Search

CN-122022754-A - Intelligent mail system and generation method based on multi-mode user value analysis

CN122022754ACN 122022754 ACN122022754 ACN 122022754ACN-122022754-A

Abstract

The invention discloses an intelligent mail system and a generation method based on multi-mode user value analysis, belongs to the field of artificial intelligence, and aims to solve the problems of serious homogenization of the marketing content of the existing mail, delay of artificial response, low utilization rate of historical communication data, insufficient personalized touch capability and the like. The system comprises a user value scoring, a mail generating engine and an automatic triggering and sending management and control subsystem, wherein multisource heterogeneous data processing, user value quantification scoring and deep intention recognition are completed through a preset five-dimensional feature system, a large language model is driven to generate personalized mails, multidimensional compliance auditing and content quality closed loop control are completed, and automatic and accurate sending of the mails is realized by combining a full-scene event triggering and dynamic priority scheduling mechanism. The invention improves the matching degree of mail content and user demands, the response time of customer behaviors, the mail reply rate and the business transformation rate, and realizes the intelligent and automatic closed-loop management of the whole mail marketing process.

Inventors

  • ZHENG HUI
  • ZHOU FUYANG
  • ZHANG YUJIA
  • CHEN HONGXI

Assignees

  • 深圳市纷享互联科技有限责任公司

Dates

Publication Date
20260512
Application Date
20260413

Claims (10)

