A method for generating, tracking, and managing viral marketing campaigns for offline brick-and-mortar stores.
By using methods for generating viral marketing campaigns in physical stores, tracking the dissemination chain, and managing conversions, we have solved the problems of high production costs, invisible dissemination chains, and difficulty in automatically segmenting customer leads in offline store marketing campaigns, and have achieved rule-based, calculable, and optimized management of campaigns.
Patent Information
- Application Number
- CN202611196244.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-07
- Publication Date
- 2026-11-10
AI Technical Summary
Marketing campaigns for existing offline physical stores are costly to produce, have invisible dissemination channels, and are difficult to automatically segment customer leads, resulting in poor effectiveness. Merchants need to rely on manual statistics and experience-based judgment.
By generating store profiles, matching templates, generating campaign scripts, validating platform rules, encoding the propagation chain, scoring customer leads, and attributing results, the campaign generation process is standardized, functionalized, and continuously optimized.
It improved the determinism of activity generation and the accuracy of template matching, reduced platform rule risks, realized the computability of the propagation chain, and formed a feedback optimization closed loop, thereby improving the effectiveness and quality of the activities.
Smart Images

Figure CN123071015A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of online marketing technology, specifically to a method for generating, tracking, and managing the dissemination chain of viral marketing campaigns for offline physical stores. Background Technology
[0002] Traditional brick-and-mortar stores have long relied on local foot traffic, referrals from existing customers, on-the-ground promotions, paper flyers, social media groups, and holiday sales to acquire customers. However, with the development of WeChat, short video platforms, mini-programs, online payments, and social sharing, store marketing is gradually shifting from point-to-point advertising to online event pages, social media viral marketing, online registration, payment-based customer retention, private domain building, and data-driven analysis.
[0003] For industries such as education and training, catering, beauty, fitness, jewelry, parenting, and photography, businesses typically need to quickly launch various promotional tools around holidays or peak / off-peak seasons, including campaigns for collecting likes, registration, raffles, group buying, distribution, coupons, questionnaires, invitations, flyers, independent stores, and data dashboards. The effectiveness of these campaigns depends not only on page design but also on the dissemination path, reward mechanism, lead quality, employee execution, and subsequent conversion. Therefore, how to automatically generate marketing campaigns suitable for specific industries and scenarios without requiring businesses to have experience in development, design, and complex marketing, and how to track dissemination levels, customer leads, order payments, employee sales, and in-store conversions in real time, has become a crucial technical challenge in the digital marketing of offline stores.
[0004] Existing offline store customer acquisition methods include manual promotion, paper flyers, WeChat group promotion, WeChat official account articles, H5 activity pages, coupon tools, group buying tools, registration forms, questionnaire tools, distribution tools, and general customer relationship management systems. Some tools can complete the creation of a single activity page or a single customer form, but there is usually no unified data structure and closed-loop control between activity template matching, page generation, social dissemination links, online payment, distribution commissions, customer segmentation, automatic follow-up, data dashboards, and multi-store execution management. Merchants still need to rely on manual statistics and experience to judge the effective dissemination links. It can be seen that the existing offline store digital marketing technology activities are costly to create, the dissemination links are invisible, customer leads are difficult to automatically segment, and the use effect is not good. Summary of the Invention
[0005] The purpose of this invention is to address the aforementioned technical problems by providing a method for generating, tracking, and managing the dissemination chain of viral marketing campaigns for offline physical stores. Through steps such as store profile generation, template matching, campaign script generation, platform rule verification, template scoring, dissemination chain encoding, dissemination graph construction, customer lead scoring, effect attribution, and parameter write-back, the method achieves the standardization, functionalization, and sustainable optimization of the campaign generation process.
[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for generating, tracking, and managing the dissemination chain of viral marketing campaigns for offline physical stores, comprising the following steps:
[0007] S1. Generate store profiles: Clean, normalize, label, and vectorize the store input data to form the profile data required for subsequent template matching and script generation;
[0008] S2. Match templates and generate activity scripts: Select candidate templates from the template library based on store profiles, and combine template fragments, store profiles and rule constraints into a structured script;
[0009] S3. Perform platform rule verification and template scoring: Perform platform rule verification on candidate scripts and select the final script based on the scoring function;
[0010] S4. Generate activity page parameters and material placement plan: Generate activity page parameters and material placement plan: Output page parameters based on TargetScript, without elaborating on specific operational activities;
[0011] S5. Generate propagation chain code: Generate traceable codes for activity entry and propagation nodes so that subsequent browsing, data entry, interaction and other events can be mapped to the propagation path;
[0012] S6. Collect events and construct a propagation graph: Transform user behavior into a propagation graph structure based on the propagation link encoding;
[0013] S7. Calculate customer lead scores: Calculate lead priority based on communication map metrics and behavioral characteristics;
[0014] S8, Effect Attribution and Parameter Update: Back-attribute campaign results to profiles, templates, script fragments, rules, and scoring functions;
[0015] S9. Generate parameters for the next round of activities: Write the updated profile status, template weight, rule weight, and scoring function parameters into the next round of generation process.
