of mixed language text. After preprocessing, the text is fed into a finely tuned cross-language pre-trained model (e.g., XLM-RoBERTa or an equivalent model). The model simultaneously performs intent classification and entity extraction tasks. Intent classification identifies the user's main request type (e.g., "create a new campaign", "optimize an existing campaign", "stop campaign"). Entity extraction identifies business objectives (e.g., "increase conversion rate", "increase brand exposure"), target regions (country / region / language / time zone), resource constraints (budget value and currency, time window, minimum exposure requirement), and promotion target information (audience profile, product ID, interest tags, etc.). The model output includes a confidence score and location information to indicate the start and end positions of the extracted items in the original text. For low-confidence or conflicting entities, the system triggers supplementary interactions or secondary parsing (e.g., multi-turn dialogue queries, candidate parsing display), and records the original text and parsing logs for auditing and subsequent model optimization.

[0048] As described in step S202 above, the instruction information is formatted into the structured task elements of a preset structure. This step maps the instruction information identified by S201 to a predefined data pattern in the system, forming machine-readable structured task elements. The mapping includes field normalization (e.g., unifying "10k Euros" and "€10,000" into numerical values ​​and standard currency codes), time and time zone standardization (converting "Next Monday" and "2026-02-02T09:00+01:00" to UTC or a unified timestamp within the system), geocoding (mapping "France" and "Paris" to ISO country codes or internal market IDs), and audience tag standardization (mapping natural language descriptions to internal audience category IDs or tag sets). At the same time, required fields are validated and completed: if key constraints are missing, default values ​​are filled in based on context or historical records, or a user confirmation process is triggered. The formatted result includes the original text, standardized value, confidence level, source, and timestamp of each element, forming a versionable structured task element object (e.g., JSON structure), and outputting it to downstream modules (rule matching, vectorization, and generation engine) to ensure semantic consistency, traceability, and error recovery capabilities in subsequent processing.

[0049] Reference Figure 3 The present invention also provides a cross-platform business promotion device, the device comprising: The receiving module 902 is used to receive natural language business instructions input by the user and multiple target business platforms applied to the promotion; Extraction module 904 is used to perform semantic and intent parsing on the natural language business instructions and extract structured task elements; The acquisition module 906 is used to acquire the platform rules of each of the target business platforms and generate a structured task vector based on the structured task elements and the platform rules. The input module 908 is used to input the structured task vector into a preset multimodal large language model generation engine to generate a multimodal business strategy scheme. The generation module 910 is used to obtain historical platform data of each of the target business platforms and generate a cross-platform delivery strategy based on the historical platform data and the multimodal business strategy scheme. The deployment module 912 is used to deploy the multimodal business strategy solution to each of the target business platforms for promotion through the cross-platform delivery strategy.

[0050] In one embodiment, deployment module 912 includes: The real-time feedback data acquisition submodule is used to acquire real-time feedback data after the execution of the multimodal business strategy scheme; The cross-platform delivery strategy acquisition submodule is used to dynamically optimize the cross-platform delivery strategy in real time based on the real-time feedback data to obtain the real-time cross-platform delivery strategy. The promotion submodule is used to deploy the multimodal business strategy to each of the target business platforms for promotion based on the real-time cross-platform delivery strategy.

[0051] In one embodiment, the cross-platform delivery strategy acquisition submodule includes: The building unit is used to build a strategy optimization model with click-through rate, conversion rate, and return on investment as optimization objectives; The update unit is used to iteratively update the policy optimization model based on the real-time feedback data using a reinforcement learning algorithm. The output unit is used to output resource allocation adjustment decisions for each of the target business platforms through the iteratively updated strategy optimization model. The adjustment unit is used to dynamically adjust the resource allocation ratio among the target business platforms based on the resource allocation adjustment decision, thereby obtaining a real-time cross-platform deployment strategy.

[0052] In one embodiment, the acquisition module 906 includes: The query submodule is used to query a pre-built platform rule knowledge base to obtain the platform rules of each target business platform; wherein, the platform rules include application programming interface specifications, content format requirements, and delivery strategy constraints; The platform rule feature acquisition submodule is used to filter and extract features from the acquired platform rules based on the structured task elements to obtain platform rule features; The fusion encoding submodule is used to fuse and encode the platform rule features with the structured task elements to generate a unified structured task vector.

