Marketing document popular formula automatic mining method based on big data analysis

Abstract

The invention discloses an automatic mining method for a marketing document popular formula based on big data analysis, and belongs to the field of digital marketing intelligent generation. Emotional expression vectors and structural design modes are obtained by analyzing historical document data, emotion generality categories are determined after clustering, and user response intensity is evaluated; and the association rule of the high-response emotion type and the structure mode is mined. A platform exclusive creation rule is determined according to a platform difference refining rule subset, copywriting templates suitable for various crowds are generated by fusing crowd data, and finally the copywriting is output and optimized by matching promotion information and adjusting parameters, so that the problems that traditional copywriting creation depends on artificial experience and is difficult to quantify and optimize are solved, and the copywriting efficiency is improved. And data-driven accurate copywriting generation is realized.

Description

Technical Field

[0001] This invention belongs to the field of digital marketing intelligent generation technology, and in particular relates to a method for automatically mining formulas for marketing copywriting hits based on big data analysis. Background Technology

[0002] In today's digital marketing landscape, copywriting serves as a crucial bridge connecting brands and consumers, directly impacting marketing effectiveness and market competitiveness. However, traditional copywriting methods rely heavily on the creator's personal experience and intuitive judgment. This approach reveals significant limitations when facing a massive user base and diverse platform and demographic needs. Human judgment struggles to systematically grasp rapidly changing user preferences and cannot adapt to the differentiated characteristics of various situations. For example, in e-commerce promotional scenarios, while experience suggests that a warm and friendly tone may attract more clicks than a formal tone, this perception cannot be quantified, and the effective pairing rules for it with specific sentence structures remain unclear. This leads to a high degree of uncertainty in copywriting effectiveness, impacting the overall performance of marketing campaigns.

[0003] A deeper technical challenge lies in the difficulty of accurately extracting the intrinsic connections and common patterns between key elements such as emotional expression and structural design from a vast number of successful and unsuccessful cases. The appeal of copywriting is the result of the combined effects of multiple dimensions, including emotion, tone, and structure; these factors are intertwined and mutually influential. Existing methods lack effective technical means to deconstruct these complex relationships. For example, they cannot quantify how subtle changes in emotional expression ultimately affect the resonance effect of structural design through tone of voice. This lack of analytical capability makes it difficult for the creative process to transcend individual experience, preventing the formation of reusable and verifiable standardized creative rules, leading to inefficiency and inconsistent quality in content production for businesses.

[0004] Therefore, the digital marketing field has long faced the core technological challenge of transforming unstructured copywriting content into computable and analyzable data models, and automatically extracting universally applicable creative formulas from them. Solving this problem requires not only overcoming the bottlenecks of natural language processing technology in deep semantic association mining, but also establishing an intelligent generation mechanism that can dynamically adapt to platform characteristics and user demographics, thereby fundamentally improving the accuracy and scalability of copywriting. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes an automatic method for mining formulas for best-selling marketing copy based on big data analysis, thereby resolving the issues present in the existing technologies.

[0006] Firstly, to achieve the above objectives, this invention provides a method for automatically mining formulas for viral marketing copy based on big data analysis, comprising the following steps: Extract historical copywriting data and user interaction records, and analyze the copywriting data to obtain emotional expression vectors and structural design patterns; Clustering sentiment expression vectors to determine common categories of sentiment expression; By associating common categories of emotional expressions with user interaction records, we can assess response intensity and identify the types of emotional expressions with high response intensity. For emotional expression types with high response intensity, we extract the association structure design pattern and obtain the association rules between emotion and structure by comparing repetition frequency; If the association rules contain platform-specific characteristics, then separate the platform rule subsets and cluster them for refinement to determine the platform-specific copywriting creation rules; By integrating platform-specific patterns and user characteristics data, a rule set is generated and user psychological change indicators are obtained to create copywriting templates suitable for diverse user groups. Match new promotional information to the copy template. If the matching degree is low, adjust the parameters and output optimized copy.

[0007] Secondly, the present invention also provides an automatic marketing copywriting best-selling formula mining system based on big data analysis, used to implement an automatic marketing copywriting best-selling formula mining method based on big data analysis, the system comprising: The data extraction and parsing module is used to extract historical copywriting data and user interaction records, and to parse the copywriting data to obtain emotional expression vectors and structural design patterns; The sentiment clustering module is used to cluster sentiment expression vectors to determine common categories of sentiment expressions; The response assessment module is used to correlate common categories of emotional expression with user interaction records, assess response intensity, and determine the type of emotional expression with high response intensity. The association rule mining module is used to extract association structure design patterns for high-response intensity emotion expression types and obtain the association rules between emotion and structure by comparing repetition frequency. The platform adaptation module is used to separate and cluster a subset of platform rules when the association rules contain platform-specific features, and to determine the platform-specific copywriting creation rules. The template generation module is used to integrate platform-specific patterns and user characteristics data to generate rule sets and obtain user psychological change indicators to obtain copywriting templates that are suitable for diverse groups of people. The copy output module is used to match new promotional information based on copy templates, adjust parameters and output optimized copy when the matching degree is low.

[0008] Thirdly, the present invention also provides a computer terminal device, comprising: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of the method for automatically mining marketing copywriting best-selling formulas based on big data analysis in the first aspect above.

[0009] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of the method for automatically mining marketing copywriting best-selling formulas based on big data analysis in the first aspect described above.

[0010] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method for automatically mining marketing copywriting best-selling formulas based on big data analysis described in the first aspect above.

[0011] Compared with the prior art, the present invention has the following advantages and technical effects: This invention provides an automatic method for mining formulas for viral marketing copy based on big data analysis. This invention overcomes the limitations of traditional copywriting, which relies on human experience, through a data-driven approach. Its technical effectiveness lies in transforming unstructured copywriting content into calculable emotional expression vectors and structural design patterns, thereby achieving systematic analysis of massive amounts of historical copywriting data. Through cluster analysis, it can automatically identify common categories of emotional expression and accurately determine high-response emotional types by associating them with user interaction records. Furthermore, by mining the correlation rules between emotional expression and structural design, quantifiable creative patterns are established. When the rules include platform-specific characteristics, it can automatically separate and refine platform-specific rule subsets, forming creative patterns for different platforms. By integrating platform-specific patterns with user demographic data and combining user psychological change indicators, copywriting templates adapted to the needs of diverse user groups are generated. Ultimately, it achieves intelligent matching and parameter optimization of newly input promotional information, significantly improving the accuracy and adaptability of copywriting and providing effective technical support for large-scale, personalized marketing content generation. Attached Figure Description

[0012] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart illustrating the overall process of automatically mining formulas for best-selling marketing copy in an embodiment of the present invention.

