Aigc digital human driven high propagation short video fission method

Abstract

The application discloses an AIGC digital human driven high propagation short video fission method, and particularly relates to the technical field of digital human propagation, which acquires original activity data and original digital human short videos under the same activity identifier, extracts and solidifies an original propagation intention anchor point set; constructs a fission graph according to propagation behavior data, identifies user fission roles and propagation circles, and determines a relay rewriting limit value and a relay rewriting range; generates a relay script scheme in the relay rewriting range, and determines a relay content boundary value according to the original propagation intention anchor point set and the relay rewriting range; and generates a relay type digital human short video and binds a relay propagation record after verification. The method is used for restraining course facts, trial listening rights and interests, applicable objects and fission guide contents in multi-level fission propagation.

Description

Technical Field

[0001] This invention relates to the field of digital human dissemination technology, and more specifically, to an AIGC digital human-driven method for high-spread short video viral marketing. Background Technology

[0002] With the application of AIGC digital human technology in scenarios such as course promotion, knowledge service lead generation, and corporate training registration, operators typically use fixed narration scripts, subtitles, course cards, and landing page entry points to generate digital human short videos, which are then disseminated through platform placement, user sharing, or private domain forwarding. While these methods meet the needs of original video release and daily promotion, in the process of multi-level viral dissemination, the video may enter different dissemination circles through different users. For each dissemination circle, the course value, usage scenarios, and reasons for forwarding are not entirely consistent. If the same version is retained, it is easy for the content to mismatch with the current dissemination circle, thus affecting viewing time, landing page clicks, trial access, and continued sharing effectiveness.

[0003] Currently, some technologies employ secondary rewriting techniques, modifying short video scripts, subtitles, or recommendation phrases. However, most of these rewrites primarily aim to increase appeal or conversion rates, lacking sustained constraints on the facts of the activity, trial benefits, registration requirements, target audience, and prohibited promises. When digital human short videos are forwarded at multiple levels, if the content is constantly adjusted at each dissemination stage, issues such as expanded course content, misrepresented trial benefits, generalized target audience, or promises exceeding the original activity rules arise. Therefore, it remains necessary to provide a digital human short video viral dissemination solution that can constrain the boundaries of the activity content while satisfying different dissemination circles. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide an AIGC digital human-driven high-propagation short video fission method to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for high-viral short video viral marketing driven by AIGC digital humans, characterized by the following steps:

[0007] Obtain raw activity data and raw digital human short videos under the same activity identifier, and extract and solidify the set of original communication intent anchor points;

[0008] A fission map is constructed based on the dissemination behavior data corresponding to the original digital human short video, the user fission role and dissemination circle corresponding to the current dissemination node are identified, and the relay rewriting limit value and the corresponding relay rewriting range are determined based on the circle deviation, node driving force and original text decay.

[0009] Within the scope of the relay rewriting, a relay script scheme is generated based on the user fission role and the propagation circle, and a list of fields that can be rewritten and a list of fields that cannot be rewritten are formed based on the scope of the relay rewriting.

[0010] The relay content boundary value is determined based on the original propagation intent anchor point set, the list of fields that can be rewritten, the list of fields that cannot be rewritten, and the relay rewriting range. The relay content boundary value is determined based on the amount of fact retention, the amount of rights retention, and the amount of rewriting exceeding the boundary.

[0011] When the boundary value of the relay content meets the preset boundary conditions and there are no mandatory interception events, a relay-style digital human short video is generated based on the relay script scheme, the relay propagation record is bound, and the current video identifier is written back to the fission graph.

[0012] In a preferred embodiment, the original activity data comes from the activity management terminal, the course management terminal, and the landing page configuration terminal, and includes at least the course name, the target audience of the course, the scope of course content, trial listening rights, registration conditions, discount rules, prohibited promises, and the official publishing entity; the original digital human short video includes at least the narration script, subtitle content, course card, landing page fields, activity rules, video identifier, release time, publishing account, and original target audience; the original set of communication intent anchor points includes at least fact anchor points, rights anchor points, target audience anchor points, prohibited expansion anchor points, and viral marketing anchor points.

[0013] In a preferred embodiment, when extracting and solidifying the original set of communication intent anchor points, the original activity data is matched with the content parsing results of the original digital human short video by field mapping, and the anchor content is determined according to the priority of the field source. When the same content has different expressions in different sources, the content in the structured fields is used as the anchor content first, and similar expressions in other sources are used as the identifiable expressions of the anchor. When any of the key fields such as course name, course content scope, trial listening rights, registration conditions, discount rules, prohibited commitments, and official release subject are empty, the corresponding activity is marked as an incomplete anchor state, and the relay-style digital human short video generation step is stopped.

[0014] In a preferred embodiment, when constructing the fission graph, the propagation behavior data is deduplicated and associated. User hash identifiers, anonymous access identifiers, video identifiers, and relay version identifiers are used as graph nodes, and sharing relationships, access relationships, audition and collection relationships, registration relationships, and secondary sharing relationships are used as graph edges. Parent nodes, child nodes, propagation levels, sharing tokens, trigger times, and corresponding video identifiers are recorded in the graph edges.

[0015] In a preferred embodiment, when identifying the propagation circle and user fission role corresponding to the current propagation node, the propagation circle is determined based on the access user tags, comment content, landing page click area, and subsequent sharing objects; when the highest circle matching value is lower than a preset matching threshold, or the difference between the highest circle matching value and the second highest circle matching value is lower than a preset difference threshold, the current propagation node is marked as a circle pending confirmation state; the user fission role is determined based on its own conversion status, number of effective second-hop visits, number of continued sharing, number of trial listening and receiving, and number of registrations, and the user fission role includes conversion users, propagation users, and bridging users.

[0016] In a preferred embodiment, determining the relay rewriting limit value includes: determining the circle deviation amount, node driving amount, and original text decay amount respectively, and weighting the circle deviation amount, node driving amount, and original text decay amount; the circle deviation amount is determined based on the matching between the current circle characteristics and the original circle characteristics; the node driving amount is determined based on the number of effective second-hop visits, the number of continued sharing, the number of trial listening and receiving, and the number of registrations; the original text decay amount is determined based on the decrease in playback completion rate, landing page click-through rate, and continued sharing rate in the current dissemination level or current dissemination circle relative to the benchmark value; when the relay rewriting limit value is lower than the first limit threshold, only the addressing method, opening hook, and sharing guidance text are allowed to be rewritten; when the relay rewriting limit value is greater than or equal to the first limit threshold and less than the second limit threshold, the pain point expression, case scenario, and digital human tone are allowed to be rewritten; when the relay rewriting limit value is greater than or equal to the second limit threshold, the expression order, scenario introduction method, and fission guidance reason are allowed to be adjusted.

[0017] In a preferred embodiment, the list of fields that can be rewritten includes the following allowed rewriting methods within the current relay rewriting scope: the way of addressing the recipient, the opening hook, the expression of pain points, the case scenario, the tone of the digital human, and the reason for the viral marketing. The list of fields that cannot be rewritten includes the course name, the scope of course content, the trial listening rights, the registration conditions, the discount rules, the official publishing entity, and the expression content corresponding to the anchor points that are prohibited from being expanded. When generating the relay script scheme, only the content in the list of fields that can be rewritten is replaced with synonyms, the order is adjusted, or the expression is contextualized. Sentences in the relay voice-over script and relay subtitle content are marked with sentence source tags and rewriting tags.

[0018] In a preferred embodiment, the fact retention amount is determined based on the hit rate of fact anchors corresponding to course name, course content scope, course service scope, and registration conditions in the relay script scheme; the rights retention amount is determined based on the hit rate of rights anchors corresponding to trial listening rights, preferential rules, group registration discounts, redemption conditions, and rights validity period in the relay script scheme; the rewrite boundary violation amount is determined based on the comparison results between the segments in the relay script scheme and the list of allowed rewrite fields, the list of unrewriteable fields, and the relay rewrite scope; when the relay content boundary value reaches or exceeds the preset boundary threshold, and there are no mandatory interception items, it is determined that the relay content boundary value meets the preset boundary conditions.