  1. 1. A multimodal user value analysis-based intelligent mail system comprising: The user value evaluation subsystem is used for processing multi-source heterogeneous user data, completing standardized preprocessing, constructing a scoring context, generating a structured scoring instruction Prompt_M based on a preset five-dimensional characteristic system, a weight configuration rule and scoring model configuration, outputting a quantized user value score containing multi-dimensional analysis through large language model reasoning, and synchronously generating and outputting a structured user intention recognition result; The mail generation engine is used for receiving the user value score and the user intention recognition result output by the user value evaluation subsystem, generating a mail composition special prompt word by combining a history communication record and a preset business rule, submitting the mail composition special prompt word to a large language model for executing reasoning, generating personalized mail content, executing automatic rule combination verification and quality check on the personalized mail content to complete content quality closed loop control, and And the automatic triggering and sending management and control subsystem is respectively in communication connection with the user value evaluation subsystem and the mail generation engine, and is used for triggering management and control and sending scheduling.
  2. 2. The system of claim 1, wherein the five-dimensional feature hierarchy includes a base attribute dimension, a user interaction dimension, a behavioral dynamics dimension, a demand urgency dimension, and a historical mail dimension, wherein: the basic attribute dimension is used for storing static attribute information of a user, wherein the static attribute information comprises a company scale, a business, a job role and a region, and the static attribute information is used for matching with a preset ideal customer portrait; The user interaction dimension is used for storing interaction behavior data of a user, wherein the interaction behavior data comprise occurrence times, occurrence frequency and session duration corresponding to page access, page downloading, telephone communication, online consultation and social media message multi-channel interaction behaviors, and the interaction behavior data are used for quantifying interaction liveness and interaction depth of the user; The behavior dynamic dimension is used for storing content labels corresponding to content browsed and downloaded by a user, and the content labels are used for identifying the interest direction, the role attribute and the behavior motivation of the user; The demand urgency dimension stores high-value interaction behavior data of a user, wherein the high-value interaction behavior data corresponds to preset strong purchase intention behaviors, the strong purchase intention behaviors comprise repeated access to product pages, repeated access to pricing pages, downloading of technical documents, addition of shopping carts and access to contact information pages, and the high-value interaction behavior data are used for identifying purchase intention urgency of the user; The historical mail dimension stores feedback data of a user on the historical mail, wherein the feedback data comprises mail opening behaviors, reply content keywords and emotion tendencies, and the emotion tendencies are judged to be positive, neutral or negative based on preset keyword rules.
  3. 3. The system of claim 2, wherein the user value scoring subsystem is configured to perform the steps of: The method comprises the steps of multi-source data aggregation and preprocessing, namely acquiring multi-source heterogeneous user data of user static information, interaction behavior sequences, historical communication records, product and target portraits and historical scoring data, completing data standardization integration, de-duplication and invalid data filtering, and constructing a complete scoring context comprising the data; The dynamic scoring instruction construction is that a structured scoring instruction Prompt_M is automatically generated by combining the input features based on a preset five-dimensional feature system, a weight configuration rule and scoring model configuration; the scoring instruction prompt_M is a special instruction for restricting a large language model to execute user value scoring reasoning, and clearly defines scoring dimension, weight rule, scoring processing logic and output format constraint, wherein the five-dimensional feature system is a basic framework of the scoring dimension, the weight configuration rule is scoring weight constraint of each dimension and subdivision behavior, and the scoring model configuration comprises scoring logic disassembling rule, large model reasoning constraint rule and output format standardization rule; Submitting the scoring instruction Prompt_M to a large language model to perform reasoning, and obtaining a user value score in a preset structural format conforming to the configuration constraint of the scoring model, wherein the user value score comprises a quantized user value total score, an original score, a weighted score and a score reason analysis based on objective facts corresponding to each dimension in the five-dimensional feature system; And identifying the user intention, namely identifying and analyzing the user key behaviors based on the scoring context to generate a structured user intention identification result, wherein the user key behaviors comprise search keywords and access or download key contents.
  4. 4. The system of claim 1, wherein the mail generation engine is configured to perform the steps of: The method comprises the steps of collecting multi-source input information, namely receiving user value scores, user intention recognition results and historical communication records of target users, wherein the user value scores comprise user value total scores, user layering grades and score details of each dimension of a five-dimensional feature system, and the user intention recognition results comprise core intention types and key topics of the users; Generating a mail communication strategy, namely generating a structured mail communication strategy prompt word based on the converged multi-source input information and combining a preset strategy generation template, submitting the mail communication strategy prompt word to a large language model to perform reasoning, and acquiring the structured mail communication strategy comprising communication actions, intention type matching and core communication strategies; and generating a final mail composition special prompt word, namely integrating the structured mail communication strategy, the converged multi-source input information and a preset mail composition rule to generate the final mail composition special prompt word, wherein the preset mail composition rule belongs to a component part of a preset business rule.
  5. 5. The system of claim 4, wherein the mail generation engine is configured to perform the steps of: analyzing the generation instruction, the user background information, the mail communication strategy and the history communication record context in the mail writing special prompt word through a large language model, and determining the core target and constraint rule of mail generation; Generating constraint contents, namely generating personalized mail contents which match the core requirements of users and attach mail communication strategies on the basis of analysis results through a large language model and follow preset content compliance requirements, format specifications and communication style constraints; And (3) standardized structured output, namely outputting a structured mail generation result according to preset format requirements agreed in the mail writing special prompt word through a large language model, wherein the mail generation result at least comprises a mail title, a mail text and a content abstract.