[0013] Figure 2 This is a schematic diagram of emotion expression vector clustering analysis according to an embodiment of the present invention. Detailed Implementation

[0014] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0015] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0016] Example 1 like Figure 1 This embodiment provides a method for automatically mining formulas for marketing copywriting bestsellers based on big data analysis, including: Extract historical copywriting data and user interaction records, and analyze the copywriting data to obtain emotional expression vectors and structural design patterns; Clustering sentiment expression vectors to determine common categories of sentiment expression; By associating common categories of emotional expressions with user interaction records, we can assess response intensity and identify the types of emotional expressions with high response intensity. For emotional expression types with high response intensity, we extract the association structure design pattern and obtain the association rules between emotion and structure by comparing repetition frequency; If the association rules contain platform-specific characteristics, then separate the platform rule subsets and cluster them for refinement to determine the platform-specific copywriting creation rules; By integrating platform-specific patterns and user characteristics data, a rule set is generated and user psychological change indicators are obtained to create copywriting templates suitable for diverse user groups. Match new promotional information to the copy template. If the matching degree is low, adjust the parameters and output optimized copy.

[0017] As one implementation method in this embodiment, the process of extracting historical copywriting data and user interaction records, and parsing the copywriting data to obtain emotional expression vectors and structural design patterns includes: Extract historical copywriting data and user interaction records; Word segmentation is used to separate vocabulary and sentence structure units, resulting in a set of emotion-related words and a set of non-emotion-related words. Based on the sentiment dictionary, sentiment-related words are matched and assigned sentiment intensity values. If the value is higher than the threshold, it is classified as a key sentiment expression element and a sentiment expression vector is formed. Analyze the sentence structure, extract the main structure and modifiers, and determine whether it conforms to common design patterns to identify the core features of the structural design pattern. By associating emotional expression vectors and structural design patterns, and then using vector mapping and fusion to extract features, a text feature description is obtained.

[0018] Step S101: By extracting historical copywriting data and corresponding user interaction records from a preset database, natural language processing technology is used to analyze the vocabulary and sentence structure features in the copywriting data to obtain emotional expression vectors and structural design patterns.

[0019] By extracting historical copywriting data and corresponding user interaction records from a pre-set database, the copywriting data is initially processed using a word segmentation tool to separate vocabulary and sentence structure units, thereby obtaining a set of emotion-related vocabulary and a set of non-emotion-related vocabulary. Based on the set of emotion-related vocabulary, a pre-established emotion dictionary is used for matching, assigning emotion intensity values ​​to the words. If the emotion intensity value is higher than a preset threshold, it is classified as a key emotion expression element, resulting in an emotion expression vector. Through structural analysis of the sentence structure units, a syntax tree generation tool is used to extract the main structure and modifying components in the copywriting, determining whether the structure conforms to common design patterns. If it does, the core features of the structural design pattern are identified. Based on the emotion expression vector and the structural design pattern, a vector mapping tool is used for correlation and fusion. Feature extraction is performed on the fused data to obtain the final copywriting feature description.

[0020] For example, in the field of e-commerce promotion, extracting historical copywriting data and corresponding user interaction records from a pre-set database can serve as a basic step in copywriting optimization.

[0021] Specifically, suppose the database stores promotional copy from e-commerce platforms, such as a high-conversion copy like "A must-have for warmth! This thermos cup is on sale for a limited time, accompanying you through the entire winter," along with user interaction records, including a 15% click-through rate, an 8% conversion rate, and numerous positive reviews. By extracting this data, we can analyze the success factors of this copy. Subsequent processing will analyze its emotional vocabulary and structural patterns, thus providing quantifiable data support for the design of new copy.

[0022] In one possible implementation, a word segmentation tool is used to perform preliminary processing on the text data. The process of separating words and sentence structures needs to be detailed. Word segmentation tools, such as machine learning-based Chinese word segmenters, break down the text into its smallest units. For example, processing "Double 11 Carnival, Limited-Time Sale!" separates words like "Double 11," "carnival," "limited-time," and "sale," as well as sentence structures like exclamatory sentences. This preliminary processing ensures the clarity of the data structure, resulting in a set of emotion-related words and a set of emotion-insensitive words. Specifically, the emotion-related word set might include "carnival" (positive emotion) and "limited-time" (sense of urgency), while the emotion-insensitive word set might include "Double 11" (event noun). This classification facilitates targeted analysis and avoids irrelevant information interfering with subsequent emotion matching.

[0023] For example, for a set of emotion-related words, a pre-established emotion dictionary is used for matching, and an emotion intensity value is assigned. The emotion dictionary here can be a database constructed based on natural language processing, containing the emotion polarity and intensity of words.

[0024] For example, "carnival" is matched with a positive emotion, and the intensity value is assigned 8 (out of 10), while "time-limited" is assigned 6, indicating a medium sense of urgency. If the preset threshold is 5, words above the threshold such as "carnival" are classified as key emotion expression elements, and finally an emotion expression vector is formed. For example, a multi-dimensional vector [8, 6, 0,...] represents the emotion distribution of the copywriting. This vector quantifies the emotion intensity and facilitates the calculation of the overall attractiveness of the copywriting.

[0025] In one possible implementation, through the structural analysis of the sentence pattern units, a grammar tree generation tool is used to extract the main structure and modifying components in the copywriting. A grammar tree generation tool such as StanfordParser will parse the sentence pattern units to generate a tree structure. For example, for "Time-limited rush purchase!", the main structure extraction is the subject-predicate-object "rush purchase", and the modifying component is "time-limited" as an adverbial of time. Determine whether the structure conforms to a common design pattern. If it conforms to the "appeal to action" pattern (ending with a verb to appeal to the user's behavior), then the core feature is determined to be "urgent appeal", which connects the emotion vector and helps to fuse and analyze the persuasiveness of the copywriting.

