AI marketing content generation method and system based on GEO trusted semantic optimization

Abstract

The application relates to the field of artificial intelligence and digital marketing technology, in particular to an AI marketing content generation method and system based on GEO trusted semantic optimization, which comprises the following steps: acquiring brand OID identification and associated full-link traceability data, and performing structured processing; constructing a GEO semantic knowledge graph containing multiple dimensions; generating a GEO trusted semantic embedding vector by adopting a multi-head attention mechanism; adaptively mapping target AI platform RAG preference features to generate platform-adapted content feature expressions; and finally generating structured marketing content by using a generative AI model and deploying the content to authoritative source channels. The OID national standard traceability system and GEO semantic optimization are deeply integrated, so that brand content has high credibility, high citation rate and cross-platform consistency in multiple generative AI platforms, the problems of lack of trusted semantic structure and poor cross-platform adaptation capability of brand content in the prior art are solved, and the AI marketing effect is improved.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and digital marketing technology, and in particular to an AI marketing content generation method and system based on GEO trusted semantic optimization. Background Technology

[0002] With the rapid development of generative AI technology, users' primary entry point for information is shifting from traditional search engines to generative AI dialogue platforms (such as DeepSeek, Doubao, Wenxin Yiyan, Tongyi Qianwen, ChatGPT, Gemini, etc.). Against this backdrop, GEO (Generative Engine Optimization) has emerged, with its core objective being to prioritize and recommend brand content within the answers generated by AI-powered large-scale models.

[0003] However, existing GEO technology has the following significant drawbacks: Lack of credible semantic structure: Traditional GEO optimization mainly relies on keyword placement and content relevance adjustments, failing to transform the product's underlying credible data into structured semantic features that AI models can understand and trust. During the Retrieval Augmentation (RAG) process, AI models struggle to distinguish between marketing claims and verifiable facts.

[0004] Weak cross-platform adaptability: There are significant differences in the RAG retrieval mechanisms, semantic matching preferences, and content parsing rules of different generative AI platforms. Existing GEO methods typically adopt a method of publishing one set of content across multiple platforms, lacking the ability to dynamically and adaptively optimize for the semantic space of each platform. This results in brand content being efficiently cited on some platforms, but ignored or ranked poorly on others.

[0005] Current GEO technology focuses on relevance optimization, such as keyword placement and content relevance adjustment, making it difficult to ensure that the same brand's content can achieve high search visibility across different AI platforms. Summary of the Invention

[0006] The purpose of this invention is to provide an AI marketing content generation method and system based on GEO trusted semantic optimization. It deeply integrates the OID national traceability standard system with GEO semantic optimization technology, and performs semantic space mapping of the RAG mechanism of the target AI platform through platform adaptation network. This enables the same brand content to obtain high search visibility and citation probability on different platforms, and allows the AI ​​model to verify the authenticity of brand information from multiple perspectives.

[0007] To achieve the above objectives, this invention provides an AI marketing content generation method based on GEO trusted semantic optimization, comprising the following steps: S1: Obtain the brand object identifier (OID) registered by the enterprise in the national OID registration center, collect the product full-chain traceability data associated with the OID identifier, and perform standardized cleaning and alignment of the traceability data to form a structured data set at the bottom layer of brand trust. Step S1: Obtain the Brand Object Identifier (OID) registered by the enterprise with the National OID Registration Center. The OID complies with the ISO / IEC 9834 standard system. Collect product traceability data associated with the OID. The traceability data includes raw material procurement information, production process parameters, quality inspection and certification data, third-party testing reports, warehousing and logistics records, and user QR code verification records. The traceability data is standardized, cleaned, and aligned to extract quantifiable and verifiable technical parameters and indicator information, forming a structured data set that forms the foundation of brand credibility.

[0008] The standardized data cleaning rules are as follows: date and unit of measurement formats are standardized sequentially; outlier detection and removal are performed; missing values ​​are marked; and duplicate data is deduplicated. All cleaning operations are logged to ensure traceability. Quantifiable verification means that the data exists in the form of specific numerical values, measurable indicators, or parameters that can be verified by an independent third party, rather than in the form of vague descriptions or subjective evaluations. After this data is stored on the blockchain, the AI ​​model can verify its authenticity through hash comparison using OID identifiers.

[0009] S2: Based on structured datasets, construct a multi-dimensional semantic network for the brand's GEO; fuse features from each dimension using graph neural networks to generate the brand's GEO semantic knowledge graph; In step S2, the GEO semantic multi-dimensional network includes at least three dimensions: technical semantic dimension, scenario semantic dimension, authority semantic dimension, and geographic origination semantic dimension. Specifically, the technical semantic dimension is configured to map production and operation data into technical feature vectors in the AI ​​semantic vector space; the scenario semantic dimension is configured to semantically label user usage scenarios adapted to brand products and associate each label with a typical user question template; the authority semantic dimension is configured to structure and encode authoritative credential information and establish a trusted association mapping with OID identifiers; and the geographic origination semantic dimension is configured to semantically express spatial location data, forming a spatial geographic semantic map.

[0010] In step S2, the graph neural network aggregates features from various dimensions through a message passing mechanism to generate a GEO semantic knowledge graph with the OID identifier as the root node.