[0053] In one embodiment, the cross-platform business promotion device further includes: The target region information acquisition module is used to acquire target region information in the structured task elements; The regional cultural feature database calling module is used to call the regional cultural feature database corresponding to the regional information; The compliance adjustment module is used to adjust the generated multimodal business strategy scheme for compliance based on the cultural taboos and compliance rules in the regional cultural feature database.

[0054] In one embodiment, the generation module 910 includes: The estimated effect achievement submodule is used to predict the estimated effect of achieving the business objectives on each target business platform based on the historical platform data. The bidding strategy setting submodule is used to allocate initial budgets, set bidding strategies, and plan delivery periods for each target business platform based on the estimated results and resource constraints in the structured task elements, thereby obtaining a cross-platform delivery strategy.

[0055] In one embodiment, the extraction module 904 includes: The recognition submodule is used to input the natural language business instructions into a preset cross-language pre-trained model for processing, so as to recognize the instruction information in the natural language business instructions; wherein, the instruction information includes business objectives, target regions, resource constraints, and promotion target information; The structured task element acquisition submodule is used to format the instruction information into the structured task elements of a preset structure.

[0056] Figure 4 An internal structural diagram of an electronic device in one embodiment is shown. This electronic device can specifically be a terminal or a server, and more specifically, a computer device. Figure 4 As shown, the electronic device includes a processor, a memory, and a network interface connected via a system bus. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and may also store a computer program. When executed by the processor, this computer program enables the processor to implement cross-platform business promotion methods. The internal memory may also store a computer program, which, when executed by the processor, enables the processor to implement cross-platform business promotion methods. Those skilled in the art will understand that... Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0057] In one embodiment, an electronic device is provided, including a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the following steps: It receives natural language business instructions from users and targets multiple business platforms for promotion. Semantic and intent parsing is performed on the natural language business instructions to extract structured task elements; Obtain the platform rules of each of the target business platforms, and generate a structured task vector based on the structured task elements and the platform rules; The structured task vector is input into a preset multimodal large language model generation engine to generate a multimodal business strategy solution; Obtain historical platform data for each of the target business platforms, and generate a cross-platform delivery strategy based on the historical platform data and the multimodal business strategy scheme; The cross-platform deployment strategy is used to deploy the multimodal business strategy solution to various target business platforms for promotion.

[0058] The system performs semantic and intent parsing on natural language commands to extract structured task elements. It then acquires rules from various target platforms and generates a unified structured task vector based on these elements and rules. This vector is input into a pre-defined multimodal large language model to generate content strategies adapted to each platform. Subsequently, it combines historical platform data to form cross-platform delivery strategies and promotes the service. This shortens the cycle from demand to content and delivery plan development, reduces the manpower costs of platform integration and material adaptation, improves cross-platform content consistency and compliance, and enhances resource allocation efficiency and promotional effectiveness by leveraging delivery strategies supported by historical data. It organically links natural language business commands with multi-target platform rules and historical data to achieve automated generation and cross-platform deployment of promotion strategies.

[0059] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, causes the processor to perform the following steps: It receives natural language business instructions from users and targets multiple business platforms for promotion. Semantic and intent parsing is performed on the natural language business instructions to extract structured task elements; Obtain the platform rules of each of the target business platforms, and generate a structured task vector based on the structured task elements and the platform rules; The structured task vector is input into a preset multimodal large language model generation engine to generate a multimodal business strategy solution; Obtain historical platform data for each of the target business platforms, and generate a cross-platform delivery strategy based on the historical platform data and the multimodal business strategy scheme; The cross-platform deployment strategy is used to deploy the multimodal business strategy solution to various target business platforms for promotion.

[0060] The system performs semantic and intent parsing on natural language commands to extract structured task elements. It then acquires rules from various target platforms and generates a unified structured task vector based on these elements and rules. This vector is input into a pre-defined multimodal large language model to generate content strategies adapted to each platform. Subsequently, it combines historical platform data to form cross-platform delivery strategies and promotes the service. This shortens the cycle from demand to content and delivery plan development, reduces the manpower costs of platform integration and material adaptation, improves cross-platform content consistency and compliance, and enhances resource allocation efficiency and promotional effectiveness by leveraging delivery strategies supported by historical data. It organically links natural language business commands with multi-target platform rules and historical data to achieve automated generation and cross-platform deployment of promotion strategies.

[0061] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. 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 can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various 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 RAMbus dynamic RAM (RDRAM), etc.