[0019] In a preferred embodiment, the mandatory interception items include course name replacement, expansion of trial listening rights, change of registration conditions, expansion of applicable objects, impersonation of official release entities, and prohibition of expanded anchor point hits; when the relay content boundary value is lower than the preset boundary threshold, but the fact retention and rights retention are not abnormal, and the rewriting over-boundary amount is too high only because the way of addressing, the opening hook, the expression of pain points, or the case scenario exceeds the current relay rewriting scope, a targeted regeneration is performed; when the fact retention or rights retention does not meet the preset boundary conditions, it falls back to the stable script template, which is based on the original digital human short video's voice-over script, subtitle content, course cards, and fission guidance anchor points, and only allows the replacement of the way of addressing and the sharing entry prompt.

[0020] In a preferred embodiment, when generating a relay-style digital human short video, the system generates spoken audio based on the relay script, determines the speech rate, pause positions, and repetition positions based on the digital human's tone identifier, generates a subtitle timeline based on the relay subtitle content, and selects scene background, screen overlays, and viral entry points based on the dissemination circle. The relay dissemination record includes at least the original video identifier, parent video identifier, current video identifier, relay version identifier, dissemination level, dissemination node identifier, dissemination circle identifier, user viral role, relay rewriting limit value, relay rewriting scope, relay content boundary value, and generation time. After the relay-style digital human short video is published, the system updates the user viral role identification rules, dissemination circle identification rules, and relay rewriting strategy based on the playback completion rate, landing page click-through rate, trial access rate, registration rate, second-click entry rate, continued sharing rate, and comment feedback.

[0021] The technical effects and advantages of this invention are as follows:

[0022] This invention extracts and solidifies the original set of communication intent anchors under the same activity identifier. It uses course name, course content scope, trial access rights, registration requirements, target audience, prohibited promises, and viral marketing guidance content as the constraint basis for the relay generation process. This ensures that digital human short videos are not simply forwarded or arbitrarily rewritten during multi-level viral dissemination. Before generating the relay script, the system first identifies the user's viral role and dissemination circle corresponding to the current dissemination node based on the viral graph. Then, it determines the scope of relay rewriting based on the relay rewriting limit value, thus providing clear boundaries for adjustments to the form of address, opening hooks, pain point expression, case scenarios, and viral marketing reasons. This ensures that the relay-style digital human short videos align with the concerns of different dissemination circles while preventing the expansion of course facts, misrepresentation of trial access rights, alteration of registration requirements, or the inclusion of promises that exceed the original activity rules.

[0023] This invention also verifies the relay script scheme through a relay content boundary value, and performs targeted regeneration or rollback to a stable script template when the preset boundary conditions are not met. This ensures that the relay-style digital human short video has a judgment, interception, and rollback processing mechanism before generation. After the relay-style digital human short video is generated, the system writes the original video identifier, parent video identifier, current video identifier, propagation level, propagation circle identifier, user fission role, relay rewriting limit value, relay rewriting scope, and relay content boundary value into the relay propagation record, and writes the current video identifier back into the fission graph, facilitating the subsequent tracing of the source and rewriting basis of each version of the video. At the same time, the system updates the user fission role identification rules, propagation circle identification rules, and relay rewriting strategy based on playback completion rate, landing page click-through rate, trial access rate, registration rate, second-click entry rate, continued sharing rate, and comment feedback, so that the short video fission propagation process can continuously adapt to the actual propagation effect while keeping the content under control. Attached Figure Description

[0024] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings;

[0025] Figure 1 This is a flowchart illustrating a method for high-propagation short video viralization driven by AIGC digital human according to the present invention.

[0026] Figure 2 This is a schematic diagram illustrating the extraction and solidification of the original propagation intent anchor points of the present invention;

[0027] Figure 3 This is a schematic diagram illustrating the fission map construction and propagation node identification of the present invention;

[0028] Figure 4 This is a schematic diagram illustrating the determination of the relay rewriting limit and relay rewriting range in this invention;

[0029] Figure 5This is a schematic diagram of the relay content boundary verification and relay-style digital human short video generation closed loop of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Example 1: This example provides a highly viral short video propagation method driven by AIGC digital humans, which is suitable for scenarios such as course promotion, knowledge service lead generation, and corporate training registration that require multi-level dissemination through digital human short videos.

[0032] In this embodiment, the system does not simply forward the original digital human short video among different users. Instead, after the short video enters different dissemination nodes and dissemination layers, it generates relay-style digital human short videos based on the viral spread. The relay rewriting boundary is limited by the original dissemination intent anchor point, allowing the relay-style digital human short videos to adapt to new dissemination layers while preventing deviations in course facts, trial access rights, applicable audiences, or viral guidance content. The overall processing flow of the method described in this embodiment is as follows: Figure 1 As shown.

[0033] Step one: The system acquires the original activity data and original digital human short videos under the same activity identifier, analyzes the original activity data and original digital human short videos, and extracts and solidifies the original set of communication intent anchor points. The original activity data includes at least the course name, target audience, course content scope, trial listening rights, registration conditions, discount rules, prohibited promises, and official publishing entity; the original set of communication intent anchor points includes at least factual anchor points, rights anchor points, target audience anchor points, prohibited expansion anchor points, and viral marketing anchor points. By solidifying the original set of communication intent anchor points in advance, the boundaries of content that needs to be retained, cannot be expanded, or cannot be rewritten in the subsequent relay script plan are clarified, providing a basis for determining the scope of relay rewriting and judging the boundary values ​​of relay content.

[0034] Step two: The system generates a fission map based on the propagation behavior data corresponding to the original digital human short video. Within the fission map, the system identifies the user fission role and propagation circle corresponding to the current propagation node. It then determines the relay rewriting limit and its corresponding range by combining the circle deviation, node-driven momentum, and original content attenuation. This process determines whether the current propagation node's propagation level, changes in the propagation circle, and subsequent propagation momentum necessitate the generation of a relay-style digital human short video. It provides information on the degree of rewriteability, including the title, opening hook, sharing prompt, pain point expression, case scenario, digital human tone, expression order, scene introduction method, and fission guidance reason.

[0035] Step 3: Within the scope of the relay rewriting, create a relay script plan based on the user fission roles and the dissemination circles, and form a list of allowed and unchangeable fields based on the relay rewriting scope. The list of allowed fields includes the salutation, opening hooks, pain point expressions, case scenarios, digital human tone, and fission guidance reasons that can be adjusted within the current relay rewriting scope. The list of unchangeable fields includes the course name, course content scope, trial listening rights, registration conditions, discount rules, official publishing entity, and the expression content corresponding to the prohibited expansion anchors. Therefore, the relay script plan can only be synonymously replaced, rearranged, or contextualized within the scope limited by the list of allowed fields, and the content in the list of unchangeable fields cannot be freely added to or changed.

[0036] Step four: The system determines the relay content boundary value based on the original propagation intent anchor point set, the list of allowed and unalterable fields, the relay rewriting range, and the relay rewriting scope. The relay content boundary value is determined based on the amount of fact retention, the amount of rights retention, and the amount of rewriting exceeding the boundary. When the relay content boundary value meets the preset boundary conditions and there are no mandatory interception events, the system generates a relay-style digital human short video based on the relay script scheme, binds the relay propagation record, and writes the current video identifier back to the fission graph. When the relay content boundary value does not meet the preset boundary conditions, the system performs a targeted regeneration or reverts to a stable script template. After the relay-style digital human short video is published, the system updates the user fission role identification rules, propagation circle identification rules, and relay rewriting strategy based on the playback completion rate, landing page click-through rate, trial access rate, registration rate, second-click entry rate, continued sharing rate, and comment feedback, forming a closed loop of relay generation, boundary verification, record writing back, and feedback update.

[0037] Specifically:

[0038] Step one: The system acquires the original activity data and the original digital human short video, and extracts the original communication intent anchor points. The original activity data comes from the activity management end, course management end, and landing page configuration end, and includes at least the course name, target audience, course content scope, trial access rights, registration requirements, discount rules, prohibited promises, and the official publishing entity. The original digital human short video is the first digital human short video released for the activity or one that is about to be released, and includes at least the spoken text, subtitles, course card, landing page fields, activity rules, video identifier, release time, publishing account, and original target audience. The system categorizes the original activity data and the original digital human short video under the same activity identifier to ensure that subsequent relay digital human short videos can be traced back to the same original activity source.