[0011] Step S2, Construction of Multidimensional Semantic Network and Knowledge Graph Based on the structured data extracted in step S1, a brand's GEO semantic multi-dimensional network is constructed; the semantic multi-dimensional network includes the following four dimensions: Technical semantic dimension: Transforming quality inspection indicators, process parameters, production capacity data, etc. into technical feature elements in the AI ​​semantic vector space; Scene semantic dimension: The user usage scenarios that the brand's products are adapted to are semantically labeled, and each label is associated with a typical user question template; Authoritative semantic dimension: The information such as third-party authoritative certifications, industry institution endorsements, and official test reports obtained by the brand are structured and encoded, and a trustworthy association mapping is established with the OID identifier; Geographic traceability semantic dimension: Semantically express data such as product origin information, raw material source, and geographical indication protection scope to form a spatial geographic semantic map; The features of the above four dimensions are fused by a graph neural network (GNN) to generate a brand's GEO semantic knowledge graph. In the knowledge graph, the OID identifier is used as the root node, and the features of each dimension are used as child nodes. The edge weights between nodes are determined by the semantic relevance strength.

[0012] S3: Encode the features of each dimension in the GEO semantic knowledge graph into semantic vectors, and use a multi-head attention mechanism to adaptively weight and fuse the semantic vectors of each dimension to generate a brand’s trusted semantic embedding vector. In step S3, the GEO trusted semantic embedding vector is generated in the following way: the feature vectors of each semantic dimension are used as input to the multi-head attention mechanism, and the outputs of each attention head are concatenated and multiplied by the output weight matrix to obtain the fused semantic embedding vector.

[0013] Step S3: Multidimensional semantic encoding and vectorization transformation. The features of each dimension in the GEO semantic knowledge graph are encoded into semantic vectors respectively. Multi-head attention mechanism is used to adaptively weight and fuse the semantic vectors of each dimension to generate the brand's GEO trust semantic embedding vector (GEO-TrustEmbedding). The structured information extracted from the brand OID traceability data is used to construct the GEO semantic knowledge graph through a multidimensional semantic network. Then, multi-head attention mechanism is used to adaptively weight and fuse the semantic vectors of each dimension to generate a multidimensional semantic vector representation that integrates trust data features. The weight parameters of the fusion process are learned by using the AI ​​citation rate in the training data as a supervision signal.

[0014] S4: Obtain the RAG semantic matching preference features of the target AI generation platform, and through a platform adaptation network containing a learnable affine transformation layer, map the GEO trusted semantic embedding vector to the retrieval semantic space of the target platform to generate content feature expressions that are compatible with the target AI platform. Step S4: Based on the dynamic adaptation of generative content of the target AI platform, obtain the RAG semantic matching preference features of the target AI generation platform. The preference features include: the type of retrieval model used by the platform, the semantic similarity measurement method, the source authority scoring rules, the content length limit, and the structured tagging preference. Based on preference features, the GEO trusted semantic embedding vector generated in step S3 is subjected to platform adaptive transformation. The adaptive transformation is implemented through a platform adaptation network, which contains at least one learnable affine transformation layer to map the general semantic vector to the retrieval semantic space of the target platform and generate content feature representations that are adapted to the target AI platform.

[0015] S5: Use the adapted content feature expression from step S4 to optimize the Top-K recall ranking of the target AI generation platform during RAG retrieval. Then, based on the adapted content feature expression, use a generative AI model to generate structured AI marketing content. Release the generated AI marketing content through authoritative information sources and associate the content with the OID identifier during the release process so that it can be automatically recognized and crawled by AI search engines and large model RAG systems.

[0016] Step S5: Credible Marketing Content Generation and Publication. Based on the adapted content feature expression, a generative AI model is used to generate structured AI marketing content. The generated content includes at least one of the following forms: structured text paragraphs, question-and-answer pairs (Q&A pairs), knowledge cards, mixed text and image content, and HTML fragments with Schema.org tags. The generated content is published through preset authoritative information source channels. During the publication process, the content is associated with OID identifiers through Schema.org structured tags to form credible information source content that can be directly referenced and verified by the AI ​​large model.

[0017] The release methods of the authoritative information sources include: For the brand's official website: Publish the generated structured content as web pages through the backend publishing function of the website content management system. At the same time, generate or update the sitemap.xml file in the website root directory, listing the URLs of all pages containing OID-related content and their last update time in accordance with the XML Sitemap protocol standard. For authoritative third-party platforms, submit content through the certified source content submission interface provided by each platform, in accordance with the JSON or XML data format required by the platform, and attach an OID identifier as a unique index of the content; for platforms that do not provide public APIs, publish through officially recognized cooperation channels of the platform, such as institutional accounts or content cooperation push.

[0018] Furthermore, the AI ​​marketing content generation method also includes the following steps: S6: Collect feedback data of the generated content on multiple AI platforms, use the feedback data as training samples, and incrementally update the multi-head attention weight parameters of step S3 and the platform adaptation network parameters of step S4 to form a closed loop of continuous optimization.

[0019] Step S6: GEO optimization effect feedback and model iteration, collect feedback data such as the actual citation rate, semantic matching score, and user click conversion of the generated content on multiple AI platforms; use the feedback data as training samples to incrementally update the multi-head attention weight parameters of step S3 and the platform adaptation network parameters of step S4, forming a closed loop of continuous optimization.

