Advertisement content intelligent analysis system and device based on machine learning

Abstract

The invention provides an intelligent advertisement content analysis system and device based on machine learning, and the system comprises the steps: obtaining advertisement content data, and constructing an initial training sample set of the intelligent advertisement content analysis system based on the advertisement content data; the initial sample set comprises advertisement feature data and advertisement category labels; performing label smoothing on the advertisement category labels in the initial sample set to obtain a final training sample set; constructing an intelligent advertisement content analysis model based on a deep learning network; training the advertisement content intelligent analysis model based on the advertisement content intelligent analysis training sample set to obtain a trained advertisement content intelligent analysis model; extracting feature data of the to-be-analyzed advertisement content; and inputting the advertisement feature data into a trained advertisement content intelligent analysis model, and performing category analysis on the advertisement content. According to the invention, the accuracy and efficiency of intelligent analysis of the advertisement content can be improved, and more reliable technical support is provided for advertisement putting and effect evaluation.

Description

Technical Field

[0001] The present invention relates to the field of machine learning technology, and more specifically, to an advertising content intelligent analysis system and device based on machine learning. Background Art

[0002] In existing advertising content analysis technologies, the classification and analysis of advertising content usually rely on traditional machine learning algorithms, which often require a lot of manual feature engineering to extract useful information, and the accuracy and efficiency of classification are limited. In addition, the diversity and complexity of advertising content make it difficult for traditional algorithms to adapt to the ever-changing advertising environment. With the development of big data and deep learning technology, people have begun to try to use deep learning networks to automatically extract advertising features and classify them, but these methods often ignore the complex relationships and conversion characteristics between advertising content, resulting in inaccurate analysis results.

[0003] In the process of implementing the embodiments of the present invention, the inventors found that there are at least the following problems or defects in the prior art: the existing intelligent analysis system for advertising content often cannot handle the dynamic changes and complexity of advertising content well when processing advertising content, especially when the advertising content switches frames, there is a lack of effective mechanism to smooth and handle the transition of advertising category labels, resulting in instability of the analysis results. At the same time, the prior art often does not take into account the confidence of the features when processing advertising features, which may cause the analysis model to be overly dependent on certain features, thereby affecting the overall classification performance. Summary of the invention

[0004] The present invention provides an advertisement content intelligent analysis system and device based on machine learning.

[0005] In a first aspect of the present invention, there is provided an advertisement content intelligent analysis system based on machine learning, comprising: Acquire advertisement content data, and construct an initial training sample set for an advertisement content intelligent analysis system based on the advertisement content data; the initial sample set includes advertisement feature data and ad category labels ; Label the advertising categories in the initial sample set Perform label smoothing to obtain the final training sample set; Building an intelligent advertising content analysis model based on deep learning network ; Based on the advertising content intelligent analysis training sample set, the advertising content intelligent analysis model Conduct training to obtain a trained advertising content intelligent analysis model ; Extract feature data of the advertising content to be analyzed ; The advertising feature data Input the trained advertising content intelligent analysis model In the process of advertising content classification analysis,

[0006] Furthermore, the advertisement category labels in the initial sample set are Perform label smoothing to obtain the final training sample set, including: all advertising category labels in the sample set Perform overall smoothing; Determine an advertisement content conversion frame in the sample set, and perform intra-group advertisement category label smoothing on a group of contents before the advertisement content conversion frame; The advertisement characteristic data Includes confidence in ad features ; Confidence based on ad features Ad category label for each content Perform confidence smoothing.

[0007] Furthermore, the following formula is used to classify all advertisement category labels in the sample set: Perform overall smoothing:

[0008] in, represents the sample ad category label, represents the number of categories, represents the smoothing parameter.

