Internet advertising delivery method and system based on artificial intelligence

Abstract

The present invention relates to an artificial intelligence-based internet advertising delivery method and system, and relates to the field of advertising delivery, including: when a first video is uploaded, classifying the first video using a classifier to obtain a first video category; configuring a first advertisement category for the first video category; when the first video is clicked and played by a first user, randomly matching pre-stored advertisements to be delivered to the first user based on the first advertisement category. This application solves the technical problem of inaccurate data recognition before delivery in the prior art, resulting in inaccurate advertising delivery, and achieves the technical effect of quickly and accurately matching video content with advertisement categories, thereby realizing precise advertising delivery.

Description

Technical Field

[0001] The present invention relates to the field of advertising delivery, and in particular to an artificial intelligence-based Internet advertising delivery method and system. Background Art

[0002] With the rapid development of internet technology, video has become a crucial information carrier on the internet. As a common advertising format, video ads are favored by advertisers for their vividness and strong reach. However, existing video advertising technologies often suffer from issues such as low relevance between ads and video content and insufficient ad placement accuracy. This is because existing technologies struggle to accurately identify the core information in video content, making it impossible to precisely match video content with ads, resulting in inaccurate ad placement. Summary of the Invention

[0003] The present invention aims to solve the technical problem in the prior art of inaccurate data recognition before delivery, which leads to inaccurate advertising delivery, and provides an artificial intelligence-based Internet advertising delivery method to solve it.

[0004] The technical solution of the present invention to solve the above technical problems is as follows:

[0005] In a first aspect, the present invention provides an artificial intelligence-based Internet advertising delivery method, comprising: when a first video is uploaded, classifying the first video through a classifier to obtain a first video category; configuring a first advertisement category for the first video category; when the first video is clicked and played by a first user, randomly matching the advertisements to be delivered from pre-stored advertisements according to the first advertisement category and delivering them to the first user.

[0006] In a second aspect, the present invention provides an artificial intelligence-based Internet advertising delivery system, comprising: a video classification module, for classifying the first video through a classifier to obtain a first video category when the first video is uploaded; an advertising configuration module, for configuring a first advertising category for the first video category; and an advertising delivery module, for randomly matching an advertisement to be delivered from pre-stored advertisements according to the first advertising category and delivering the advertisement to the first user when the first video is clicked to play by the first user.

[0007] In a third aspect, the present application provides an electronic device comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is used to execute an artificial intelligence-based Internet advertising delivery method provided in the present application.

[0008] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program, which is used to execute an artificial intelligence-based Internet advertising delivery method provided by the present application.

[0009] The beneficial effects of the present invention are:

[0010] When the first video is uploaded, the first video is classified by a classifier to obtain a first video category, and the content category of the video is automatically identified to provide a basis for subsequent advertising matching; a first advertising category is configured for the first video category, so that videos of the same category can correspond to advertisements of the same category, thereby improving the relevance of advertising delivery; when the first video is clicked and played by the first user, advertisements to be delivered are randomly matched from pre-stored advertisements according to the first advertising category and delivered to the first user, quickly matching advertisements related to the video category and delivering them immediately when the user clicks on the video, thereby improving the timeliness and accuracy of advertising delivery. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 A flowchart of an artificial intelligence-based Internet advertising delivery method provided by the present invention;

[0012] Figure 2 A schematic diagram of the structure of an artificial intelligence-based Internet advertising delivery system provided by the present invention;

[0013] Figure 3 A schematic structural diagram of the electronic device provided by the present invention;

[0014] Figure 4 A schematic structural diagram of a computer-readable storage medium provided by the present invention.

[0015] In the accompanying drawings, the components represented by each number are listed as follows: video classification module 11, advertisement configuration module 12, advertisement delivery module 13, electronic device 500, memory 510, processor 520, computer program 511, computer readable storage medium 600, computer program 611. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0017] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the specified features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0018] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or illustration". Any embodiment of the present invention described as "for example" is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any person skilled in the art to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed herein.

[0019] Example 1:

[0020] like Figure 1 As shown, an embodiment of the present invention provides an Internet advertising delivery method based on artificial intelligence, including:

[0021] S1: When a first video is uploaded, the first video is classified by a classifier to obtain a first video category.