[0020] Furthermore, the GEO semantic multi-dimensional network also includes a clinical evidence semantic dimension and a safety semantic dimension. The semantic dimension of clinical evidence is configured to transform clinical trial conclusions into semantic vectors; encode the level of evidence-based medicine; and incorporate clinical guideline citations into the semantic graph. The safety semantic dimension is configured to compare the incidence of adverse reactions with similar drugs to generate safety comparison semantic features; and to assign higher weight to high-risk information. The system includes a clinical evidence semantic dimension extension submodule and a security guardrail submodule for the healthcare industry. These modules ensure that generated medical marketing content does not violate drug advertising review regulations and automatically filter out illegal promotional content for prescription drugs to the public. The clinical evidence semantic dimension and security semantic dimension are integrated with the technical semantic dimension, scenario semantic dimension, authority semantic dimension, and geographic traceability semantic dimension through node expansion and edge relationship updates of a graph neural network, forming an extended GEO semantic knowledge graph specifically for the healthcare industry.

[0021] This invention also discloses an AI marketing content generation system based on GEO trusted semantic optimization, comprising: The OID parsing and verification module is used to communicate with the OID registry center or its authorized mirror nodes to parse and verify the validity of OID identifiers in real time, and pass the parsing results as input to the semantic network construction module. The OID data acquisition and processing module is used to acquire and process brand OID trusted traceability data and output a structured data set. The semantic network construction module is used to build a multi-dimensional GEO semantic knowledge graph, including technical semantics, scenario semantics, authoritative semantics, and geographic origination semantics. The semantic vector generation module is used to encode features of each dimension into semantic vectors and generate GEO trusted semantic embedding vectors through a multi-head attention mechanism. The platform adaptive module is used to perform semantic space mapping and content feature adaptation for RAG preference features of different AI platforms. The content generation and publishing module is used to generate structured AI marketing content and deploy it to authoritative information sources. The feedback and iteration module is used to collect actual recommendation performance data and optimize model parameters.

[0022] The ID data acquisition and processing module is used to obtain the brand OID logo registered by the enterprise in the National OID Registration Center, collect the product full-chain traceability data associated with the OID, including raw material procurement information, production process parameters, quality inspection and certification data, third-party testing reports, warehousing and logistics records, and user QR code verification records. The traceability data is standardized, cleaned and aligned, and a structured data set is output. The semantic network construction module is used to build a GEO semantic multi-dimensional network based on structured data, which includes four dimensions: technical semantics, scene semantics, authority semantics, and geographic source semantics. It uses graph neural networks to fuse features of each dimension and generate a GEO semantic knowledge graph with OID identifier as the root node. The semantic vector generation module encodes the features of each dimension in the GEO semantic knowledge graph into semantic vectors, and uses a multi-head attention mechanism to adaptively weight and fuse the semantic vectors of each dimension to generate GEO trusted semantic embedding vectors. The platform adaptive module is used to obtain the RAG semantic matching preference features of the target AI generation platform. Through the platform adaptation network containing a learnable affine transformation layer, the general GEO trusted semantic embedding vector is mapped to the retrieval semantic space of the target platform. The content generation and publishing module is used to generate structured AI marketing content based on the adapted content feature expression, using a generative AI model, associating the content with OID identifiers through Schema.org structured markup, and deploying it to authoritative information source channels; The feedback and iteration module is used to collect feedback data such as the actual citation rate, semantic matching score, and user click conversion of the generated content across multiple AI platforms, and to incrementally update the multi-head attention weight parameters and platform-adaptive network parameters.

[0023] Furthermore, the AI ​​marketing content generation system based on GEO trusted semantic optimization also includes: The clinical evidence semantic dimension extension submodule is used to transform clinical trial data and evidence-based medicine evidence into semantic vectors, and then incorporate these semantic vectors into the clinical evidence semantic dimension of the GEO semantic knowledge graph. The security guardrail submodule is used to apply weight enhancement and mandatory inclusion control to security-related content during the content generation process. The weight enhancement refers to applying a preset multiple of enhancement factor to the attention weight of the security semantic dimension during the multi-head attention mechanism fusion process. The mandatory inclusion control refers to ensuring that security-related content is not omitted or weakened in the final generated structured AI marketing content.

[0024] The evidence-based medicine evidence includes clinical trial conclusions, the level of evidence-based medicine evidence, and the citation of clinical guidelines; the safety-related content includes at least contraindications, adverse reactions, and warning information. The enhancement factor, denoted as λsafe, is the core adjustable parameter of the safety guardrail submodule. It is used to apply a multiplicative amplification effect to the output weights of the safety semantic dimension (i.e., the attention heads that contain information such as contraindications, black box warnings, and adverse reactions) during the multi-head attention fusion stage.

[0025] This invention discloses an AI marketing content generation method and system based on GEO trusted semantic optimization. It uses OID as the unique digital identity of brand trusted data, transforming data assets into structured semantic features that can be directly identified, referenced, and trusted by AI large models. By constructing a semantic knowledge graph containing multiple dimensions, the AI ​​model can verify the authenticity of brand information from multiple perspectives. Through platform adaptation networks, semantic space mapping is performed on the RAG mechanism of the target AI platform, thereby enabling the same brand content to obtain high search visibility and citation probability on different AI platforms. Attached Figure Description

[0026] 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.

[0027] Figure 1 This is a flowchart illustrating an AI marketing content generation method based on GEO trusted semantic optimization according to the present invention.

[0028] Figure 2 This is a schematic diagram of the composition architecture of an AI marketing content generation system based on GEO trusted semantic optimization according to the present invention.