[0016] In the aforementioned method for generating, tracking, and managing viral marketing campaigns for offline physical stores, step S1 specifically includes the following steps:
[0017] S11. Collect basic store data: The fields to be collected include store industry, store location, product category, service radius, target customer group, business characteristics, historical activity titles, historical visit data, historical inquiry keywords, available material descriptions and prohibited expressions;
[0018] S12. Perform data cleaning: process null values, duplicate fields, abnormal characters, synonyms, and non-standard numbers;
[0019] S13. Generate rule tags: Rule tags should include at least industry tags, customer group tags, region tags, category tags, conversion target tags, material condition tags, expression style tags, and prohibited expression tags;
[0020] S14. Generate numerical features: Numerical features include historical average visit value, historical effective lead rate, historical dissemination depth, historical text length, number of materials, target customer matching radius, etc.
[0021] S15. Generate semantic vector: Perform word segmentation, noise reduction, and text encoding on the store introduction, product category selling points, historical activity titles, and customer inquiry keywords to obtain the store semantic vector V_store;
[0022] S16. Generate store profile: Profile={RuleTags,NumericFeatures,V_store,NegativeRules,HistoryState}.
[0023] In the aforementioned method for generating, tracking, and managing viral marketing campaigns for offline physical stores, step S2 specifically includes the following steps:
[0024] S21. Construct template features: Each template is preprocessed as Template={IndustryTags,SceneTags,GoalTags,TextBlocks,MaterialSlots,RuleTags,V_template,EffectState};
[0025] S22. Perform hard filtering: If the template industry tag does not match the store industry tag, the template requires more materials than the store material conditions, the template risk tag hits the store's prohibited expression, or the template target tag conflicts with the store's target tag, then the template will be removed.
[0026] S23. Calculate semantic similarity: For templates that have not been eliminated, calculate Sim=cosine(V_store,V_template); when Sim is less than the preset threshold T_sim, the template will not be included in the candidate set.
[0027] S24. Calculate the initial selection score for the template: Initial selection score for the template:
[0028] MatchScore = 0.35 × RuleMatch + 0.25 × Sim + 0.20 × GoalMatch + 0.10 × MaterialFit + 0.10 × HistoryFit, where RuleMatch is the rule tag matching score, GoalMatch is the target matching score, MaterialFit is the material fit score, and HistoryFit is the historical effect fit score.
[0029] S25. Select candidate templates: Sort the templates from highest to lowest according to MatchScore and select the top K templates as the candidate template set CandidateTemplates;
[0030] S26. Generate structured activity scripts: For each candidate template, fill the store profile field into the template TextBlocks and MaterialSlots to obtain the script candidate ScriptCandidate.
[0031] In the aforementioned method for generating, tracking, and managing viral marketing campaigns for offline physical stores, step S3 specifically includes the following steps:
[0032] S31. Establish the platform rule set: Platform rule set PlatformRules = {RequiredRules, ForbiddenRules, RiskRules, FormatRules};
[0033] RequiredRules represent fields that must exist.
[0034] ForbiddenRules indicates that an expression is prohibited.
[0035] RiskRules indicates a high-risk expression.
[0036] FormatRules specifies formatting restrictions such as title length, button text length, total script length, and the number of elements to be placed.
[0037] S32. Specific platform rules include:
[0038] The title does not contain absolute terms;
[0039] Scripts must not contain any unprovable promises of results;
[0040] The time, location, benefits, and restrictions in the script must be consistent with the store profile fields;
[0041] The same rights field in the same script must not contain contradictory expressions;
[0042] The action guidance must not use induced expressions explicitly prohibited by the platform;
[0043] S33. Perform mandatory rule verification: If the script lacks any core segment of Hook, Offer, Action, or Restriction, mark it as RequiredFail, and enter the completion or elimination process;
[0044] S34. Perform prohibition rule verification: If the script text hits a word, phrase or regular expression in ForbiddenRules, mark it as ForbiddenFail, and do not enter the final sorting;
[0045] S35. Perform risk rule verification: If the script hits RiskRules, calculate RiskScore; RiskScore=Σ(w_i×hit_i), where hit_i indicates whether the i-th risk rule is hit, and w_i indicates the risk weight of this rule;
[0046] S36. Perform consistency verification: Compare the fact fields in the script with the fact fields in the Profile item by item. If the field values conflict, the store profile field overrides the script field; if the override cannot be performed, eliminate the candidate script;
[0047] S37. Calculate the final template score:
[0048] FinalScore=0.25×MatchScore+0.20×TextCompleteness+0.20×ExpectedEffect+0.15×Novelty+0.10×FormatScore-0.10×RiskScore;
[0049] Where: TextCompleteness represents the script completeness,
[0050] ExpectedEffect represents the estimated value based on historical effects,
[0051] Novelty represents the degree of difference from recent activity texts,
[0052] FormatScore represents the format score;
[0053] S38. Select the final script: Select the script with the highest FinalScore and RequiredFail=false, ForbiddenFail=false, RiskScore<T_risk as the target activity script TargetScript.
[0054] In the aforementioned method for generating, tracking, and managing viral marketing campaigns for offline physical stores, step S3 specifically includes the following steps:
[0055] S41. Parse MaterialSlot in TargetScript to determine the material type, size, theme, and replacement fields required for each material slot;
[0056] S42. Sort available materials according to their quality score: MaterialScore = 0.40 × Sharpness + 0.25 × Subject Relevance + 0.20 × Size Fit + 0.15 × Historical Performance Value.
[0057] S43. Fill the high-scoring material into the material placeholder, forming PageParams={Title,Blocks,MaterialMap,StyleParams,ConstraintText,TrackingConfig};
[0058] S44. Perform format validation on PageParams to ensure that the title length, number of paragraphs, number of materials, button text, and restriction descriptions meet the platform's format rules.
