Search

CN-121981781-A - Online advertisement delivery optimization system and method based on big data analysis

CN121981781ACN 121981781 ACN121981781 ACN 121981781ACN-121981781-A

Abstract

The invention relates to the technical field of online advertisement delivery, in particular to an online advertisement delivery optimizing system and method based on big data analysis, wherein an emotion type detecting unit extracts video space-time micro-expression characteristics and image-text semantic emotion characteristics according to video image-text contents browsed by users, generates cross-modal emotion vectors, outputs fine-granularity emotion labels with intensity grading, and calculating a dynamic risk value through an accumulated overload prediction model based on the acquired continuous emotion sequence and the using time-length real-time data of the user equipment, judging to trigger early warning, after the advertisement putting optimization unit receives the early warning, matching and shielding the similar advertisements by emotion fingerprints, selecting opposite advertisements according to emotion wheel polarity mapping space and user acceptance thermodynamic diagram, searching by using an intensity attenuation algorithm when the advertisements can not be replaced, generating low-intensity advertisements, solving the problem that the conventional putting ignores emotion accumulation, and balancing advertisement putting and user experience.

Inventors

  • WU YAJUN
  • LI PEIGANG
  • XU KEYUN

Assignees

  • 众智嘉能(北京)信息科技有限公司

Dates

Publication Date
20260505
Application Date
20260112

Claims (10)