[0029] Figure 3This is a multi-dimensional network diagram illustrating an AI marketing content generation method based on GEO trusted semantic optimization, as described in this invention. Detailed Implementation

[0030] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0031] Please see Figures 1 to 3 , The specific implementation method is as follows: GEO Trusted Semantic Optimization: Taking a Geographical Indication Agricultural Product Brand as an Example Step S1 Implementation: Obtain trusted OID data Taking a certain tea brand as an example, the company has completed OID registration (OID: 1.2.156.3300.131) with the National OID Registry Center. Its products are connected to the national OID traceability system, enabling full-chain data collection and traceability from tea planting and processing to final sales. The company holds a legal food business license, and its products meet national food safety standards. Furthermore, the company has a centuries-old tea-making history, and its products have gained a certain degree of market recognition. This credible data serves as the brand's underlying data. Using the method of this invention, a GEO semantic knowledge graph is constructed, and AI marketing content optimization is performed. Raw material procurement information: The fresh tea leaves are produced in Xiamei Village, Wuyishan City, Fujian Province. The tea gardens are distributed on the seventy-two peaks of Wuyishan Mountain at an altitude of about 1,500-1,700 meters (represented by Dawangshan Tea Garden). The planting model of one variety per mountain is adopted. The harvesting time is spring 2025. Production process parameters: fixation temperature 260-280℃, rolling time 25 minutes, drying temperature 80-90℃; Quality inspection and certification data: The company's products comply with national food safety standards (GB 2763-2021 "National Food Safety Standard Maximum Residue Limits for Pesticides in Food" and GB 2762-2022 "National Food Safety Standard Limits for Contaminants in Food"); according to third-party testing institutions, the sensory, physicochemical and other indicators of the products meet the relevant standard requirements; in addition, the company owns multiple registered trademarks and has complete brand qualifications; Warehousing and logistics records: stored in a constant temperature environment of 15-18℃ and humidity of 50%-60%, with traceable cold chain transportation routes.

[0032] After cleaning the above data, a structured data set is formed.

[0033] Step S2 Implementation: Constructing a four-dimensional semantic knowledge graph Technical semantic dimension: Convert the fixation temperature of 260-280℃ into a technical feature vector and label it with a high temperature fixation semantic tag; convert the SGS pesticide residue not detected into a safety feature and generate a semantic node with the highest safety level; convert Wuyi rock tea into a variety feature and label it with a semantic tag for oolong tea (semi-fermented).

[0034] Scene semantic dimension: Four scene tags are labeled: family tea drinking, business tea gift, office tea break, and tea ceremony enthusiasts, and typical user question templates are generated for each tag: Recommendations for premium organic teas suitable for gifting. What are some refreshing teas to drink in the office? Which brand of Wuyi Rock Tea is the best? Which organic tea brands are trustworthy? Authoritative semantic dimension: The authoritative credentials obtained by the brand are structured and encoded, and a trusted association mapping is established with the OID identifier: OID traceability authority: The registration information of the National OID Registration Center (OID: 1.2.156.3300.131) is encoded as a semantic node for certification in the national traceability system, serving as the root node for brand trusted data; Geographical Indication Association: Historical and cultural landmarks such as the starting point of the Wuyishan Tea Road and the Zou Family Residence in Xiamei Village are encoded as semantic nodes of cultural geographical indications. Brand Qualification Authority: The registered trademarks held by the company are coded as semantic nodes for brand qualification certification and associated with OID identifiers.

[0035] Geographical origin semantic dimension: Construct a semantic map of the place of origin, including geographical feature tags such as Wuyi Mountain, Xiamei Village, the starting point of the Tea Road, seventy-two peaks and rocks, one variety per mountain, Danxia landform, weathered rock soil, rich in minerals, mountain flavor, century-old tea-making family, and Zoufu Dafu Residence.

[0036] By fusing the above four dimensions through a graph neural network, a brand knowledge graph is generated.

[0037] Step S3: Generate GEO trusted semantic embedding vectors A semantic encoder pre-trained using a BERT-based domain adaptation model encodes features of each dimension into a 768-dimensional vector. A 4-head attention mechanism is employed, with each of the four dimensions corresponding to an attention head, to calculate the final GEO-trusted semantic embedding vector after weighted fusion. Experiments show that this vector achieves a semantic similarity of 0.92 (cosine similarity) with frequently recommended AI tags such as safety, mineral nourishment, and gift-giving.

[0038] Step S4 Implementation: Dynamic Adaptation to Multiple Platforms For the DeepSeek platform: The platform's RAGs tend to prioritize citing paragraphs containing specific numerical values ​​and authoritative sources. The adaptation network strengthens the GEO trusted semantic embedding vectors with technical parameters such as temperature, altitude, and the number of detection items, as well as authoritative credential fields, generating content feature representations suitable for DeepSeek.

[0039] For the Doubao platform: This platform prefers structured question-and-answer (Q&A) and contextual descriptions. The adaptation network expands the contextual labels in the semantic vectors to generate preset question-and-answer pairs: "Q: Is this tea suitable as a gift? A: It has obtained geographical indication protection and organic certification, and the gift box packaging is elegant..." etc.

[0040] For the Wenxin Yiyan platform: This platform has a strong preference for Chinese cultural background and origin stories. Adapting the network increases the weight of the semantic dimension of geographical traceability, generating content expressions with cultural characteristics such as "nourished by Wuyi Mountain minerals" and "traditional tea-making techniques."