[0026] For example, according to the emotion expression vector and the structural design pattern, a vector mapping tool is used for associative fusion, and feature extraction is performed on the fused data to obtain the final copywriting feature description. A vector mapping tool such as an embedding model maps the emotion vector and the structural features to a unified space. For example, by fusing the high intensity of "carnival" and the "appeal to action" pattern, features such as "high-emotion-driven promotional copywriting" are extracted, and finally described as "This copywriting enhances user interaction with positive emotion and urgent structure, and is expected to increase the click-through rate by 15%". This fusion brings the technical effect of copywriting optimization, such as accurately predicting user responses in marketing operations.

[0027] As an implementation method in this embodiment, the process of clustering emotion expression vectors to determine the common emotion expression categories includes: Calculate the distance between emotion expression vectors. If the distance is less than the threshold, they are grouped into the same category to obtain a preliminary emotion expression grouping; Calculate the clustering center for the grouping result, extract the representative emotion feature values, and determine the core characteristics of emotion expression; Obtain the distribution range of the feature values, compare the similarity between groups. If the overlapping of the distribution ranges exceeds the threshold, the groups are merged to obtain an optimized emotion expression category division; Assign a unique identification label to the optimized category to obtain the result of common emotion expression categories, and judge the typical classification pattern.

[0028] like Figure 2 As shown, in step S102, based on the obtained emotion expression vector, the vector is processed by a clustering algorithm. If the similarity of the vectors exceeds a preset threshold during the clustering process, they are classified into the same group, thus determining the common category of emotion expression.

[0029] Based on the data content of the emotion expression vectors, a preset vector distance calculation tool is used to measure the distance between the vectors. If the distance between two vectors is less than a preset threshold, they are grouped into the same group, resulting in the preliminary emotion expression grouping results. For the preliminary emotion expression grouping results, a cluster center calculation tool is used to extract the center points of the vectors in each group, obtaining representative emotion feature values ​​for each group and determining the core characteristics of the emotion expression. The feature value distribution range of each group is obtained from the core characteristics, and a feature value comparison tool is used to judge the similarity of feature values ​​between different groups. If the feature value distribution ranges of two groups overlap by more than a preset threshold, the two groups are merged to obtain the optimized emotion expression category classification. For the optimized emotion expression category classification, a category label generation tool is used to assign a unique identifier label to each category, obtaining the final common category results of emotion expression and determining the typical classification pattern of emotion expression.

[0030] For example, when processing emotion expression vectors, the first step is to measure the distance between these vectors. Suppose there is historical copywriting data in a pre-existing database, which has been transformed into emotion vectors. For instance, one vector represents positive emotions such as "joy" (0.8 intensity) and "excitement" (0.6 intensity), while another vector represents "trust" (0.7 intensity) and "urgency" (0.5 intensity). ... For example, all vectors dominated by joy and excitement can be grouped into a "passionate promotion" group, while vectors dominated by trust and reassurance can be grouped into a "quality assurance" group. This grouping helps to initially identify emotional patterns in the copywriting, providing a foundation for subsequent analysis.

[0031] In one possible implementation, a cluster center calculation tool is used to extract the center point of each group for these initial groupings.

[0032] Specifically, for all vectors in the positive sentiment group, the tool calculates the average value for each dimension. For example, the average intensity of the "joy" dimension is 0.75, thus obtaining the representative sentiment feature value for the group. This central point represents the core characteristic of the group. For instance, the core characteristic might show that the positive group is dominated by "joy," with an intensity distribution between 0.6 and 0.9. This extraction process is achieved through averaging or weighted averaging methods, ensuring that the core characteristic captures the most typical expressions within the group, thereby helping to identify the sentiment types with high user interaction in copywriting optimization.

[0033] For example, starting with core characteristics, the tool obtains the feature value distribution range for each group. For instance, the "joy" range for the positive group is 0.5-0.9, and the "sadness" range for the negative group is 0.4-0.8. Then, a feature value comparison tool is used to calculate the similarity between different groups. By checking the overlap ratio, if the overlap exceeds a preset threshold such as 50%, the two groups are merged. For example, if the positive and neutral groups overlap by 60%, the tool will determine that they have similar emotional transitions and merge them to obtain an optimized category classification. This merging reduces redundant classifications and, in actual copywriting generation, improves the accuracy of emotional categories, avoiding confusion in user interaction records.

[0034] In one possible implementation, for the optimized category division, a unique identifier label is assigned using a category label generation tool.

[0035] For example, a label like "POS_NEUT_01" is assigned to the merged positive-neutral category, and "NEG_01" to the negative category. This process involves automatically generating feature-based labels, ensuring each category has a unique identifier, thereby obtaining the final common category results for sentiment expression. These labels allow for the identification of typical classification patterns; for instance, positive patterns are often used for promotional interactions, while negative patterns are used for warning content. This approach standardizes sentiment classification in business contexts, facilitating the rapid invocation of corresponding patterns during subsequent copywriting design and improving user interaction efficiency.

[0036] As one implementation method in this embodiment, the process of associating common categories of emotional expression with user interaction records, assessing response intensity, and determining the type of emotional expression with high response intensity includes: The interaction frequency data under the emotion expression category is obtained from the user interaction record database, and the emotion category association mapping table is obtained by classifying and organizing the data. Based on the mapping table, sentiment analysis tools are used to quantitatively evaluate the response intensity of sentiment expression categories under different user preferences. If the response intensity is higher than the threshold, it is marked as a high-response sentiment type, forming a set of high-response sentiment types. We obtain user feedback data corresponding to the set of highly responsive emotional types from user interaction records, compare and analyze the matching degree, and determine the most representative combination of emotional expressions. Based on the most representative combinations, user behavior feature data is obtained, and in-depth analysis is conducted to determine the emotion expression recommendation scheme suitable for different user preferences.

[0037] Step S103: Obtain the correspondence between the determined common categories of emotional expression and historical user interaction records, use a sentiment analysis model to evaluate the response intensity of the category under different user preferences, and determine the emotional expression type with high response intensity.