[0009] Further, the advertisement category label smoothing is performed on a group of contents before the advertisement content conversion frame, including: for each advertisement content conversion frame, determining the active index of the content group before the conversion according to the label value of the content of the previous frame before the advertisement content conversion frame and the label value of the advertisement content conversion frame and the target index ; According to the formula , calculate the advertising content before switching frames In the content tag The corresponding label value; According to the formula , calculate the advertising content before switching frames In the content tag The corresponding label value; Among them, The frame is the advertisement content conversion frame. Indicates In the ad category label of the frame content The corresponding label value; Indicates In the ad category label of the frame content The corresponding label value, It is The index of the maximum value of the label in the ad category label of the frame content. It is The index of the maximum value of the ad category label in the frame content.

[0010] Furthermore, based on the confidence of the advertising features Ad category label for each content Confidence smoothing is performed, including: for each frame of content, setting the confidence of the main advertisement feature to 1, and calculating the mean of the confidence of all advertisement features; The mean of the confidences and the advertising category label of the frame content Multiply them together to get a confidence-based smoothed label for the frame content.

[0011] Furthermore, an intelligent analysis model for advertising content is constructed based on a deep learning network. , including: constructing an advertising feature graph with advertising features as nodes and correlations between advertising features as connections ; Construct a deep learning network comprising multiple sequentially connected convolutional layers and fully connected layers ; Each of the convolutional layers Used to perform convolution on the input features to extract the advertising feature map The fully connected layer Used to classify the output of the convolutional layer.

[0012] Furthermore, the convolutional layer Used to perform convolution on the input features to extract the advertising feature map The invention relates to a method for partitioning the neighborhood of each node in the advertisement feature graph into subsets based on correlation; and constructing an adjacency matrix of each node based on the subsets after partitioning. ; Convolutional Layer According to the formula Perform convolution operation to extract features; in, represents the input features of the convolutional layer, represents the output features of the convolutional layer, represents the convolution kernel weight, represents the bias term, represents the activation function, Represents a convolution operation.

[0013] Furthermore, the convolutional layer It includes a feature enhancement unit for adaptively adjusting the feature importance of each node to other adjacent nodes; The feature enhancement unit includes batch normalization layers connected in sequence , Relu layer , dropout layer , convolutional layer and Sigmoid layer ; The normalization layer Used to make the feature enhancement matrix asymmetric; Layers are used for nonlinear transformations; The convolutional layer is used to prevent overfitting. The convolution kernel is 1×1, which is used to make the enhancement matrix consistent with the dimension of the corresponding convolution layer; The layer is used to map the output to the range [0,1].

[0014] In a second aspect of the present invention, there is provided an advertisement content intelligent analysis device based on machine learning, comprising: The training set construction module is used to obtain advertising content data and construct an initial training sample set for the advertising content intelligent analysis system based on the advertising content data; the initial sample set includes advertising feature data and ad category labels ; Label the advertising categories in the initial sample set Perform label smoothing to obtain the final training sample set; Model training module, used to build an intelligent analysis model for advertising content based on deep learning network ; Based on the advertising content intelligent analysis training sample set, the advertising content intelligent analysis model Conduct training to obtain a trained advertising content intelligent analysis model ; Advertisement content intelligent analysis module, used to extract characteristic data of the advertisement content to be analyzed ; The advertising feature data Input the trained advertising content intelligent analysis model In the process of advertising content classification analysis,

[0015] Furthermore, the training set construction module includes: an overall smoothing module for smoothing all advertisement category labels in the sample set. Perform overall smoothing; An intra-group smoothing module is used to determine an advertisement content conversion frame in a sample set and perform intra-group advertisement category label smoothing on a group of contents before the advertisement content conversion frame; Confidence smoothing module, used for the advertisement feature data Includes confidence in ad features ; Confidence based on ad features Ad category label for each content Perform confidence smoothing.

[0016] The above-mentioned embodiments of the present invention have at least the following beneficial effects: the present invention can effectively extract features in the advertisement feature graph by using a deep learning network to construct an intelligent analysis model for advertisement content, and use a combination of a convolutional layer and a fully connected layer to perform accurate category analysis on advertisement content. In addition, by performing label smoothing on advertisement category labels during training, including overall smoothing, intra-group smoothing, and confidence smoothing, the adaptability of the model to changes in advertisement content can be enhanced, and the accuracy and robustness of classification can be improved.

