CN-121981759-A - Cross-modal alignment type intelligent marketing strategy generation method and system based on strategy intention guidance

CN121981759ACN 121981759 ACN121981759 ACN 121981759ACN-121981759-A

Abstract

The invention relates to the technical field of marketing strategies, in particular to a cross-modal alignment intelligent marketing strategy generation method and system based on strategy intention guidance, which are used for acquiring a task bulletin field submitted by a user when a platform creates a marketing task, preprocessing the task bulletin field to generate sentence vectors, and inputting the sentence vectors into a strategy encoder to generate strategy intention vectors; the method comprises the steps of constructing an image representation vector, a video representation vector and a text representation vector, fusing the image representation vector, the video representation vector and the text representation vector into a content representation vector, constructing a fusion matching function, outputting a structure alignment score through the vector matching function, inputting the structure alignment score into a scoring function, outputting a final combination score through the scoring function, sorting and screening based on the final combination score, and outputting a candidate combination set, calculating platform configuration for each pair of combinations in the candidate combination set, calculating a preferred time period and a release frequency coefficient after determining the platform configuration, and generating a structured deployment plan based on the combination, the platform configuration, the preferred time period and the release frequency coefficient.

Inventors

DU DALIANG
WANG HUAHUA
GU SHENGLONG

Assignees

广州云智达创科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260408

Claims (10)

1. The cross-modal alignment type intelligent marketing strategy generation method based on strategy intention guidance is characterized by comprising the following steps of: Acquiring a task bulletin field submitted by a user when a platform creates a marketing task, preprocessing the task bulletin field to generate a sentence vector, inputting the sentence vector into a strategy encoder to generate a strategy intention vector, extracting sub-vectors related to keywords from the word vector, and performing label space projection to generate a tonal supplementary vector; Extracting image representation vectors, video representation vectors and text representation vectors from image, video and text information through a visual coding network, an image frame averaging network and a language modeling network, fusing the image representation vectors, the video representation vectors and the text representation vectors into content representation vectors through a cross-modal dynamic fusion model, wherein the cross-modal dynamic fusion model completes dynamic weight distribution of single-modal features through a weighting function, a vector construction function with strategy offset compensation is introduced to realize structural alignment of modal features and strategy space, and a tonality maintenance regular term is added at the same time, the tonality maintenance regular term is used for ensuring that an image style does not deviate from a tonality supplement vector in a strategy target, and comprises features of an image tonality extraction projection matrix, the image representation vectors, tonality supplement vectors, tonality dimensions, two norms and a regular coefficient; Constructing a fusion matching function based on the strategy intention vector and the content representation vector, outputting a structure alignment score through the vector matching function, inputting the structure alignment score into a scoring function with a key dimension deviation penalty, outputting a final combination score through the scoring function, introducing a content diversity balance item based on the maximum mean value difference, sorting and screening based on the final combination score and the content diversity balance item, and outputting a candidate combination set; after the platform configuration is determined, the maximum active segment of the target audience is positioned through a sliding time window clustering algorithm to serve as a preferred time segment of the combination delivery, and then personalized delivery frequency coefficients are obtained through combination scoring and frequency control function calculation of a cross-vector deviation distance, and a structured deployment plan is generated based on the combination, the platform configuration, the preferred time segment and the delivery frequency coefficients.
2. The method for generating the cross-modal alignment intelligent marketing strategy based on the strategy intention guidance according to claim 1, wherein the preprocessing of the task presentation field generates sentence vectors, and specifically comprises the following steps: Performing character standardization and clear format on the task briefing field, removing common punctuation, abnormal redundant labels and platform template sentences, completing dictionary-based error correction and unified replacement, and generating a standard field text; the standard field text is input into a language coding model based on a transformer structure, and a context vector representation is extracted through the language coding model to generate sentence vectors.
3. The strategic intent guided cross-modality aligned intelligent marketing strategy generation method of claim 1, wherein the visual coding network comprises a convolution layer comprising a convolution kernel, a normalization model, and a maximum pool operation model.
4. The policy intent directed based cross-modality aligned intelligent marketing policy generation method of claim 1, wherein the expression of the vector constructor is: ; Wherein the method comprises the steps of The content represents a vector of contents and, For the content mapping matrix to be a mapping matrix, For the modal fusion vector, In order to be a policy intent vector, The output dimension of the fusion mode; For a mapping matrix of modalities to a policy space, Is a constant term.
5. The policy intent directed cross-modality aligned intelligent marketing policy generation method of claim 1, wherein the fusion matching function comprises a dimension component of a policy intent vector, a dimension component corresponding to a content representation vector, and a weight ratio of task policy factors.
6. The policy intent guidance based cross-modality aligned intelligent marketing policy generation method according to claim 1, wherein the scoring functions comprise penalty factors, dimension level gating flags, policy intent vectors, and content representation vectors.
7. The policy intent guided cross-modality aligned intelligent marketing policy generation method according to claim 1, wherein the candidate combination set comprises a final combination score, a content diversity balance item, a content ID, and a damerson ID.
8. The policy intent guidance based cross-modality aligned intelligent marketing policy generation method according to claim 1, wherein the content diversity balance item comprises a candidate content quantity and a diversity control parameter.
9. The policy intent guidance based cross-modality aligned intelligent marketing strategy generation method according to claim 1, wherein the frequency control function comprises a final combined score, an absolute deviation distance between a policy intent vector and a content representation vector, and an adjustment term coefficient.
10. The cross-modal alignment type intelligent marketing strategy generation system based on strategy intention guidance is characterized by comprising the following components: The strategy intention vector construction module is used for acquiring a task bulletin field submitted by a user when the platform creates a marketing task, preprocessing the task bulletin field to generate a sentence vector, inputting the sentence vector into the strategy encoder to generate a strategy intention vector, extracting subvectors related to keywords from the word vector, and performing label space projection to generate a tonal supplementary vector; The content representation vector construction module is used for extracting image representation vectors, video representation vectors and text representation vectors from image, video and text information through a visual coding network, an image frame averaging network and a language modeling network, fusing the image representation vectors, the video representation vectors and the text representation vectors into content representation vectors through a cross-modal dynamic fusion model, the cross-modal dynamic fusion model completes dynamic weight distribution of single-modal features through a weighting function, a vector construction function with strategy offset compensation is introduced to realize structural alignment of the modal features and a strategy space, and a tonality maintenance regular term is added at the same time, the tonality maintenance regular term is used for ensuring that an image style does not deviate from a tonality complementary vector in a strategy target, and comprises features of an image tonality extraction projection matrix, the image representation vectors, tonality complementary vectors, tonality dimensions, two norms and a regular coefficient; The vector fusion combination module is used for constructing a fusion matching function based on the strategy intention vector and the content representation vector, outputting a structure alignment score through the vector matching function, inputting the structure alignment score into a scoring function with key dimension deviation penalty, outputting a final combination score through the scoring function, introducing a content diversity balance item based on the maximum mean value difference, sorting and screening based on the final combination score and the content diversity balance item, and outputting a candidate combination set; the deployment plan generation module is used for calculating platform configuration for each pair of combinations in the candidate combination set, after the platform configuration is determined, positioning the maximum active segment of the target audience to serve as a preferred time segment of combination delivery through a sliding time window clustering algorithm based on the full-period time sequence active behavior data of the user, calculating through fusion of combination scores and a frequency control function of a cross-vector deviation distance to obtain personalized delivery frequency coefficients, and generating a structured deployment plan based on the combination, the platform configuration, the preferred time segment and the delivery frequency coefficients.