[0059] In the aforementioned method for generating, tracking, and managing viral marketing campaigns for offline physical stores, step S3 specifically includes the following steps:
[0060] S51. Generate activity-level code activity_id, which includes store profile version, template number, script number and generation timestamp;
[0061] S52. Generate an entry-level code source_id to distinguish different delivery entry points, different employee entry points, or different initial propagation nodes;
[0062] S53. When a propagation relationship occurs, generate node_id and parent_node_id, and record Edge=(parent_node_id, node_id, source_id, time);
[0063] S54. The propagation encoding generation function can be node_id=Hash(activity_id+source_id+parent_node_id+user_seed+timestamp).
[0064] Here, user_seed is the de-identified user seed.
[0065] In the aforementioned method for generating, tracking, and managing viral marketing campaigns for offline physical stores, step S3 specifically includes the following steps:
[0066] S6. Collect events and construct a propagation graph: Transform user behavior into a propagation graph structure based on the propagation link encoding;
[0067] S61. Collect event Event={activity_id,node_id,event_type,event_time,event_value}, where event_type can be an abstract behavior type such as browsing, clicking, dwelling, submitting, forwarding, or consulting;
[0068] S62. Deduplicat events with the same node_id, reduce the weight of abnormally high-frequency events, and assign events with missing parent_node_id to the direct entry point.
[0069] S63. Construct a propagation graph G=(N,E), where N represents the set of propagation nodes and E represents the set of propagation edges;
[0070] S64. Calculate propagation graph metrics: Depth represents the maximum propagation level, Breadth represents the number of first-level diffusions, ValidRate represents the percentage of valid events, and NodeContribution represents the node contribution.
[0071] In the aforementioned method for generating, tracking, and managing viral marketing campaigns for offline physical stores, step S3 specifically includes the following steps:
[0072] S7. Calculate customer lead scores: Calculate lead priority based on communication map metrics and behavioral characteristics;
[0073] S71. Extract LeadFeatures = {Information completeness, behavior intensity, dissemination contribution, path depth, source quality, historical similarity, text intention keyword score};
[0074] S72. Calculate BehaviorScore = 0.35 × Number of valid clicks + 0.25 × Normalized dwell time + 0.20 × Number of key actions + 0.20 × Number of return visits;
[0075] S73. Calculate the propagation contribution SpreadScore = 0.40 × direct spread number + 0.30 × indirect spread number + 0.20 × propagation depth + 0.10 × source quality;
[0076] S74. Calculate the total lead score: LeadScore = 0.30 × BehaviorScore + 0.25 × SpreadScore + 0.20 × Information Completeness + 0.15 × Intended Keyword Score + 0.10 × Historical Similarity;
[0077] S75. Based on the LeadScore threshold, cues are classified into high-intention, medium-intention, low-intention, or invalid cues.
[0078] In the aforementioned method for generating, tracking, and managing viral marketing campaigns for offline physical stores, step S3 specifically includes the following steps:
[0079] S8, Effect Attribution and Parameter Update: Back-attribute campaign results to profiles, templates, script fragments, rules, and scoring functions;
[0080] S81. Aggregate activity results by activity_id to obtain Result={number of visits, number of valid events, number of valid leads, average propagation depth, average lead quality, and number of risk hits};
[0081] S82. Attribute the Result according to template number, script fragment, material placeholder and entry code to obtain Attribution={template_id:block_id:score_delta};
[0082] S83. Update the template history effect ExpectedEffect_new=α×ExpectedEffect_old+(1-α)×ResultScore, where α is the smoothing coefficient;
[0083] S84. Update Risk Rule Weights: If a certain type of script fragment triggers risk or low-quality results multiple times, increase the weight of the corresponding RiskRules; if no risk is triggered for a long time, decrease the weight.
[0084] In the aforementioned method for generating, tracking, and managing viral marketing campaigns for offline physical stores, step S3 specifically includes the following steps:
[0085] S91. Update the store profile HistoryState to write high-quality script snippets, high-quality material types, high-quality dissemination entry points, and low-quality expressions into the profile state.
[0086] S92. Update the template weight (TemplateWeight) to prioritize the recall of high-scoring templates under similar store profiles.
[0087] S93. Update the scoring function parameters. If historical data shows that text completeness has a higher impact on the result, increase the weight of TextCompleteness; if risk hit has a higher impact on the result, increase the penalty weight of RiskScore.
[0088] S94. Output the next round of generation parameters NextParams={ProfileUpdated,TemplateWeight,PlatformRulesUpdated,ScoringWeightsUpdated}.
[0089] Compared with the prior art, the advantages of the present invention are as follows:
[0090] 1. Increased the certainty of activity generation. By using store profiles, template features, platform rules, and scoring functions to jointly constrain script generation, the instability caused by relying solely on the experience of operations personnel is reduced.
[0091] 2. Improved template matching accuracy. Through hard filtering of rule tags, semantic vector similarity, and MatchScore comprehensive scoring, the template selection is made more closely aligned with store characteristics and target constraints.
[0092] 3. Reduced platform rule risks. Pre-validation using RequiredRules, ForbiddenRules, RiskRules, and FormatRules prevents incomplete, conflicting, or high-risk scripts from entering the release process.
[0093] 4. Computable propagation paths have been achieved. A propagation graph is constructed using activity_id, source_id, node_id, and parent_node_id, enabling quantitative analysis of propagation paths, node contributions, and clue sources.
