CN-122022923-A - Advertisement marketing material generation method and device based on time sequence diagram neural network

CN122022923ACN 122022923 ACN122022923 ACN 122022923ACN-122022923-A

Abstract

The application provides a method and a device for generating advertising marketing materials based on a time sequence diagram neural network, and relates to the field of intelligent content generation. The method comprises the steps of obtaining multi-source material information, carrying out differentiated representation on the multi-source material information, forming mutually independent but associable basic feature sets, constructing a time sequence diagram association structure comprising user association, material association and environment association based on the basic feature sets, executing feature propagation processing of a time sequence diagram neural network on the time sequence diagram association structure, applying differentiated information update constraint on different association relations, constructing an advertisement marketing material generation model in an improved anti-facts association modeling mode, introducing information update constraint in a model training process, and outputting advertisement marketing materials to be put corresponding to target users through the advertisement marketing material generation model. The application solves the problem that the advertising marketing materials generated by the existing advertising marketing material generation method have poor effect in cross-scene delivery.

Inventors

CHEN CHENG
BAO ZHIHUI

Assignees

苏州卓尔数科信息科技有限公司

Dates

Publication Date: 20260512
Application Date: 20251225

Claims (10)

1. A method for generating advertising marketing materials based on a neural network of a timing diagram, the method comprising: the method comprises the steps of obtaining multi-source material information, wherein the multi-source material information comprises user behavior information, user attribute information, advertisement material information and external environment related information; distinguishing and representing the multi-source material information, and forming mutually independent but associable basic feature sets; constructing a time sequence diagram association structure comprising user association, material association and environment association based on the basic feature set; Executing characteristic propagation processing of the time sequence diagram neural network on the time sequence diagram association structure, and applying differentiated information updating constraint to different association relations; Constructing an advertisement marketing material generation model in an improved counter-fact association modeling mode, and introducing the information updating constraint in a model training process; and outputting the advertisement marketing materials to be put corresponding to the target user through the advertisement marketing material generation model.
2. The method according to claim 1, wherein said distinguishing the multi-source material information and forming a basic feature set that is independent but associable with each other, specifically comprises: Carrying out time-series analysis on the user behavior information, and extracting user behavior characteristics reflecting the behavior change trend, behavior frequency distribution and behavior persistence of the user in different time windows; performing stability modeling on the user attribute information to form user attribute features representing long-term attribute states of users, and maintaining a associable but unmixed representation relationship with the user behavior features; carrying out content structure analysis on the advertisement material information, forming material characteristics representing the material expression mode, the content composition and the version difference, and enabling the material characteristics to be updated independently of the user behavior change; the external environment related information is subjected to conditional expression, and environment characteristics for describing a throwing period, a flow distribution state, an activity strategy or external disturbance factors are formed; And on the premise of keeping the user behavior characteristics, the user attribute characteristics, the material characteristics and the environment characteristics independently represented, establishing an associatable mapping relation among various characteristics to form the basic characteristic set.
3. The method according to claim 2, wherein the constructing a timing diagram association structure including user association, material association and environment association based on the basic feature set specifically includes: Constructing user nodes based on the user behavior characteristics and the user attribute characteristics, constructing material nodes based on the material characteristics, and constructing environment condition nodes based on the environment characteristics; Establishing a behavior association edge between the user node and the material node, establishing an environment association edge between the user node and the environment condition node, and establishing a condition association edge between the material node and the environment condition node; And giving time sequence weight to each associated side according to the time sequence, and forming the time sequence diagram association structure capable of reflecting the relation between the behavior occurrence sequence and the environment change.
4. The method of claim 3, wherein after said constructing a timing diagram association structure comprising user associations, material associations, and environmental associations based on said set of base features, said method further comprises: And performing differentiated message propagation on each associated edge through a time sequence diagram neural network, and introducing conditional modulation or gating constraint into the environmental condition nodes, so that the time sequence diagram neural network decouples user behavior change and external environment change in the characteristic updating process.
5. A method according to claim 3, wherein the performing feature propagation processing of the timing diagram neural network on the timing diagram association structure and applying differentiated information update constraints to different association relations specifically comprises: Performing feature aggregation processing based on time sequence on the behavior association edges to enable user behavior features and user attribute features associated with the user nodes to be updated in a joint mode on a time sequence dimension so as to add time sequence consistency constraints for describing user behavior continuity and behavior change trend; Performing condition-aware information modulation processing on the environment-related edges, so that the user behavior characteristics and the user attribute characteristics are limited by the conditions of the environment characteristics in the updating process, and environment condition constraints for limiting the influence range of external environment changes are added; Introducing a gating control information updating mechanism to the condition-associated sides, so that the material characteristics adjust the characteristic transmission intensity according to the matching degree of the environmental characteristics in the updating process, and adding a condition suppression constraint for suppressing excessive interference of environmental changes on the material characteristics; and taking the time sequence consistency constraint, the environment condition constraint and the condition suppression constraint as the information updating constraint.
6. The method of claim 3, wherein the constructing an advertising marketing material generation model by the improved anti-facts correlation modeling method specifically comprises: based on the corresponding user behavior characteristics, user attribute characteristics and environment characteristics of the same user node in the same time window, constructing a counterfactual association relationship between the user node and at least one material node, and forming a virtual association edge different from the behavior association edge; Executing feature propagation processing based on the behavior association edge and the virtual association edge, and acquiring a corresponding feature updating result; based on the difference between the different feature updating results, constructing inverse fact difference guiding information for representing the influence degree of the change of the material features on the user behavior; And introducing the inverse fact difference guiding information, and updating model parameters under the action of the information updating constraint.
7. The method of claim 1, wherein outputting, by the advertising marketing material generation model, the advertising marketing material to be delivered corresponding to the target user, comprises: determining a characteristic state associated with the target user in the time sequence diagram association structure based on user behavior information, user attribute information and external environment related information corresponding to the target user; Under the information updating constraint action, inputting the characteristic state into the advertising marketing material generation model, and outputting material characteristic representation matched with the target user; and constructing the corresponding advertising marketing materials to be put on the basis of the material characteristic representation.
8. An advertising marketing material generating device based on a time sequence diagram neural network is characterized by comprising an acquisition module and a processing module, wherein, The acquisition module is used for acquiring multi-source material information, wherein the multi-source material information comprises user behavior information, user attribute information, advertisement material information and external environment related information, distinguishing and representing the multi-source material information, and forming a mutually independent but associable basic feature set; The processing module is used for executing feature propagation processing of the time sequence diagram neural network on the time sequence diagram association structure, applying differentiated information update constraint to different association relations, constructing an advertisement marketing material generation model through an improved anti-facts association modeling mode, introducing the information update constraint in a model training process, and outputting advertisement marketing materials to be put corresponding to target users through the advertisement marketing material generation model.
9. An electronic device comprising a processor, a communication bus, a user interface, a network interface, and a memory, the memory for storing instructions, the user interface and the network interface for communicating to other devices, the processor for executing instructions stored in the memory to cause the electronic device to perform the method of any of claims 1-7.
10. A non-transitory computer readable storage medium storing instructions which, when executed, perform the method of any one of claims 1 to 7.