[0022] Specifically, when a user or the video platform backend uploads the first video, a pre-trained classifier is used to analyze and process the first video. By obtaining the video frame information of the first video and performing feature extraction on the video frame, a feature vector of the first video is obtained. The extracted feature vector is then input into the classifier for classification judgment, thereby obtaining the category to which the first video belongs, that is, obtaining the first video category. Among them, the classifier adopts a video classification algorithm, such as a convolutional neural network model based on deep learning. The classifier is trained with a large amount of video data so that it can quickly and accurately determine the category to which the video belongs. In actual applications, common video categories are defined and labeled in advance, such as entertainment, news, sports, education, etc., and the classifier is trained based on this. After obtaining the first video category, it is stored for subsequent advertising matching and delivery processes. By identifying and recording video categories, a data foundation is laid for achieving accurate advertising delivery.

[0023] S2: Configure a first advertisement category for the first video category.

[0024] Specifically, a database of correspondences between video categories and advertising categories is pre-established. This database records one or more advertising categories corresponding to each video category. For example, sports videos are assigned advertisements related to sporting goods and sports drinks; educational videos are assigned advertisements related to online courses and supplementary teaching materials. When the video category of the first video is obtained, the correspondence database is automatically searched to find one or more advertising categories that match the first video category, which are used as the first advertising category. The first advertising category is semantically associated with the first video category, which improves the targeting and accuracy of advertising delivery.

[0025] In another embodiment, the advertiser or administrator may manually specify the first advertisement category corresponding to the first video category. Through manual intervention, the configuration relationship between the video category and the advertisement category can be adjusted more flexibly and dynamically.

[0026] Regardless of whether automatic matching or manual configuration is used, establishing a correspondence between the first video category and the first advertisement category and configuring the first advertisement category for the first video category can provide a basis for subsequent advertisement delivery, so as to quickly respond and deliver when an advertisement request is received.

[0027] S3: When the first video is clicked by the first user to play, an advertisement to be delivered is randomly matched from pre-stored advertisements according to the first advertisement category and delivered to the first user.

[0028] Specifically, a large amount of candidate advertising data is pre-stored, and each piece of advertising data includes information such as advertising content and advertising category. These candidate advertisements can be provided and uploaded by advertisers, or obtained from advertising platforms or advertising networks. When the first user clicks to play the first video on the video platform, the play request is received, and the first video category of the first video and the corresponding configured first advertising category are obtained. Then, one or more advertisements that match the first advertising category are screened out from the pre-stored candidate advertisements to form a set of alternative advertisements. Next, an advertisement is randomly selected from the set of alternative advertisements as the advertisement to be delivered. The random selection method can, on the one hand, avoid the frequent recurrence of the same advertisement, which may cause user aesthetic fatigue; on the other hand, it can also give different advertisements equal opportunities to be displayed, thereby increasing the participation and satisfaction of advertisers. Afterwards, the randomly selected advertisement to be delivered is pushed to the first user, so that the user receives relevant advertising information while watching the first video.

[0029] By dynamically matching and delivering relevant advertisements based on video categories when the first user watches the first video, the timeliness and accuracy of advertisement delivery can be improved through personalized advertisement recommendation.

[0030] Furthermore, classifying the first video using a classifier to obtain a first video category includes:

[0031] Step S110: performing feature preprocessing on the first video to obtain a first video feature vector;

[0032] Step S120: Processing the first video feature vector through a classifier to obtain the first video category.

[0033] In one feasible implementation, feature preprocessing is first performed on the first video to convert the video data into a feature vector representation suitable for input into the classifier. Various video feature extraction algorithms, such as frame difference, optical flow, HOG (Histogram of Oriented Gradients), and SIFT (Scale-Invariant Feature Transform), are employed to capture temporal and spatial information between video frames, generating high-dimensional feature data. Next, the extracted feature data undergoes preprocessing, such as normalization, dimensionality reduction, and filtering, to remove redundancy and noise and improve feature robustness and discriminability. After preprocessing, a feature vector representation of the first video is obtained, denoted as the first video feature vector.

[0034] Afterwards, the first video feature vector is input into a pre-trained classifier, and the classifier calculates and judges to obtain the category label to which the video belongs, i.e., the first video category. The classifier can use machine learning or deep learning algorithms, such as support vector machines, random forests, convolutional neural networks, long short-term memory networks, etc. It is trained through a large amount of labeled video sample data to learn the characteristic patterns and discrimination rules of videos of different categories. When the first video feature vector is input, the classifier calculates and classifies the features based on the learned model parameters, and outputs the category with the highest probability as the judgment result. For example, if the classifier outputs the result of "Sports: 0.8, Entertainment: 0.1, News: 0.1", then "Sports" with the largest probability value is taken as the category label of the first video, as the first video category.