[0039] The system standardizes the structured fields in the original activity data. For course names, course content scope, trial access rights, registration requirements, and discount rules, the system retains the original field values ​​and records the field source, update time, and validity period. For prohibited promises, the system reads prohibited expressions and prohibited promise phrases, including guarantees of employment, guaranteed salary increases, unconditional refunds, guaranteed exam pass rates, and semantically similar promises, and uses these as boundary content that cannot be generated in subsequent relay scripts. For official publishing entities, the system records the publishing entity name, publishing entity identifier, and authorized activity scope to limit the identity descriptions that can be used in subsequent relay-style digital human short videos.

[0040] The system parses the original digital human short videos. For videos with existing text scripts, the system directly reads the spoken text and subtitles. For videos with only finished video files, the system obtains the spoken text through speech recognition and the subtitles through subtitle recognition or video frame text recognition, and records the appearance time of each sentence of spoken text and subtitles according to the video timeline. The system also reads course card and landing page fields. The course card includes at least the course name, trial access, registration access, and discount information. The landing page fields include at least the course introduction, target audience, benefits description, registration requirements, and activity rules. The parsed content is written into the original content parsing table. Each record in the original content parsing table includes the content text, content source, location of appearance, appearance time, and activity identifier.

[0041] The system maps the raw activity data to the fields of the raw content parsing table to extract the original communication intent anchor points. The process of extracting, merging, and solidifying the original communication intent anchor points is as follows: Figure 2As shown. The original dissemination intent anchor point refers to the boundary of the dissemination content that should be preserved, not expanded, or not rewritten during the subsequent relay-style digital human short video generation process. The system extracts fields according to their source priority, with structured fields from the activity management and course management ends having higher priority than the spoken text and subtitles in the original digital human short video, and landing page fields having higher priority than user comments or subsequent dissemination feedback. If the same content is expressed differently in different sources, the system prioritizes the content in the structured fields as the anchor content, and uses similar expressions from other sources as the identifiable expressions of that anchor point.

[0042] The original communication intent anchors include at least factual anchors, benefit anchors, target audience anchors, prohibition of expansion anchors, and referral guidance anchors. Factual anchors are formed from fields related to the facts of the course, such as the course name, course content scope, course service scope, and registration conditions. They are used to prevent the relay script scheme from expanding the course content to include services not provided in the original activity. Benefit anchors are formed from trial class benefits, discount rules, group discounts, redemption conditions, and benefit validity periods. They are used to prevent the relay script scheme from rewriting trial classes, coupons, or group discounts into benefits not provided in the original activity. Target audience anchors are formed from the course's target audience and the original target demographic. They are used to prevent the relay script scheme from improperly expanding the course's scope of application to groups not declared in the original activity. Prohibition of expansion anchors are formed from prohibited promises and expressions semantically similar to prohibited promises. They are used to restrict the relay script scheme from making guarantees, absolutes, or promises exceeding the activity rules. The viral marketing anchor points consist of the sharing entry point, invitation rules, methods for claiming benefits, and registration entry point, which are used to define the boundaries of expression regarding sharing, invitation, claiming, and registration in the relay script scheme.

[0043] When extracting anchor points, the system merges different expressions with the same meaning. For example, if the original activity data states "free trial lesson," the original digital human short video subtitles state "claim trial lesson first," and the landing page field states "claim trial lesson rights," the system will merge these contents into the same rights anchor point and retain the three source expressions, but the system will not expand this rights anchor point to "free access to the entire course." Similarly, if the original activity data states that the course is applicable to users with no prior experience in office software and working users who want to improve their work efficiency, the system can use "users with no prior experience" and "users who want to improve their work efficiency" as identifiable expressions for the applicable target anchor point, but it will not expand this to all job seekers or all company employees.

[0044] An anchor record is generated for each original dissemination intent anchor. The anchor record includes at least the anchor identifier, anchor type, anchor content, source field, source text, source time, validity period, whether it can be rewritten, and the reason why it cannot be rewritten. The anchor type distinguishes between factual anchors, rights anchors, applicable object anchors, anchors prohibited from expansion, and viral marketing anchors. Whether it can be rewritten indicates whether the anchor can be replaced with a synonym during the relay-style digital human short video generation process. The reason why it cannot be rewritten refers to the basis for the anchor not being rewritten, such as involving course facts, rights conditions, registration rules, or prohibited promises.

[0045] To facilitate stable comparison of the original communication intent anchor set in subsequent relay script schemes, the system also establishes an anchor identifiable expression table for each original communication intent anchor. This anchor identifiable expression table includes at least the original field expression, synonym expression, abbreviation expression, prohibited substitution expression, and numerical constraint information. For fact anchors and rights anchors, the system uses the original field expression in the structured fields as hard-matching content, and uses expressions in the original digital human short video, course card, and landing page fields that have the same meaning and do not change the substantive scope as synonym expressions. For applicable object anchors, the system only records the applicable object scope declared in the original activity data and its identifiable expression, and does not include groups not declared in the original activity data in the applicable object anchors. For prohibited expansion anchors, the system records prohibited expression words, prohibited promise phrases, and promise expressions with similar semantics.

[0046] When determining whether a relay script scheme hits the original communication intent anchor point, the system first performs a hard field match. If the hard field match fails, it then performs a synonym matching based on the anchor point identifiable expression table. For segments containing the number of trial sessions, discount amount, validity period of benefits, registration conditions, or applicable scope, the system simultaneously compares whether the corresponding values, periods, and scopes are consistent with the original activity data. If the relay script scheme uses similar expressions but expands the trial benefits, relaxes the discount rules, changes the registration conditions, or expands the applicable scope, it is not considered a valid hit, and the corresponding segment is written into the out-of-bounds segment record. For prohibited expansion anchor points, as long as a segment in the relay script scheme hits a prohibited expression word, a prohibited promise sentence, or is similar to the promise expression corresponding to the prohibited expansion anchor point, it is determined that the prohibited expansion anchor point has been hit.

[0047] To avoid missing anchor points, the system performs an integrity check after generating the original set of communication intent anchor points. If any of the key fields—course name, course content scope, trial access rights, registration requirements, discount rules, prohibited promises, or official publishing entity—is empty, the system marks the activity as having incomplete anchor points and stops proceeding to the relay-style digital human short video generation step. If all key fields exist, but the spoken text or subtitles in the original digital human short video are inconsistent with the structured fields, the system generates the original communication intent anchor points based on the structured fields and records the inconsistencies in the anchor point difference record for subsequent manual review or regeneration of the original digital human short video. After completing this step, the system outputs the original set of communication intent anchor points, which serves as the basis for determining the scope of relay rewriting and judging whether the relay script scheme exceeds the limits.

[0048] Step two: The system constructs a fission graph, determines the relay rewriting limit, and accordingly defines the relay rewriting scope. After obtaining the original set of propagation intent anchor points, the system collects propagation behavior data of the original digital human short video after its release. This propagation behavior data includes viewing behavior, sharing behavior, landing page access behavior, audition / receive behavior, registration behavior, and continued sharing behavior. Viewing behavior includes at least the video identifier, user hash identifier or anonymous access identifier, playback start time, playback completion status, and playback duration; sharing behavior includes at least the sharing user, sharing time, sharing entry point, sharing token, and the shared video identifier; landing page access behavior includes at least the access source, landing page dwell time, clicked area, and access user tags; audition / receive behavior and registration behavior are used to determine whether the propagation path has brought about substantial conversion; continued sharing behavior is used to determine whether users who received the video continue to propagate it to the next level.

[0049] The system performs deduplication and correlation processing on the propagation behavior data. For playback, click, or access events repeatedly triggered by the same user hash identifier within a short period of time, the system merges them according to the same video identifier, the same sharing token, and the same access time window. For access records where the user hash identifier cannot be obtained, the system retains the anonymous access identifier and establishes a temporary correlation based on the sharing token, access source, landing page path, and activity identifier. The processed behavior records are written into the fission graph, where the user hash identifier, anonymous access identifier, video identifier, and relay version identifier serve as graph nodes, and sharing relationships, access relationships, trial access relationships, registration relationships, and secondary sharing relationships serve as graph edges. The organization of propagation nodes, propagation levels, and various propagation relationships in the fission graph is as follows: Figure 3 As shown.