  6. 6. The system of claim 5, wherein the mail generation engine is configured to perform the following automated compliance audit and content quality closed loop control steps: Based on preset content auditing rules, executing multi-dimensional verification on personalized mail content generated by a large language model, wherein the multi-dimensional verification at least comprises key information matching verification, brand specification compliance verification, content compliance verification and format normalization verification; the verification is passed, namely if personalized mail content passes through all verification, the personalized mail content is judged to be qualified content, and the mail is pushed to a sending queue; And (3) checking the closed loop control which is not passed, if the personalized mail content is not passed through verification, analyzing, positioning and verifying reasons which are not passed through, and executing the following branch processing: An automatic regeneration branch, which is to generate an optimization constraint instruction based on the reasons that the verification is not passed, write special prompt words in combination with the original mails to regenerate personalized mail contents, and execute automatic verification again until the contents pass the verification or reach the preset maximum retry times; and (3) manually inserting branches, namely if the maximum retry times are not checked, or the content relates to a preset complex processing scene, routing the content to a manual processing link, receiving a manually corrected qualified mail content or a manually supplemented generation instruction, and completing the quality control of the mail content.
  7. 7. The system of claim 1, wherein the automated trigger and transmit control subsystem is configured to perform the trigger control steps of: the rule receiving comprises continuously receiving all-link behavior events of a user and user state changes based on a preset multi-dimensional triggering rule, wherein the multi-dimensional triggering rule at least comprises a user behavior event triggering rule, a timing task triggering rule, a user state change triggering rule and a historical communication response triggering rule; Triggering judgment, namely completing triggering validity verification and locking a target user to be followed when a received event or state change is matched with a triggering condition of any triggering rule; And starting the process, namely sending a trigger instruction to the user value evaluation subsystem, triggering the user value evaluation subsystem to pull multi-source heterogeneous data of a target user, starting the user value evaluation and user intention recognition process, synchronously sending a start instruction to the mail generation engine, and triggering the personalized mail generation process.
  8. 8. The system of claim 1, wherein the automated trigger and transmit control subsystem is configured to perform the following transmit scheduling steps: receiving and enqueuing the qualified mail content with the closed loop control of the content quality, which is output by the mail generating engine, and storing the mail to be sent into a preset priority queue; The priority dynamic judgment comprises the steps of distributing a sending priority to each mail to be sent based on a preset priority rule, wherein the judgment dimension of the priority rule at least comprises a user layering grade corresponding to a user value score, the emergency degree of a triggering event, the aging requirement of mail follow-up and the response condition of historical communication; scheduling and sending management and control, namely performing scheduling management on mails to be sent according to the allocated priority, supporting three sending modes of real-time sending, preset time sending and optimal reaching time sending; and (3) sending closed-loop management, namely recording the sending state and the sending result of each mail, and completing the closed-loop management and the data archiving of the whole mail sending process.
  9. 9. A method for generating intelligent mail based on multi-mode user value analysis, applied to the system as set forth in any one of claims 1-8, comprising the steps of: the triggering control subsystem is used for receiving user behavior events and user state changes based on preset triggering rules through automatic triggering and sending control subsystems, and triggering the whole flow of user value evaluation and mail generation when triggering conditions are met; Processing multi-source heterogeneous user data through a user value evaluation subsystem, completing standardized preprocessing, constructing a scoring context, generating a structured scoring instruction Prompt_M based on a preset five-dimensional feature system, a weight configuration rule and scoring model configuration, submitting the scoring instruction Prompt_M to a large language model for reasoning, outputting quantized user value scores containing multi-dimensional analysis, and synchronously generating and outputting a structured user intention recognition result; The mail prompt word is generated by receiving the user value scores and the user intention recognition results through a mail generation engine and generating a mail writing special prompt word by combining a historical communication record and a preset business rule; personalized mail content generation, namely submitting the mail writing special prompt word to a large language model to perform reasoning, and generating personalized mail content; Performing automatic compound rule checking and quality checking on the generated personalized mail content to finish the content quality closed loop control, and outputting qualified mail content passing the checking; And E, mail scheduling and automatic sending, namely receiving the qualified mail content through an automatic triggering and sending management and control subsystem, and carrying out real-time or timed scheduling management on the mail to be sent through a preset priority queue based on a preset scheduling rule to complete automatic sending of the mail.
  10. 10. The method according to claim 9, wherein the five-dimensional feature hierarchy comprises a basic attribute dimension, a user interaction dimension, a behavior dynamic dimension, a demand urgency dimension, and a historical mail dimension, in particular: The basic attribute dimension corresponds to static attribute information of a user, wherein the static attribute information comprises a company scale, a business, a job role and a region, and is used for matching a preset ideal customer portrait; The user interaction dimension corresponds to interaction behavior data of a user, wherein the interaction behavior data comprise occurrence times, occurrence frequency and session duration corresponding to page access, page downloading, telephone communication, online consultation and social media message multi-channel interaction behaviors, and the interaction dimension is used for quantifying interaction liveness and interaction depth of the user; the behavior dynamic dimension corresponds to a content tag corresponding to content browsed and downloaded by a user and is used for identifying the interest direction, the character attribute and the behavior motivation of the user; The demand urgency dimension corresponds to high-value interaction behavior data of a user, the high-value interaction behavior data corresponds to preset strong purchase intention behaviors, and the strong purchase intention behaviors comprise repeated access to product pages, repeated access to pricing pages, downloading of technical documents, addition of shopping carts and access to contact information pages, and are used for identifying purchase intention urgency of the user; The dimension of the historical mail corresponds to feedback data of the user on the historical mail, wherein the feedback data comprises mail opening behaviors, reply content keywords and emotion tendencies, and the emotion tendencies are judged to be positive, neutral or negative based on preset keyword rules.

