Intelligent short message content generation and accurate reaching method based on AI semantic analysis

Abstract

The invention relates to the technical field of short message pushing, in particular to an intelligent short message content generation and accurate reaching method based on AI semantic analysis, and the method comprises the steps: obtaining user portrait data containing user tags, preferences and behavior tracks, and generating content theme tags based on the user preferences; the system intelligently judges whether a user tag is included or not by analyzing a user Tida password; if yes, the corresponding personalized template is directly called; if not, extracting a marketing intention, matching user portraits, screening a core user group, and combining to generate target short message content; performing compliance verification on the basis of content theme tags, filtering sensitive words to generate standard short message content, analyzing semantic features through a content strategy library constructed by AI, and matching an optimal touch scheme; according to the method, a standard content injection reach scheme is dynamically optimized, a reach instruction is generated, accurate pushing and real-time effect evaluation are executed, an intelligent marketing closed loop is formed, and whole-process automation of personalized generation and effect optimization of short message contents is realized.

Description

Technical Field

[0001] This invention relates to the field of SMS push technology, and in particular to a method for intelligent SMS content generation and precise delivery based on AI semantic analysis. Background Technology

[0002] With the rapid development of mobile internet and digital marketing, SMS, as an efficient and direct communication method, is widely used in scenarios such as customer outreach, marketing promotion, and service notifications. Traditional SMS sending methods often use bulk sending and fixed templates, resulting in highly homogenized content and a lack of personalization and scenario adaptability. This leads to low user open rates, poor conversion rates, and even user complaints and unsubscriptions due to sensitive content or frequent pushes, seriously affecting user experience and brand image.

[0003] Existing SMS systems have explored intelligent features to some extent, with some platforms introducing simple user segmentation and template variable replacement mechanisms to achieve basic personalized content generation. However, these methods still rely on manually preset rules and lack the ability to understand users' deep behavioral preferences and semantic intent, making it difficult to achieve truly personalized content generation. Furthermore, SMS delivery strategies are mostly based on fixed times or frequencies, failing to optimize for dynamic factors such as user activity cycles and historical response behavior, leading to inappropriate delivery timing and low dissemination efficiency. In addition, with the strengthening of national regulation of communication and information services, compliance requirements for SMS content are becoming increasingly stringent, with sensitive word filtering, marketing frequency control, and user authorization becoming essential considerations. Traditional keyword matching review mechanisms struggle to handle semantically variable and contextually complex marketing language, easily resulting in missed detections or misjudgments. Therefore, there is an urgent need for an intelligent SMS generation and delivery method that can deeply integrate user profiles, understand semantic intent, automatically generate compliant content, and achieve accurate delivery and a closed-loop feedback loop for results.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide an intelligent SMS content generation and precise delivery method based on AI semantic analysis, which aims to solve the technical problems of existing SMS marketing systems, such as content homogenization, rigid delivery strategies, lack of semantic understanding and compliance verification capabilities, making it difficult to achieve personalized generation, precise delivery and efficient conversion.

[0006] To achieve the above objectives, this invention provides a method for intelligent SMS content generation and precise delivery based on AI semantic analysis, the method comprising: Acquire user profile data, which includes user tags, user preferences, and user behavior patterns, and generate content topic tags based on the user preferences; Set up SMS generation and outreach strategies, receive and parse user outreach messages, determine whether the user outreach message contains the user tag, if yes, retrieve the personalized SMS template corresponding to the user tag based on the generation and outreach strategy, and set it as the target SMS content; otherwise, extract the marketing intent of the user outreach message, match the marketing intent with the user profile data, filter the core user group that matches the marketing intent from all users, combine the personalized SMS template of the core user group based on the generation and outreach strategy, and set it as the target SMS content; The compliance of the target SMS content is verified based on the content topic tags, sensitive words and redundant information are filtered to generate standard SMS content, a content strategy library is built based on AI technology, the semantic features of the standard SMS content are analyzed, and the optimal outreach scheme is matched in the content strategy library based on the semantic features and the user outreach message. The standard SMS content is injected into the optimal outreach scheme for dynamic optimization, and outreach instructions are generated based on the user outreach phrases. Based on the outreach instructions, precise SMS outreach and real-time effect evaluation are performed to complete the intelligent marketing closed loop corresponding to the user outreach phrases.

[0007] Optionally, the SMS generation and outreach strategy can be set based on the following steps: The user associated with the user profile data is set as the target user. A preset number of SMS template groups are configured for the target user. The personalized content of the target user is stored in any template group. The target user and all the corresponding template groups are integrated to construct a status scheduling table. The status scheduling table updates the content usage rate of all template groups in real time. The status scheduling table dynamically switches the content generation status of any template group based on the content usage rate to complete the SMS content generation process. After parsing the user outreach message, a content retrieval request is initiated to the template group based on the user tag. Based on the content retrieval request, all personalized content fragments stored in the template group are extracted sequentially to complete the dynamic assembly process of the SMS content. The generation process and the assembly process are integrated and set as the SMS generation and outreach strategy.

[0008] Optionally, generating the target SMS content includes the following steps: The user tag or the core user group in the user outreach message is set as the outreach target. A content retrieval request is sent to the template group corresponding to the outreach target. The content retrieval request includes the content theme tag and timeliness requirement. The matching degree of all template groups is verified based on the content retrieval request. If the template group matches the content retrieval request, the corresponding stored content fragment is output. The content fragments of all template groups are merged based on the timeliness requirement and set as the target SMS content.

[0009] Optionally, the user reach message may be determined to contain the user tag based on the following steps: The user reach message is deconstructed using a deep learning semantic parsing model to generate extended demand semantics. Parsing tags are set, which include reach object identifiers and marketing intents. Based on the parsing tags, the target description statement corresponding to the reach object identifier and the scenario description statement corresponding to the marketing intent are extracted from the extended demand semantics. Based on text embedding similarity, determine whether there is a semantic match for the user tag in the target description statement. If there is, determine that the user outreach message contains the user tag. Otherwise, redefine the extended requirement semantics as the marketing intent corresponding to the user outreach message.

[0010] Optionally, the process of selecting a core user group that aligns with the marketing intent from all users includes the following steps: The textual features of the marketing intent are semantically matched with the user behavior trajectory and the keywords of the user preference of any user. The cross-modal similarity between the two is calculated by a pre-trained language model and set as a first value. If the first value is greater than a first threshold, then any user is included in the core user group.

[0011] Optionally, matching the optimal outreach scheme in the content strategy library based on the semantic features and the user outreach message includes the following steps: Define strategy tags, and the AI ​​technology generates multiple content generation strategies based on the strategy tags. Organize all the content generation strategies hierarchically according to the strategy tags to build the content strategy library. The strategy tags include strategy scenarios, conversion paths, compliance risks, and content iteration logic. Based on the SMS compliance verification process, sensitive semantics in the target SMS content corresponding to the same content theme tags are identified and filtered, and the valid content retained is set as the standard SMS content. Based on the natural language generation algorithm, the feature extraction rules corresponding to the content topic tags are configured, and the semantic features of the standard SMS content are parsed through the feature extraction rules; In the content iteration logic, strategies that match the semantic features and the content topic tags are selected to form a candidate strategy set. The semantic association strength between the user outreach message and the strategy scenario of any candidate strategy is calculated and set as a second value. The strategy with the highest second value is set as the optimal outreach solution.