[0050] In the fission graph, the publishing node corresponding to the original digital human short video serves as the root node. Access nodes generated by direct sharing or distribution from the root node are recorded as the first propagation level. Access nodes generated by further sharing from users at the first propagation level are recorded as the second propagation level, and so on. The system records the parent node, child node, propagation level, sharing token, trigger time, and corresponding video identifier in each graph edge. Through this graph structure, the system determines which propagation path the current propagation node originates from, which propagation level it is at, and whether the propagation node has subsequently generated second-hop access, preview access, registration, or further sharing.

[0051] The system identifies the propagation circle and user roles corresponding to the current propagation node based on the fission map. The system first reads the sharing entry point, access user tags, comment content, landing page click area, dwell time, subsequent sharing targets, trial listening behavior, registration behavior, and continued sharing behavior of the current propagation node, and uses the same statistical window under the same activity identifier as the judgment period.

[0052] When identifying dissemination circles, we match the user tags, comment content, landing page click areas, and subsequent sharing targets of the current dissemination node with the predefined circle features. The preset circle features include tags, comment keywords, key click areas, and sharing target types corresponding to the job-seeking / career-change circle, corporate training circle, administrative office circle, and university student employment circle. We obtain tag matching values, comment matching values, click area matching values, and sharing target matching values, and finally determine the circle matching value based on these matching values.

[0053] Specifically, the system calculates a matching value for each candidate propagation layer. For any candidate propagation layer, the system reads the corresponding tags, comment keywords, key click areas, and sharing object types, and compares them with the current propagation node's user tags, comment content, landing page click areas, and subsequent sharing objects. The tag matching value is determined based on the ratio between the weight of the user tags hit by the current propagation node and the total weight of the tags corresponding to the candidate propagation layer; the comment matching value is determined based on the number of times the comment content hits the comment keywords corresponding to the candidate propagation layer and the weight of the hit keywords; the click area matching value is determined based on the ratio between the number of clicks generated by the current propagation node in the key click area corresponding to the candidate propagation layer and the total number of valid clicks; the sharing object matching value is determined based on the ratio between the number of subsequent sharing objects falling into the sharing object type corresponding to the candidate propagation layer and the total number of subsequent sharing objects.

[0054] The system weights the tag matching value, comment matching value, click area matching value, and share object matching value to obtain the layer matching value of the candidate propagation layer. All matching values ​​are limited to between 0 and 1. When the current propagation node lacks one type of data, the system does not record that item as an anomaly, but instead reassigns its weight to the remaining existing matching items for normalization. The system determines the candidate propagation layer with the highest layer matching value as the current propagation layer. If the highest layer matching value is lower than a preset matching threshold, or if the difference between the highest and second-highest layer matching values ​​is lower than a preset difference threshold, the current propagation node is marked as a layer pending confirmation, and the relay rewriting range is not expanded due to changes in the layer when determining the relay rewriting limit.

[0055] The preset matching threshold can be set to 0.50, and the preset difference threshold can be set to 0.10. When the current propagation node is marked as being in a state of pending confirmation at the next concentric circle, the system will not expand the relay rewriting range due to changes in the concentric circle.

[0056] When identifying user referral roles, the system tracks the current propagation node's own conversion status, number of effective second-hop visits, number of continued shares, number of trial listening / received items, and number of registrations. The own conversion status is recorded when the current propagation node engages in trial listening / received items or registration. Effective second-hop visits refer to visits triggered by the current propagation node's sharing behavior, where the landing page dwell time reaches the preset effective dwell time, or where any of the following behaviors occurs: click, trial listening / received items, registration, or continued sharing. The preset effective dwell time can be set to 5 to 10 seconds. If the own conversion status is established, and the number of effective second-hop visits is lower than the first second-hop threshold, and the number of continued shares is lower than the first sharing threshold, then the current propagation node is identified as a conversion-oriented user. If the number of effective second-hop visits reaches the first second-hop threshold, or the number of continued shares reaches the first sharing threshold, or the number of trial listening / received items triggered by the current propagation node reaches one or more times, then the current propagation node is identified as a propagation-oriented user. If the current propagation circle differs from the parent propagation circle, and the circle deviation reaches the preset bridging threshold, then the current propagation node is identified as a bridging-oriented user. The thresholds for the first and second hops can be set to 2 to 5 times, the first sharing threshold can be set to 1 to 3 times, and the preset bridging threshold can be set to 0.45 to 0.60. When the same propagation node meets multiple role conditions simultaneously, it is prioritized as a bridging user; if it does not meet the bridging user condition but meets the propagation user condition, it is identified as a propagation user; the rest, if they meet the conditions for their own conversion status to be established, are identified as conversion users.

[0057] After determining the current dissemination node, user roles in viral marketing, and the dissemination circle, the system determines the relay rewriting limit. This limit indicates whether a relay-style digital human short video needs to be generated at the current dissemination node and the extent to which the original script can be rewritten. The system determines the scope of relay rewriting based on this limit. When the limit is low, it means the original digital human short video is still suitable for continued dissemination at the current node. In this case, only minor rewriting of the salutation, opening hook, and sharing prompts is allowed; changes to course content, trial access rights, registration conditions, and target audience are not permitted. When the limit is in the middle range, the system allows adaptation and rewriting of pain point expressions, case scenarios, and digital human tone, while preserving the original dissemination intent anchor. When the relay rewriting limit value is high, it indicates that the current propagation circle, propagation node, or propagation level has significantly differed from the original release environment. This allows for substantial adjustments to the expression order, scenario introduction method, and viral guidance reasons of the relay script solution, but the factual anchor, benefit anchor, applicable target anchor, prohibited expansion anchor, and viral guidance anchor obtained in Step 1 must not be altered. After completing this step, the system outputs the user's viral role, propagation level, propagation circle, relay rewriting limit value, and relay rewriting scope.

[0058] Step 3: The system generates a relay script scheme based on the user's fission role, propagation layer, and relay rewriting range, and performs verification based on the relay content boundary value. After obtaining the user's fission role, propagation level, propagation layer, relay rewriting limit value, and relay rewriting range of the current propagation node in Step 2, the system uses the relay rewriting range as the generation boundary of the relay script scheme and calls the script generation module to generate the relay script scheme.

[0059] Before generating the relay script, the script generation module first creates a list of allowed and prohibited fields based on the relay rewriting scope. The list of allowed fields includes permitted salutations, opening hooks, pain point expressions, case scenarios, digital human tone, and reasons for viral marketing within the current relay rewriting scope. The list of prohibited fields includes the course name, course content scope, trial access benefits, registration requirements, discount rules, official publishing entity, and the content corresponding to prohibited anchor points. When generating the relay voiceover script and subtitles, the script generation module only performs synonym replacements, order adjustments, or contextualized expressions on the content in the allowed fields list, without adding any course services, trial access benefits, registration requirements, or promised results not recorded in the original activity data.

[0060] When generating relay script solutions, the script generation module uses the list of fields that can be modified as the scope of content that can be generated, and the list of fields that cannot be modified as the scope of content that cannot be freely generated. For the course name, course content scope, trial listening rights, registration conditions, discount rules, official publishing entity, and expressions corresponding to anchors that cannot be expanded in the list of fields that cannot be modified, the script generation module can only reference the original field expressions or identifiable expressions in the original set of propagation intent anchors, and cannot expand them on its own based on propagation circles, user fission roles, or comment feedback. For the salutation methods, opening hooks, pain point expressions, case scenarios, digital human tone, and fission guidance reasons in the list of fields that can be modified, the script generation module can only modify them within the levels allowed by the current relay modification scope.

[0061] The relay script solution adopts a segmented structure for output. Each relay segment includes a segment identifier, segment content, segment source marker, rewriting marker, the original propagation intent anchor referenced, whether it belongs to a field that can be rewritten or not, and an initial risk marker. The segment source marker records that the relay segment originates from the original spoken text, subtitle content, course card, landing page field, or original propagation intent anchor. The rewriting marker records whether the relay segment is unrewritten, synonymous rewritten, scenario rewritten, ordered, or rewritten for viral marketing purposes. After generating the relay script solution, the system calculates the fact retention amount, rights retention amount, and rewriting overreach amount sentence by sentence according to the above segmented structure.

[0062] The relay script solution is not a complete rewrite of the original digital human short video script. Instead, within the relay rewriting scope determined in step two, it makes partial adjustments to the original script's salutation, opening hooks, pain point expressions, case scenarios, digital human tone, and reasons for viral marketing. During the generation process, the system retains the original set of communication intent anchors obtained in step one and uses factual anchors, rights anchors, applicable target anchors, prohibited expansion anchors, and viral marketing guidance anchors as constraints of the relay script solution.