[0094] 5. A feedback optimization loop has been formed. By using Attribution, ExpectedEffect_new, and NextParams, the activity results are written back to the parameters of the profile, template, rules, and scoring function, improving the quality of the next round of generation. Attached Figure Description
[0095] Figure 1 This is a flowchart of the present invention; Detailed Implementation
[0096] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.
[0097] like Figure 1 As shown, a method for generating, tracking, and managing the dissemination chain of viral marketing campaigns for offline physical stores includes the following steps:
[0098] S1. Generate store profiles: Clean, normalize, label, and vectorize the store input data to form the profile data required for subsequent template matching and script generation;
[0099] S2. Match templates and generate activity scripts: Select candidate templates from the template library based on store profiles, and combine template fragments, store profiles and rule constraints into a structured script;
[0100] S3. Perform platform rule verification and template scoring: Perform platform rule verification on candidate scripts and select the final script based on the scoring function;
[0101] S4. Generate activity page parameters and material placement plan: Generate activity page parameters and material placement plan: Output page parameters based on TargetScript, without elaborating on specific operational activities;
[0102] S5. Generate propagation chain code: Generate traceable codes for activity entry and propagation nodes so that subsequent browsing, data entry, interaction and other events can be mapped to the propagation path;
[0103] S6. Collect events and construct a propagation graph: Transform user behavior into a propagation graph structure based on the propagation link encoding;
[0104] S7. Calculate customer lead scores: Calculate lead priority based on communication map metrics and behavioral characteristics;
[0105] S8, Effect Attribution and Parameter Update: Back-attribute campaign results to profiles, templates, script fragments, rules, and scoring functions;
[0106] S9. Generate parameters for the next round of activities: Write the updated profile status, template weight, rule weight, and scoring function parameters into the next round of generation process.
[0107] Step S1 specifically includes the following steps:
[0108] S11. Collect basic store data: The fields to be collected include store industry, store location, product category, service radius, target customer group, business characteristics, historical activity titles, historical visit data, historical inquiry keywords, available material descriptions and prohibited expressions;
[0109] Data can be collected from forms, historical activity records, text imports, or API synchronization.
[0110] S12. Perform data cleaning: process null values, duplicate fields, abnormal characters, synonyms, and non-standard numbers;
[0111] For example, "children's art" and "children's art studio" are merged into a unified industry label, the address field is converted into administrative region and geographical coordinates, and meaningless stop words in the text are deleted.
[0112] S13. Generate rule tags: Rule tags should include at least industry tags, customer group tags, region tags, category tags, conversion target tags, material condition tags, expression style tags, and prohibited expression tags;
[0113] Rule tags are used for subsequent template hard filtering.
[0114] S14. Generate numerical features: Numerical features include historical average visit value, historical effective lead rate, historical dissemination depth, historical text length, number of materials, target customer matching radius, etc.
[0115] Numerical features are used in subsequent scoring function calculations.
[0116] S15. Generate semantic vector: Perform word segmentation, noise reduction, and text encoding on the store introduction, product category selling points, historical activity titles, and customer inquiry keywords to obtain the store semantic vector V_store;
[0117] V_store is used to calculate the similarity with the template semantic vector V_template.
[0118] S16. Generate store profile: Profile={RuleTags,NumericFeatures,V_store,NegativeRules,HistoryState}.
[0119] NegativeRules stores prohibited expressions and platform risk constraints, while HistoryState stores historical effect states.
[0120] Step S2 specifically includes the following steps:
[0121] S21. Construct template features: Each template is preprocessed as Template={IndustryTags,SceneTags,GoalTags,TextBlocks,MaterialSlots,RuleTags,V_template,EffectState};
[0122] Among them, TextBlocks are replaceable script fragments, MaterialSlots are placeholder materials, and EffectState is a historical effect statistics.
[0123] S22. Perform hard filtering: If the template industry tag does not match the store industry tag, the template requires more materials than the store material conditions, the template risk tag hits the store's prohibited expression, or the template target tag conflicts with the store's target tag, then the template will be removed.
[0124] S23. Calculate semantic similarity: For templates that have not been eliminated, calculate Sim=cosine(V_store,V_template); when Sim is less than the preset threshold T_sim, the template will not be included in the candidate set.
[0125] S24. Calculate the initial selection score for the template: Initial selection score for the template:
[0126] MatchScore = 0.35 × RuleMatch + 0.25 × Sim + 0.20 × GoalMatch + 0.10 × MaterialFit + 0.10 × HistoryFit, where RuleMatch is the rule tag matching score, GoalMatch is the target matching score, MaterialFit is the material fit score, and HistoryFit is the historical effect fit score.
[0127] S25. Select candidate templates: Sort the templates from highest to lowest according to MatchScore and select the top K templates as the candidate template set CandidateTemplates;
[0128] S26. Generate structured activity scripts: For each candidate template, fill the store profile field into the template TextBlocks and MaterialSlots to obtain the script candidate ScriptCandidate.
[0129] The script fragment source, replacement fields, template number, and image version number are retained during generation.
[0130] The script generates the following pseudocode:
[0131] fortemplateinCandidateTemplates:
[0132] blocks=select_blocks(template.TextBlocks,Profile.RuleTags,Profile.HistoryState)
[0133] script=fill(blocks,Profile.RuleTags,Profile.NumericFeatures)
[0134] script=attach_constraints(script,Profile.NegativeRules,template.RuleTags)
[0135] outputScriptCandidate(script,template.id,profile.version).