[0012] Optionally, the step of performing precise SMS outreach and real-time effect evaluation based on the outreach instruction includes the following steps: The optimal outreach scheme is trained using reinforcement learning based on the standard SMS content to generate a prediction model. The prediction model outputs an outreach response probability sequence based on the semantic features. The timing and channel for SMS push are determined based on the outreach response probability sequence and set as the outreach instruction. An effect tracking table is constructed. Based on the reach response probability sequence and real-time SMS conversion data, an operational intervention strategy is matched in the effect tracking table and set as the intelligent marketing closed loop. The operational intervention strategy includes a secondary reach threshold, dynamic content adjustment rules, and a user churn early warning mechanism.

[0013] Furthermore, to achieve the above objectives, the present invention also provides an intelligent SMS content generation and precise delivery device based on AI semantic analysis. The device includes: a memory, a processor, and an AI semantic analysis-based intelligent SMS content generation and precise delivery program stored in the memory and executable on the processor. The AI ​​semantic analysis-based intelligent SMS content generation and precise delivery program is configured to implement the steps of the AI ​​semantic analysis-based intelligent SMS content generation and precise delivery method as described above.

[0014] Furthermore, to achieve the above objectives, the present invention also provides a medium storing an AI semantic analysis-based intelligent SMS content generation and precise delivery program, wherein when the AI ​​semantic analysis-based intelligent SMS content generation and precise delivery program is executed by a processor, it implements the steps of the AI ​​semantic analysis-based intelligent SMS content generation and precise delivery method as described above.

[0015] This invention provides an intelligent SMS content generation and precise delivery method based on AI semantic analysis. The method integrates multi-dimensional profile data such as user tags, user preferences, and behavioral patterns, and combines this with AI semantic analysis to generate content theme tags. This achieves refined content generation, moving from personalized recommendations to tailored strategies for each user, significantly enhancing the match between SMS content and user interests, and increasing user open rates and reading intentions. It introduces a user delivery command parsing mechanism, which can automatically identify marketing intent and supports two generation paths: precise template calling for explicit user tags, or automatic filtering of core user groups and combination of personalized templates for ambiguous commands, improving the system's adaptability and flexibility in diverse marketing scenarios. Semantic-level compliance verification is performed based on content theme tags, combined with sensitive word filtering and redundant information... SMS cleanup not only effectively mitigates policy risks but also enhances the conciseness and professionalism of SMS content, reducing user complaints and unsubscription rates, and safeguarding brand image. By building an AI-driven content strategy library, it matches the optimal outreach solution based on semantic features and user engagement phrases, upgrading strategies from experience-driven to data-driven and semantic-driven, thereby improving outreach success rate and conversion efficiency. It organically integrates content generation, strategy matching, outreach execution, and real-time performance evaluation to form a complete intelligent marketing closed loop, supporting dynamic optimization and operational intervention based on response data, enabling continuous iteration and performance improvement of marketing campaigns. The fully automated process significantly reduces the burden of manual template configuration, strategy setting, and content review, improving enterprise marketing efficiency and making it suitable for large-scale, high-frequency digital marketing scenarios. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating an embodiment of the intelligent SMS content generation and precise delivery method based on AI semantic analysis of the present invention.

[0017] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] Reference Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the intelligent SMS content generation and precise delivery method based on AI semantic analysis of the present invention, which presents an embodiment of the intelligent SMS content generation and precise delivery method based on AI semantic analysis of the present invention.

[0020] In one embodiment, the AI-based semantic analysis-based intelligent SMS content generation and precise delivery method includes: Step S100: Obtain user profile data, which includes user tags, user preferences, and user behavior patterns. Generate content topic tags based on user preferences.

[0021] User profile data can be a multi-dimensional set of user features consisting of user tags, user preferences, and user behavior trajectories. This data can serve as the foundational input for content generation and outreach decisions, supporting personalized content matching and user intent inference. In this embodiment, user profile data can be continuously collected and aggregated from event logs, transaction records, interaction behaviors, and user attribute databases to form dynamic feature vectors. For example, user profile data can include, but is not limited to, one or more of static tags, dynamic preferences, and time-series behavioral trajectories.

[0022] Content topic tags can be semantic classification identifiers that reflect content interest tendencies and are abstracted from user preference semantics. They can be used to achieve semantic alignment between SMS content and user interests, replacing the coarse-grained distribution of traditional keyword matching. In this embodiment, content topic tags can utilize natural language processing models to perform topic modeling on user preference text, extract high-frequency semantic clusters, and map them into structured tags.

[0023] Generating content topic tags based on user preferences can involve semantically encoding user preference text, extracting high-dimensional semantic features, and clustering them to generate tags. Furthermore, generating content topic tags based on user preferences can be achieved by using the BERT model to contextually encode users' historical browsing and click text, generating topic tags through LDA clustering, or by using Word2Vec+K-Means to perform vector space clustering of user search keyword sequences, outputting semantic topic identifiers. This allows for an abstract transformation from raw behavioral data to semantically guided content.

[0024] Step S200: Set SMS generation and outreach strategy, receive and parse user outreach messages, determine whether the outreach message contains user tags, if so, retrieve the personalized SMS template corresponding to the user tags based on the generation and outreach strategy, and set it as the target SMS content; otherwise, extract the marketing intent of the outreach message, match the marketing intent with user profile data, filter the core user group that matches the marketing intent among all users, combine the personalized SMS template of the core user group based on the generation and outreach strategy, and set it as the target SMS content.

[0025] The reach commands can be natural language instructions input by marketers expressing marketing intent. These instructions can serve as semantic entry points for triggering system behavior, driving the selection of content generation paths and strategy matching. In this embodiment, reach commands can receive structured or unstructured text input through a front-end interface or API, which is then parsed by a word segmentation and intent classification module. For example, reach commands can include, but are not limited to, one or more of the following: precise instruction type, fuzzy intent type, and scenario combination type.

[0026] Parsing out messaging and matching marketing intent to filter core user groups can be achieved by classifying messaging intent and then performing semantic similarity matching between the classification results and user profile data. Further, this can be done by using text classification models to identify messaging intent categories and then using cosine similarity ranking based on user tag vectors for filtering, or by constructing a semantic graph and performing path reasoning between messaging keywords and user behavior graph nodes to locate potential target groups. This allows for the automatic mapping from vague marketing instructions to precise user groups.

[0027] Step S300: Verify the compliance of the target SMS content based on the content topic tags, filter sensitive words and redundant information to generate standard SMS content, build a content strategy library based on AI technology, analyze the semantic features of the standard SMS content, and match the optimal outreach plan in the content strategy library based on the semantic features and outreach commands.

[0028] The AI-driven content strategy library can be an intelligent database built based on historical outreach data and semantic features, storing the mapping relationship between outreach plans and their effects. It can be used to achieve semantic matching and optimal selection of outreach strategies, supporting automated strategy iteration. In this embodiment, the AI-driven content strategy library can perform semantic vector modeling and clustering storage of historical SMS content, outreach time, and user response results through supervised learning. For example, the AI-driven content strategy library may include, but is not limited to, one or more of the following: time window strategy, user response weight strategy, and channel collaboration strategy.