[0063] The relay script should include at least the relay spoken text, relay subtitles, digital human tone markers, scene expression content, and viral marketing prompts. The system determines the focus of the relay script based on the user's role in viral marketing. If the current dissemination node is a dissemination-oriented user, the system will emphasize why the course is worth sharing with similar users; if the current dissemination node is a conversion-oriented user, the system will emphasize why the current user is suitable for a trial or enrollment; if the current dissemination node is a bridging user, the system will emphasize which dissemination circle the video is suitable to be shared with. For example, when the original digital human short video targets job seekers transitioning to new careers, the original script can focus on learning AI office skills from scratch and enhancing resume competitiveness. When step two identifies the current target audience as corporate training, the system can, within the scope of rewriting, adjust the pain point expression to "team members frequently handle spreadsheets, notifications, and reports, and can offer a trial lesson on AI office automation." When the target audience is university students seeking employment, the system can adjust the expression to suit students who want to include AI office skills in their resumes. These adjustments only change the expression scenario and perspective; they do not change the course name, course content scope, trial lesson rights, registration requirements, or prohibited promises.

[0064] After the relay script scheme is generated, the system performs a relay content boundary check on it. The relay content boundary value is used to determine whether the relay script scheme retains the original propagation intent anchor point and whether it exceeds the relay rewriting range. Based on the relay content boundary value, the system determines whether the relay script scheme can proceed to the relay-style digital human short video generation step. If the relay content boundary value meets the preset boundary conditions, the system allows the relay script scheme to proceed to the next generation step; if the relay content boundary value does not meet the preset boundary conditions, the system refuses to use the relay script scheme and triggers regeneration or reverts to a stable script template.

[0065] Step four: The system generates a relay-style digital human short video, binds it to the relay dissemination record, and updates the relay rewriting strategy based on dissemination feedback. After the relay script scheme that passes the relay content boundary check is output in Step three, the system uses the relay script scheme as input for video generation to generate a relay-style digital human short video. The relay script scheme includes the relay voice-over script, relay subtitle content, digital human tone markers, scene expression content, viral guidance, the referenced original dissemination intent anchor points, and the corresponding relay content boundary values. Before generating the video, the system reads the relay rewriting scope and the set of original dissemination intent anchor points corresponding to the relay script scheme again to confirm that the relay script scheme is still in a generateable state; if the trial rights, registration conditions, or preferential rules in the original activity data change during the waiting period for generation, the system pauses generation and returns to Step one to re-solidify the original dissemination intent anchor points.

[0066] The system generates spoken audio based on the relay script and determines the speech rate, pause positions, and emphasis positions based on digital human tone identifiers. After the audio is generated, the system generates a phoneme timeline or equivalent audio time alignment information based on the spoken audio and passes this information to the digital human lip-reading driver module to synchronize the digital human's lip movements, facial expressions, and basic actions with the relay script. The system generates a subtitle timeline based on the relay subtitle content and matches the subtitle timeline with the time alignment information of the spoken audio. If the subtitle appearance time deviates from the corresponding sentence segment of the spoken audio by more than a preset time deviation, such as more than 300 milliseconds, the system automatically adjusts the subtitle timeline; if it still cannot match after adjustment, it returns to the relay script scheme for sentence segmentation.

[0067] The system selects the corresponding scene background, visual overlay, and referral entry point based on the target audience. For example, when the target audience is a corporate training group, the visual background can be an office scene, and the referral entry point can prompt the user to share it with the colleague responsible for team training. When the target audience is a college student job-seeking group, the visual background can be a learning or resume preparation scene, and the referral entry point can prompt the user to share it with classmates who are also preparing for job hunting. The above visual content is only used to demonstrate scene adaptation and does not change the factual anchor, rights anchor, applicable target anchor, prohibition of expansion anchor, and referral guidance anchor that were established in Step 1.

[0068] After generating the spoken audio, digital lip movements, subtitles, and scene visuals, the system synthesizes a relay-style digital human short video. This relay-style digital human short video includes at least the digital spoken audio, relay subtitle content, a course or activity entry point, a referral guide entry point, and an AIGC identifier. The AIGC identifier can be overlaid on a preset position in the video frame or synchronously written into the video file metadata or bypass metadata. After video synthesis, the system generates a video verification value. This verification value can be determined based on the video file hash, script scheme identifier, digital human identifier, and generation time, and is used to trace the generation source of the relay-style digital human short video.

[0069] After generation, the system binds a relay propagation record to the relay-style digital human short video. The relay propagation record includes at least the original video identifier, parent video identifier, current video identifier, relay version identifier, propagation level, propagation node identifier, propagation circle identifier, user referral role, relay rewriting limit value, relay rewriting scope, relay content boundary value, and generation time. After generating the relay propagation record, the system writes the current video identifier back to the propagation graph and establishes a relay generation relationship between the parent video node and the current video node. This relay generation relationship differs from a regular sharing relationship; it indicates that the current video is not a direct forward, but rather a regeneration based on the parent video's propagation, according to the user referral role, propagation circle, and relay rewriting scope.

[0070] After the relay-style digital human short video is released, the system continues to collect dissemination feedback data. This feedback data includes playback completion rate, landing page click-through rate, preview and claim rate, registration rate, second-click entry rate, continued sharing rate, and comment feedback. Based on this feedback data, the system determines whether the relay-style digital human short video is suitable for continued dissemination within the corresponding dissemination circle, and updates the user fission role identification rules, dissemination circle identification rules, and relay rewriting strategy accordingly. The updated relay rewriting strategy is used to determine the subsequent relay rewriting limit value, enabling the system to adjust the relay rewriting scope based on actual dissemination feedback; the relay script scheme is still verified based on the relay content boundary value. Thus, the system forms a closed loop from original digital human short video parsing, fission map construction, relay rewriting limit judgment, relay script generation, relay content boundary verification, relay-style digital human short video generation to dissemination feedback updates.

[0071] Example 2 further explains the method for determining the relay rewriting limit value. The relay rewriting limit value is used to determine whether the original digital human short video needs to generate a relay-style digital human short video during the viral spread process, and to limit the rewriteable scope of the relay script scheme. The relay rewriting limit value is not determined solely based on a single dissemination indicator such as play count or click-through rate, but rather comprehensively determined by considering the deviation of the current dissemination circle from the original target circle, the driving force of the current dissemination node on subsequent dissemination, and the attenuation of the original digital human short video in the current dissemination environment.

[0072] The system obtains the current propagation node and its parent node from the fission graph, and reads the corresponding user tags, sharing entry points, comment content, landing page click areas, dwell time, number of second-hop visits, number of continued sharing, number of free trials, as well as the playback completion rate, landing page click-through rate, and continued sharing rate at the current propagation level. All of the above data comes from the fission graph and propagation behavior data, without introducing new data definitions.

[0073] The circle deviation C represents the degree of deviation of the current propagation circle from the original target circle. The system generates original circle features based on the original target circle in the original activity data, and generates current circle features based on the user tags, comment content, landing page click areas, and subsequent sharing objects under the current propagation node. The system matches the current circle features with the original circle features to obtain the circle matching value. and in accordance with Determine the concentric circle deviation C. Concentric circle matching value. The matching value is obtained by weighting tag matching value, comment matching value, click area matching value, and shared object matching value, with each matching value limited to between 0 and 1. When the current circle characteristics are still concentrated on job seeking / career change, beginner learning, resume improvement, etc., the circle matching value is... The circle matching value is relatively high, while the circle deviation C is relatively low; when the current circle characteristics shift to corporate training, team training, administrative office work, group invoices, or learning progress management, the circle matching value is relatively high. The deviation C increases as the propagation node is reduced. If the current propagation node is marked as being in a state of pending confirmation, the deviation C is set to no more than 0.40 to avoid excessive amplification of the relay rewriting scope due to a small number of access tags or short-term comments.