[0136] Step S3 specifically includes the following steps:
[0137] S31. Establish the platform rule set: Platform rule set PlatformRules = {RequiredRules, ForbiddenRules, RiskRules, FormatRules};
[0138] RequiredRules represent fields that must exist.
[0139] ForbiddenRules indicates that an expression is prohibited.
[0140] RiskRules indicates a high-risk expression.
[0141] FormatRules specifies formatting restrictions such as title length, button text length, total script length, and the number of elements to be placed.
[0142] S32. Specific platform rules include:
[0143] The title does not contain absolute terms;
[0144] Scripts must not contain any unprovable promises of results;
[0145] The time, location, benefits, and restrictions in the script must be consistent with the store profile fields;
[0146] The same rights field in the same script must not contain contradictory expressions;
[0147] Action guidance must not use suggestive language that is expressly prohibited by the platform;
[0148] S33. Perform required rule validation: If the script is missing any of the core segments Hook, Offer, Action, or Restriction, it will be marked as RequiredFail and will enter the completion or removal process.
[0149] S34. Perform forbidden rule validation: If the script text matches a word, phrase, or regular expression in ForbiddenRules, it is marked as ForbiddenFail and will not be included in the final sorting.
[0150] S35. Perform risk rule verification: If the script hits RiskRules, calculate RiskScore; RiskScore=Σ(w_i×hit_i), where hit_i indicates whether the i-th risk rule is hit, and w_i represents the risk weight of said rule;
[0151] S36. Perform consistency verification: Compare the fact fields in the script with the fact fields in the Profile item by item. If field values conflict, the script field is overwritten by the store profile field; if overwriting is not possible, the candidate script is eliminated;
[0152] S37. Calculate the final template score:
[0153] FinalScore=0.25×MatchScore+0.20×TextCompleteness+0.20×ExpectedEffect+0.15×Novelty+0.10×FormatScore-0.10×RiskScore;
[0154] where: TextCompleteness represents the script completeness,
[0155] ExpectedEffect represents the estimated value based on historical effects,
[0156] Novelty represents the degree of difference from recent activity texts,
[0157] FormatScore represents the format score;
[0158] S38. Select the final script: Select the script with the highest FinalScore and satisfying RequiredFail=false, ForbiddenFail=false, and RiskScore<T_risk as the target activity script TargetScript.
[0159] Step S3 specifically includes the following steps:
[0160] S41. Parse the MaterialSlot in the TargetScript, and determine the material type, size, theme and replacement field required for each material slot;
[0161] S42. Sort available materials according to the material quality score MaterialScore=0.40×clarity+0.25×subject relevance+0.20×size adaptability+0.15×historical performance value;
[0162] S43. Fill the high-scoring material into the material placeholder, forming PageParams={Title,Blocks,MaterialMap,StyleParams,ConstraintText,TrackingConfig};
[0163] S44. Perform format validation on PageParams to ensure that the title length, number of paragraphs, number of materials, button text, and restriction descriptions meet the platform's format rules.
[0164] Step S3 specifically includes the following steps:
[0165] S51. Generate activity-level code activity_id, which includes store profile version, template number, script number and generation timestamp;
[0166] S52. Generate an entry-level code source_id to distinguish different delivery entry points, different employee entry points, or different initial propagation nodes;
[0167] S53. When a propagation relationship occurs, generate node_id and parent_node_id, and record Edge=(parent_node_id, node_id, source_id, time);
[0168] S54. The propagation encoding generation function can be node_id=Hash(activity_id+source_id+parent_node_id+user_seed+timestamp).
[0169] Here, user_seed is the de-identified user seed.
[0170] Step S3 specifically includes the following steps:
[0171] S6. Collect events and construct a propagation graph: Transform user behavior into a propagation graph structure based on the propagation link encoding;
[0172] S61. Collect event Event={activity_id,node_id,event_type,event_time,event_value}, where event_type can be an abstract behavior type such as browsing, clicking, dwelling, submitting, forwarding, or consulting;
[0173] S62. Deduplicat events with the same node_id, reduce the weight of abnormally high-frequency events, and assign events with missing parent_node_id to the direct entry point.
[0174] S63. Construct a propagation graph G=(N,E), where N represents the set of propagation nodes and E represents the set of propagation edges;
[0175] S64. Calculate propagation graph metrics: Depth represents the maximum propagation level, Breadth represents the number of first-level diffusions, ValidRate represents the percentage of valid events, and NodeContribution represents the node contribution.
[0176] Step S3 specifically includes the following steps:
[0177] S7. Calculate customer lead scores: Calculate lead priority based on communication map metrics and behavioral characteristics;
[0178] S71. Extract LeadFeatures = {Information completeness, behavior intensity, dissemination contribution, path depth, source quality, historical similarity, text intention keyword score};
[0179] S72. Calculate BehaviorScore = 0.35 × Number of valid clicks + 0.25 × Normalized dwell time + 0.20 × Number of key actions + 0.20 × Number of return visits;
[0180] S73. Calculate the propagation contribution SpreadScore = 0.40 × direct spread number + 0.30 × indirect spread number + 0.20 × propagation depth + 0.10 × source quality;
[0181] S74. Calculate the total lead score: LeadScore = 0.30 × BehaviorScore + 0.25 × SpreadScore + 0.20 × Information Completeness + 0.15 × Intended Keyword Score + 0.10 × Historical Similarity;
[0182] S75. Categorize leads into high-intention, medium-intention, low-intention, or invalid leads based on the LeadScore threshold. The threshold can be dynamically adjusted based on historical campaign performance.