[0035] By extracting features of the first video, preprocessing and classifier judgment, the category information of the first video is obtained, which lays the foundation for subsequent advertisement matching.

[0036] Furthermore, performing feature preprocessing on the first video to obtain a first video feature vector includes:

[0037] Step S111: obtaining the video frame at time t and the video frame at time t+1 of the first video;

[0038] Step S112: performing feature fusion on the video frame at time t and the video frame at time t+1 to obtain a first feature vector at time t+1;

[0039] Step S113: performing time feature fusion on the first feature vector at time t+1 to obtain a second feature vector at time t+1;

[0040] Step S114: until the second feature vector at time t+N is obtained, where N represents the total number of video frames minus one;

[0041] Step S115: Using a fully connected layer, sequentially concatenate the first feature vector at time t+1 to the second feature vector at time t+N to obtain the first video feature vector.

[0042] In a preferred embodiment, first, two adjacent video frames from a first video are acquired in chronological order, denoted as the video frame at time t and the video frame at time t+1, respectively, where t is the frame timestamp. Feature extraction and fusion processing are then performed on the video frames at time t and time t+1. Specifically, algorithms such as convolutional neural networks are employed to extract feature vectors representing appearance information from each of the two video frames. These feature vectors are then fused according to a specific fusion strategy (such as splicing, summing, or averaging) to obtain a fused feature vector, denoted as the first feature vector at time t+1. By fully leveraging the temporal continuity and visual correlation between adjacent video frames, a more comprehensive and robust feature representation is achieved through feature fusion.

[0043] Next, the first time t+1 feature vector undergoes further temporal feature fusion. Specifically, using algorithms such as recurrent neural networks (such as LSTM, GRU, etc.) or temporal convolutional networks, the first time t+1 feature vector is sequentially modeled and fused with feature vectors from several previous time points to capture the long-term dependencies between video frames. This fused feature vector is then recorded as the second time t+1 feature vector. This process is repeated, processing each pair of adjacent frames in the first video. A series of second time feature vectors is obtained, including the second time t+1 feature vector, until the second time t+N feature vector is obtained, totaling N feature vectors, where N is the total number of video frames in the first video minus one. The processing flow for the feature vector at time t+N is the same as for the first time t+1 feature vector. Subsequently, through a fully connected layer, the feature vectors from time t+1 to time t+N are sequentially concatenated or fused to form the first video feature vector. This feature vector contains the spatiotemporal feature information of the entire video and is used for subsequent video classification tasks.

[0044] Furthermore, feature fusion is performed on the video frame at time t and the video frame at time t+1 to obtain a first feature vector at time t+1, including:

[0045] Step S1121: performing optical flow analysis on the video frame at time t and the video frame at time t+1 to obtain an optical flow image;

[0046] Step S1122: performing pixel difference analysis on the video frame at time t and the video frame at time t+1 to obtain a pixel difference image;

[0047] Step S1123: stitching the optical flow image and the pixel difference image to obtain a stitched image;

[0048] Step S1124: performing feature extraction on the spliced ​​image to obtain the first feature vector at time t+1.

[0049] In a preferred embodiment, when performing feature fusion on the video frame at time t and the video frame at time t+1 to obtain the first feature vector at time t+1, an optical flow analysis algorithm, such as the Lucas-Kanade algorithm or the Horn-Schunck algorithm, is first used to calculate the pixel motion information between the video frame at time t and the video frame at time t+1. Optical flow is a vector field that describes the pixel motion pattern in an image sequence, reflecting the correspondence and motion trend between video frames. Through optical flow analysis, an optical flow image between the two video frames is obtained, in which the value of each pixel represents the displacement vector of the point between the two frames. The optical flow image can effectively capture the motion information in the video and provide useful prior knowledge for subsequent feature extraction. Next, a pixel difference image is obtained by calculating the difference between the corresponding pixel points in the video frame at time t and the video frame at time t+1. Specifically, the pixel value of the video frame at time t is subtracted from the pixel value of the video frame at time t+1 to obtain a difference image, and then the absolute value is taken to obtain the pixel difference image. The pixel difference image reflects the pixel changes between two video frames, highlighting the moving areas and foreground objects to capture the changes in video content.