[0029] Verifying the compliance of target SMS content based on content topic tags can involve comparing the semantic consistency of the SMS content with the content topic tags to identify sensitive expressions that deviate from the topic. Furthermore, this compliance verification can also utilize semantic similarity models to detect the presence of promotional terms unrelated to the topic, or construct a topic-sensitive word association graph. When the SMS content deviates from the topic tags, a hidden violation warning can be triggered, thereby improving the compliance verification's ability to identify context-dependent violations.

[0030] Matching the optimal outreach strategy within a content strategy library based on semantic features and outreach commands can be achieved by jointly inputting the semantic vector of the standard SMS content and the intent vector of the outreach command into the strategy library and retrieving historical best matches. Furthermore, this matching can be accomplished by employing a dual-tower neural network to jointly encode the SMS content and commands, matching historical best outreach combinations through vector retrieval, or by using a graph neural network to model the relationships between outreach strategies in the strategy library and recommending suitable solutions through path reasoning. This allows for the migration of outreach strategies from manual experience to a semantic data-driven approach.

[0031] Step S400: Inject the standard SMS content into the optimal outreach plan for dynamic optimization, generate outreach instructions based on the outreach phrases, and execute precise SMS outreach and real-time effect evaluation based on the outreach instructions to complete the intelligent marketing closed loop corresponding to the outreach phrases.

[0032] Dynamically optimizing standard SMS content by injecting it into the optimal outreach plan can involve combining the generated standard SMS with the matching outreach plan, injecting it into the execution engine, and collecting response feedback in real time. Furthermore, this dynamic optimization can also collect open, click, and unsubscribe behaviors within 10 minutes of SMS sending, feeding this feedback to the strategy library to update weight parameters, or re-encoding the response data with semantic features to trigger the strategy library's online incremental learning mechanism. This allows for a closed-loop optimization process from generation to outreach to feedback.

[0033] Taking personalized promotional outreach during e-commerce promotional events as an example, the AI-based semantic analysis-based intelligent SMS content generation and precise outreach method in this embodiment can be as follows: marketers input outreach commands to push limited-time discounts to female users who have recently browsed but not placed orders. The system parses the commands as fuzzy intent-type instructions, combines user profile data to filter out female users who have browsed beauty products and are currently active but have not purchased, generates exquisite lifestyle theme tags based on their preferences, automatically combines semantically adapted templates including exclusive privileges and limited-time exclusives, removes illegal words such as "cheapest" and "absolute" after semantic compliance verification, matches the strategy library with the priority outreach plan for 20:00+ APP push notifications with historically high conversion rates, and collects opening and clicking behaviors in real time after sending, updating the weight of the intent-theme combination in the strategy library.

[0034] This embodiment provides an intelligent SMS content generation and precise delivery method based on AI semantic analysis. It achieves semantic abstraction of content by acquiring user profile data and generating content theme tags based on user preferences; it filters core user groups by parsing delivery commands and matching them with marketing intentions, driving intelligent decision-making paths; it enhances the ability to identify context-dependent violations by verifying the compliance of target SMS content based on content theme tags; it achieves automated strategy recommendation by matching the optimal delivery plan in the content strategy library based on semantic features and delivery commands; and it dynamically optimizes the optimal delivery plan by injecting standard SMS content into it, completing a closed-loop feedback of generation, delivery, and evaluation. This achieves the technical effect of improving content matching accuracy and response efficiency.

[0035] In some embodiments, the SMS generation and outreach strategy is set based on the following steps: users associated with user profile data are set as target users; a preset number of SMS template groups are configured for target users; personalized content of target users is stored in any template group; target users and all corresponding template groups are integrated to construct a state scheduling table; the state scheduling table updates the content usage rate of all template groups in real time; the state scheduling table dynamically switches the content generation state of any template group based on the content usage rate, thus completing the SMS content generation process; after parsing the outreach command, a content retrieval request is initiated to the template group based on the user tag; personalized content fragments stored in all template groups are extracted sequentially based on the content retrieval request, thus completing the dynamic assembly process of the SMS content; the generation process and the assembly process are integrated and set as the SMS generation and outreach strategy.

[0036] The status scheduling table can be a dynamic mapping structure that records multiple template groups associated with a target user and their content usage rates. It can be used to implement resource scheduling and priority management of personalized content fragments, supporting adaptive activation and switching of template groups. In this embodiment, the status scheduling table constructs an independent scheduling unit for each user based on user profile data, continuously calculating the call frequency and response feedback of content fragments in each template group, and updating the usage rate weights. For example, the status scheduling table can include, but is not limited to, one or more of the following: high-frequency activation groups, low-frequency optimization groups, and cold-start verification groups.

[0037] Based on user profile data, a status scheduling table is constructed and the usage rate of template group content is updated in real time. This can be achieved by creating a set of template groups for each target user, statistically analyzing the number of times content segments within each group are called and user response behavior, and dynamically adjusting the usage rate weights. Furthermore, this operation can be implemented by using a sliding time window to statistically analyze the call frequency and click-through rate of each template group over the past 7 days, and then calculating the usage rate using a weighted average. Alternatively, it can be achieved by dynamically assigning different weight decay coefficients to template groups based on user lifecycle stages. This allows for dynamic priority management of template group resources, preventing inefficient content from occupying system resources for extended periods.

[0038] Initiating a content retrieval request to a template group based on the reach command can be achieved by generating a structured retrieval instruction based on user tags and content topic tags after parsing the reach command, thus triggering the template group content extraction process. Furthermore, this operation can be implemented by encoding the command intent as a tag combination key and mapping it to the template group index list in the state scheduling table, or by automatically preloading candidate fragments of corresponding template groups for high-frequency commands in conjunction with historical retrieval patterns. This enables semantic linkage between marketing intent and content resources, improving the response efficiency of dynamic assembly.

[0039] Content retrieval requests can be structured instructions triggered by a notification, used to extract personalized content fragments from a template group. These instructions can drive the dynamic assembly of SMS content, enabling a paradigm shift from complete templates to semantic fragment splicing. In this embodiment, after parsing the notification, the content retrieval request generates a unique retrieval identifier based on user tags and topic tags, and then searches the status scheduling table for available content fragments corresponding to the template group. For example, content retrieval requests can include, but are not limited to, single-fragment retrieval, multi-fragment combination retrieval, priority-ordered retrieval, or one or more of these.

[0040] Extracting personalized content fragments from all template groups based on the content call request order can be achieved by extracting and concatenating content fragments that meet the conditions from multiple template groups in a preset order according to the tag matching rules carried in the content call request. Furthermore, this operation can be implemented by sorting by theme tag consistency, prioritizing the extraction of fragments with the highest matching degree with the content theme tags, or by prioritizing the call of content fragments from high-frequency activation groups based on the usage rate weight in the status scheduling table. This allows for the transformation of SMS content from a fixed template to a semantically driven fragment combination mode.

[0041] Taking the tiered reactivation promotion of maternal and infant brand members as an example, the AI ​​semantic analysis-based intelligent SMS content generation and precise outreach method in this embodiment can target users who have recently browsed baby food but have not purchased it. Marketers can input outreach commands to recommend highly rated organic rice cereal. The system parses the commands into precise instruction-type commands, generates parenting and health-related tags based on user profiles, and triggers content call requests. The status scheduling table displays three template groups associated with the user: high-frequency purchase group (usage rate 0.85), new product trial group (usage rate 0.32), and promotion sensitive group (usage rate 0.18). The system extracts organic certification segments from the high-frequency purchase group and limited-time trial segments from the new product trial group according to priority, and splices them into a complete SMS. After semantic compliance verification, the optimal outreach plan is injected and sent. After the response data feedback, the usage rate of the new product trial group in the scheduling table is updated to 0.41.