Description

Advertisement marketing material generation method and device based on time sequence diagram neural network Technical Field The application relates to the field of intelligent content generation, in particular to an advertising marketing material generation method and device based on a time sequence diagram neural network. Background The generation of advertisement marketing materials is a core link in an advertisement delivery system, and is widely dependent on data driving and intelligent means at present, and proper advertisement contents are automatically generated or selected according to user portraits and historical feedback so as to improve click rate and conversion rate. With the improvement of data scale and computing power, advertisement marketing material generation is gradually driven by a model from manual experience leading, and especially in a large-scale delivery scene, the automatic adaptation of different users and different delivery environments through algorithms has become a mainstream technical route. The existing advertising marketing material generation method is characterized in that the association relation among the behavior, the attribute and the advertising content of a user is analyzed through a time sequence diagram neural network, so that the interest evolution rule of the user is mined, and the advertising material is generated or recommended according to the interest evolution rule. However, by the method, since the clicking, staying, purchasing and other actions of the user in a certain time period are often simultaneously influenced by external factors such as activity subsidy, inventory change, channel position adjustment, bid item throwing strength, social hot spots and the like, the time sequence diagram neural network easily misjudges the external common factors as action change evidence caused by the materials, so that the wrong association relation is solidified in the model, the effect of the generated advertisement materials is obviously reduced when the external environment is recovered to be normal or the throwing scene is changed, and the effect of the generated materials is poor in throwing across scenes. Therefore, there is a need for a method and apparatus for generating advertising marketing materials based on a neural network of timing diagrams. Disclosure of Invention The application provides a method and a device for generating advertisement marketing materials based on a time sequence diagram neural network, which solve the problem that the advertisement marketing materials generated by the existing method for generating advertisement marketing materials have poor effect in cross-scene delivery. The application provides an advertising marketing material generation method based on a time sequence diagram neural network, which comprises the steps of obtaining multi-source material information, wherein the multi-source material information comprises user behavior information, user attribute information, advertising material information and external environment related information, distinguishing and representing the multi-source material information, forming mutually independent but associable basic feature sets, constructing a time sequence diagram association structure comprising user association, material association and environment association based on the basic feature sets, executing feature propagation processing of the time sequence diagram neural network on the time sequence diagram association structure, applying differentiated information updating constraint on different association relations, constructing an advertising material generation model in an improved anti-facts association modeling mode, introducing information updating constraint in a model training process, and outputting advertising marketing materials to be put corresponding to target users through the advertising marketing material generation model. The method comprises the steps of carrying out time-series analysis on user behavior information, extracting user behavior characteristics reflecting behavior change trend, behavior frequency distribution and behavior persistence of users in different time windows, carrying out stability modeling on user attribute information to form user attribute characteristics representing long-term attribute states of the users and keeping a associable but unmixed representation relation with the user behavior characteristics, carrying out content structure analysis on advertisement material information, forming material characteristics representing material expression modes, content constitution and version differences, enabling the material characteristics to be updated independently of user behavior changes, carrying out conditional representation on external environment related information, forming environment characteristics for describing throwing periods, flow distribution states, activity strategies or external disturbance factors, and e