[0017] Furthermore, by constructing a feature enhancement unit, the present invention can adaptively adjust the feature importance of each node to other adjacent nodes, thereby enhancing the model's ability to express advertising features. At the same time, by adopting a correlation-based partitioning method and constructing an adjacency matrix, the complex relationship between advertising features can be captured more accurately, further improving the performance of intelligent analysis of advertising content. The application of these technologies enables the advertising content intelligent analysis system to achieve higher classification accuracy and faster processing speed when processing large-scale and diverse advertising data. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood by reading the detailed description below with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, in which: Figure 1 A schematic diagram of a process flow of an advertising content intelligent analysis system based on machine learning provided by an embodiment of the present invention; Figure 2 A schematic diagram of the structure of an advertising content intelligent analysis device based on machine learning provided by an embodiment of the present invention; Figure 3 The schematic diagram schematically shows the structure of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0020] Those skilled in the art know that the embodiments of the present invention can be implemented as a system, device, apparatus, method or computer program product. Therefore, the present invention can be specifically implemented in the following forms, namely: complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0021] It should be noted that any number of elements in the drawings is for illustration rather than limitation, and any naming is only for distinction and does not have any limiting meaning.

[0022] Reference below Figure 1 , Figure 1 The following is a flow chart of an advertising content intelligent analysis system based on machine learning provided by an embodiment of the present invention. Figure 1 As shown, an advertisement content intelligent analysis system 100 based on machine learning includes: Step 101, obtaining advertisement content data, and constructing an initial training sample set for an advertisement content intelligent analysis system based on the advertisement content data; the initial sample set includes advertisement feature data and ad category labels ; Label the advertising categories in the initial sample set Perform label smoothing to obtain the final training sample set; Step 102: Building an advertising content intelligent analysis model based on a deep learning network ; Based on the advertising content intelligent analysis training sample set, the advertising content intelligent analysis model Conduct training to obtain a trained advertising content intelligent analysis model ; Step 103: Extracting feature data of the advertisement content to be analyzed ; The advertising feature data Input the trained advertising content intelligent analysis model In the process of advertising content classification analysis,

[0023] It should be noted that this embodiment relates to an advertising content intelligent analysis system based on machine learning, which implements intelligent analysis of advertising content by executing a series of steps, including obtaining advertising content data, constructing an initial training sample set, and performing label smoothing on advertising category labels to form a final training sample set.

[0024] Specifically, obtaining advertising content data involves collecting various forms of advertisements, such as videos, images, texts, etc. When constructing the initial training sample set, it is necessary to extract advertising feature data and advertising category labels. Advertising feature data may include but is not limited to the visual elements, text content, audio information, etc. of the advertisement. Advertising category labels are labels for classifying advertising content, such as food advertisements, car advertisements, etc.

[0025] Preferably, when performing label smoothing on the advertising category labels in the initial sample set, a specific algorithm can be used to adjust the distribution of the labels to reduce the model's bias towards minority categories. For example, a label smoothing technique with a temperature parameter of 2 can be used to control the degree of label smoothing by adjusting the smoothing parameter, thereby improving the generalization ability of the model.

[0026] Furthermore, the present embodiment also includes constructing an advertising content intelligent analysis model based on a deep learning network. The model is trained through a training sample set to identify and classify advertising content.

[0027] Specifically, the deep learning network may include multiple sequentially connected convolutional layers and fully connected layers. The convolutional layers are used to extract features from the ad feature map, while the fully connected layers are used to classify these features. Each convolutional layer can be configured with different convolution kernel sizes and numbers to meet different feature extraction requirements.