[0042] This embodiment sets users associated with user profile data as target users, constructs a status scheduling table, and updates the usage rate of template group content in real time. It initiates content call requests to template groups based on reach commands, and extracts personalized content fragments stored in all template groups according to the content call request order. This achieves resource popularity modeling and dynamic priority management of personalized content fragments, breaking through the rigid structure of traditional static templates. It transforms marketing intent into structured call instructions, drives fine-grained semantic fragment splicing, and upgrades SMS generation from complete template replacement to a semantically driven combination mode. This mechanism works in conjunction with content theme tags to embed compliance verification into fragment-level processing. While ensuring personalization and compliance, it significantly improves template reuse efficiency and system response elasticity, supporting automated closed-loop operation of one policy per person in large-scale scenarios.

[0043] In some embodiments, generating target SMS content includes the following steps: setting user tags or core user groups in the outreach message as outreach targets, sending a content retrieval request to the template group corresponding to the outreach target, wherein the content retrieval request includes content theme tags and timeliness requirements, verifying the matching degree of all template groups based on the content retrieval request, if the template group matches the content retrieval request, outputting the corresponding stored content fragment, and merging the content fragments of all template groups based on the timeliness requirements and setting them as target SMS content.

[0044] Content topic tags can be semantic classification identifiers based on user preference semantics, reflecting content interest tendencies. They can be used as a semantic verification dimension for content call requests, aligning content fragments with user interest intentions. In this embodiment, content topic tags can utilize natural language processing models to perform topic modeling on user preference text, extracting high-frequency semantic clusters and mapping them into structured tags. For example, content topic tags can include, but are not limited to, one or more of consumption intent categories, sentiment categories, and scenario trigger categories. Content call requests can be structured query instructions triggered by a touch command, containing content topic tags and timeliness requirements. They can be used to drive semantic matching and timeliness filtering of content fragments in a template group, supporting the accuracy and real-time nature of dynamic assembly. In this embodiment, after parsing the touch command, the content call request can bind content topic tags according to user tags or core user groups, and attach timeliness constraints to form an executable query parameter set. For example, content call requests can include, but are not limited to, one or more of immediate-effective calls, time-limited calls, and periodically reused calls.

[0045] Sending a content retrieval request containing content theme tags and timeliness requirements to the template group corresponding to the target audience can be achieved by constructing a structured request carrying content theme tags and timeliness constraints based on the parsing results of the outreach message, and then sending it to the target template group set. Furthermore, sending this content retrieval request to the template group corresponding to the target audience can be accomplished by encoding the content theme tags into vectors and combining them with the timeliness window to form a joint query key to retrieve the template group index, or by using a multi-condition filter to prioritize matching content fragments with theme tag similarity higher than a threshold and creation time within the timeliness window. This allows for a shift in content extraction from full-volume retrieval to precise filtering based on both semantic and timeliness dimensions.

[0046] Verifying the matching degree of all template groups based on content call requests can be achieved by calculating semantic similarity and judging timeliness validity of content fragments in each template group, and filtering candidate fragments that meet the request conditions. Furthermore, verifying the matching degree of all template groups based on content call requests can be done by using a semantic encoding model to compare the vector cosine similarity between content fragments and content topic tags, retaining items above a threshold, or by combining timestamps and timeliness requirements to filter out fragments that have exceeded their validity period or have expired, retaining only valid candidate implementations. This allows for the verification of the meaning of content fragments. Figure 1 Dual verification of consistency and timeliness improves the accuracy and timeliness of assembled content.

[0047] To integrate content fragments from all template groups based on timeliness requirements, the content fragments that pass matching verification can be prioritized and semantically concatenated according to timeliness requirements to generate complete target SMS content. Furthermore, this integration can be achieved by prioritizing content fragments from all template groups based on their urgency, inserting high-urgency fragments first and low-urgency fragments as supplements to avoid information redundancy, or by using a semantic deduplication model to merge semantically overlapping fragments and retain the most timely and representative expressions. This allows for the temporal coordination and semantic integration of multi-source content fragments, ensuring that the output content is concise, timely, and free of redundancy.

[0048] Taking the instant outreach of near-expiration promotions on a travel platform as an example, the AI-based semantic analysis-based intelligent SMS content generation and precise outreach method in this embodiment can be as follows: a user browses Sanya hotels but has not booked, and the marketer inputs a message reminding the user who has not booked within 3 days to enjoy a final discount. The system parses this message as a time-sensitive message, identifies the user as the target audience, and generates a theme tag for travel decision-making. A content call request is sent to the associated template group, carrying the theme tag and the requirement of validity within 72 hours. The system verifies three content fragments in the template group: limited-time offer (created 1 hour ago) is valid, free pick-up and drop-off (created 5 days ago) has expired and is removed, and discount stacking (created 12 hours ago) is valid. The limited-time offer and discount stacking fragments are merged according to time priority, duplicate promotional words are removed, and the target SMS content is generated: the limited-time offer for the hotel you are interested in is about to expire, and the discount stacking makes it even more cost-effective. This completes the dynamic assembly.

[0049] This embodiment sets user tags or core user groups in the outreach message as the outreach targets, sends a content retrieval request containing content theme tags and timeliness requirements to the template group corresponding to the outreach targets, verifies the matching degree of all template groups based on the content retrieval request, merges content fragments from all template groups based on the timeliness requirements, and sets them as the target SMS content. This achieves a transformation in content extraction from full-volume retrieval to precise filtering based on both semantic and timeliness dimensions, realizing the semantic meaning of content fragments. Figure 1 Dual verification of consistency and time validity enhances the accuracy and timeliness of assembled content, enabling temporal collaboration and semantic integration of multi-source content fragments. This ensures that the output content is concise, timely, and free of redundancy, achieving an intelligent generation chain of semantic triggering, fragment selection, and temporal fusion. This gives SMS content scene adaptability, compliance accuracy, and time sensitivity, achieving refined and real-time operation with a personalized approach without the need for manual template maintenance.

[0050] In one embodiment, the determination of whether a reach message contains user tags is based on the following steps: deconstructing the reach message using a deep learning semantic parsing model to generate extended demand semantics, setting parsing tags, which include reach object identifiers and marketing intents; extracting the target description statement corresponding to the reach object identifier and the scenario description statement corresponding to the marketing intent from the extended demand semantics based on the parsing tags; determining whether there is a semantic match for the user tag in the target description statement based on text embedding similarity; if so, determining that the reach message contains user tags; otherwise, redefining the extended demand semantics as the marketing intent corresponding to the reach message.

[0051] The deep learning semantic parsing model can be a neural network model capable of performing structured semantic decomposition of natural language instructions. It can be used to transform unstructured access commands into structured semantic expressions with semantic roles, supporting subsequent accurate semantic extraction. In this embodiment, the deep learning semantic parsing model can be trained based on the Transformer architecture. It takes the access command text as input and outputs an embedding vector sequence containing semantic roles, identifying semantic component boundaries through an attention mechanism. For example, the deep learning semantic parsing model can include, but is not limited to, one or more of the following: an intent recognition sub-model, an entity extraction sub-model, and a semantic role annotation sub-model.