[0041] Step S5 Implementation: Content Generation and Publishing This embodiment uses Wenxin ERNIE 4.0 to call a generative AI model, based on the adapted content feature expression, and generates the following structured content displayed in JSON-LD format: The adjustments are as follows: { "@context": "https: / / schema.org", "@type": "Product", "name": "Zou Family Tea · Wuyi Rock Tea", "identifier": "oid:1.2.156.3300.131", "brand": { @type": "Brand", "name": "Zou Family Tea" }, "description": "Produced in Xiamei Village, Wuyishan City, Fujian Province. The tea gardens are distributed across the 72 peaks of Wuyishan Mountain, at an altitude of 1500-1700 meters, and adopt a 'one variety per mountain' planting model. The product meets the national standards GB 2763-2021 and GB2762-2022." "origin": { "@type": "Place", "name": "Xiamei Village, Wuyishan City, Fujian Province", "containedIn": "Seventy-two peaks and rocks of Wuyishan", "additionalProperty": [ {"@type": "PropertyValue", "name": "Altitude", "value": "1500-1700 meters"}, {"@type": "PropertyValue", "name": "Planting Mode", "value": "One variety per mountain"} ] }, "productionProcess": [ {"@type": "DefinedTerm", "name": "Firing Temperature","description": "260-280℃"}, {"@type": "DefinedTerm", "name": "Kneading Time","description": "25 minutes"}, {"@type": "DefinedTerm", "name": "Drying Temperature","description": "80-90℃"} ], "certification": [ { "@type": "Certification", "name": "National OID Traceability System Certification", "identifier": "oid:1.2.156.3300.131" }, { "@type": "Certification", "name": "Compliance with National Food Safety Standards", "standard": ["GB 2763-2021", "GB 2762-2022"] } ], "heritage": { "@type": "Thing", "name": "The Historical Heritage of the Zou Family Tea Making", }, "intendedUse": ["Home Tea", "Business Tea Gifts", "Office Tea Breaks", "Tea Ceremony Appreciation"] "typicalQuestions": [ { Question: Which brand of Wuyi Rock Tea is the most authentic? "Answer": "Zoufu Family Tea is produced in Xiamei Village, the core production area of ​​Wuyishan, and its OID (Original ID) is traceable." }, { Question: What kind of tea is a classy gift choice? "Answer": "Zou Family Tea - Wuyi Rock Tea, a century-old tradition, an excellent gift." } ] } The content was published on authoritative sources such as the brand's official website, the National Agricultural Product Quality and Safety Traceability Platform, and Xinhua News Agency's Traceability China Channel, and was also notified to major AI platforms to crawl it through sitemaps and structured data submission tools.

[0042] For the brand's official website: Publish the generated structured content as web pages through the backend publishing function of the website content management system (CMS), embed structured data in JSON-LD format into the web pages, and generate or update the sitemap.xml file in the website root directory, listing the URLs of all pages containing OID-related content and their last update time in accordance with the XML Sitemap protocol standard. For authoritative third-party platforms: Submit content through the certified source content submission interface provided by each platform, in accordance with the JSON or XML data format required by the platform, and attach an OID identifier as a unique index of the content; for platforms that do not provide public APIs, publish through the platform's officially recognized cooperative channels.

[0043] S6 Performance Verification Before implementing the method of this invention, when the brand searched for organic tea recommendations on the DeepSeek platform, the top 5 recommendations generated by AI did not include the brand. The citation rate on the Doubao platform is 0%. Monitoring data one month after implementation shows: DeepSeek: This brand appeared as the second most cited answer in organic tea recommendations, with its citation rate increasing to 23%. Doubao: In the Q&A section on tea suitable for gifting, the brand's relevant Q&A pair was fully quoted, resulting in a 3.2-fold increase in user click-through rate; ERNIE Bot: In the responses to "what are the high mountain teas", the brand's origin description is fully cited, and the search index increases by 210%. Specific Example 2 This example takes a certain prescription drug brand for cardiovascular diseases as an example, and specifies the application process of the method of the present invention in the medical and health field.

[0045] Implementation of Step S1: Medical-specific data collection This drug has been approved by the National Medical Products Administration (NMPA), with the approval number of National Medicine Approval Z12020025. The Tianjin Pharmaceutical Darentang Group to which this drug belongs has registered an OID identifier at the National OID Registration Center, with the code: 1.2.156.3300.143. Medical-specific data associated with this OID is collected through the system interface: NMPA approval information: including basic access information such as drug registration certificate (for example, the certificate number is in the format of "National Medicine Certificate Z + 8 digits"), drug approval number (for example, the format is "National Medicine Approval Z + 8 digits"), and the GMP certificate number of the manufacturer (for example, the format is "TJ + 8 digits"). All the above information is cross-verified through the official website of the National Medical Products Administration to ensure that the data is authentic and valid.

[0046] The objects include the full text of the drug specification, and the specific steps of structured processing include: extracting the chapters of the drug specification such as indications, usage and dosage, contraindications, adverse reactions, and drug interactions into independent fields respectively, performing field mapping and format standardization according to the preset data model to form a key-value pair structure. The indications include promoting qi and activating blood circulation, removing blood stasis and relieving pain, increasing coronary blood flow, and relieving angina pectoris; the usage and dosage include sublingual administration: 4-6 pills at a time, 3 times a day; during acute attack: 10-15 pills at a time; the drug specification is 40mg per pill; the contraindications include that those who are allergic to this product and pregnant women are prohibited from using it; the adverse reactions include that the incidence of gastrointestinal discomfort is about 5%-10%; the drug interactions include that the dosage needs to be adjusted when combined with diuretics; The clinical trial data includes that the clinical trial data is derived from the publicly disclosed Phase III clinical trial result announcement on the enterprise's official website and published medical literature: the sample size of the Phase III clinical trial is 1200 cases, the main angina pectoris relief compliance rate is 78.5%, and the safety data summary includes the incidence of adverse events, in which the incidence of grade ≥3 adverse events is 2.1%, the list of serious adverse events, and the withdrawal rate due to adverse events, etc.