[0183] Step S3 specifically includes the following steps:
[0184] S8, Effect Attribution and Parameter Update: Back-attribute campaign results to profiles, templates, script fragments, rules, and scoring functions;
[0185] S81. Aggregate activity results by activity_id to obtain Result={number of visits, number of valid events, number of valid leads, average propagation depth, average lead quality, and number of risk hits};
[0186] S82. Attribute the Result according to template number, script fragment, material placeholder and entry code to obtain Attribution={template_id:block_id:score_delta};
[0187] S83. Update the template history effect ExpectedEffect_new=α×ExpectedEffect_old+(1-α)×ResultScore, where α is the smoothing coefficient;
[0188] S84. Update Risk Rule Weights: If a certain type of script fragment triggers risk or low-quality results multiple times, increase the weight of the corresponding RiskRules; if no risk is triggered for a long time, decrease the weight.
[0189] Step S3 specifically includes the following steps:
[0190] S91. Update the store profile HistoryState to write high-quality script snippets, high-quality material types, high-quality dissemination entry points, and low-quality expressions into the profile state.
[0191] S92. Update the template weight (TemplateWeight) to prioritize the recall of high-scoring templates under similar store profiles.
[0192] S93. Update the scoring function parameters. If historical data shows that text completeness has a higher impact on the result, increase the weight of TextCompleteness; if risk hit has a higher impact on the result, increase the penalty weight of RiskScore.
[0193] S94. Output the next round of generation parameters NextParams={ProfileUpdated,TemplateWeight,PlatformRulesUpdated,ScoringWeightsUpdated}. This is for use in step S2.
[0194] This method can execute pseudocode.
[0195] profile=build_profile(store_data)
[0196] candidates=retrieve_templates(profile,template_library)
[0197] scripts=[generate_script(profile,t)fortincandidates]
[0198] valid_scripts=[sforsinscriptsifcheck_required(s)andnotcheck_forbidden(s)]
[0199] target=argmax(valid_scripts,score=final_score(profile,script,platform_rules))
[0200] page_params=build_page_params(target,profile)
[0201] graph=update_spread_graph(events,tracking_code)
[0202] lead_scores=compute_lead_scores(graph,events)
[0203] next_params=update_parameters(profile,target,graph,lead_scores,results).
[0204] Example 1: Store profile generation.
[0205] Assume the input data includes store industry text, product category text, regional description, historical activity titles, historical visit data, and customer inquiry keywords. First, the text is segmented, stop words are removed, and synonyms are merged. Then, industry tags, customer group tags, regional tags, and target tags are generated. Next, the store introduction, product category selling points, and inquiry keywords are text-encoded to obtain V_store. Finally, RuleTags, NumericFeatures, V_store, NegativeRules, and HistoryState are combined to form the store profile.
[0206] Example 2: Template matching and script generation.
[0207] Assuming each template in the template library already possesses IndustryTags, SceneTags, GoalTags, TextBlocks, MaterialSlots, RuleTags, V_template, and EffectState, the method first performs hard filtering based on industry tags, target tags, and prohibited expressions. Then, it calculates the cosine similarity between V_store and V_template, and finally selects candidate templates using the MatchScore function. For each candidate template, the method fills the TextBlocks with the rule tags and numerical features from the Profile, forming a structured ScriptCandidate.
[0208] Example 3: Rule verification and scoring.
[0209] For the generated ScriptCandidate, the method first checks whether the core fragments such as Hook, Offer, Action, and Restriction are complete. Then, it filters out prohibited expressions using ForbiddenRules and calculates the RiskScore using RiskRules. If a candidate script passes both the required and prohibited rules, the final score is calculated using the FinalScore function, and the script with the highest score is selected as the TargetScript.
[0210] Example 4: Propagation graph and clue scoring.
[0211] After the activity runs, the method records abstract events based on activity_id, source_id, node_id, and parent_node_id, and constructs a propagation graph G=(N,E). Then, it extracts behavior intensity, propagation contribution, path depth, source quality, and intent keyword scores from the events, calculates the LeadScore, and classifies the leads into different priorities.
[0212] Example 5: Parameter write-back.
[0213] After the event ends, the method performs effect attribution based on template number, script fragment, material placeholder, and propagation entry point to obtain Attribution; then, it updates the template effect value according to ExpectedEffect_new=α×ExpectedEffect_old+(1-α)×ResultScore, and adjusts the RiskRules weights according to the risk hit situation, and finally outputs NextParams for the next round of event generation.
[0214] The principle of this embodiment lies in the following: through a series of steps including store profile generation, template matching, script formation, platform rule verification, template scoring, propagation link encoding, lead scoring, effect attribution, and post-event parameter updates, the activity is transformed from generation to optimization into a computable, reusable, and verifiable process. Unlike methods that merely describe specific operational actions such as activity pages, registration, and sharing, this invention focuses on how to use algorithmic rules to transform store characteristics, activity templates, platform limitations, propagation events, and result feedback into an executable activity generation method. The steps in this method can be executed by software programs or collaboratively by servers, terminals, or cloud services.
[0215] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.