[0050] After obtaining the optical flow image and pixel difference image, the two images are stitched together to form a stitched image. The stitching method can be horizontal stitching, vertical stitching, or channel stitching, etc., selected according to the specific features. By stitching the optical flow image and pixel difference image, motion information and pixel change information are simultaneously utilized to obtain a more comprehensive feature representation. Subsequently, feature extraction algorithms such as convolutional neural networks are used to learn and extract features from the stitched image. Through multi-layer convolution and pooling operations, convolutional neural networks automatically learn the hierarchical features in the image, from low-level edges and textures to high-level semantic concepts. The stitched image is input into a pre-trained convolutional neural network, and after forward propagation calculation, a fixed-length feature vector is obtained, namely the first feature vector at time t+1. This feature vector integrates the information of the optical flow image and the pixel difference image, and can better represent the content changes and semantic information between adjacent video frames.

[0051] Furthermore, feature extraction is performed on the spliced ​​image to obtain the first feature vector at time t+1, including:

[0052] Step S11241: inputting the stitched image into three parallel convolutional feature extraction networks to obtain three stitched image feature vectors;

[0053] Step S11242: flattening the three stitched image feature vectors to obtain three flattened feature vectors;

[0054] Step S11243: splicing the three flattened feature vectors according to RGB channels to obtain a spliced ​​feature vector;

[0055] Step S11244: fusing the concatenated feature vectors through a fully connected neural network to obtain the first feature vector at time t+1, wherein the fully connected neural network includes multiple fully connected layers, and the sizes of the multiple fully connected layers decrease in sequence.

[0056] In a preferred embodiment, when extracting features from a stitched image, first, the stitched image is simultaneously input into three parallel convolutional feature extraction networks. The three networks have the same structure but independent parameters, and their purpose is to extract feature information of the stitched image from different scales and angles. Specifically, each convolutional feature extraction network consists of multiple convolutional layers, pooling layers, and activation functions, and maps the stitched image to a low-dimensional feature space through layer-by-layer feature transformation and abstraction. The three parallel convolutional feature extraction networks calculate in parallel to obtain three stitched image feature vectors of the same dimension, which are respectively denoted as stitched image feature vector 1, stitched image feature vector 2, and stitched image feature vector 3. Next, the three stitched image feature vectors obtained are flattened. Specifically, the multi-dimensional tensor form of each feature vector is converted into a one-dimensional vector form, keeping the order of the eigenvalues ​​unchanged. After flattening, three flattened feature vectors are obtained, which are respectively denoted as flattened feature vector 1, flattened feature vector 2, and flattened feature vector 3. The flattening operation can unify feature vectors from different sources into the same representation, preparing for subsequent feature fusion.

[0057] Next, flattened feature vectors 1, 2, and 3 are concatenated in the order of their RGB channels to produce a concatenated feature vector that is three times longer than the original. This aggregates and integrates the feature information extracted by the three parallel convolutional feature extraction networks, resulting in a more comprehensive and high-dimensional feature representation. The concatenated feature vector contains semantic information about the stitched image at different scales and angles, providing richer prior knowledge for subsequent feature fusion. The resulting concatenated feature vector is then fed into a pre-designed fully connected neural network, where a forward propagation calculation yields the first feature vector at time t+1. A fully connected neural network consists of multiple fully connected layers, each containing several neurons, which perform nonlinear transformations and high-level abstraction on the input features. Unlike convolutional neural networks, each neuron in a fully connected layer is connected to all neurons in the previous layer, enabling global integration and fusion of features. To improve feature fusion and reduce the risk of overfitting, the fully connected neural network adopts a decreasing size design, with the number of neurons in each fully connected layer decreasing. This structural design gradually compresses the feature dimensionality and extracts the most discriminative and representative high-level features. In the last layer of the fully connected neural network, the output feature vector is the feature vector at the first time t+1, which integrates the semantic information of the spliced ​​image on different convolutional networks, different flattened feature vectors and different fully connected layers, and can better represent the content association and action characteristics between adjacent video frames.

[0058] Furthermore, performing time feature fusion on the first feature vector at time t+1 to obtain a second feature vector at time t+1 includes:

[0059] Step S1131: Obtain the first feature vector at time t+1 to the first feature vector at time t+L, where L represents a preset time step;

[0060] Step S1132: Input the first feature vector at time t+1 to the first feature vector at time t+L into the LSTM neural network for processing to obtain the second feature vector at time t+1.