[0052] Parsing tags can be structured meta-tags used to identify semantic components in outreach statements, including two semantic roles: outreach object identifiers and marketing intent. These tags can provide a semantic role classification framework for extended demand semantics, supporting the targeted extraction of target description statements and scenario description statements. In this embodiment, parsing tags can be generated by a deep learning semantic parsing model at the output layer, annotating the components of the input statement using a predefined set of semantic roles. For example, parsing tags can include, but are not limited to, one or more of outreach object identifiers, marketing intent identifiers, and time constraint identifiers.

[0053] Extended demand semantics can be a semantic expression structure generated by a deep learning semantic parsing model, containing structured semantic roles and semantic components. It can be used as the semantic reconstruction result of the reach command, carrying the complete intent and object information after semantic parsing for subsequent semantic matching. In this embodiment, extended demand semantics can be reorganized into a key-value pair semantic structure based on the embedded vector sequence and parsed labels output by the semantic parsing model, preserving the contextual association of the original semantics. For example, extended demand semantics can include, but is not limited to, one or more of explicit instruction semantics, implicit intent semantics, and contextual modification semantics. User tags can be semantic identifiers derived from user profile data, used to represent user identity or group attributes. They can be used as a reference benchmark for semantic matching to determine whether the target description statement points to a specific user group. In this embodiment, user tags can be generated through user behavior trajectories and preference text clustering, and stored as structured semantic vectors or semantic graph nodes. For example, user tags can include, but are not limited to, one or more of static attribute tags, dynamic behavior tags, and interest clustering tags.

[0054] By deconstructing access commands using a deep learning semantic parsing model to generate extended demand semantics, one can input the access command into a pre-trained semantic parsing model and output a structured semantic expression with semantic role annotations. Furthermore, generating extended demand semantics by deconstructing access commands using a deep learning semantic parsing model can be achieved by employing a BERT-SeqLabel architecture to perform word-level semantic role annotation on the commands and concatenating them into a key-value structured semantic graph, or by using a T5 model for semantic restatement and role extraction to generate standardized semantic triple sequences. This enables end-to-end conversion of access commands from natural language to structured semantic expression.

[0055] Extracting target descriptions corresponding to reachable object identifiers and scenario descriptions corresponding to marketing intents from extended demand semantics based on parsing tags can be achieved by separating text fragments of corresponding semantic components from extended demand semantics based on the semantic role classification of parsing tags. Furthermore, this extraction can be accomplished by retrieving the corresponding value field in the semantic structure based on the key value of the parsing tag, extracting its text unit, or by using an attention weight matrix to locate words strongly associated with the parsing tag in the utterance and concatenating them into semantic clauses. This allows for semantic decoupling of marketing intent and reachable object, supporting independent processing and matching.

[0056] Determining whether a semantic match for a user tag exists in a target description based on text embedding similarity can be achieved by calculating the cosine similarity between the semantic vectors of the target description and the user tags, and then determining if there are any matches exceeding a threshold. Furthermore, this can be accomplished by using Sentence-BERT to encode and compare the target description with all user tag vectors, taking the highest similarity value, or by constructing a user tag semantic graph and calculating the path similarity between the target statement and tag nodes using a graph neural network. This allows for semantic-level recognition of user tags, supporting accurate matching of non-literal and fuzzy expressions.

[0057] Taking cross-channel user reactivation for maternal and infant brands as an example, the AI ​​semantic analysis-based intelligent SMS content generation and precise outreach method in this embodiment can be as follows: marketers input the outreach phrase "send a new product trial coupon to mothers who frequently buy diapers," and the deep learning semantic parsing model deconstructs it into extended demand semantics, generating parsing tags {user_target: "mothers who frequently buy diapers", campaign_intent: "new product trial promotion"}; the system extracts the target description phrase "mothers who frequently buy diapers" and calculates the similarity with the semantic vector of "high-frequency maternal and infant consumer users" in the user tag library to confirm semantic matching; the system determines that the outreach phrase contains user tags and directly calls the preset "maternal and infant new product trial" template, avoiding full user screening and improving response efficiency.

[0058] This embodiment transforms outreach commands into structured extended demand semantics through a deep learning semantic parsing model. Based on parsing tags, it achieves semantic separation between the target audience and marketing intent, enabling the system to independently extract target descriptions. By determining the semantic matching between target descriptions and user tags through text embedding similarity, it breaks through the dependence of traditional keyword matching on expression form. Even if users use colloquial or non-standardized descriptions, it can accurately identify the target group, thereby supporting the dual-path logic of accurate template invocation and automatic screening of core user groups. This upgrades outreach command parsing from a rule engine to a semantic understanding system, achieving deep analysis of natural language intent and semantic-level positioning of user groups. It is a key technical link in achieving personalized marketing and adaptive marketing scenarios.

[0059] In one embodiment, the process of selecting a core user group that matches the marketing intent from all users includes the following steps: semantically matching the textual features of the marketing intent with the user behavior trajectory and user preference keywords of any user; calculating the cross-modal similarity between the two through a pre-trained language model and setting it as a first value; if the first value is greater than a first threshold, then any user is included in the core user group.

[0060] The textual features of marketing intent can be semantic text representations of marketing objectives obtained from parsing the outreach commands. These representations can be used as semantic query conditions for user filtering, driving matching calculations with individual user characteristics. In this embodiment, the textual features of marketing intent can extract core intent phrases from outreach commands using a semantic parsing model and encode them into vector form via an embedding layer. For example, the textual features of marketing intent can include, but are not limited to, one or more of action-oriented features, object-oriented features, and context-constrained features. User behavior trajectories can be sequences of user interactions generated on digital platforms, arranged chronologically. These sequences can reflect the dynamic evolution of user interests and serve as a structured behavioral basis for semantic matching. In this embodiment, user behavior trajectories can be collected through a tracking system, collecting behavior logs such as clicks, browsing, dwell time, and adding items to cart. These logs are aggregated into timestamp sequences and encoded into embedding vectors. For example, user behavior trajectories can include, but are not limited to, one or more of high-frequency interaction sequences, silent period behavior patterns, and pre-churn behavior patterns.

[0061] User preference keywords can be a set of words extracted from a user's historical interaction content that represents their stable interest tendencies. These keywords can serve as semantic proxies for users' static interests, supplementing the short-term insufficiency of behavioral trajectories. In this embodiment, user preference keywords can be extracted from user comments, search terms, and favorited content based on TF-IDF or topic models. For example, user preference keywords can include, but are not limited to, one or more of the following: category preference words, sentiment preference words, and scene trigger words.

[0062] The pre-trained language model can be a deep neural network model pre-trained on a large-scale text corpus, possessing contextual semantic understanding capabilities. It can be used to uniformly encode multi-source heterogeneous semantic inputs (marketing intent, behavioral trajectories, preference keywords) into a dense representation in a semantic vector space. In this embodiment, the pre-trained language model can employ a Transformer architecture to perform masked language modeling and contrastive learning training on general and industry-specific corpora, outputting fixed-dimensional semantic embeddings. For example, the pre-trained language model can include, but is not limited to, one or more of the following: a general semantic encoder, a domain-adaptive encoder, and a multi-task joint encoder.