Claims
1. A method for generating, tracking, and managing the dissemination chain of viral marketing campaigns for offline physical stores, characterized in that, Includes the following steps: S1. Generate store profiles: Clean, normalize, label, and vectorize the store input data to form the profile data required for subsequent template matching and script generation; S2. Match templates and generate activity scripts: Select candidate templates from the template library based on store profiles, and combine template fragments, store profiles and rule constraints into a structured script; S3. Perform platform rule verification and template scoring: Perform platform rule verification on candidate scripts and select the final script based on the scoring function; S4. Generate activity page parameters and material placement plan: Generate activity page parameters and material placement plan: Output page parameters based on TargetScript, without elaborating on specific operational activities; S5. Generate propagation chain code: Generate traceable codes for activity entry and propagation nodes so that subsequent browsing, data entry, interaction and other events can be mapped to the propagation path; S6. Collect events and construct a propagation graph: Transform user behavior into a propagation graph structure based on the propagation link encoding; S7. Calculate customer lead scores: Calculate lead priority based on communication map metrics and behavioral characteristics; S8, Effect Attribution and Parameter Update: Back-attribute campaign results to profiles, templates, script fragments, rules, and scoring functions; S9. Generate parameters for the next round of activities: Write the updated profile status, template weight, rule weight, and scoring function parameters into the next round of generation process.
2. The method for generating, tracking, and managing the dissemination chain of viral marketing activities for offline physical stores according to claim 1, characterized in that, Step S1 specifically includes the following steps: S11. Collect basic store data: The fields to be collected include store industry, store location, product category, service radius, target customer group, business characteristics, historical activity titles, historical visit data, historical inquiry keywords, available material descriptions and prohibited expressions; S12. Perform data cleaning: process null values, duplicate fields, abnormal characters, synonyms, and non-standard numbers; S13. Generate rule tags: Rule tags should include at least industry tags, customer group tags, region tags, category tags, conversion target tags, material condition tags, expression style tags, and prohibited expression tags; S14. Generate numerical features: Numerical features include historical average visit value, historical effective lead rate, historical dissemination depth, historical text length, number of materials, target customer matching radius, etc. S15. Generate semantic vector: Perform word segmentation, noise reduction, and text encoding on the store introduction, product category selling points, historical activity titles, and customer inquiry keywords to obtain the store semantic vector V_store; S16. Generate store profile: Profile={RuleTags,NumericFeatures,V_store,NegativeRules,HistoryState}.
3. The method for generating, tracking, and managing the dissemination chain of viral marketing activities for offline physical stores according to claim 2, characterized in that, Step S2 specifically includes the following steps: S21. Construct template features: Each template is preprocessed as Template={IndustryTags,SceneTags,GoalTags,TextBlocks,MaterialSlots,RuleTags,V_template,EffectState}; S22. Perform hard filtering: If the template industry tag does not match the store industry tag, the template requires more materials than the store material conditions, the template risk tag hits the store's prohibited expression, or the template target tag conflicts with the store's target tag, then the template will be removed. S23. Calculate semantic similarity: For templates that have not been eliminated, calculate Sim=cosine(V_store,V_template); when Sim is less than the preset threshold T_sim, the template will not be included in the candidate set. S24. Calculate the initial selection score for the template: Initial selection score for the template: MatchScore = 0.35 × RuleMatch + 0.25 × Sim + 0.20 × GoalMatch + 0.10 × MaterialFit + 0.10 × HistoryFit, where RuleMatch is the rule tag matching score, GoalMatch is the target matching score, MaterialFit is the material fit score, and HistoryFit is the historical effect fit score. S25. Select candidate templates: Sort the templates from highest to lowest according to MatchScore and select the top K templates as the candidate template set CandidateTemplates; S26. Generate structured activity scripts: For each candidate template, fill the store profile field into the template TextBlocks and MaterialSlots to obtain the script candidate ScriptCandidate.
4. The method for generating, tracking, and managing the dissemination chain of viral marketing activities for offline physical stores according to claim 3, characterized in that, Step S3 specifically includes the following steps: S31. Establish the platform rule set: Platform rule set PlatformRules = {RequiredRules, ForbiddenRules, RiskRules, FormatRules}; RequiredRules represent fields that must exist. ForbiddenRules indicates that an expression is prohibited. RiskRules indicates a high-risk expression. FormatRules specifies formatting restrictions such as title length, button text length, total script length, and the number of elements to be placed. S32. Specific platform rules include: The title does not contain absolute terms; Scripts must not contain any unprovable promises of results; The time, location, benefits, and restrictions in the script must be consistent with the store profile fields; The same rights field in the same script must not contain contradictory expressions; Action guidance must not use suggestive language that is expressly prohibited by the platform; S33. Perform required rule validation: If the script is missing any of the core segments Hook, Offer, Action, or Restriction, it will be marked as RequiredFail and will enter the completion or removal process. S34. Perform forbidden rule verification: if the script text hits a word, phrase or regular expression in ForbiddenRules, it is marked as ForbiddenFail and will not enter the final ranking; S35. Perform risk rule verification: if the script hits RiskRules, calculate RiskScore; RiskScore=Σ(w_i×hit_i), where hit_i indicates whether the i-th risk rule is hit, and w_i indicates the risk weight of this rule; S36. Perform consistency verification: compare the fact fields in the script with the fact fields in the Profile item by item. If the field values conflict, the script field is overwritten by the store profile field; if overwriting is not possible, the candidate script is eliminated; S37. Calculate the final template score: FinalScore=0.25×MatchScore+0.20×TextCompleteness+0.20×ExpectedEffect+0.15×Novelty+0.10×FormatScore-0.10×RiskScore; Where: TextCompleteness represents the script completeness, ExpectedEffect represents the estimated value based on historical effects, Novelty represents the degree of difference from recent activity texts, FormatScore represents the format score; S38. Select the final script: select the script with the highest FinalScore, RequiredFail=false, ForbiddenFail=false, and RiskScore<T_risk as the target activity script TargetScript.