[0047] Medical insurance information: Class A Catalogue of the National Medical Insurance (2025 Edition), the selected price in centralized procurement The above data is cleaned and de-identified to protect patient privacy, forming a structured data set.

[0048] Implementation of Step S2 — Construction of a medical-specific multi-dimensional semantic network Based on the four general dimensions of technical semantics, contextual semantics, authoritative semantics, and geographic source semantics, a fifth dimension—clinical evidence semantics—is added. Semantic dimension of clinical evidence: The conclusion of the clinical trial, that the effect of relieving angina pectoris is better than that of the control drug, and the difference is statistically significant (P<0.01), is converted into a semantic vector; the level of evidence-based medicine, such as Class IA evidence, is coded; the reference of clinical guidelines, such as the "Chinese Guidelines for the Prevention of Cardiovascular Diseases (2025 Edition)", is included in the semantic map; Safety semantic dimension, also known as extended authority semantic dimension: compare the incidence of adverse reactions and gastrointestinal discomfort rate of 5%-10% with similar drugs, which is 10%-15% lower than similar drugs, and generate safety comparison semantic features; give higher weight to high-risk information such as black box warnings and contraindications; Expanding the authoritative semantic dimension: linking drug compliance statements and "This advertisement is for medical professionals only" with OID.

[0049] The clinical evidence semantic dimension and the safety semantic dimension are added to the GEO semantic knowledge graph through node expansion of the graph neural network: with the OID identifier as the root node, new clinical evidence sub-nodes and safety sub-nodes are added, wherein there are supporting edges between the clinical evidence sub-nodes and the authoritative semantic sub-nodes, and there are constraint edges between the safety sub-nodes and the technical semantic sub-nodes. The graph neural network uses a message passing mechanism to jointly aggregate the features of the newly added dimensions with the original four-dimensional features to generate a five-dimensional GEO semantic knowledge graph specifically for the healthcare industry.

[0050] Step S3 Implementation – Generate Trusted Semantic Embedding Vectors for Medical GEO A medical domain semantic encoder, pre-trained using a language model based on BioBERT or finely tuned in the medical field, encodes features of each dimension into vectors. Specifically, an attention head weight enhancement mechanism is employed for the safety semantic dimension, giving safety-related information a higher weight in the vector; in this embodiment, the weight of contraindication information is three times that of ordinary information. A 5-head attention mechanism is used, with one attention head corresponding to each dimension, to generate the final medical GEO trusted semantic embedding vector.

[0051] In this invention, the enhancement factor (denoted as λsafe) is the core adjustable parameter of the safety guardrail submodule, which is used to apply a multiplicative amplification effect to the output weight of the safety semantic dimension (i.e., the attention head containing information such as contraindications, black box warnings, and adverse reactions) during the multi-head attention fusion stage.

[0052] Initial value setting rules The initial value of the enhancement factor is automatically preset by the system based on the security risk level of the target industry: For low-risk industries (such as ordinary food and daily necessities): λsafe=1.5, in which case safety information is only used as a general semantic feature in the fusion process; Medium-risk industries (such as health products and medical devices): λsafe=2.5, the weight of safety information is significantly higher than that of ordinary marketing descriptions; High-risk industries (such as prescription drugs and infant formula): λsafe=4.0~5.0, ensuring that safety semantics dominate the embedding vector.

[0053] Taking the quick-acting heart-saving pill in Example 2 as an example, after the system identifies the drug type (prescription drug) and NMPA approval number in the OID-related data, it automatically sets the enhancement factor to λsafe=4.0. During the multi-head attention calculation process, the original attention weight αsafe of the safety semantic dimension (calculated by the model's self-attention mechanism, initially about 0.15) is multiplied by the enhancement factor λsafe to become 0.60, and then normalized and redistributed, which increases the contribution of safety features in the final generated GEO trusted semantic embedding vector by 4 times.

[0054] (ii) Dynamic optimization mechanism (feedback closed loop) The enhancement factor is not fixed but dynamically optimized through the feedback and iteration module in step S6. The specific optimization algorithm employs Bayesian optimization or a multi-armed bandit strategy, using a combination of content compliance approval rate and AI platform citation rate as reward signals. When the system detects that a platform's content is marked as non-compliant or low-trust due to a lack of security information, the reinforcement learning agent automatically increments λsafe (step size +0.5). When excessive redundancy of security information leads to a significant decrease in AI citation rate, the agent should appropriately reduce λsafe (step size -0.2). The optimization process seeks the Pareto optimal balance between the two objectives mentioned above.

[0055] Step S4 Implementation – Dynamic Adaptation for Medical AI Platforms Medical content preferences for different AI platforms: For DeepSeek Healthcare: This platform prefers to cite clinical guidelines and authoritative database information. It adapts its network to enhance the semantic dimensions of clinical evidence and authoritative sources, generating structured content that includes guideline citations and clinical trial data. Regarding Baidu Health AI Assistant: This platform has a preference for patient education content. The adapted network will expand information of concern to patients, such as usage, dosage, and precautions, into a question-and-answer format. For Haodf.com's AI-powered online consultation: the platform needs to adapt to the information needs of doctors. This includes adapting the network output to include a high-level content view containing detailed pharmacological parameters and clinical research data. For medical professional platforms such as Medlive and DXY: Adapt to the network and strengthen the compliance semantic dimension to ensure that the output content meets the regulatory requirements for the dissemination of drug information.