[0061] In a preferred embodiment, when performing time feature fusion on the first t+1 moment feature vector to obtain the second t+1 moment feature vector, first, from the series of first moment feature vectors obtained, a time window of length L starting from the first t+1 moment feature vector is selected, and L continuous feature vectors are extracted therein to obtain the first t+1 moment feature vector until the first t+L moment feature vector. Among them, the time step L is a pre-set parameter that determines the length of the context information considered each time the time feature fusion is performed. By sliding the time window, local time series features are extracted at different time positions and used as the input sequence of the LSTM neural network.

[0062] Next, the L first-time feature vectors are sequentially input into a pre-trained LSTM neural network in chronological order. A forward propagation calculation yields the second feature vector at time t+1. LSTM neural networks are recurrent neural networks that effectively learn and store long-term dependencies by incorporating gating mechanisms and memory units. Compared to traditional recurrent neural networks, LSTM can alleviate the problems of vanishing and exploding gradients. During the temporal feature fusion process, the LSTM neural network gradually receives the L first-time feature vectors and accumulates and updates temporal information in its internal state. Controlled by the gating units, the LSTM adaptively determines when to forget past states, when to update the current state, and when to output the final feature representation. After processing by the LSTM neural network, the feature vector output at time t+1 is the second feature vector at time t+1. Compared to the first feature vector at time t+1, the second feature vector at time t+1 not only contains the current video frame information but also incorporates contextual information from the previous L-1 time steps, better capturing the temporal continuity and motion evolution patterns in the video.

[0063] By integrating time window extraction and LSTM neural network, we achieve information integration on the time dimension of the first t+1 moment feature vector, and obtain the second t+1 moment feature vector containing richer temporal semantics. This fully utilizes the temporal continuity and motion evolution laws of video data, providing important prior knowledge for subsequent video classification tasks.

[0064] Furthermore, the first video feature vector is processed by a classifier to obtain the first video category, including:

[0065] Step S121: obtaining a video feature vector dataset and a video category identification dataset;

[0066] Step S122: supervising a classifier with the video category identification dataset, and using the video feature vector dataset as classifier input to train the classifier.

[0067] In a feasible implementation, when the first video feature vector is processed by a classifier to obtain the first video category, first, a data set for training the classifier is prepared. The data set includes two parts: a video feature vector data set and a video category identification data set. The video feature vector data set is obtained by performing a feature preprocessing process on a large number of video samples. After feature extraction and fusion, each video sample forms a feature vector of fixed length. The feature vectors of all video samples are aggregated to form a video feature vector data set; the video category identification data set is label information that corresponds one-to-one to the video feature vector data set. For each video sample, there is a predefined category identification that indicates the semantic category to which the video belongs, for example, sports, news, entertainment, education, etc. The category identifications of all video samples are aggregated to form a video category identification data set. The acquisition of these two data sets can be completed through various methods such as manual annotation, web crawling, and public data sets. In order to ensure the generalization performance of the classifier, the video samples in the data set should cover as many different shooting scenes, action types, time lengths, etc. as possible, and the category distribution should be relatively balanced.

[0068] Next, the classifier is trained using the obtained video feature vector dataset and video category identifier dataset. The classifier can employ machine learning algorithms such as support vector machines, decision trees, and naive Bayes, or deep learning algorithms such as multilayer perceptrons and convolutional neural networks. During training, the classifier uses the video feature vector dataset as input and the video category identifier dataset as a supervisory signal. Through iterative optimization, the classifier continuously adjusts internal parameters and weights to minimize the error between the predicted and true categories. To improve the classifier's generalization performance and avoid overfitting, cross-validation is used for model evaluation and hyperparameter selection. The dataset is divided into training, validation, and test sets. Model training is performed on the training set, model selection is performed on the validation set, and the final model performance is evaluated on the test set. After sufficient training and fine-tuning, the resulting classifier can effectively map video feature vectors to corresponding video category identifiers, enabling automatic classification of new, unknown videos.

[0069] By constructing a video feature vector dataset and a video category identification dataset and using them to conduct supervised learning on the classifier, a video classification model with superior performance is obtained to provide support for subsequent advertising delivery.