[0062] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0063] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A cross-platform business promotion method, characterized in that, The method includes: It receives natural language business instructions from users and targets multiple business platforms for promotion. Semantic and intent parsing is performed on the natural language business instructions to extract structured task elements; Obtain the platform rules of each of the target business platforms, and generate a structured task vector based on the structured task elements and the platform rules; The structured task vector is input into a preset multimodal large language model generation engine to generate a multimodal business strategy solution; Obtain historical platform data for each of the target business platforms, and generate a cross-platform delivery strategy based on the historical platform data and the multimodal business strategy scheme; The cross-platform deployment strategy is used to deploy the multimodal business strategy solution to various target business platforms for promotion.

2. The cross-platform business promotion method according to claim 1, characterized in that, The step of deploying the multimodal business strategy solution to various target business platforms for promotion through the cross-platform delivery strategy includes: Obtain real-time feedback data after the execution of the multimodal service strategy scheme; The cross-platform delivery strategy is dynamically optimized in real time based on the real-time feedback data to obtain a real-time cross-platform delivery strategy. Based on the real-time cross-platform deployment strategy, the multimodal business strategy solution is deployed to each of the target business platforms for promotion.

3. The cross-platform business promotion method according to claim 2, characterized in that, The step of dynamically optimizing the cross-platform delivery strategy in real time based on the real-time feedback data to obtain a real-time cross-platform delivery strategy includes: Construct a strategy optimization model with click-through rate, conversion rate, and return on investment as optimization objectives; Based on the real-time feedback data, the policy optimization model is iteratively updated using a reinforcement learning algorithm. The resource allocation adjustment decisions for each target business platform are output through the iteratively updated strategy optimization model. Based on the resource allocation adjustment decision, the resource allocation ratio among the target business platforms is dynamically adjusted to obtain a real-time cross-platform deployment strategy.

4. The cross-platform business promotion method according to claim 1, characterized in that, The step of obtaining the platform rules of each of the target business platforms and generating a structured task vector based on the structured task elements and the platform rules includes: The system queries a pre-built platform rules knowledge base to obtain the platform rules for each of the target business platforms; wherein, the platform rules include application programming interface specifications, content format requirements, and delivery strategy constraints. Based on the structured task elements, the acquired platform rules are filtered and feature extracted to obtain platform rule features; The platform rule features and the structured task elements are fused and encoded to generate a unified structured task vector.

5. The cross-platform business promotion method according to claim 1, characterized in that, After the step of inputting the structured task vector into a preset multimodal large language model generation engine to generate a multimodal business strategy solution, the method further includes: Obtain target region information from the structured task elements; Call the regional cultural feature database corresponding to the aforementioned regional information; Based on the cultural taboos and compliance rules in the regional cultural feature database, the generated multimodal business strategy scheme is adjusted for compliance.

6. The cross-platform business promotion method according to claim 1, characterized in that, The step of generating a cross-platform delivery strategy based on the historical platform data and the multimodal business strategy scheme includes: Based on the historical platform data, predict the estimated effect of achieving the business objectives on each target business platform; Based on the estimated results and the resource constraints in the structured task elements, an initial budget is allocated to each target business platform, a bidding strategy is set, and a delivery period is planned to obtain a cross-platform delivery strategy.

7. The cross-platform business promotion method according to claim 1, characterized in that, The step of performing semantic and intent parsing on the natural language business instructions and extracting structured task elements includes: The natural language business instructions are input into a preset cross-language pre-trained model for processing to identify the instruction information in the natural language business instructions; wherein, the instruction information includes business objectives, target regions, resource constraints, and promotion target information; The instruction information is formatted into the structured task elements of a preset structure.

8. A cross-platform business promotion device, characterized in that, The device includes: The receiving module is used to receive natural language business instructions input by the user and to respond to multiple target business platforms applied for promotion. The extraction module is used to perform semantic and intent parsing on the natural language business instructions and extract structured task elements; The acquisition module is used to acquire the platform rules of each of the target business platforms and generate a structured task vector based on the structured task elements and the platform rules. The input module is used to input the structured task vector into a preset multimodal large language model generation engine to generate a multimodal business strategy scheme; The generation module is used to acquire historical platform data of each of the target business platforms and generate a cross-platform delivery strategy based on the historical platform data and the multimodal business strategy scheme. The deployment module is used to deploy the multimodal business strategy solution to each of the target business platforms for promotion through the cross-platform delivery strategy.

9. A computer-readable storage medium, characterized in that, The system contains a computer program that, when executed by a processor, causes the processor to perform the steps of the cross-platform business promotion method as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, The device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the cross-platform business promotion method as described in any one of claims 1 to 7.

Images