[0038] By using a pre-established user interaction record database, historical user interaction frequency data under different emotional expression categories is obtained. The correspondence between these emotional expression categories and user consumption preferences (such as "pursuing cost-effectiveness," "emphasizing quality," and "being enthusiastic about new products") is categorized and organized to obtain a preliminary emotional category association mapping table. Based on this mapping table, sentiment analysis tools are used to quantitatively evaluate the response intensity of each emotional expression category under different user preferences. If the response intensity exceeds a preset threshold, it is marked as a high-response emotional type, thus determining the set of high-response emotional types. For this set of high-response emotional types, corresponding user feedback data is obtained from historical user interaction records. Data comparison tools are used to analyze the matching degree between the user feedback data and the emotional types, identifying the most representative emotional expression combinations. Through these most representative emotional expression combinations, relevant user behavior feature data is obtained. Feature extraction tools are used to deeply mine this user behavior feature data to determine emotional expression recommendation schemes suitable for different user preferences.

[0039] In one possible implementation, when obtaining historical user data through a pre-established database of user interaction records, one can envision a scenario for an e-commerce platform where the database stores users' likes, comments, and shares over the past year.

[0040] Specifically, this database might include fields such as user ID, emotion category (e.g., "happy," "sad," or "angry"), interaction frequency (e.g., number of likes per month), and user preference tags (e.g., "likes inspirational content," "enthusiastic about home goods," "beauty influencer"). During the classification and organization process, this data is first filtered, mapping emotion categories to preferences. For example, users with high-frequency interactions under the "happy" category are categorized as "positive," thus forming a preliminary emotion category association mapping table. This table might be presented in tabular form, with each row representing a category and each column displaying the number of users with the corresponding preference and frequency statistics. In this way, the mapping table ensures that it captures users' interaction patterns under different emotions, providing a foundation for subsequent analysis.

[0041] In one possible implementation, when using a sentiment analysis tool for quantitative evaluation based on this sentiment category association mapping table, the sentiment analysis tool can be understood as a software module based on natural language processing and machine learning that evaluates category effectiveness by calculating response strength.

[0042] Specifically, this tool might use metrics such as average interaction duration or sentiment score for quantification. For example, it might calculate the response intensity of the "happiness" category for users with an "inspirational preference." If the average interaction duration exceeds a preset threshold, such as 30 seconds, it's marked as a high-response sentiment type. The process involves inputting mapping table data, running analysis algorithms such as sentiment polarity calculation, and then outputting a high-response set. This set might include the "happiness" and "surprise" types because they show an intensity above the threshold under specific preferences. In this way, the tool's evaluation not only identifies the set but also reveals the intrinsic relationship between sentiment and user responses, improving the accuracy of recommendations in business operations.

[0043] For example, when obtaining feedback data from historical user interaction records for a set of high-response sentiment types, a data comparison tool can be interpreted as a comparison algorithm framework that analyzes the degree of matching through similarity calculations such as cosine similarity.

[0044] Specifically, in the case of a comprehensive e-commerce platform, feedback such as product ratings, post-purchase reviews, and inquiry volume are extracted from records. A comparison tool then matches the feedback vectors with sentiment type vectors. For example, if the match between the "surprise" type and high-rated feedback exceeds 80%, it is considered a representative combination. The analysis process involves data cleaning, vector transformation, and similarity threshold determination, ultimately yielding the most representative combinations such as "happiness + motivation," which helps in understanding the core patterns of user sentiment preferences.

[0045] In one possible implementation, when user behavior feature data is obtained through the most representative combination of emotional expressions and then deeply mined using feature extraction tools, the feature extraction tools can be regarded as a data mining module that uses techniques such as principal component analysis to mine behavioral patterns.

[0046] Specifically, in e-commerce recommendation systems, behavioral data such as browsing time and purchase rate are extracted from combinations like "happiness + motivation." The tool then performs in-depth analysis of this data, identifying features such as "peak-hour interaction" to determine recommendation strategies. For example, it might recommend promotional content with a "motivational preference" to users. The data mining process includes data aggregation, feature selection, and pattern recognition to ensure the strategy is applicable to different preferences, optimizing personalized emotional expression in business operations and improving user satisfaction.

[0047] As one implementation method in this embodiment, the process of extracting association structure design patterns for high-response intensity emotion expression types and comparing repetition frequencies to obtain association rules between emotion and structure includes: Structural design patterns associated with high-response intensity emotional expression types were extracted from copywriting data, and the frequency of pattern repetition was statistically analyzed to obtain preliminary correspondence data. Filter the initial correspondence data. If the repetition frequency of the structural design pattern is higher than the threshold, it is judged as a strong correlation, and the filtered correlation pattern set is obtained. By comparing the matching degree between emotional expression types and structural design patterns, the main structural design patterns are determined after sorting, and a classification mapping table is formed. Based on the classification mapping table, obtain the association rules between emotion expression type and structural design pattern, store and determine the most suitable combination relationship.

[0048] Step S104: For the identified high-response intensity emotional expression type, extract the associated structural design pattern from the text data, and obtain the association rule between structural design and emotional expression by comparing the recurrence frequency between the patterns.

[0049] Structural design patterns associated with high-response emotional expression types are extracted from the text data. The recurrence frequency of these patterns is statistically analyzed using a comparison tool to obtain preliminary correspondence data between the structural design patterns and emotional expression types. This preliminary correspondence data is then filtered using a preset threshold. If the recurrence frequency of a structural design pattern exceeds the preset threshold, a strong correlation is determined between the pattern and the emotional expression type, resulting in a filtered set of associated patterns. Based on this set of associated patterns, a data comparison tool is used to rank the matching degree between each emotional expression type and the structural design pattern, identifying the primary structural design pattern corresponding to each emotional expression type and forming a classification mapping table. Using this classification mapping table, association rules between each emotional expression type and structural design pattern are obtained. Data is stored based on these rules to determine the most suitable combination of structural design patterns and emotional expression types.

[0050] In one possible implementation, extracting structural design patterns associated with high-response intensity emotional expression types from text data can be understood as performing pattern recognition on text content in user interaction records.

[0051] Specifically, structural design patterns here refer to recurring organizational forms in copywriting, such as the "problem-solution" structure, the "feature-advantage-benefit" structure, or "scarcity" prompts. These patterns are associated with highly responsive emotions such as joy or resonance. The frequency of recurrence among these patterns is statistically analyzed using a comparison tool.