[0028] More specifically, when building a deep learning network, you can use a pre-trained model as a basis, such as ResNet or VGG, and then fine-tune it according to the characteristics of the advertising content. In addition, you can also introduce dropout technology to prevent overfitting and improve the generalization ability of the model.

[0029] Furthermore, the present embodiment also involves extracting feature data of the advertisement content to be analyzed, and inputting the feature data into a trained advertisement content intelligent analysis model to perform category analysis.

[0030] Specifically, extracting feature data of the advertisement content to be analyzed includes preprocessing the advertisement content, such as resizing, normalization, etc., and feature extraction, such as extracting features using a convolutional layer in a deep learning network.

[0031] Preferably, before inputting the advertisement feature data into the model, further data enhancement processing can be performed, such as rotation, scaling, cropping, etc., to improve the adaptability of the model to different advertisement forms. In addition, a multi-task learning strategy can be adopted to simultaneously train the model for advertisement classification and other related tasks, such as sentiment analysis, to further improve the performance of the model.

[0032] In some embodiments, the advertisement category labels in the initial sample set are Perform label smoothing to obtain the final training sample set, including: all advertising category labels in the sample set Perform overall smoothing; Determine an advertisement content conversion frame in the sample set, and perform intra-group advertisement category label smoothing on a group of contents before the advertisement content conversion frame; The advertisement characteristic data Includes confidence in ad features ; Confidence based on ad features Ad category label for each content Perform confidence smoothing.

[0033] It should be noted that this embodiment describes in detail the specific steps of label smoothing of the advertisement category labels in the initial sample set to obtain the final training sample set. This includes overall smoothing of all advertisement category labels in the sample set, and determining the advertisement content conversion frame in the sample set, and performing intra-group advertisement category label smoothing on a group of content before the advertisement content conversion frame.

[0034] Specifically, overall smoothing can reduce the model's bias towards minority categories by adjusting the distribution of labels. For example, a label smoothing technique with a temperature parameter of 2 can be used to control the degree of label smoothing by adjusting the smoothing parameter. Intra-group smoothing targets a group of content before the advertisement content conversion frame, and determines the active index and target index of the content group before conversion based on the label value of the previous frame content and the label value of the current frame.

[0035] Preferably, intra-group smoothing can be achieved by calculating the label values ​​corresponding to the active index and the target index. For example, for each advertisement content conversion frame, the label value of the content group before conversion can be calculated using a specific formula based on the label value of the previous frame content and the label value of the current frame.

[0036] Furthermore, the advertisement feature data includes the confidence of the advertisement feature, and the advertisement category label of each content is confidence smoothed based on the confidence of the advertisement feature. Confidence smoothing can be performed by calculating the confidence of the main advertisement feature of each frame content, setting it to 1, and then calculating the mean of the confidence of all advertisement features. This mean is then multiplied by the advertisement category label of the frame content to obtain a smoothed label based on the confidence.

[0037] Specifically, different confidence calculation methods can be used in the confidence smoothing process, such as weighted average or median, to adapt to different advertising content characteristics. In addition, a confidence threshold can be introduced, and only when the confidence of the feature exceeds this threshold, it will be included in the smoothing calculation to improve the accuracy of the smoothing effect.

[0038] More specifically, this embodiment also includes using a specific formula to perform overall smoothing on all advertisement category labels in the sample set. Overall smoothing can be achieved by the following formula:

[0039] in, represents the sample ad category label, represents the number of categories, represents the smoothing parameter.

[0040] Preferably, the smoothing parameter You can adjust this to achieve the best smoothing effect based on the characteristics of your dataset. For example, if the distribution of classes in your dataset is very uneven, you can increase In addition, adaptive smoothing technology can be used to dynamically adjust the The value of .

[0041] In some embodiments, the following formula is used to classify all advertisement category labels in the sample set: Perform overall smoothing:

[0042] in, represents the sample ad category label, represents the number of categories, represents the smoothing parameter.