[0063] Cross-modal similarity can be a numerical metric that measures the semantic consistency of different data modalities (such as text, behavioral sequences, and keyword sets) in a unified semantic space. It can be used to achieve quantitative matching between marketing intent and individual user behavioral preferences, supporting user screening without label dependence. In this embodiment, cross-modal similarity can be calculated using cosine similarity or dot product similarity to measure the semantic vectors of marketing intent text features and user behavioral trajectories / preference keywords. For example, cross-modal similarity can include, but is not limited to, one or more of text-sequence similarity, text-word set similarity, and text-behavior graph similarity.

[0064] Semantic matching of the textual features of marketing intent with any user's behavior trajectory and preferred keywords can be achieved by encoding the three types of inputs into semantic vectors and performing alignment calculations in a unified semantic space. In this embodiment, semantic matching can be achieved by using a multimodal contrastive learning framework, calculating the similarity between the marketing intent vector, the user behavior trajectory vector, and the keyword set mean vector, and then weighted and fused. This allows for semantic-level alignment between unstructured intents and structured user features.

[0065] Calculating cross-modal similarity between two entities using a pre-trained language model can be achieved by using the semantic vector output by the pre-trained language model to calculate the similarity score between the marketing intent vector and the user feature vector. In this embodiment, this calculation process can be implemented by using Sentence-BERT to encode the marketing intent and the user preference keyword set separately, and calculating the cosine similarity as the first value, thereby achieving a direct mapping from distance metrics in the semantic space to user selection decisions.

[0066] Taking the precise re-engagement of high-end skincare brands as an example, the AI ​​semantic analysis-based intelligent SMS content generation and precise outreach method in this embodiment can be an outreach message such as "Send a limited-time discount to users who have browsed the anti-aging serum page in the last two weeks but have not placed an order." The system parses the marketing intent text feature as "users who browsed the anti-aging serum page but did not purchase it," and the pre-trained language model encodes it into a semantic vector. At the same time, it extracts any user's user behavior trajectory (browsing the anti-aging serum page 3 times in the last two weeks) and preferred keywords (anti-aging, ingredient-conscious, luxury). It calculates the cross-modal similarity between the intent vector and the behavior trajectory vector. If the score is higher than the threshold, the user is included in the core user group. Even if the user does not have the "anti-aging" tag, the system still identifies its conversion potential based on behavioral semantic matching.

[0067] This embodiment semantically matches the textual features of marketing intent with any user's behavioral trajectory and preferred keywords. A pre-trained language model calculates the cross-modal similarity between the two, setting it as a first value. If the first value is greater than a first threshold, any user is included in the core user group. The pre-trained language model achieves a unified semantic representation of the textual features of marketing intent, user behavioral trajectory, and preferred keywords, constructing a cross-modal semantic space and breaking through the dependence of traditional tag matching on explicit attributes. By quantifying the potential matching degree between individual users and marketing intent through cross-modal similarity measurement, the system can identify conversion intentions behind implicit behavioral patterns such as "browsing but not purchasing." It can automatically filter high-potential user groups without manually pre-setting grouping rules, supporting the automatic combination of personalized templates under fuzzy instructions. This achieves a leap from group segmentation to individual intent alignment, directly solving the problem of insufficient understanding of deep behavioral preferences and promoting the evolution of outreach strategies from experience-based rules to semantic-driven approaches.

[0068] In one embodiment, matching the optimal outreach scheme in the content strategy library based on semantic features and outreach commands includes the following steps: Define strategy tags. AI technology generates multiple content generation strategies based on strategy tags. All content generation strategies are hierarchically organized according to strategy tags to build a content strategy library. Strategy tags include strategy scenarios, conversion paths, compliance risks, and content iteration logic.

[0069] The strategy tags can be structured semantic dimension tags used to identify the attributes of content generation strategies, including strategy scenarios, conversion paths, compliance risks, and content iteration logic. These tags can be used to achieve semantic classification and computable organization of marketing strategies, supporting hierarchical retrieval and intelligent matching of the strategy library. In this embodiment, strategy tags can be automatically generated and continuously evolved by the system based on historical strategy execution results and manual annotation rules, embedded as meta-information of strategy entries. For example, strategy tags can include, but are not limited to, one or more of scenario-driven strategies, conversion path strategies, risk avoidance strategies, and iterative optimization strategies.

[0070] A content strategy library can be a structured knowledge base organized by strategy tags, storing content generation strategies and their semantic mapping relationships. It can be used to support semantic matching and dynamic selection of outreach solutions, achieving interpretability and iterability of strategies. In one specific embodiment, the content strategy library can use natural language generation algorithms to abstract strategies from historical outreach behavior and response data, and store them hierarchically according to the strategy tag dimension. Furthermore, generating multiple types of content generation strategies based on strategy tags and hierarchically organizing them to construct the content strategy library can be achieved by semantically clustering and structurally encoding historical outreach strategies according to the four dimensions of the strategy tags, constructing a tree-like hierarchical storage structure. Further, this operation can be achieved by using a decision tree model to branch strategy samples according to strategy scenarios, then refining them layer by layer according to conversion paths and compliance risks, and finally updating the weights of the leaf nodes using content iteration logic. This allows marketing strategies to upgrade from discrete templates to a semantic, reasonable network structure.

[0071] Based on the SMS compliance verification process, sensitive semantics in target SMS content corresponding to the same content theme tags are identified and filtered, and the valid content retained is set as standard SMS content.

[0072] Content topic tags can be semantic classification identifiers that reflect content interest tendencies and are abstracted from user preference semantics. They can be used as semantic anchors for content generation and compliance verification, supporting feature extraction and strategy matching of standard SMS content. In an exemplary embodiment, content topic tags can utilize natural language processing models to perform topic modeling on user preference text, extracting high-frequency semantic clusters and mapping them into structured tags. For example, content topic tags can include one or more of the following: consumption intent, sentiment tendency, and scenario trigger.

[0073] Based on the natural language generation algorithm, feature extraction rules are configured for the content topic tags, and the semantic features of standard SMS content are parsed through the feature extraction rules.

[0074] Furthermore, based on natural language generation algorithms, feature extraction rules corresponding to content topic tags are configured to parse the semantic features of standard SMS content. This can be achieved by pre-setting semantic feature extraction templates for each content topic tag, dynamically generating matching feature extraction logic through an NLG model, and applying it to the standard SMS content. Further, this operation can be implemented by using a template filling mechanism to automatically combine keyword weights, sentence patterns, and semantic role constraints based on topic tags to generate feature extraction rules. This enables semantic feature extraction to have topic-adaptive capabilities, overcoming the generalization bottleneck of fixed rules.

[0075] In the content iteration logic, strategies that match semantic features and content topic tags are selected to form a candidate strategy set. The semantic association strength between the outreach message and the strategy scenario of any candidate strategy is calculated and set as the second value. The strategy with the highest second value is set as the optimal outreach solution.

[0076] Furthermore, calculating the semantic association strength between the outreach message and the policy scenarios of the candidate strategies and selecting the highest value as the optimal outreach solution can be achieved by calculating the similarity between the semantic vector of the outreach message and the policy scenario label vector of the candidate strategies, and taking the strategy corresponding to the highest value as the optimal solution. Further, this operation can be achieved by using a contrastive learning model to align the semantic spaces of the outreach message and the policy scenarios, and by ranking the candidate strategies using cosine similarity. This allows for fine-grained semantic alignment between marketing intent and policy scenarios, improving the accuracy of strategy selection and scenario adaptability.