5. The method for generating, tracking, and managing the dissemination chain of viral marketing activities for offline physical stores according to claim 4, characterized in that, Step S3 specifically includes the following steps: S41. Parse the MaterialSlot in TargetScript to determine the material type, size, theme and replacement field required for each material slot; S42. Sort the available materials according to the material quality score MaterialScore=0.40×clarity+0.25×subject relevance+0.20×size adaptability+0.15×historical performance value; S43. Fill the high-score materials into the material placeholders to form PageParams={Title,Blocks,MaterialMap,StyleParams,ConstraintText,TrackingConfig}; S44. Perform format verification on PageParams to ensure that the title length, number of paragraphs, number of materials, button text and restriction description meet the platform format rules.
6. The method for generating, tracking, and managing the dissemination chain of viral marketing activities for offline physical stores according to claim 5, characterized in that, Step S3 specifically includes the following steps: S51. Generate an activity-level code activity_id, the content of which includes the store profile version, template number, script number and generation timestamp; S52. Generate an entry-level code source_id, which is used to distinguish different delivery entries, different employee entries or different initial propagation nodes; S53. When a propagation relationship occurs, generate node_id and parent_node_id, and record Edge=(parent_node_id, node_id, source_id, time); S54. The propagation encoding generation function can be node_id=Hash(activity_id+source_id+parent_node_id+user_seed+timestamp). Here, user_seed is the de-identified user seed.
7. The method for generating, tracking, and managing the dissemination chain of viral marketing activities for offline physical stores according to claim 6, characterized in that, Step S3 specifically includes the following steps: S6. Collect events and construct a propagation graph: Transform user behavior into a propagation graph structure based on the propagation link encoding; S61. Collect event Event={activity_id,node_id,event_type,event_time,event_value}, where event_type can be an abstract behavior type such as browsing, clicking, dwelling, submitting, forwarding, or consulting; S62. Deduplicat events with the same node_id, reduce the weight of abnormally high-frequency events, and assign events with missing parent_node_id to the direct entry point. S63. Construct a propagation graph G=(N,E), where N represents the set of propagation nodes and E represents the set of propagation edges; S64. Calculate propagation graph metrics: Depth represents the maximum propagation level, Breadth represents the number of first-level diffusions, ValidRate represents the percentage of valid events, and NodeContribution represents the node contribution.
8. The method for generating, tracking, and managing the dissemination chain of viral marketing activities for offline physical stores according to claim 7, characterized in that, Step S3 specifically includes the following steps: S7. Calculate customer lead scores: Calculate lead priority based on communication map metrics and behavioral characteristics; S71. Extract LeadFeatures = {Information completeness, behavior intensity, dissemination contribution, path depth, source quality, historical similarity, text intention keyword score}; S72. Calculate BehaviorScore = 0.35 × Number of valid clicks + 0.25 × Normalized dwell time + 0.20 × Number of key actions + 0.20 × Number of return visits; S73. Calculate the propagation contribution SpreadScore = 0.40 × direct spread number + 0.30 × indirect spread number + 0.20 × propagation depth + 0.10 × source quality; S74. Calculate the total lead score: LeadScore = 0.30 × BehaviorScore + 0.25 × SpreadScore + 0.20 × Information Completeness + 0.15 × Intended Keyword Score + 0.10 × Historical Similarity; S75. Based on the LeadScore threshold, cues are classified into high-intention, medium-intention, low-intention, or invalid cues.
9. A method for generating, tracking, and managing the dissemination chain of viral marketing activities for offline physical stores, as described in claim 8, is characterized in that... Step S3 specifically includes the following steps: S8, Effect Attribution and Parameter Update: Back-attribute campaign results to profiles, templates, script fragments, rules, and scoring functions; S81. Aggregate activity results by activity_id to obtain Result={number of visits, number of valid events, number of valid leads, average propagation depth, average lead quality, and number of risk hits}; S82. Attribute the Result by template number, script fragment, material placeholder and entry code to obtain Attribution={template_id:block_id:score_delta}; S83. Update the template history effect ExpectedEffect_new=α×ExpectedEffect_old+(1-α)×ResultScore, where α is the smoothing coefficient; S84. Update Risk Rule Weights: If a certain type of script fragment triggers risk or low-quality results multiple times, increase the weight of the corresponding RiskRules; if no risk is triggered for a long time, decrease the weight.
10. A method for generating, tracking, and managing the dissemination chain of viral marketing activities for offline physical stores, as described in claim 9, is characterized in that... Step S3 specifically includes the following steps: S91. Update the store profile HistoryState to write high-quality script snippets, high-quality material types, high-quality dissemination entry points, and low-quality expressions into the profile state. S92. Update the template weight (TemplateWeight) to prioritize the recall of high-scoring templates under similar store profiles. S93. Update the scoring function parameters. If historical data shows that text completeness has a higher impact on the result, increase the weight of TextCompleteness; if risk hit has a higher impact on the result, increase the penalty weight of RiskScore. S94. Output the next round of generation parameters NextParams={ProfileUpdated, TemplateWeight,PlatformRulesUpdated, ScoringWeightsUpdated}.