[0052] For example, using a text analysis engine such as a comparison algorithm based on natural language processing, the text data is first segmented and processed, and then the frequency of occurrence of a specific pattern, such as the "question-answer" structure, is calculated. If the pattern is repeated more than 50 times in 1,000 texts, preliminary correspondence data can be generated, which helps to establish a mapping foundation between sentiment types and patterns.

[0053] In one possible implementation, the preliminary correspondence data is filtered using a preset threshold, for example, a threshold of 30% of the repetition frequency. If the repetition frequency of a certain structural design pattern is higher than this threshold, it is determined that the pattern has a strong correlation with the emotional expression type, thus obtaining a set of filtered association patterns. This filtering process involves a data filtering algorithm.

[0054] Specifically, the corresponding relationship data is first imported into the database, and then the query script is run to compare the frequency value of each pattern. If the frequency of the "emotional climax progression" pattern under the emotion of joy reaches 40%, which exceeds the threshold, it is marked as a strong association. This step ensures that only the set of highly correlated patterns is retained, providing refined input for subsequent analysis.

[0055] For example, based on the filtered set of association patterns, the matching degree between each type of emotional expression and the structural design pattern can be ranked using a data comparison tool, and the matching quantification can be achieved through cosine similarity calculation.

[0056] Specifically, in business scenarios, such as targeting "motivational" emotions in user preferences, we compare them with patterns like "storytelling". If the matching score is above 0.8, it is ranked higher. Finally, we determine the main structural design pattern corresponding to each type of emotional expression and form a classification mapping table. This table uses emotional type as the row and pattern as the column, and marks the matching score for easy and quick lookup.

[0057] In one possible implementation, the association rules between each type of emotional expression and the structural design pattern are obtained through the classification mapping table. For example, the rule is in the form of "if the emotion is resonance, then the contrasting structural pattern is preferred". The data is stored according to the rule, such as in a relational database, and the most suitable combination relationship between the structural design pattern and the emotional expression type is determined.

[0058] Specifically, in actual business operations, the validity of the rules is verified by verifying historical interaction data. If the combination of "resonance + comparison structure" increases the response intensity in user feedback by 20%, it is determined to be the most suitable. This not only optimizes the copywriting generation strategy, but also improves the targeting of user interaction.

[0059] As one implementation method in this embodiment, if the association rules contain platform-specific characteristics, the process of separating and clustering a subset of platform rules to determine the platform-specific copywriting patterns includes: The dataset containing platform-specific features is extracted from association rules, and rule subsets for specific platforms are separated by classification and filtering. The resulting rule grouping set is then processed by labeling. Clustering tools were used to refine the rule grouping set, platform characteristics were divided in multiple dimensions, and the copywriting creation pattern characteristics within the subset were determined. If the characteristics of copywriting patterns meet the threshold, the core pattern elements within the subset are extracted, and the degree of matching with the platform-specific attributes is judged to obtain the initial selection set of platform-specific copywriting patterns. By comparing and verifying the initial selection set, we can obtain copywriting patterns that are highly relevant to the platform's unique characteristics. We can then integrate the verification results to determine the final set of copywriting patterns specific to the platform.

[0060] Step S105: If the obtained association rules contain platform difference features, then separate the rule subset of the specific platform, and use a clustering algorithm to further refine the subset to determine the copywriting rules specific to the platform.

[0061] A dataset containing platform-specific features is extracted from the association rules. A pre-established classification tool is used to initially filter the dataset, separating rule subsets specific to each platform. These subsets are then labeled to obtain a set of rule groups for each platform. Based on these rule groups, a clustering tool is used to refine the subsets, performing multi-dimensional segmentation based on platform characteristics to determine the copywriting patterns within each subset. If these copywriting patterns meet a preset threshold, a feature extraction tool is used to obtain the core pattern elements within the subset, assessing their matching degree with platform-specific attributes to obtain a preliminary set of platform-specific copywriting patterns. This preliminary set is then compared and verified to identify copywriting patterns highly correlated with platform-specific features. The verification results are then processed using a data integration tool to determine the final set of platform-specific copywriting patterns.

[0062] For example, when processing association rules for copywriting, the first step is to extract datasets containing platform-specific characteristics from these rules, such as rule data for social media platforms, including Twitter's preference for short copy and Instagram's visual guidance features. A pre-built classification tool, such as a machine learning-based classifier, is then used to initially filter the dataset. This tool analyzes keywords and structural labels to separate platform-specific rule subsets; for example, the Twitter-related rule subset is labeled as "short and interactive," thus obtaining platform-specific rule groups. These grouped sets help distinguish copywriting patterns across different platforms, ensuring more targeted subsequent processing.

[0063] In one possible implementation, the subsets are grouped according to the aforementioned rules, and then refined using clustering tools such as the K-means algorithm. During this process, multi-dimensional segmentation is performed based on platform characteristics, such as text length, emotional intensity, and interactive elements. This identifies the text creation patterns within each subset; for example, in the Twitter subset, the pattern of frequently used hashtags is identified. This refinement helps reveal subtle differences between platforms, providing a foundation for text optimization.

[0064] Specifically, if these copywriting patterns meet preset thresholds, such as a repetition frequency exceeding 80%, then core pattern elements within a subset are obtained using feature extraction tools such as TF-IDF. The degree of matching with platform-specific attributes is then assessed, such as evaluating whether hashtag usage highly matches Twitter's real-time interactive attributes, thus obtaining an initial selection set of platform-specific copywriting patterns. This initial selection set summarizes the core elements initially screened, facilitating further verification.

[0065] For example, by comparing and validating the initial selection, patterns in copywriting highly correlated with platform-specific characteristics can be identified. For instance, comparing the patterns on Twitter and Instagram reveals that Twitter emphasizes concise appeals while Instagram emphasizes descriptive narratives. The validation results are then processed using data integration tools such as database merging scripts to determine the final set of platform-specific copywriting patterns. This set ultimately forms actionable guidelines, such as establishing a pattern for Twitter to "start with a question to stimulate interaction," thereby improving the responsiveness of copywriting on specific platforms.

[0066] In one possible implementation, the entire process, from extracting the dataset to forming the final collection, ensures platform adaptability for copywriting.

[0067] For example, in actual business, when extracting copywriting rules for e-commerce platforms, the classification tool first filters out subsets of Taobao and JD.com, then clusters and divides them to reveal Taobao's promotional orientation characteristics. If the threshold is met, core elements such as "limited-time discount phrases" are extracted, and finally integrated into a set of exclusive rules. This not only optimizes copywriting design but also improves user conversion rates.