Description

Cross-modal alignment type intelligent marketing strategy generation method and system based on strategy intention guidance Technical Field The invention relates to the technical field of marketing strategies, in particular to a cross-mode alignment type intelligent marketing strategy generation method and system based on strategy intention guidance. Background In the background that new media content marketing is mainstream, promotion of brands on platforms such as tremble, reddish books and the like has become an important growth mode. However, as content ecology expands rapidly, the formulation of marketing strategies is evolving from relatively simple dado selection and material selection into a complex decision task that relies on multi-source information, cross-modal content, and real-time user behavior. Branding Brief often contains multiple elements such as product selling points, expression styles, platform preferences, etc., but its text expression lacks standardization, making marketing intent difficult to systematically parse and quantify. On the content side, the modes of creation of the rater are various, different expression modes such as pictures, short videos, mouth broadcasting and the like are covered, and the same rater can present completely different adjustability in different materials, so that the traditional screening mode depending on labels and historical indexes cannot accurately understand the fit degree of the content style and marketing targets. Meanwhile, the active behavior of the platform user has obvious timeliness and stage, and the user has obvious difference in different conversion stages for the same type of content, but the conventional system is difficult to match and relate the user stage change with the content strategy, so that the delivery time and the content presentation mode are difficult to uniformly manage. The existing SaaS marketing tools are mostly remained on the presentation and regularization screening level of static data, and lack the capability of integrally fusing Brief intention, david content characteristics, user behavior patterns and platform rhythm, so that an automatic closed loop from strategy generation to executable deployment cannot be realized. With the continuous evolution of content forms, increasingly abundant platform mechanisms and accelerated changes of user behavior patterns, the traditional manual operation mode is difficult to support high-quality, systematic and accurate marketing strategy generation, so an intelligent strategy generation method capable of uniformly analyzing Brief intention, constructing multi-mode content characterization, completing task-oriented content matching and generating an executable delivery plan is urgently needed, and the problems that strategies are unquantifiable, content is unjustifiable and rhythm is uncontrollable in the prior art are solved. Disclosure of Invention In order to solve the problems, the invention provides a cross-modal alignment type intelligent marketing strategy generation method and system based on strategy intention guidance. In order to achieve the above purpose, the present invention provides a cross-modal alignment intelligent marketing strategy generation method based on strategy intention guidance, comprising: Acquiring a task bulletin field submitted by a user when a platform creates a marketing task, preprocessing the task bulletin field to generate a sentence vector, inputting the sentence vector into a strategy encoder to generate a strategy intention vector, extracting sub-vectors related to keywords from the word vector, and performing label space projection to generate a tonal supplementary vector; Extracting image representation vectors, video representation vectors and text representation vectors from image, video and text information through a visual coding network, an image frame averaging network and a language modeling network, fusing the image representation vectors, the video representation vectors and the text representation vectors into content representation vectors through a cross-modal dynamic fusion model, wherein the cross-modal dynamic fusion model completes dynamic weight distribution of single-modal features through a weighting function, a vector construction function with strategy offset compensation is introduced to realize structural alignment of modal features and strategy space, and a tonality maintenance regular term is added at the same time, the tonality maintenance regular term is used for ensuring that an image style does not deviate from a tonality supplement vector in a strategy target, and comprises features of an image tonality extraction projection matrix, the image representation vectors, tonality supplement vectors, tonality dimensions, two norms and a regular coefficient; Constructing a fusion matching function based on the strategy intention vector and the content representation vector, outputting a structure alignment score