[0070] The embodiment of the present invention provides an artificial intelligence-based Internet advertising method, which has at least the following technical effects:

[0071] When the first video is uploaded, the classifier is used to classify the first video to obtain the first video category. The classifier is used to identify the video content and obtain the video category, laying the foundation for subsequent ad matching. The first ad category is configured for the first video category. By configuring the corresponding ad category for each video category, it is ensured that videos of the same category are matched with similar ads with high relevance, thereby improving the targeted advertising. When the first video is clicked to play by the first user, the pre-stored ads are randomly matched to be delivered to the first user based on the first ad category. Ads of the corresponding category are quickly matched based on the ad category, and the ads are pushed immediately when the user clicks the video, thereby improving the timeliness of ad delivery and achieving precise advertising delivery.

[0072] Example 2:

[0073] like Figure 2 As shown, based on the same inventive concept as the artificial intelligence-based Internet advertising delivery method provided in Example 1, an embodiment of the present invention further provides an artificial intelligence-based Internet advertising delivery system, including:

[0074] A video classification module 11 is configured to classify a first video using a classifier when the first video is uploaded to obtain a first video category;

[0075] an advertisement configuration module 12, configured to configure a first advertisement category for the first video category;

[0076] The advertisement delivery module 13 is configured to deliver advertisements to the first user by randomly matching pre-stored advertisements according to the first advertisement category when the first video is clicked by the first user for playback.

[0077] Furthermore, the video classification module 11 includes the following execution steps:

[0078] Performing feature preprocessing on the first video to obtain a first video feature vector;

[0079] The first video feature vector is processed by a classifier to obtain the first video category.

[0080] Furthermore, the video classification module 11 further includes the following execution steps:

[0081] Obtaining a video frame at time t and a video frame at time t+1 of the first video;

[0082] Performing feature fusion on the video frame at time t and the video frame at time t+1 to obtain a first feature vector at time t+1;

[0083] Performing time feature fusion on the first feature vector at time t+1 to obtain a second feature vector at time t+1;

[0084] Until the second feature vector at time t+N is obtained, N represents the total number of video frames minus one;

[0085] The first video feature vector is obtained by sequentially concatenating the first feature vector at time t+1 to the second feature vector at time t+N through a fully connected layer.

[0086] Furthermore, the video classification module 11 further includes the following execution steps:

[0087] Performing optical flow analysis on the video frame at time t and the video frame at time t+1 to obtain an optical flow image;

[0088] Performing pixel difference analysis on the video frame at time t and the video frame at time t+1 to obtain a pixel difference image;

[0089] Stitching the optical flow image and the pixel difference image to obtain a stitched image;

[0090] Feature extraction is performed on the spliced ​​image to obtain the first feature vector at time t+1.

[0091] Furthermore, the video classification module 11 further includes the following execution steps:

[0092] Inputting the spliced ​​image into three parallel convolutional feature extraction networks to obtain three spliced ​​image feature vectors;

[0093] Flattening the three stitched image feature vectors to obtain three flattened feature vectors;

[0094] Concatenate the three flattened feature vectors according to RGB channels to obtain a concatenated feature vector;

[0095] The concatenated feature vectors are fused through a fully connected neural network to obtain the first feature vector at time t+1, wherein the fully connected neural network includes multiple fully connected layers, and the sizes of the multiple fully connected layers decrease in sequence.

[0096] Furthermore, the video classification module 11 further includes the following execution steps:

[0097] Obtaining the first feature vector at time t+1 up to the first feature vector at time t+L, where L represents a preset time step;

[0098] The first feature vector at time t+1 to the first feature vector at time t+L are input into the LSTM neural network for processing to obtain the second feature vector at time t+1.

[0099] Furthermore, the video classification module 11 further includes the following execution steps:

[0100] Obtain a video feature vector dataset and a video category identification dataset;

[0101] The classifier is supervised by the video category identification dataset, and the video feature vector dataset is used as the classifier input to train the classifier.

[0102] Example 3:

[0103] See also Figure 3 , Figure 3 Schematic diagram of an embodiment of an electronic device provided by an embodiment of the present invention. Figure 3 As shown, an electronic device 500 provided by an embodiment of the present invention includes a memory 510, a processor 520, and a computer program 511 stored in the memory 510 and executable on the processor 520. When the processor 520 executes the computer program 511, an artificial intelligence-based Internet advertising delivery method is implemented.

[0104] Example 4:

[0105] See also Figure 4 , Figure 4 Schematic diagram of an embodiment of a computer-readable storage medium provided in an embodiment of the present invention. Figure 4 As shown, this embodiment provides a computer-readable storage medium 600 on which a computer program 611 is stored. When the computer program 611 is executed by a processor, an artificial intelligence-based Internet advertising delivery method is implemented.