[0068] Specifically, the refinement of clustering tools involves the principle of multi-dimensional partitioning, that is, by calculating the distance between feature vectors, similar patterns are clustered into one category. For example, in copywriting data, the dimensions include sentiment type and structural pattern. After partitioning, the significance of the pattern features is judged to ensure the accuracy of the initial selection set.

[0069] For example, feature extraction tools operate based on statistical analysis. The process of obtaining core elements includes calculating the weight of the elements in the subset and comparing the matching degree with the platform attributes. If the value is higher than 0.9, it is considered a strong match, thereby constructing an initial selection set.

[0070] In one possible implementation, the comparative validation phase uses cross-validation to compare the correlation of patterns across different platforms. For example, correlation coefficients can be used to calculate the differences between Twitter and Instagram patterns. The integration tool then merges the validation data into structured tables to form the final set, which is easy for the copywriting team to use.

[0071] Step S106: By integrating the determined platform-specific copywriting creation rules and user characteristic data, an expanded rule set is generated. During the generation process, user psychological change indicators are obtained to acquire copywriting templates that are suitable for diverse user groups.

[0072] By leveraging a pre-established database of copywriting patterns, platform-specific copywriting style characteristics are acquired. Combined with the target audience's interest and preference data, data fusion tools are used to integrate these two types of data, resulting in a preliminary set of copywriting rules. Based on this preliminary set of rules, data on users' psychological changes under different copywriting content are obtained. For at least one of these psychological change indicators, in-depth analysis is performed using data mining tools to determine a set of psychological trigger points suitable for diverse audiences. If the data for a trigger point in the set exceeds a preset threshold, a content generation tool generates a corresponding copywriting template fragment for that trigger point, which is then integrated into a complete draft copywriting template. A text optimization tool performs semantic consistency detection and sentiment calibration on the draft copywriting template to obtain the final copywriting template, thus adapting it to the needs of the target audience.

[0073] For example, in the actual copywriting process, the first step is to extract style characteristic data for specific platforms from a pre-established database of copywriting patterns. This database is typically a structured storage system containing copy samples and analysis results from various platforms such as social media or e-commerce platforms.

[0074] Specifically, for an e-commerce platform like Taobao, the database might store style characteristics such as concise promotional slogans, emphasis on discounts, and patterns in user reviews. This is then combined with target audience interest and preference data—for example, data on young consumers' preferences might include their liking for fashion trends and interactive elements—using data fusion tools. Such tools can be machine learning-based algorithmic frameworks that overlay style characteristics with preference data through matching algorithms, such as calculating similarity vectors. This merges Taobao's promotional style with the fashion interests of young people, generating a preliminary set of copywriting rules, such as the rule to "use vivid adjectives to highlight product features," thereby ensuring that the copywriting better aligns with the platform and user needs.

[0075] In one possible implementation, based on this initial set of copywriting rules, further data on users' psychological changes under different copywriting content can be obtained. This data comes from user feedback systems, such as click-through rates, dwell time, and emotional feedback scores collected through A / B testing. For at least one psychological change indicator, in-depth analysis is performed using data mining tools, often variants of the Apriori algorithm or neural network models. These tools uncover data patterns; for example, analyzing the excitement level of young users when seeing interactive copywriting. If the excitement level is higher than average, it is identified as a psychological trigger point. The analysis process includes data cleaning, feature extraction, and clustering. For example, noisy data is first filtered, then keyword frequencies are extracted, and finally, clustering is done into a set of trigger points, such as "limited-time scarcity triggers purchase impulse" or "user testimonials trigger trust," thus adapting to the psychological needs of diverse groups.

[0076] For example, when the data for a certain trigger point in the set of psychological trigger points exceeds a preset threshold, such as the score for the curiosity trigger point exceeding 80%, a corresponding copywriting template fragment is generated through a content generation tool. This tool can be a generative model based on natural language processing, such as a variant of GPT. It creates fragments for trigger points, for example, generating a fragment like "Imagine how this trendy jacket will make you the center of attention on the street?" for fashion copywriting, and then integrates these fragments into a complete template draft, ensuring logical coherence.

[0077] In one possible implementation, the draft copy template is ultimately subjected to semantic consistency checks and sentiment calibration using text optimization tools. These tools include semantic analysis engines, such as the BERT model, to check logical consistency between sentences, for example, verifying whether the promotional section and the introductory section semantically match. Simultaneously, sentiment is calibrated to ensure an appropriate proportion of positive emotions, ultimately resulting in a copy template that is tailored to the needs of the target audience, such as young consumers. This optimization enhances the copy's appeal and leads to higher conversion rates in the business.

[0078] Step S107: Based on the obtained copywriting template adapted to diverse audiences, the newly input promotional information is matched. If the matching degree is lower than the preset threshold, the template parameters are adjusted to obtain optimized copywriting output to improve marketing effectiveness.

[0079] Based on the content of the promotional information, at least one matching template is retrieved from a pre-established copywriting template library. The keywords of the promotional information are compared with the keywords of the template to determine if the matching degree reaches a preset threshold, thus obtaining a preliminary matching result. If the preliminary matching result is lower than the preset threshold, the parameters of the template are adjusted to obtain adjusted copywriting content. The adjusted copywriting content is then compared again with the promotional information to determine if it meets the matching requirements. Based on the adjusted copywriting content, the preference characteristics of the target audience are retrieved from a pre-established diverse audience preference database. The tone of the copywriting is optimized based on these preference characteristics to determine if it conforms to the habits of the target audience, resulting in an optimized copywriting version. By comparing the optimized copywriting version with preset marketing effectiveness evaluation rules, the final output content is obtained. The format of the final output content is adjusted to determine the presentation format suitable for different channels.

[0080] For example, when processing promotional information, the system first retrieves at least one matching template from a pre-built copywriting template library. This library is a database storing various marketing copywriting styles; for example, for promotional activities on e-commerce platforms, the library might contain templates for limited-time discounts, buy-one-get-one-free offers, etc. Suppose the promotional information is about a limited-time offer on a smartwatch. The system will search for similar templates in the library based on keywords such as "smartwatch," "discount," and "limited-time." If a template emphasizing "technology product discount" is found, it will be extracted for further processing.