  1. 1. An online advertisement delivery optimization system based on big data analysis, comprising: The emotion type detection unit (1) is used for collecting video image-text contents currently browsed by a user in real time and outputting emotion type labels through a pre-trained emotion classification model; The emotion accumulation effect monitoring unit (2) acquires an emotion type tag sequence of the video image-text content continuously browsed by a user through a time window sliding mechanism, calculates an accumulated overload risk value of a current emotion sequence in real time based on an accumulated overload prediction model trained by historical user behavior data, establishes based on a nonlinear mapping relation between the length and the strength of the emotion sequence and an emotion tolerance threshold of the user, and generates an accumulated overload early warning signal when the risk value exceeds a dynamic threshold; The advertisement putting optimization unit (3) receives the accumulated overload early warning signal and the real-time emotion state of the user, forcibly shields candidate advertisements with the same category as the early warning emotion under the accumulated overload early warning state, screens substitute advertisements from advertisement libraries with opposite emotion polarities according to the historical preference data of the user and the real-time emotion state, and selects advertisements with at least two levels of reduced emotion intensity in the same category when the advertisements cannot be replaced, so as to block an emotion accumulation chain and realize putting based on the emotion tolerance threshold value of the user.
  2. 2. The online advertisement delivery optimization system based on big data analysis of claim 1, wherein the pretrained emotion classification model realizes emotion classification label output through a multi-modal hierarchical feature extraction architecture, and specifically comprises the following steps: Extracting space-time micro-expression features from video content by adopting a three-dimensional convolutional neural network, extracting semantic emotion features from the video content by adopting a pre-trained language model architecture, generating cross-modal emotion vectors by two types of features through a gating fusion module, and outputting fine granularity classification results containing emotion intensity classification labels through a full-connection layer, wherein an intensity classification standard is established based on physiological response thresholds calibrated by large-scale user investigation data.
  3. 3. The online advertisement delivery optimization system based on big data analysis according to claim 2, wherein the training process of the cumulative overload prediction model adopts a dynamic sequence mapping technology, and specifically comprises: Based on emotion sequences and follow-up behavior labels in historical user behavior data, a long-term and short-term memory neural network framework is constructed, an input layer receives a double-channel coding vector composed of emotion type label sequences and corresponding strength grading labels, a hidden layer draws attention mechanisms to give weights to emotion at the tail of the sequence, an output layer generates a nonlinear mapping relation through a user portrait embedding layer, and finally a cumulative overload prediction model capable of dynamically updating the weights is formed.
  4. 4. The online advertisement delivery optimization system based on big data analysis of claim 3, wherein the real-time calculation of the cumulative overload risk value of the current emotion sequence is realized by a space-time state perception engine, and the online advertisement delivery optimization system specifically comprises: Inputting the real-time emotion sequence obtained by the emotion accumulation effect monitoring unit (2) into a trained long-period and short-period memory neural network, synchronously loading the use duration, the ambient light intensity and the heart rate variability of the current equipment of the user, carrying out real-time feature fusion through a user portrait embedding layer in an accumulation overload prediction model, and outputting a dynamic risk value reflecting the emotion tolerance state of the user, wherein the value continuously increases along with the sequence and has an exponentially rising trend.
  5. 5. The online advertisement delivery optimization system based on big data analysis of claim 4, wherein the calculation process of the dynamic risk value is additionally provided with a critical strengthening module, and the system specifically comprises: When the content of the continuous same emotion category exceeds the preset number, activating a sequence tail emotion intensity multiplication algorithm to automatically promote the numerical weight of the intensity classification label corresponding to the tail emotion label in the input vector, and dynamically adjusting forgetting door parameters of a long-short-period memory neural network hidden layer based on the real-time physiological index data of the user to accelerate the sensitivity of cumulative effect modeling.
  6. 6. The online advertisement delivery optimization system based on big data analysis of claim 5, wherein the generation of the accumulated overload early warning signal adopts a double-threshold trigger mechanism, and specifically comprises the following steps: The basic static threshold value and the dynamic floating threshold value are preset, a yellow early warning signal is generated when the dynamic risk value exceeds the basic static threshold value and three continuous calculation periods keep rising trend, the dynamic risk value is updated to a red early warning signal when the dynamic risk value further breaks through the dynamic floating threshold value, the dynamic floating threshold value is automatically adjusted according to the current pressure sensitivity coefficient of a user, and the red early warning signal triggers the forced shielding function of the advertisement putting optimizing unit (3).
  7. 7. The online advertisement delivery optimization system based on big data analysis of claim 6, wherein the forced shielding of candidate advertisements with the same category as the early warning emotion is realized by emotion fingerprint matching technology, and the system specifically comprises: And establishing an advertisement emotion feature matrix library, wherein each advertisement generates emotion fingerprint codes containing user main emotion categories and intensity grades through an emotion category detection unit (1), and when a red early warning signal is received, the emotion fingerprint codes of advertisements to be put are compared with emotion category labels in the early warning signal through a bloom filter, so that real-time interception is implemented on the advertisements matched with the first three codes.
  8. 8. The online advertisement delivery optimization system based on big data analysis of claim 7, wherein the selection of the alternative advertisement from the advertisement library with opposite emotion polarity adopts an emotion vector space navigation technique, and the system specifically comprises the following steps: And constructing a high-dimensional emotion polarity mapping space based on emotion wheel theory, mapping emotion fingerprint codes of all advertisements in an advertisement library with opposite emotion polarities into space vectors, generating personalized emotion receptivity thermodynamic diagrams according to user history preference data, and preferentially selecting advertisement vector generation candidate sets which are located in opposite polarity quadrants of the mapping space and in a thermodynamic diagram high-receptivity area of the early-warning emotion vectors when a red early-warning signal is triggered.
  9. 9. The online advertisement delivery optimization system based on big data analysis of claim 8, wherein selecting advertisements with at least two levels of reduced emotional intensity in the same category is realized by an intensity decay chain algorithm, and specifically comprises: And when the advertisement cannot be replaced, analyzing the emotion intensity grading labels in the early warning signals, searching advertisements of the same emotion category but with at least two grades of reduced intensity grading labels in the advertisement emotion feature matrix library, and if the matching fails, starting a cross-grade attenuation mechanism, namely, reserving the main emotion category and removing all high-intensity elements, and generating a new advertisement instance for putting.
  10. 10. A method for implementing an online advertising delivery optimization system comprising a big data analysis based on any of claims 1-9, comprising the steps of: S1, extracting space-time micro-expression features from video content by collecting video image-text content currently browsed by a user, extracting semantic emotion features from the image-text content by adopting a pre-trained language model, generating a cross-modal emotion vector by a gating fusion module, and outputting a fine granularity classification result with an emotion intensity classification label; S2, capturing a continuous emotion category label sequence based on a time window sliding mechanism, forming a double-channel coding vector by the continuous emotion category label sequence and a corresponding intensity grading label, inputting the double-channel coding vector into a long-short-period memory neural network, synchronously fusing data of the use duration, the ambient light intensity and the heart rate variability of user real-time equipment, outputting a dynamic risk value through an accumulated overload prediction model, activating a critical strengthening module when the sequence length exceeds a preset number, improving the tail emotion weight, adjusting forgetting door parameters, and generating a yellow or red early warning signal by combining a double-threshold triggering mechanism; S3, if a red early warning signal is triggered, advertisements with the same class as the early warning class of three matching bits before emotion fingerprint coding are forcedly shielded through an emotion fingerprint matching technology, alternative advertisements are screened from opposite quadrants of emotion polarity based on a polarity mapping space constructed by emotion wheel theory and a user emotion acceptance thermodynamic diagram, and when the advertisements can not be replaced, an intensity attenuation chain algorithm is started to search advertisements with the same class of intensity reduced by at least two stages, or a new advertisement instance is generated through cross-stage attenuation; S4, only starting an intensity gradient adaptation mechanism aiming at the yellow early warning signal, reserving advertisements of the same emotion type but forcibly degrading intensity labels by at least two stages, simultaneously refluxing user follow-up behavior data to a historical user behavior database, and dynamically updating accumulated overload prediction model weights.