[0056] At the same time, it filters according to the medical compliance requirements of different countries / regions: it mandates the inclusion of a compliance statement that this advertisement is for medical professionals only; when outputting to public platforms, it automatically filters out illegal content that promotes prescription drugs directly to patients, and only allows the display of drug instruction leaflet information.

[0057] Step S5 Implementation – Credible Medical Content Generation and Publishing This embodiment uses a generative AI model, specifically Baidu Wenxin ERNIE Medical Edition, to generate structured medical marketing content (displayed in JSON-LD format) based on the adapted content feature representation. { "@context": "https: / / schema.org", "@type": "Drug", "name": "Quick-acting Heart-saving Pills", "identifier": "oid:1.2.156.3300.143", "manufacturer": { "@type": "Organization", "name": "Jin Yao Darentang Group Co., Ltd." }, "approvalNumber": "National Drug Approval Number Z12020025", "Indications": "Promotes the flow of qi and blood, removes blood stasis and relieves pain, increases coronary blood flow and alleviates angina pectoris. Used for coronary heart disease and angina pectoris of the qi stagnation and blood stasis type." "dosageForm": "droplets", "strength": "40mg / pill", Dosage: Sublingual administration of 4-6 pills three times a day; for acute attacks, 10-15 pills at a time. "Contraindications": "This product is contraindicated in pregnant women and those with known hypersensitivity to it." "adverse reactions": "The incidence of gastrointestinal discomfort is approximately 5%-10%", "clinicalTrial": { "@type": "MedicalStudy", "studyType": "Phase III clinical trial", "sampleSize": 1200, "primaryEndpoint": "Angina relief rate: 78.5%", "significanceLevel": "P<0.01", "evidenceLevel": "Class IA" }, "guideline Recommendation": "Medications recommended in the Chinese Guidelines for the Prevention of Cardiovascular Diseases (2025 Edition)" "medicalInsurance": "National Medical Insurance Category A Catalog (2025 Edition)", "compliance": This information is for medical professionals only or for healthcare professionals only. } The generated content is published to the National Drug Traceability Collaboration Platform and authoritative medical media. In this embodiment, authoritative medical information sources such as China Medical Network and hospital internal systems are used.

[0058] Among them, the National Drug Traceability Collaboration Platform: Drug marketing authorization holders log in to the National Drug Traceability Collaboration Service Platform and manually upload data through the web page input interface provided by the platform, or automatically upload structured drug information data containing OID identifiers through the API interface in accordance with the "Manufacturing Enterprise-Drug Consumables Traceability Information Interface Connection Scheme"; the uploaded data, after being reviewed by the collaboration platform, becomes part of the drug traceability data system; For authoritative medical media: For medical professional platforms that support independent publication by institutional accounts (such as Medlive, DXY, etc.), log in to the backend through the platform's enterprise-verified account and submit medical information content containing the OID identifier according to the content format required by the platform; for academic journal platforms that only accept manuscript submissions (such as China Medical Network), submit through the platform's online submission system and attach the OID identifier and credible data proof in the manuscript. Internal hospital systems: Through the data distribution mechanism of the National Drug Traceability Collaboration Platform, structured drug information is pushed to various drug traceability systems. The hospital information system downloads drug data from the traceability systems via interfaces. For hospitals that have established data integration partnerships, structured drug information (including OID identifiers) is pushed to the hospital's drug information management system through the standardized API interface provided by the hospital information platform (following the FHIR technical specifications and the WS / T 846 series hospital information platform interaction standards). Step S6: Monitoring data three months after implementation shows: In professional information verification scenarios: When medical professionals use DeepSeek Medical Edition to query the NMPA approval information of a drug, the structured instruction manual data for that drug appears as the second result in the search results, compared to the ninth result before optimization, significantly improving the efficiency of medical professionals' access to information. In Baidu Health AI Assistant, when patients scan a code to verify drug information, the drug instructions provided by the AI ​​are 99.2% consistent with the instructions approved by the National Medical Products Administration (NMPA) (compared to 78.5% before optimization), effectively reducing potential medication safety risks caused by inconsistencies in information.

[0059] Regarding compliant dissemination: All medical and health content generated and published by the system on the National Drug Traceability Collaboration Platform achieved a 100% compliance pass rate within three months. By mandating the inclusion of a compliance statement that this information is for medical professionals only, and by automatically filtering prescription drug promotion content on public platforms, no compliance risk events occurred during the entire AI content generation and publication process.

[0060] In terms of information traceability and consistency: After binding information through the OID identifier (1.2.156.3300.143), the third-party traceability verification system can compare the AI-generated content with the original drug registration certificate in real time, realizing the verifiability and traceability of the content, and providing a credible solution that meets regulatory requirements for the digital marketing of prescription drugs.