[0081] In one possible implementation, the keywords in the promotional information are compared with the keywords in the template to determine whether the matching degree reaches a preset threshold. This matching degree can be calculated by determining the keyword overlap rate. For example, if the preset threshold is 80%, and the promotional information has 10 keywords, and the template matches 7 of them, the matching degree is 70%, which is below the threshold, resulting in a preliminary unsatisfactory matching result. This comparison process involves natural language processing tools, such as using a cosine similarity algorithm to quantify the similarity between keyword vectors.

[0082] Specifically, the keywords are first converted into vector form, such as "intelligent" corresponding to a numerical vector. Then, the cosine of the angle between the two vectors is calculated. If the value is greater than 0.8, it is considered to have reached the threshold; otherwise, it enters the adjustment stage.

[0083] For example, if the initial matching result is lower than the preset threshold, the template parameters are adjusted to obtain the adjusted copy content. This adjustment may include replacing specific words or changing the sentence structure. For example, the original template is "Seize the opportunity to buy products and enjoy discounts", which is adjusted to "Take action now and buy smartwatches to enjoy limited-time offers". Then, the adjusted copy is compared with the promotional information again to determine whether it meets the matching requirements. Here, the comparison may use a semantic analysis engine to check the overall consistency. If the matching degree rises to 85% after adjustment, it is considered to meet the requirements and the next step is continued.

[0084] In one possible implementation, based on the adjusted copy content, the system obtains the preference characteristics of the target audience from a pre-established database of diverse audience preferences. This database collects interest data of users of different age groups. For example, for young white-collar workers, the preference characteristics might be "fashionable" and "convenient". The system will query the corresponding entries in the database, and then optimize the tone of the copy based on these characteristics to determine whether it conforms to the habits of the target audience, thus obtaining an optimized copy version.

[0085] Specifically, the optimization process can use a sentiment analysis model to first analyze whether the tone words in the copy, such as "immediately," are too urgent. If the target audience prefers a gentler expression, it can be changed to "Come and pick this favorite item!" to ensure that the copy is more in line with the relaxed and lively style that urban youth like.

[0086] For example, by comparing the optimized copy version with preset marketing performance evaluation rules, the final output content is obtained. These rules may include click-through rate prediction criteria, such as rules requiring the copy to contain a call to action and be no more than 100 characters long. If these rules are met, the final content is output. Then, the format of the final output content is adjusted to determine the presentation format suitable for different channels. For example, for WeChat pushes, it is adjusted to a short article with images, while for web pages, it is adjusted to a long paragraph with buttons. This adjustment can improve the applicability of the copy across multiple channels and achieve better marketing communication.

[0087] In one possible implementation, the entire process is seamlessly connected by the output of each step serving as the input for the next. For example, the result of matching the template directly influences the selection of adjustment parameters, thereby ensuring logical continuity from the initial promotional information to the final output.

[0088] Specifically, if a mismatch in tone is found during preference optimization, the process will go back to the copywriting adjustment stage for re-optimization, thus ensuring the relevance and effectiveness of the copywriting.

[0089] Based on this, this invention provides an automatic method for mining formulas for viral marketing copywriting based on big data analysis. This invention overcomes the limitations of traditional copywriting creation, which relies on human experience, through a data-driven approach. Its technical effectiveness lies in transforming unstructured copywriting content into calculable emotional expression vectors and structural design patterns, thereby achieving systematic analysis of massive amounts of historical copywriting data. Through cluster analysis, it can automatically identify common categories of emotional expression and accurately determine high-response emotional types by associating them with user interaction records. Furthermore, by mining the correlation rules between emotional expression and structural design, quantifiable creative rules are established. When the rules contain platform-specific characteristics, it can automatically separate and refine platform-specific rule subsets, forming creative rules for different platforms. By integrating platform-specific rules with user characteristic data and combining user psychological change indicators, copywriting templates adapted to the needs of diverse user groups are generated. Ultimately, it achieves intelligent matching and parameter optimization of newly input promotional information, significantly improving the accuracy and adaptability of copywriting creation, and providing effective technical support for large-scale, personalized marketing content generation.

[0090] Example 2 In this embodiment, a computer terminal device is provided, including: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of the above-described method for automatically mining marketing copywriting best-selling formulas based on big data analysis.

[0091] In this embodiment, a computer-readable storage medium is also provided, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the above-described method for automatically mining marketing copywriting best-selling formulas based on big data analysis.

[0092] In this embodiment, an electronic device is also provided, including a memory and a processor. The memory stores a computer program, and the processor is configured to run the computer program to perform the steps of the above-described method for automatically mining marketing copywriting best-selling formulas based on big data analysis.

[0093] In this embodiment, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the above-described method for automatically mining marketing copywriting best-selling formulas based on big data analysis.

[0094] The aforementioned program can run on a processor or be stored in memory (or a computer-readable medium). Computer-readable media includes both permanent and non-permanent, removable and non-removable media, and information storage can be achieved by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random-access memory (SRAM), dynamic random-access memory (DRAM), other types of random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0095] These computer programs may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes can be implemented by different modules for different steps.

[0096] This embodiment provides such a device or system. The system, referred to as an automatic marketing copywriting formula mining system based on big data analysis, includes: The data extraction and parsing module is used to extract historical copywriting data and user interaction records, and to parse the copywriting data to obtain emotional expression vectors and structural design patterns; The sentiment clustering module is used to cluster sentiment expression vectors to determine common categories of sentiment expressions; The response assessment module is used to correlate common categories of emotional expression with user interaction records, assess response intensity, and determine the type of emotional expression with high response intensity. The association rule mining module is used to extract association structure design patterns for high-response intensity emotion expression types and obtain the association rules between emotion and structure by comparing repetition frequency. The platform adaptation module is used to separate and cluster a subset of platform rules when the association rules contain platform-specific features, and to determine the platform-specific copywriting creation rules. The template generation module is used to integrate platform-specific patterns and user characteristics data to generate rule sets and obtain user psychological change indicators to obtain copywriting templates that are suitable for diverse groups of people. The copy output module is used to match new promotional information based on copy templates, adjust parameters and output optimized copy when the matching degree is low.