Description

Online advertisement delivery optimization system and method based on big data analysis Technical Field The invention relates to the technical field of online advertisement delivery, in particular to an online advertisement delivery optimizing system and method based on big data analysis. Background On-line advertisement delivery is an important technology, and in the background of improving the current digital marketing refinement demand, the technology is a key support for balancing advertisement delivery effect and user experience, and can analyze user preference by means of big data, improve advertisement click rate and conversion efficiency, avoid resource waste caused by blind delivery, pay attention to user emotion change, prevent user dislike caused by improper advertisement stimulation, maintain platform user retention rate, adapt to short video platform, information APP and electronic commerce platform high-frequency advertisement delivery scenes, and meet double demands of advertisement marketing targets and user experience optimization; The conventional online advertisement delivery technology based on big data is faced with the core problems that the user emotion accumulation effect is not considered in practical application, the user emotion overload and experience deterioration are easy to cause, the conventional system pushes advertisements based on the user history preference, the emotion characteristics of the current browsing content of the user are not collected in real time, the accumulated influence of continuous emotion sequences on the user is not monitored, meanwhile, an emotion overload risk assessment mechanism is lacked, whether the user reaches an emotion tolerance threshold value due to continuous reception of the same type of emotion content cannot be judged, advertisements with the same type or high intensity as the current user emotion are still continuously pushed during delivery, the defect formation causes the user to continuously receive the same type of emotion advertisements, the emotion accumulation effect is aggravated, the emotion accumulation effect gradually exceeds the emotion tolerance range, the dysphoria and contradiction psychology are generated, secondly, the user with emotion overload frequently skips advertisements and even exits from a platform, the click rate and the completion rate of advertisements are reduced, the marketing goal of an advertiser is difficult to reach, the advertisement delivery resource waste is also caused, finally, the emotion experience of the user is ignored for a long time, the user is reduced, the user liveness and the user activity and the retention rate of the platform is reduced, the advertisement is more formed, the advertisement experience is poor, the lower the user experience is the more, the flow is lower, the longer the life is, the longer the circulation is satisfied, the service is the user is better, the user experience is the service life is better, the user experience, the user is difficult to be more the user experience is the user experience and the user experience is difficult to be better the user experience, and the user experience and the user life is based on the user, and the user. Disclosure of Invention The invention aims to provide an online advertisement delivery optimization system and method based on big data analysis, so as to solve the problems in the background technology. 1. Because the user emotion accumulation and early warning overload cannot be monitored in real time, the case collects the content emotion through the emotion type detection unit and outputs a label, the emotion accumulation effect monitoring unit calculates a risk value by using a prediction model, and the early warning is triggered by double-threshold judgment, so that the emotion overload risk can be timely identified. 2. Because the similar or high-intensity advertisements are still thrown when the emotion is overloaded, the advertisement throwing optimization unit is used for shielding the similar advertisements, selecting the opposite advertisements or reducing the intensity advertisements, the emotion accumulation can be blocked, and the balance between the user experience and the advertisement throwing is ensured. To achieve the above object, there is provided an online advertisement delivery optimization system based on big data analysis, comprising: The emotion type detection unit is used for collecting video image-text contents currently browsed by a user in real time and outputting emotion type labels through a pre-trained emotion classification model; The emotion accumulation effect monitoring unit acquires an emotion type tag sequence of the video image-text content continuously browsed by a user through a time window sliding mechanism, calculates an accumulated overload risk value of a current emotion sequence in real time based on an accumulated overload prediction model trained by historical user behavior data,