[0106] It should be noted that, in the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0107] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0108] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A system that specifies the functions of a box or boxes.

[0109] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction system that is implemented in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0110] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0111] Although preferred embodiments of the present invention have been described, additional changes and modifications to these embodiments may occur to those skilled in the art once the basic inventive concepts become known.

[0112] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the present invention and its equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An artificial intelligence-based Internet advertising method, characterized in that: include: When a first video is uploaded, the first video is classified by a classifier to obtain a first video category; configuring a first advertisement category for the first video category; When the first video is clicked by the first user to play, randomly matching the advertisement to be delivered from the pre-stored advertisements according to the first advertisement category and delivering the advertisement to the first user; The step of classifying the first video by a classifier to obtain a first video category includes: Performing feature preprocessing on the first video to obtain a first video feature vector; Processing the first video feature vector using a classifier to obtain the first video category; The step of performing feature preprocessing on the first video to obtain a first video feature vector includes: Obtaining a video frame at time t and a video frame at time t+1 of the first video; Performing feature fusion on the video frame at time t and the video frame at time t+1 to obtain a first feature vector at time t+1; Performing time feature fusion on the first feature vector at time t+1 to obtain a second feature vector at time t+1; Until the second feature vector at time t+N is obtained, N represents the total number of video frames minus one; Through a fully connected layer, sequentially concatenate the first feature vector at time t+1 to the second feature vector at time t+N to obtain the first video feature vector; The step of performing feature fusion on the video frame at time t and the video frame at time t+1 to obtain a first feature vector at time t+1 includes: Performing optical flow analysis on the video frame at time t and the video frame at time t+1 to obtain an optical flow image; Performing pixel difference analysis on the video frame at time t and the video frame at time t+1 to obtain a pixel difference image; Stitching the optical flow image and the pixel difference image to obtain a stitched image; Performing feature extraction on the spliced ​​image to obtain the first feature vector at time t+1; The step of performing feature extraction on the spliced ​​image to obtain the first feature vector at time t+1 includes: Inputting the spliced ​​image into three parallel convolutional feature extraction networks to obtain three spliced ​​image feature vectors; Flattening the three stitched image feature vectors to obtain three flattened feature vectors; Concatenate the three flattened feature vectors according to RGB channels to obtain a concatenated feature vector; The concatenated feature vectors are fused through a fully connected neural network to obtain the first feature vector at time t+1, wherein the fully connected neural network includes multiple fully connected layers, and the sizes of the multiple fully connected layers decrease in sequence.

2. The method according to claim 1, wherein Performing time feature fusion on the first feature vector at time t+1 to obtain a second feature vector at time t+1 includes: Obtaining the first feature vector at time t+1 up to the first feature vector at time t+L, where L represents a preset time step; The first feature vector at time t+1 to the first feature vector at time t+L are input into the LSTM neural network for processing to obtain the second feature vector at time t+1.

3. The method according to claim 1, wherein Processing the first video feature vector by a classifier to obtain the first video category includes: Obtain a video feature vector dataset and a video category identification dataset; The classifier is supervised by the video category identification dataset, and the video feature vector dataset is used as the classifier input to train the classifier.

4. An artificial intelligence-based Internet advertising system, characterized in that: A method for delivering Internet advertisements based on artificial intelligence according to any one of claims 1 to 3, comprising: a video classification module, configured to classify a first video using a classifier to obtain a first video category when the first video is uploaded; an advertisement configuration module, configured to configure a first advertisement category for the first video category; an advertisement delivery module, configured to, when the first video is clicked by the first user to play, randomly match an advertisement to be delivered from pre-stored advertisements according to the first advertisement category and deliver the advertisement to the first user; The step of classifying the first video by a classifier to obtain a first video category includes: Performing feature preprocessing on the first video to obtain a first video feature vector; The first video feature vector is processed by a classifier to obtain the first video category.

5. An electronic device, characterized in that: include: Memory for storing computer software programs; A processor is used to read and execute the computer software program, thereby implementing an artificial intelligence-based Internet advertising delivery method as described in any one of claims 1-3.

6. A non-transitory computer-readable storage medium, characterized in that The storage medium stores a computer software program, which, when executed by a processor, implements an artificial intelligence-based Internet advertising delivery method as described in any one of claims 1 to 3.

Images