[0097] The system or apparatus is used to implement the functions of the methods in the above embodiments. Each module in the system or apparatus corresponds to each step in the method, as has been described in the method and will not be repeated here.

[0098] The above implementation method solves the problem of automatically mining marketing copywriting formulas based on big data analysis in related technologies, thereby ensuring that the problems existing in the prior art are resolved.

[0099] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for automatically mining formulas for best-selling marketing copy based on big data analysis, characterized in that, Includes the following steps: Extract historical copywriting data and user interaction records, and analyze the copywriting data to obtain emotional expression vectors and structural design patterns; Clustering sentiment expression vectors to determine common categories of sentiment expression; By associating common categories of emotional expressions with user interaction records, we can assess response intensity and identify the types of emotional expressions with high response intensity. For emotional expression types with high response intensity, we extract the association structure design pattern and obtain the association rules between emotion and structure by comparing repetition frequency; If the association rules contain platform-specific characteristics, then separate the platform rule subsets and cluster them for refinement to determine the platform-specific copywriting creation rules; By integrating platform-specific patterns and user characteristics data, a rule set is generated and user psychological change indicators are obtained to create copywriting templates suitable for diverse user groups. Match new promotional information to the copy template. If the matching degree is low, adjust the parameters and output optimized copy.

2. The method according to claim 1, characterized in that, The process of extracting historical copywriting data and user interaction records, and analyzing the copywriting data to obtain emotional expression vectors and structural design patterns includes: Extract historical copywriting data and user interaction records; Word segmentation is used to separate vocabulary and sentence structure units, resulting in a set of emotion-related words and a set of non-emotion-related words. Based on the sentiment dictionary, sentiment-related words are matched and assigned sentiment intensity values. If the value is higher than the threshold, it is classified as a key sentiment expression element and a sentiment expression vector is formed. Analyze the sentence structure, extract the main structure and modifiers, and determine whether it conforms to common design patterns to identify the core features of the structural design pattern. By associating emotional expression vectors and structural design patterns, and then using vector mapping and fusion to extract features, a text feature description is obtained.

3. The method according to claim 1, characterized in that, The process of clustering sentiment expression vectors to determine common categories of sentiment expression includes: Calculate the distance between sentiment expression vectors; if the distance is less than a threshold, they are grouped into the same group to obtain preliminary sentiment expression grouping. Calculate cluster centers for the grouping results, extract representative emotional feature values, and determine the core characteristics of emotional expression; Obtain the distribution range of feature values, compare the similarity between groups, and merge groups if the distribution ranges overlap beyond a threshold to obtain the optimized sentiment expression category classification. A unique identifier label is assigned to the optimized category to obtain the common category results of sentiment expression and to determine the typical classification pattern.

4. The method according to claim 1, characterized in that, The process of correlating common categories of sentiment expression with user interaction records, assessing response intensity, and identifying high-intensity sentiment expression types includes: The interaction frequency data under the emotion expression category is obtained from the user interaction record database, and the emotion category association mapping table is obtained by classifying and organizing the data. Based on the mapping table, sentiment analysis tools are used to quantitatively evaluate the response intensity of sentiment expression categories under different user preferences. If the response intensity is higher than the threshold, it is marked as a high-response sentiment type, forming a set of high-response sentiment types. We obtain user feedback data corresponding to the set of highly responsive emotional types from user interaction records, compare and analyze the matching degree, and determine the most representative combination of emotional expressions. Based on the most representative combinations, user behavior feature data is obtained, and in-depth analysis is conducted to determine the emotion expression recommendation scheme suitable for different user preferences.

5. The method according to claim 1, characterized in that, For high-response emotional expression types, the process of extracting association structure design patterns and comparing repetition frequencies to obtain the association rules between emotion and structure includes: Structural design patterns associated with high-response intensity emotional expression types were extracted from copywriting data, and the frequency of pattern repetition was statistically analyzed to obtain preliminary correspondence data. Filter the initial correspondence data. If the repetition frequency of the structural design pattern is higher than the threshold, it is judged as a strong correlation, and the filtered correlation pattern set is obtained. By comparing the matching degree between emotional expression types and structural design patterns, the main structural design patterns are determined after sorting, and a classification mapping table is formed. Based on the classification mapping table, obtain the association rules between emotion expression type and structural design pattern, store and determine the most suitable combination relationship.

6. The method according to claim 1, characterized in that, If the association rules include platform-specific characteristics, the process of separating and clustering subsets of platform rules to determine platform-specific copywriting patterns includes: The dataset containing platform-specific features is extracted from association rules, and rule subsets for specific platforms are separated by classification and filtering. The resulting rule grouping set is then processed by labeling. Clustering tools were used to refine the rule grouping set, platform characteristics were divided in multiple dimensions, and the copywriting creation pattern characteristics within the subset were determined. If the characteristics of copywriting patterns meet the threshold, the core pattern elements within the subset are extracted, and the degree of matching with the platform-specific attributes is judged to obtain the initial selection set of platform-specific copywriting patterns. By comparing and verifying the initial selection set, we can obtain copywriting patterns that are highly relevant to the platform's unique characteristics. We can then integrate the verification results to determine the final set of copywriting patterns specific to the platform.

7. A system for automatically mining formulas for best-selling marketing copy based on big data analysis, characterized in that, The system for implementing the method of any one of claims 1-6 comprises: The data extraction and parsing module is used to extract historical copywriting data and user interaction records, and to parse the copywriting data to obtain emotional expression vectors and structural design patterns; The sentiment clustering module is used to cluster sentiment expression vectors to determine common categories of sentiment expressions; The response assessment module is used to correlate common categories of emotional expression with user interaction records, assess response intensity, and determine the type of emotional expression with high response intensity. The association rule mining module is used to extract association structure design patterns for high-response intensity emotion expression types and obtain the association rules between emotion and structure by comparing repetition frequency. The platform adaptation module is used to separate and cluster a subset of platform rules when the association rules contain platform-specific features, and to determine the platform-specific copywriting creation rules. The template generation module is used to integrate platform-specific patterns and user characteristics data to generate rule sets and obtain user psychological change indicators to obtain copywriting templates that are suitable for diverse groups of people. The copy output module is used to match new promotional information based on copy templates, adjust parameters and output optimized copy when the matching degree is low.

8. A computer terminal device, characterized in that, include: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors perform the steps of the method as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-6.

Images