[0077] Taking the compliant outreach optimization of financial product recommendations as an example, the AI ​​semantic analysis-based intelligent SMS content generation and precise outreach method in this embodiment can be as follows: marketers input outreach commands to push new stable wealth management products to high-net-worth clients. The system identifies this as a fuzzy intent command, filters out user groups with high spending power and low risk preference, and generates a stable value-added content theme tag. Based on this tag, the system calls preset NLG feature extraction rules to extract semantic features such as principal protection, long-term, and stable returns from the SMS to be sent. The system then searches the strategy library for matching scenarios such as wealth management, path: APP pop-up + SMS secondary outreach, risk: avoid promising returns, and iteration: weaken the return figures. The system calculates the semantic association strength between the command and the candidate strategies, finds that it is highly consistent with the wealth management scenario, and finally selects this strategy as the optimal outreach solution to ensure content compliance and precise path.

[0078] This embodiment constructs a content strategy library by defining strategy tags and organizing them hierarchically. It dynamically configures feature extraction rules using a natural language generation algorithm to analyze the semantic features of standard SMS content. By calculating the semantic correlation strength between the outreach message and the strategy scenario, it selects the optimal outreach plan. This achieves four-dimensional structured modeling of marketing strategies through strategy tags, enabling the content strategy library to possess semantic reasoning capabilities and realizing a leap from static templates to dynamic semantic networks. Dynamically configuring feature extraction rules for content topic tags based on a natural language generation algorithm gives the semantic parsing of standard SMS content topic adaptability, overcoming the rigidity of keyword matching. By calculating the semantic correlation strength between the outreach message and the strategy scenario, a fine-grained intent-policy matching mechanism is established, so that the selection of the optimal outreach plan no longer relies on manual rules but is driven by semantic similarity. The synergy of these three elements allows strategy generation, content understanding, and outreach decision-making to operate entirely in a closed loop using an AI semantic mechanism. This improves the scenario adaptability of strategies, the compliance accuracy of content, and the intelligent selection capability of outreach paths without increasing human intervention.

[0079] In some embodiments, the precise delivery of SMS messages and real-time effect evaluation based on the delivery instruction includes the following steps: training the optimal delivery scheme through reinforcement learning based on standard SMS content to generate a prediction model; the prediction model outputs a delivery response probability sequence based on semantic features; and the timing and channel for SMS delivery are determined based on the delivery response probability sequence and set as the delivery instruction.

[0080] The prediction model can be a machine learning model trained using reinforcement learning to predict the probability of a user's response to specific SMS content. It can output a sequence of reach response probabilities, driving dynamic decisions regarding reach timing and channels, replacing the crude reach approach based on fixed times or frequencies. In this embodiment, the prediction model can take the semantic features of standard SMS content and historical reach response data as input, and learn the optimal reach action and state value mapping through iterative optimization via a policy network and a value network. Furthermore, the prediction model can include, but is not limited to, one or more of the following: a time-series response prediction model, a channel preference prediction model, and a user activity window prediction model.

[0081] A predictive model is generated by training the optimal outreach strategy based on standard SMS content using reinforcement learning. This can be achieved by using SMS semantic features as state input, outreach timing and channel as action space, and user response as reward signal, then optimizing model parameters using a policy gradient algorithm. Furthermore, this operation can be implemented by constructing a policy network using the PPO algorithm, with SMS subject tags and user profiles as state, push time window and channel type as action, and click behavior as positive reward. This allows outreach decisions to shift from static rules to dynamic policies based on feedback learning.

[0082] A performance tracking table is constructed. Based on the reach response probability sequence and real-time SMS conversion data, operational intervention strategies are matched in the performance tracking table and set as an intelligent marketing closed loop. The operational intervention strategies include secondary reach thresholds, dynamic content adjustment rules, and user churn early warning mechanisms.

[0083] The performance tracking table can be a structured log table that records the reach response probability sequence and real-time conversion data. It serves as a feedback mechanism for mapping operational intervention strategies, enabling quantitative tracking of reach effectiveness and automatic matching of intervention strategies, supporting the operation of a smart marketing closed loop. In this embodiment, the performance tracking table can collect data on actions such as opening, clicking, unsubscribing, and conversion in real time after the SMS message is sent, and store this data in time alignment with the probability sequence output by the prediction model. Furthermore, the performance tracking table can include, but is not limited to, one or more of the following: a response delay record table, a conversion path tracking table, and a user behavior decay table.

[0084] Operational intervention strategies can be a set of automated response rules triggered by performance tracking tables to optimize outreach effectiveness. These rules can be used to automatically execute secondary outreach, content adjustments, or churn warnings when specific response patterns are detected, enabling adaptive correction of marketing activities. In this embodiment, operational intervention strategies can predefine rule templates and match them with probability thresholds and behavioral patterns in the performance tracking table to trigger corresponding intervention actions. Furthermore, operational intervention strategies may include, but are not limited to, secondary outreach triggering strategies, dynamic content adjustment strategies, and user churn warning strategies.

[0085] Matching operational intervention strategies in the performance tracking table based on the reach response probability sequence and real-time SMS conversion data can be achieved by comparing real-time conversion data with the probability sequence output by the predictive model to identify abnormal response patterns and trigger preset intervention rules. Furthermore, this operation can be implemented by setting probability thresholds and time windows. When a high response is predicted but the message is not actually sent, rules for secondary outreach and content weakening adjustments can be activated, thus creating a closed-loop conversion from passive performance recording to proactive operational intervention.

[0086] Taking the real-time optimization of course promotion on an education platform as an example, the intelligent SMS content generation and precise delivery method based on AI semantic analysis in this embodiment can be as follows: the system pushes a Python introductory course SMS to the user, the prediction model outputs a high response probability based on the user's historical learning behavior and semantic features, but the actual open rate is 0; the effect tracking table records this anomaly, triggers the secondary reach threshold rule, and a simplified title is pushed through the WeChat service account after 2 hours; at the same time, it is detected that the user has opened the course page multiple times in the past 7 days but has not registered, the system activates the content dynamic adjustment rule, adjusts the system learning in the original SMS to a 15-minute course, and starts the churn warning mechanism. If there is still no conversion within 24 hours, the user is automatically added to the exclusive consultant reach queue.

[0087] This embodiment generates a predictive model by using reinforcement learning to train the optimal outreach plan based on standard SMS content. It transforms semantic features and historical response data into a learnable outreach response probability sequence, enabling the timing and channel selection to shift from manual preset to dynamic optimization based on feedback. By matching operational intervention strategies with the outreach response probability sequence and real-time conversion data in the effect tracking table, it constructs an automatically triggered secondary outreach, content adjustment, and churn warning mechanism, upgrading single outreach to a continuously responding adaptive operational process. The synergy of these two aspects enables the system to have online learning and real-time intervention capabilities, achieving continuous evolution of outreach strategies and proactive guidance of user behavior without relying on manual configuration, thus achieving a qualitative leap from one-time push notifications to intelligent closed-loop operations.