[0061] The above-disclosed embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of the invention. Those skilled in the art will understand that implementing all or part of the above-described embodiments and making equivalent changes in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. An AI marketing content generation method based on GEO trusted semantic optimization, characterized in that, Includes the following steps: S1: Obtain the Brand Object Identifier (OID) registered by the enterprise in the National OID Registry Center, collect the full-chain traceability data of the product associated with the OID, and perform standardized cleaning and alignment of the traceability data to form a structured data set at the bottom layer of brand trust. S2: Based on structured datasets, construct a multi-dimensional semantic network for the brand's GEO; fuse features from each dimension using graph neural networks to generate the brand's GEO semantic knowledge graph; S3: Encode the features of each dimension in the GEO semantic knowledge graph into semantic vectors, and use a multi-head attention mechanism to adaptively weight and fuse the semantic vectors of each dimension to generate a brand’s trusted semantic embedding vector. S4: Obtain the RAG semantic matching preference features of the target AI generation platform, and through a platform adaptation network containing a learnable affine transformation layer, map the GEO trusted semantic embedding vector to the retrieval semantic space of the target platform to generate content feature expressions that are compatible with the target AI platform. S5: Use the adapted content feature expression from step S4 to optimize the Top-K recall ranking of the target AI generation platform during RAG retrieval. Then, based on the adapted content feature expression, use a generative AI model to generate structured AI marketing content. Release the generated AI marketing content through authoritative information sources and associate the content with the OID identifier during the release process so that it can be automatically recognized and crawled by AI search engines and large model RAG systems.

2. The AI ​​marketing content generation method based on GEO trusted semantic optimization as described in claim 1, characterized in that, The AI ​​marketing content generation method also includes the following steps: S6: Collect feedback data of the generated content on multiple AI platforms, use the feedback data as training samples, and incrementally update the multi-head attention weight parameters of step S3 and the platform adaptation network parameters of step S4 to form a closed loop of continuous optimization.

3. The AI ​​marketing content generation method based on GEO trusted semantic optimization as described in claim 1, characterized in that, In step S1, the traceability data includes raw material procurement information, production process parameters, quality inspection and certification data, third-party testing reports, warehousing and logistics records, and user QR code verification records.

4. The AI ​​marketing content generation method based on GEO trusted semantic optimization as described in claim 1, characterized in that, In step S2, the GEO semantic multidimensional network includes at least three dimensions: technical semantic dimension, scene semantic dimension, authority semantic dimension, and geographic tracing semantic dimension.

5. The AI ​​marketing content generation method based on GEO trusted semantic optimization as described in claim 4, characterized in that, The technical semantic dimension is configured to map production operation data into technical feature vectors in the AI ​​semantic vector space; the scenario semantic dimension is configured to semantically label user usage scenarios adapted to brand products and associate each label with a typical user question template; the authority semantic dimension is configured to structure and encode authoritative credential information and establish a trusted association mapping with OID identifiers; the geographic traceability semantic dimension is configured to semantically express spatial location data to form a spatial geographic semantic map.

6. The AI ​​marketing content generation method based on GEO trusted semantic optimization as described in claim 1, characterized in that, In step S2, the graph neural network aggregates features from various dimensions through a message passing mechanism to generate a GEO semantic knowledge graph with the OID identifier as the root node.

7. The AI ​​marketing content generation method based on GEO trusted semantic optimization as described in claim 1, characterized in that, In step S3, the GEO trusted semantic embedding vector is generated in the following way: the feature vectors of each semantic dimension are used as input to the multi-head attention mechanism, and the outputs of each attention head are concatenated and multiplied by the output weight matrix to obtain the fused semantic embedding vector.

8. The AI ​​marketing content generation method based on GEO trusted semantic optimization as described in any one of claims 1 to 7, characterized in that, The GEO semantic multidimensional network also includes a clinical evidence semantic dimension and a safety semantic dimension. The semantic dimension of clinical evidence is configured to transform clinical trial conclusions into semantic vectors; encode the level of evidence-based medicine; and incorporate clinical guideline citations into the semantic graph. The safety semantic dimension is configured to compare the incidence of adverse reactions with similar drugs to generate safety comparison semantic features. Give higher weight to high-risk information.

9. An AI marketing content generation system based on GEO trusted semantic optimization, characterized in that, include: The OID parsing and verification module is used to communicate with the OID registry center or its authorized mirror nodes to parse and verify the validity of OID identifiers in real time, and pass the parsing results as input to the semantic network construction module. The OID data acquisition and processing module is used to acquire and process brand OID trusted traceability data and output a structured data set. The semantic network construction module is used to build a multi-dimensional GEO semantic knowledge graph, including technical semantics, scenario semantics, authoritative semantics, and geographic origination semantics. The semantic vector generation module is used to encode features of each dimension into semantic vectors and generate GEO trusted semantic embedding vectors through a multi-head attention mechanism. The platform adaptive module is used to perform semantic space mapping and content feature adaptation for RAG preference features of different AI platforms. The content generation and publishing module is used to generate structured AI marketing content and deploy it to authoritative information sources. The feedback and iteration module is used to collect actual recommendation performance data and optimize model parameters.

10. The AI ​​marketing content generation system based on GEO trusted semantic optimization as described in claim 9, characterized in that, The AI ​​marketing content generation system based on GEO trusted semantic optimization also includes: The clinical evidence semantic dimension extension submodule is used to transform clinical trial data and evidence-based medicine evidence into semantic vectors, and then incorporate these semantic vectors into the clinical evidence semantic dimension of the GEO semantic knowledge graph. The security guardrail submodule is used to apply weight enhancement and mandatory inclusion control to security-related content during the content generation process. The weight enhancement refers to applying a preset multiple of enhancement factor to the attention weight of the security semantic dimension during the multi-head attention mechanism fusion process. The mandatory inclusion control refers to ensuring that security-related content is not omitted or weakened in the final generated structured AI marketing content.

Images