Description

Intelligent mail system and generation method based on multi-mode user value analysis Technical Field The invention belongs to the technical field of intersection of artificial intelligence and Customer Relationship Management (CRM), and particularly relates to an intelligent mail system based on multi-mode user value analysis and a generation method thereof. Background The prior art has the following defects: The mail is seriously homogenized, the traditional mail template can not dynamically adjust the content according to the behavior characteristics of the user, and the reply pertinence is poor. Traditional template mail is static and ubiquitous and cannot dynamically adjust content based on user real-time behavior (e.g., browsing specific product pages). This results in the pushed information being severely disjointed from the user's current interests and roles, lacking in relevance and pertinence. The customer cannot feel the understanding, and is very easy to treat as junk mail, so that the recovery rate and the conversion rate are low. Response delay, namely, the average time consumed by manually writing the mail can not respond to the key user behaviors in real time. The core of this pain spot is missing the marketing "golden window". Studies have shown that potential customers are linked within the first few minutes after interest, with the highest conversion. However, the manual mail writing process is cumbersome, and takes 15-30 minutes on average, at which time the user's attention is diverted or preempted by the competitor. This delay not only wastes time, but also directly results in business opportunity loss, making the acquisition of the investment by the potential customers in the earlier stage unable to honor the value. The information utilization rate is low, and a large amount of effective information in the historical mail is not systematically extracted and multiplexed. Enterprise historical mail contains large amounts of high-value "dark data" (e.g., customer demand, pain spots, project phases), but they are scattered in unstructured text form in personal mailboxes, forming "data islands". The lack of automated tool extraction is extremely inefficient by manual scrolling alone, resulting in a lack of historical context for each communication, severely compromising customer experience and sales efficiency. The individuation is insufficient, the user value is evaluated by lacking a quantitative model, and the degree of differentiation of the mail language and gas/content structure is insufficient. Mail policies are highly dependent on sales person subjective experiences and cannot quantitatively distinguish customer value. The high potential customers are similar to the communication resources obtained by the common clues, resulting in failure to provide a good experience to the most important customers. Therefore, the industry is in urgent need of an intelligent mail automatic solution scheme which can not only construct a full-dimensional quantitative evaluation system based on multi-mode user behavior data to realize accurate user value layering and deep intention recognition, fundamentally solve the problems of manual experience, serious homogenization and insufficient personalized matching of traditional mail marketing, but also realize mail content automatic generation and compliance quality closed loop control highly fitting user demands by means of AI generation capacity, fully mine the core value of multiplexing historical communication data, thoroughly break the industry dilemma of customer data island and low information utilization rate, realize intelligent triggering and dynamic priority marketing gold window based on the real-time behavior and life cycle state of users, and break the core problem of manual response delay and business machine loss. Disclosure of Invention Aiming at the problems of serious content homogenization, delayed manual response, missing of a marketing gold window, low utilization rate of high-value information caused by islanding of historical communication data and insufficient personalized touch capability caused by lack of a quantized user value evaluation system in the conventional enterprise sales development and customer relationship management scene, the invention provides an intelligent mail system and a generation method based on multi-mode user value analysis. Through constructing a multi-mode user value evaluation subsystem based on a five-dimensional characteristic system, an AI mail generation engine with content quality closed loop control capability and a real-time event driven automatic triggering and sending management and control subsystem, a full-flow automatic closed loop of user value quantitative evaluation, personalized mail intelligent generation and accurate touch scheduling is formed, the technical effects of poor pertinence of static template mails, low response hysteresis of manual writing efficiency, insufficient user d