[0074] The node-driven quantity B represents the current propagation node's ability to drive subsequent propagation. The system reads the number of effective second-hop visits, continued sharing, trial access, and registrations generated after the current propagation node shares, and forms two-hop visit components, continued sharing components, trial access components, and registration components, respectively. Each component is determined according to the ratio of its corresponding quantity to a preset normalization upper limit, and the maximum value does not exceed 1. The normalization upper limit for the number of effective second-hop visits can be set to 5 to 20 times, the normalization upper limit for the number of continued sharing can be set to 2 to 10 times, the normalization upper limit for the number of trial accesses can be set to 1 to 5 times, and the normalization upper limit for the number of registrations can be set to 1 to 3 times; when historical propagation data already exists for the same activity, the upper quartile value of the corresponding indicator within the historical statistics window of the same activity can also be used as the normalization upper limit. The system performs weighted processing on the above components to obtain the node-driven quantity B. Through this processing, the node-driven quantity B can reflect the different roles of propagating users, converting users, and bridging users in the fission link, rather than solely relying on whether the current propagation node itself has registered.

[0075] The original video decay factor D represents the degree of decline in the dissemination effect of the original digital human short video within the current dissemination level or dissemination circle. The system uses the completion rate, landing page click-through rate, and continued sharing rate from the previous dissemination level or original target circle as benchmark values, and the corresponding indicators in the current dissemination level or dissemination circle as the current value. If the current value is higher than or equal to the corresponding benchmark value, the corresponding decline component is recorded as 0; if the current value is lower than the corresponding benchmark value, the corresponding decline component is determined according to the decline ratio between the benchmark value and the current value. The system determines the original video decay factor D based on the decline components of the completion rate, landing page click-through rate, and continued sharing rate, and limits the original video decay factor D to between 0 and 1. If the number of valid plays in the current statistical window is lower than the preset sample size, this statistical window is not used to increase the original video decay factor D, but only the corresponding dissemination data is retained as observation data; the preset sample size can be set to 100 to 300 valid plays.

[0076] The system determines the relay rewriting limit value M based on the layer deviation C, the node-driven amount B, and the original manuscript attenuation D. The correspondence between the relay rewriting limit value and different relay rewriting ranges is as follows: Figure 4As shown. Specifically, the following formula can be used: ;

[0077] Where M represents the relay rewriting limit; C represents the circle deviation; B represents the node driving force; D represents the original text attenuation; a, b, and c represent the weights corresponding to the circle deviation, node driving force, and original text attenuation, respectively, and a, b, and c are all greater than 0, a + b + c = 1. In one implementation, a is 0.35, b is 0.35, and c is 0.30. For activities that emphasize cross-circle propagation, the value of a can be increased; for activities that emphasize the driving force of fission nodes, the value of b can be increased; for activities that emphasize the recognition of original video effect attenuation, the value of c can be increased. C, B, and D are all normalized to between 0 and 1. The larger the value, the more the current propagation node needs to generate an adapted relay-style digital human short video.

[0078] The system determines the scope of relay rewriting based on the relay rewriting limit value M. When M is less than the first limit threshold, the system only allows the relay script to rewrite the salutation, opening hook, and sharing prompts, but not the course content, trial access rights, registration conditions, or target audience. When M is greater than or equal to the first limit threshold and less than the second limit threshold, the system allows the relay script to rewrite the pain point expression, case scenario, and digital human tone without changing the original communication intention anchor. When M is greater than or equal to the second limit threshold, the system allows the relay script to adjust the order of expression, the way the scenario is introduced, and the reason for viral marketing, but it still cannot change the factual anchor, benefit anchor, target audience anchor, prohibited expansion anchor, or viral marketing prompt anchor. The first limit threshold can be set to 0.35, and the second limit threshold can be set to 0.65, or it can be adjusted based on historical activity data and manual review results.

[0079] For example, if the original digital human short video targets job seekers and those transitioning to new careers, and its current dissemination still originates from job-seeking groups, with no significant decrease in completion rate and continued sharing rate, then the rewriting limit is relatively low. The system only allows rewriting "Students can listen first" to "Friends who want to change careers can listen first." If the video enters a company's administrative group through bridging users, and these users focus on group registration instructions and invoice instructions, and the original video's completion rate decreases within that group, then the group deviation and original text decay increase. The system can allow rewriting the pain point expression as "When the team frequently processes spreadsheets, notifications, and reports, they can watch this trial lesson first," but it still cannot rewrite the trial lesson rights as free access to the entire course, nor can it expand the improvement of office efficiency into guaranteed employment or guaranteed salary increases.

[0080] In this embodiment, the system first determines the relay rewriting limit value using the layer deviation, node-driven amount, and original text attenuation amount, and then determines the relay rewriting range based on the relay rewriting limit value. This relay rewriting range serves as the basis for generating the relay script scheme and judging the boundary value of the relay content. If there is no relay rewriting range output by the relay rewriting limit value, it is impossible to determine whether the relay script scheme has exceeded the rewriting boundary.

[0081] Example 3 specifically explains the method for determining the relay content boundary value and the feedback update method after the relay-style digital human short video is generated. The relay content boundary value is used to determine whether the relay script scheme still maintains the original activity content boundaries after circle adaptation and partial rewriting. The relay content boundary value depends on the relay rewriting range, where the amount of rewriting exceeding the boundary must be determined based on the relay rewriting range.

[0082] After the relay script is generated, the system reads the relay spoken text, relay subtitles, scene expression content, and viral guidance from the script, and also reads the original set of dissemination intent anchor points and the scope of relay rewriting. The system determines the relay content boundary values ​​based on the amount of factual retention, the amount of rights retention, and the amount of rewriting exceeding the boundaries. The closed-loop relationship between relay content boundary verification, relay-style digital human short video generation, and dissemination feedback updates is as follows: Figure 5 As shown. The fact retention amount is used to indicate the degree to which the relay script scheme retains the course name, course content scope, service scope, and registration conditions; the rights retention amount is used to indicate the degree to which the relay script scheme retains the trial listening rights, discount rules, group registration discounts, and redemption conditions; the rewrite out-of-bounds amount is used to indicate whether the relay script scheme exceeds the relay rewrite scope.

[0083] In one specific implementation, the relay content boundary value N is determined according to the following formula:

[0084] ;

[0085] Where N represents the relay content boundary protection value; F represents the fact retention amount; E represents the equity retention amount; V represents the rewriting boundary violation amount; x, y, and z represent the weights corresponding to the fact retention amount, equity retention amount, and rewriting boundary violation amount, respectively, and x, y, and z are all greater than 0, x + y + z = 1. For ease of implementation, x can be 0.40, y can be 0.35, and z can be 0.25; for activities with higher requirements for equity expression, the value of y can be increased; for activities with higher requirements for rewriting boundaries, the value of z can be increased. F, E, and V are all normalized to between 0 and 1 before being included in the calculation. The higher the fact retention amount and equity retention amount, the higher the relay content boundary protection value; the higher the rewriting boundary violation amount, the lower the relay content boundary protection value.

[0086] The fact retention quantity F is determined through field comparison and semantic matching. The system counts the number of fact anchors that are effectively hit and the total number of fact anchors that should be retained in the relay script scheme to obtain the fact anchor hit ratio. The fact anchors include the anchors fixed in step one in the course name, course content scope, course service scope, and registration conditions. Expressions that are completely consistent with fact anchors, as well as identifiable expressions recorded in step one, are considered to be effective hits. The system also determines the fact anomaly deduction quantity based on the hit situation of newly added course content, newly added course service scope, changes in registration conditions, course name replacement, and prohibited expansion anchors. If the relay script scheme only uses the original field expressions or identifiable expressions recorded in the original propagation intent anchor set, and does not add course facts not recorded in the original activity data, the fact anomaly deduction quantity is 0; if course name replacement, changes in registration conditions, or prohibited expansion anchor hits occur, it is determined that the fact retention quantity F does not meet the preset boundary conditions, and the relay script scheme is rejected.

[0087] The retained benefit amount E is determined based on the benefit anchor hit rate and the benefit abnormal deduction amount. The system counts the number of effectively hit benefit anchors and the total number of benefit anchors that should be retained in the relay script scheme to obtain the benefit anchor hit rate; the benefit anchors include the anchors fixed in step one among trial benefit, discount rules, group discount, redemption conditions, and benefit validity period. Redemption of trial class, listening to a trial class first, and other identifiable expressions recorded in step one are all considered to have hit the same trial benefit anchor. The system also determines the benefit abnormal deduction amount based on the expansion of trial benefit, change of discount rules, expansion of group discount, relaxation of redemption conditions, and extension of benefit validity period. If the relay script scheme only describes trial benefit, discount rules, group discount, redemption conditions, and benefit validity period within the identifiable expressions recorded in the original communication intent anchor set, the benefit abnormal deduction amount is 0; if the substantive content of trial benefit, discount rules, or group discount changes, the benefit retained amount E is determined to not meet the preset boundary conditions.