[0088] Furthermore, to achieve the above objectives, the present invention also provides an intelligent SMS content generation and precise delivery device based on AI semantic analysis. The device includes: a memory, a processor, and an AI semantic analysis-based intelligent SMS content generation and precise delivery program stored in the memory and executable on the processor. The AI ​​semantic analysis-based intelligent SMS content generation and precise delivery program is configured to implement the steps of the AI ​​semantic analysis-based intelligent SMS content generation and precise delivery method as described above.

[0089] Furthermore, to achieve the above objectives, the present invention also provides a medium storing an AI semantic analysis-based intelligent SMS content generation and precise delivery program, wherein when the AI ​​semantic analysis-based intelligent SMS content generation and precise delivery program is executed by a processor, it implements the steps of the AI ​​semantic analysis-based intelligent SMS content generation and precise delivery method as described above.

[0090] Other embodiments or specific implementations of the AI ​​semantic analysis-based intelligent SMS content generation and precise delivery device described in this invention can be referred to the above-mentioned method embodiments, and will not be repeated here.

[0091] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for intelligent SMS content generation and precise delivery based on AI semantic analysis, characterized in that, The method includes: Acquire user profile data, which includes user tags, user preferences, and user behavior patterns, and generate content topic tags based on the user preferences; Set up SMS generation and outreach strategies, receive and parse user outreach messages, determine whether the user outreach message contains the user tag, if yes, retrieve the personalized SMS template corresponding to the user tag based on the generation and outreach strategy, and set it as the target SMS content; otherwise, extract the marketing intent of the user outreach message, match the marketing intent with the user profile data, filter the core user group that matches the marketing intent from all users, combine the personalized SMS template of the core user group based on the generation and outreach strategy, and set it as the target SMS content; The compliance of the target SMS content is verified based on the content topic tags, sensitive words and redundant information are filtered to generate standard SMS content, a content strategy library is built based on AI technology, the semantic features of the standard SMS content are analyzed, and the optimal outreach scheme is matched in the content strategy library based on the semantic features and the user outreach message. The standard SMS content is injected into the optimal outreach scheme for dynamic optimization, and outreach instructions are generated based on the user outreach phrases. Based on the outreach instructions, precise SMS outreach and real-time effect evaluation are performed to complete the intelligent marketing closed loop corresponding to the user outreach phrases.

2. The method for intelligent SMS content generation and precise delivery based on AI semantic analysis as described in claim 1, characterized in that, The SMS generation and delivery strategy is set based on the following steps: The user associated with the user profile data is set as the target user. A preset number of SMS template groups are configured for the target user. The personalized content of the target user is stored in any template group. The target user and all the corresponding template groups are integrated to construct a status scheduling table. The status scheduling table updates the content usage rate of all template groups in real time. The status scheduling table dynamically switches the content generation status of any template group based on the content usage rate to complete the SMS content generation process. After parsing the user outreach message, a content retrieval request is initiated to the template group based on the user tag. Based on the content retrieval request, all personalized content fragments stored in the template group are extracted sequentially to complete the dynamic assembly process of the SMS content. The generation process and the assembly process are integrated and set as the SMS generation and outreach strategy.

3. The method for intelligent SMS content generation and precise delivery based on AI semantic analysis as described in claim 2, characterized in that, Generating the target SMS content includes the following steps: The user tag or the core user group in the user outreach message is set as the outreach target. A content retrieval request is sent to the template group corresponding to the outreach target. The content retrieval request includes the content theme tag and timeliness requirement. The matching degree of all template groups is verified based on the content retrieval request. If the template group matches the content retrieval request, the corresponding stored content fragment is output. The content fragments of all template groups are merged based on the timeliness requirement and set as the target SMS content.

4. The method for intelligent SMS content generation and precise delivery based on AI semantic analysis as described in claim 1, characterized in that, The following steps are used to determine whether the user outreach message contains the user tag: The user reach message is deconstructed using a deep learning semantic parsing model to generate extended demand semantics. Parsing tags are set, which include reach object identifiers and marketing intents. Based on the parsing tags, the target description statement corresponding to the reach object identifier and the scenario description statement corresponding to the marketing intent are extracted from the extended demand semantics. Based on text embedding similarity, determine whether there is a semantic match for the user tag in the target description statement. If there is, determine that the user outreach message contains the user tag. Otherwise, redefine the extended requirement semantics as the marketing intent corresponding to the user outreach message.

5. The method for intelligent SMS content generation and precise delivery based on AI semantic analysis as described in claim 4, characterized in that, The process of selecting a core user group that aligns with the marketing intent from all users includes the following steps: The textual features of the marketing intent are semantically matched with the user behavior trajectory and the keywords of the user preference of any user. The cross-modal similarity between the two is calculated by a pre-trained language model and set as a first value. If the first value is greater than a first threshold, then any user is included in the core user group.

6. The method for intelligent SMS content generation and precise delivery based on AI semantic analysis as described in claim 1, characterized in that, The process of matching the optimal outreach scheme in the content strategy library based on the semantic features and the user outreach message includes the following steps: Define strategy tags, and the AI ​​technology generates multiple content generation strategies based on the strategy tags. Organize all the content generation strategies hierarchically according to the strategy tags to build the content strategy library. The strategy tags include strategy scenarios, conversion paths, compliance risks, and content iteration logic. Based on the SMS compliance verification process, sensitive semantics in the target SMS content corresponding to the same content theme tags are identified and filtered, and the valid content retained is set as the standard SMS content. Based on the natural language generation algorithm, the feature extraction rules corresponding to the content topic tags are configured, and the semantic features of the standard SMS content are parsed through the feature extraction rules; In the content iteration logic, strategies that match the semantic features and the content topic tags are selected to form a candidate strategy set. The semantic association strength between the user outreach message and the strategy scenario of any candidate strategy is calculated and set as a second value. The strategy with the highest second value is set as the optimal outreach solution.

7. The method for intelligent SMS content generation and precise delivery based on AI semantic analysis as described in claim 6, characterized in that, The process of accurately reaching customers via SMS and evaluating its real-time effectiveness based on the reach command includes the following steps: The optimal outreach scheme is trained using reinforcement learning based on the standard SMS content to generate a prediction model. The prediction model outputs an outreach response probability sequence based on the semantic features. The timing and channel for SMS push are determined based on the outreach response probability sequence and set as the outreach instruction. An effect tracking table is constructed. Based on the reach response probability sequence and real-time SMS conversion data, an operational intervention strategy is matched in the effect tracking table and set as the intelligent marketing closed loop. The operational intervention strategy includes a secondary reach threshold, dynamic content adjustment rules, and a user churn early warning mechanism.

8. A device for intelligent SMS content generation and precise delivery based on AI semantic analysis, characterized in that, The device includes: a memory, a processor, and an AI semantic analysis-based intelligent SMS content generation and precise delivery program stored in the memory and executable on the processor, wherein the AI ​​semantic analysis-based intelligent SMS content generation and precise delivery program is configured to implement the steps of the AI ​​semantic analysis-based intelligent SMS content generation and precise delivery method as described in any one of claims 1 to 7.

9. A medium, characterized in that, The medium stores an AI-based intelligent SMS content generation and precise delivery program, which, when executed by a processor, implements the steps of the AI-based intelligent SMS content generation and precise delivery method as described in any one of claims 1 to 7.

Images