[0088] The rewrite boundary violation quantity V is determined based on the matching relationship between the relay script scheme and the relay rewrite scope. In one specific implementation, the rewrite boundary violation quantity V is determined based on the proportion of out-of-bounds segments, the boundary violation status of key fields, and the hit status of prohibited expansion anchors. The proportion of out-of-bounds segments refers to the ratio between the number of segments judged to be out-of-bounds in the relay script scheme and the total number of segments in the relay script scheme; the boundary violation status of key fields is used to indicate whether the course name, course content scope, trial listening rights, registration conditions, preferential rules, applicable objects, or official publishing entities have been substantially rewritten, and is 0 when no substantial rewriting has occurred, and is 1 when any substantial rewriting has occurred; the hit status of prohibited expansion anchors is used to indicate whether the relay script scheme hits prohibited expansion anchors or their semantically similar expressions, and is 0 when not hit, and is 1 when hit.

[0089] Specifically, the rewritten out-of-bounds amount V can be determined according to the following formula:

[0090] V = r1V1 + r2V2 + r3V3;

[0091] Where V represents the amount of rewriting out of bounds, V1 represents the proportion of out-of-bounds segments, V2 represents the out-of-bounds status of key fields, V3 represents the state where anchor points are prohibited from being expanded, and r1, r2, and r3 represent the weights corresponding to the proportion of out-of-bounds segments, the out-of-bounds status of key fields, and the state where anchor points are prohibited from being expanded, respectively, and r1, r2, and r3 are all greater than 0, and r1 + r2 + r3 = 1. In one implementation, r1 is 0.40, r2 is 0.35, and r3 is 0.25.

[0092] If a relay script solution encounters a mandatory interception event, the mandatory interception event takes precedence over the numerical judgment of the relay content boundary value. The system determines whether a relay script solution can enter the relay-style digital human short video generation process based on the relay content boundary value N. When N reaches or exceeds the preset boundary threshold, and the relay script solution does not have any mandatory interception events, the system determines that the relay script solution meets the preset boundary conditions and allows it to enter the video generation process. The preset boundary threshold can be set to 0.75; for activities with high requirements for committed content, such as education and training, financial services, and healthcare, the preset boundary threshold can be increased to 0.80 to 0.90. The mandatory interception events include course name replacement, expansion of trial listening rights, change of registration conditions, expansion of applicable objects, impersonation of official publishing entities, and hitting of prohibited expanded anchor points. If N reaches the preset boundary threshold, but any mandatory interception event exists, the system still refuses to use the relay script solution and writes the corresponding segment into the out-of-bounds segment record.

[0093] If N is lower than the preset boundary threshold, but the factual retention amount F and the equity retention amount E are not abnormal, and the rewriting out-of-bounds amount V is high only because the way of addressing, the opening hook, the expression of pain points, or the case scenario exceeds the current relay rewriting range, the system allows the script generation module to perform a targeted regeneration. During targeted regeneration, the system will pass the out-of-bounds segment record, the list of allowed rewrite fields, and the list of non-rewrite fields to the script generation module.

[0094] Targeted regeneration only applies to sentences within the out-of-bounds segment records that fall under the categories of addressing style, opening hooks, pain point expressions, or case scenarios. It does not allow free rewriting of sentences corresponding to course names, course content scope, trial access rights, registration requirements, discount rules, applicable audiences, official publishing entities, or prohibited expansion anchors. During targeted regeneration, the system retains sentences that have not exceeded the boundaries and uses the list of allowed rewrite fields, the current relay rewrite scope, the original dissemination intent anchor, and the list of non-rewrite fields corresponding to the out-of-bounds sentences as regeneration constraints. After targeted regeneration, the system recalculates the fact retention amount, rights retention amount, rewrite out-of-bounds amount, and relay content boundary protection value. If the regenerated relay script scheme still does not meet the preset boundary protection conditions, the system will not continue automatic generation but will revert to the stable script template.

[0095] If the factual retention amount F or the equity retention amount E does not meet the preset boundary conditions, the system will not perform automatic expansion correction, but will revert to the stable script template. The stable script template is based on the original digital human short video script, subtitle content, course cards and fission guidance anchors. It can only change the way of addressing and the sharing entry prompt, but will not change the factual anchors, equity anchors, applicable target anchors, prohibited expansion anchors and fission guidance anchors.

[0096] After passing the relay content boundary verification, the relay script scheme generates a relay-style digital human short video and binds a relay propagation record to this video. The relay propagation record includes at least the original video identifier, parent video identifier, current video identifier, relay version identifier, propagation level, propagation node identifier, propagation circle identifier, user fission role, relay rewriting limit value, relay rewriting scope, relay content boundary value, and generation time. The system writes the current video identifier back to the fission graph and establishes a relay generation relationship between the parent video node and the current video node. This relay generation relationship indicates that the current video is not a direct forwarding but is regenerated based on the parent video propagation, according to the user fission role, propagation circle, and relay rewriting scope.

[0097] After the relay-style digital human short video is released, the system collects data on playback completion rate, landing page click-through rate, preview and claim rate, registration rate, second-click entry rate, continued sharing rate, and comment feedback. Based on the above dissemination feedback data, the relay dissemination feedback value Q is determined. Specifically, it can be expressed using the following formula:

[0098] ;

[0099] In this system, Q represents the relay propagation feedback value; R represents the playback completion rate, reflecting the retention rate of the relay-style digital human short video; L represents the landing page click-through rate, reflecting the number of viewers who enter the course or activity page; S represents the trial listening and registration rate, reflecting the actual conversion rate; T represents the second-click entry rate or continued sharing rate, reflecting the subsequent viral propagation; m, n, o, and p represent the weights corresponding to the playback completion rate, landing page click-through rate, trial listening and registration rate, and second-click entry rate or continued sharing rate, respectively, and m, n, o, and p are all greater than 0, and m + n + o + p = 1. R, L, S, and T all use data within the same statistical window and are normalized to between 0 and 1. The statistical window can be set to 30 minutes, 2 hours, and 24 hours after the video is published; for videos whose exposure does not reach the preset sample size, the system does not directly adjust the relay rewriting strategy based on short-term data, but instead keeps them in an observation state.

[0100] The system compares the relay propagation feedback value Q with the baseline feedback value within the same propagation sphere. The baseline feedback value is determined by the historical feedback data of the original digital human short video within that propagation sphere, or by the average feedback data of stable relay versions within the same activity. Each propagation sphere, user referral role, and scene expression content corresponds to a relay rewriting strategy weight W. If the relay propagation feedback value Q is higher than the baseline feedback value, and both the second-click entry rate and the trial listening rate meet preset requirements, the system increases the corresponding relay rewriting strategy weight W by a preset upward adjustment step size. If the relay propagation feedback value Q is lower than the baseline feedback value, or if abnormal comments such as irrelevant content, misunderstanding of rights, or not being the required course appear in the comments, and the proportion of abnormal comments reaches a preset abnormal proportion, the system decreases the corresponding relay rewriting strategy weight W by a preset downward adjustment step size. The preset upward adjustment step size can be set to 0.03 to 0.08, the preset downward adjustment step size can be set to 0.03 to 0.10, and the updated relay rewriting strategy weight W is limited to between 0 and 1.

[0101] The proportion of abnormal comments is determined by the ratio of the number of abnormal comments to the number of valid comments in the current statistics window. When the number of valid comments is lower than the preset comment sample size, the comment feedback is only written into the correction sample of the propagation circle identification rule and does not trigger a separate reduction in the relay rewriting strategy weight W. The preset comment sample size can be set to 10 to 30 comments. When updating the relay rewriting strategy, the system does not directly change the original propagation intent anchor point, but adjusts the mapping relationship between the user fission role identification rule, the propagation circle identification rule, and the relay rewriting scope. For example, if a bridging user brings the original video into the corporate training circle, and the relay-style digital human short video using the team report processing scenario obtains a high relay propagation feedback value, the system increases the relay rewriting strategy weight W corresponding to the corporate training circle, bridging user, and team report processing scenario; if the feedback of the relay-style digital human short video using the personal resume improvement scenario in the same circle is low, the system reduces the relay rewriting strategy weight W corresponding to that scenario. The above update results are used to influence the determination of the limit value of the next round of relay rewriting and the selection of the relay rewriting scope, but must not cause subsequent relay script schemes to exceed the fact anchor point, the rights anchor point, the applicable object anchor point, the prohibited expansion anchor point, and the fission guidance anchor point.

[0102] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0103] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0104] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0105] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0106] The above description is merely a specific embodiment of this application, but the scope of protection of this application 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 this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for high-spread short video viral marketing driven by AIGC digital humans, characterized in that, Includes the following steps: Obtain raw activity data and raw digital human short videos under the same activity identifier, and extract and solidify the set of original communication intent anchor points; A fission map is constructed based on the propagation behavior data corresponding to the original digital human short video, the user fission role and propagation circle corresponding to the current propagation node are identified, and the relay rewriting limit value and the corresponding relay rewriting range are determined based on the circle deviation, node driving force and original text decay. Within the scope of the relay rewriting, a relay script scheme is generated based on the user fission role and the propagation circle, and a list of fields that can be rewritten and a list of fields that cannot be rewritten are formed based on the scope of the relay rewriting. The relay content boundary value is determined based on the original propagation intent anchor point set, the list of fields that can be rewritten, the list of fields that cannot be rewritten, and the relay rewriting range. The relay content boundary value is determined based on the amount of fact retention, the amount of rights retention, and the amount of rewriting exceeding the boundary. When the boundary value of the relay content meets the preset boundary conditions and there are no mandatory interception events, a relay-style digital human short video is generated based on the relay script scheme, the relay propagation record is bound, and the current video identifier is written back to the fission graph.

2. The AIGC digital human-driven high-propagation short video fission method according to claim 1, characterized in that: The original activity data comes from the activity management terminal, course management terminal, and landing page configuration terminal, and includes at least the course name, target audience, course content scope, trial listening rights, registration requirements, discount rules, prohibited promises, and official publishing entity; the original digital human short video includes at least the narration script, subtitle content, course card, landing page fields, activity rules, video identifier, release time, publishing account, and original target audience; the original set of communication intent anchor points includes at least fact anchor points, rights anchor points, target audience anchor points, prohibited expansion anchor points, and viral marketing anchor points.

3. The AIGC digital human-driven high-propagation short video fission method according to claim 2, characterized in that: When extracting and solidifying the original set of communication intent anchor points, the original activity data is matched with the content parsing results of the original digital human short video, and the anchor content is determined according to the priority of the field source. When the same content has different expressions in different sources, the content in the structured fields is used as the anchor content first, and similar expressions in other sources are used as the identifiable expressions of the anchor. When any of the key fields such as course name, course content scope, trial listening rights, registration conditions, discount rules, prohibited commitments, and official release subject are empty, the corresponding activity is marked as an incomplete anchor state, and the relay-style digital human short video generation step is stopped.

4. The AIGC digital human-driven high-propagation short video fission method according to claim 1, characterized in that: When constructing the fission graph, the propagation behavior data is deduplicated and associated. User hash identifiers, anonymous access identifiers, video identifiers, and relay version identifiers are used as graph nodes. Sharing relationships, access relationships, trial listening and receiving relationships, registration relationships, and secondary sharing relationships are used as graph edges. Parent nodes, child nodes, propagation levels, sharing tokens, trigger times, and corresponding video identifiers are recorded in the graph edges.

5. The AIGC digital human-driven high-propagation short video fission method according to claim 4, characterized in that: When identifying the propagation circle and user fission role corresponding to the current propagation node, the propagation circle is determined based on the access user tags, comment content, landing page click area, and subsequent sharing objects. When the highest circle matching value is lower than the preset matching threshold, or the difference between the highest circle matching value and the second highest circle matching value is lower than the preset difference threshold, the current propagation node is marked as a circle pending confirmation state. The user fission role is determined based on its own conversion status, number of effective second-hop visits, number of continued sharing, number of trial listening and receiving, and number of registrations. The user fission role includes conversion users, propagation users, and bridging users.

6. The AIGC digital human-driven high-propagation short video fission method according to claim 5, characterized in that: The determination of the relay rewriting limit includes: determining the circle deviation, node-driven volume, and original text decay, and weighting these values; the circle deviation is determined based on the matching between the current circle characteristics and the original circle characteristics; the node-driven volume is determined based on the number of effective second-hop visits, the number of continued sharing, the number of trial listenings, and the number of registrations; the original text decay is determined based on the decrease in playback completion rate, landing page click-through rate, and continued sharing rate relative to the baseline value in the current dissemination level or current dissemination circle; when the relay rewriting limit is lower than the first limit threshold, only the salutation, opening hook, and sharing prompt are allowed to be rewritten; when the relay rewriting limit is greater than or equal to the first limit threshold and less than the second limit threshold, rewriting pain point expressions, case scenarios, and digital human tone is allowed; when the relay rewriting limit is greater than or equal to the second limit threshold, adjusting the expression order, scenario introduction method, and viral guidance reason is allowed.

7. The AIGC digital human-driven high-propagation short video fission method according to claim 6, characterized in that: The list of fields that can be rewritten includes the following: the way of addressing the recipient, the opening hook, the expression of pain points, the case scenario, the tone of the digital human, and the reason for the viral marketing within the current relay rewriting scope. The list of fields that cannot be rewritten includes the course name, the scope of course content, the trial listening rights, the registration conditions, the discount rules, the official publishing entity, and the expression content corresponding to the anchor points that are prohibited from being expanded. When generating the relay script, only the content in the list of fields that can be rewritten is replaced with synonyms, the order is adjusted, or the expression is contextualized. Sentences in the relay voice-over text and relay subtitles are marked with sentence source tags and rewriting tags.

8. The AIGC digital human-driven high-propagation short video fission method according to claim 7, characterized in that: The fact retention amount is determined based on the hit rate of fact anchors corresponding to course name, course content scope, course service scope, and registration conditions in the relay script scheme; the rights retention amount is determined based on the hit rate of rights anchors corresponding to trial listening rights, discount rules, group registration discounts, redemption conditions, and rights validity period in the relay script scheme; the rewriting boundary violation amount is determined based on the comparison results between the segments in the relay script scheme and the list of allowed rewrite fields, the list of unrewriteable fields, and the relay rewriting scope; when the relay content boundary value reaches or exceeds the preset boundary threshold, and there are no mandatory interception items, it is determined that the relay content boundary value meets the preset boundary conditions.

9. The AIGC digital human-driven high-propagation short video fission method according to claim 8, characterized in that: The mandatory interception items include course name replacement, expansion of trial listening rights, changes in registration conditions, expansion of applicable objects, impersonation of official release entities, and prohibited expansion of anchor points; when the relay content boundary value is lower than the preset boundary threshold, but the fact retention and rights retention are not abnormal, and the rewriting over-limit is too high only because the way of addressing, the opening hook, the expression of pain points, or the case scenario exceeds the current relay rewriting scope, a targeted regeneration is performed; when the fact retention or rights retention does not meet the preset boundary conditions, it reverts to the stable script template. The stable script template is based on the original digital human short video's voice-over script, subtitle content, course cards, and fission guidance anchor points, and only allows the replacement of the way of addressing and the sharing entry prompt.

10. A method for high-propagation short video fission driven by AIGC digital human according to claim 8 or 9, characterized in that: When generating relay-style digital human short videos, the system generates spoken audio based on the relay script, determines the speaking speed, pause positions, and emphasis positions based on the digital human's tone identifiers, generates a subtitle timeline based on the relay subtitle content, and selects scene backgrounds, video overlays, and viral entry points based on the dissemination circles. The relay dissemination record includes at least the original video identifier, parent video identifier, current video identifier, relay version identifier, dissemination level, dissemination node identifier, dissemination circle identifier, user viral roles, relay rewriting limit value, relay rewriting scope, relay content boundary value, and generation time. After the relay-style digital human short video is published, the system updates the user viral role identification rules, dissemination circle identification rules, and relay rewriting strategy based on the playback completion rate, landing page click-through rate, trial access rate, registration rate, second-click entry rate, continued sharing rate, and comment feedback.

Images