[0043] It should be noted that this embodiment describes in detail the specific operation of overall smoothing of all advertisement category labels in the sample set, which is achieved by adopting a specific formula. The purpose of overall smoothing is to reduce the model's excessive reliance on any single category, thereby improving the model's generalization ability.

[0044] Specifically, the overall smoothing operation includes each advertising category label in the sample set Apply the following formula:

[0045] In this formula, represents the sample ad category label, represents the number of categories, represents the smoothing parameter. This formula is obtained by A small amount of noise is added to achieve smoothing. The size of the noise is determined by control, and This ensures that label smoothing is uniform in the multi-class case.

[0046] Preferably, the smoothing parameter The choice of can be adjusted according to the characteristics of the dataset and the performance requirements of the model. For example, if some categories in the dataset have very few samples, a larger value to increase the label weights of these categories, thereby improving the model's ability to identify these minority categories. In addition, The value of can also be determined by cross-validation to find the best smoothing effect.

[0047] Specifically, this embodiment also includes smoothing the intra-group advertisement category labels for a group of content before the advertisement content conversion frame in the sample set. The intra-group smoothing operation involves determining the label value of the content of the frame before the advertisement content conversion frame and the label value of the advertisement content conversion frame, and then determining the active index and target index of the content group before the conversion based on these label values. This step is to capture the dynamic changes of the advertisement content in the time series, so as to more accurately reflect the characteristics of the advertisement content.

[0048] Preferably, intra-group smoothing can be achieved by calculating the label values ​​corresponding to the active index and the target index. For example, for each advertising content conversion frame, a specific formula can be used to calculate the label value of the content group before the conversion based on the label value of the previous frame content and the label value of the current frame. This method can better capture the dynamic changes of advertising content in time series, thereby improving the sensitivity of the model to changes in advertising content.

[0049] More specifically, during the intra-group smoothing process, different calculation methods can be used to determine the active index and the target index. For example, the index can be determined by calculating the similarity of the contents of adjacent frames, or a machine learning algorithm can be used to predict which frames have active contents. In addition, a time decay factor can be introduced so that the content farther from the current frame has less influence on the current frame.

[0050] In some embodiments, smoothing the advertisement category label within a group of content before the advertisement content conversion frame includes: for each advertisement content conversion frame, determining an active index of the content group before the conversion according to the label value of the content of the previous frame before the advertisement content conversion frame and the label value of the advertisement content conversion frame and the target index ; According to the formula , calculate the advertising content before switching frames In the content tag The corresponding label value; According to the formula , calculate the advertising content before switching frames In the content tag The corresponding label value; Among them, The frame is the advertisement content conversion frame. Indicates In the ad category label of the frame content The corresponding label value; Indicates In the ad category label of the frame content The corresponding label value, It is The index of the maximum value of the label in the ad category label of the frame content. It is The index of the maximum value of the ad category labels in the frame content.

[0051] It should be noted that this embodiment describes in detail the specific steps of smoothing the intra-group advertisement category labels for a group of content before the advertisement content conversion frame. This includes determining the label value of the content of the previous frame before the advertisement content conversion frame and the label value of the advertisement content conversion frame, and then determining the active index and target index of the content group before the conversion based on these label values.

[0052] Specifically, the intra-group smoothing operation involves determining the active index of the content group before the conversion based on the label value of the content of the previous frame of the advertisement content conversion frame and the label value of the advertisement content conversion frame for each advertisement content conversion frame. and the target index The active index is the index with the largest label value in the content group before the conversion frame, and the target index is the index with the largest label value in the content of the conversion frame.

[0053] Preferably, the intra-group smoothing can be calculated by the following formula before the advertisement content conversion frame: In the content tag The corresponding tag values ​​are:

[0054] And calculate the advertising content before switching frames In the content tag The corresponding tag values ​​are:

[0055] in, It is The index of the maximum value of the label in the ad category label of the frame content. It is The index of the maximum value of the ad category label in the frame content.

[0056] It should be noted that this embodiment also includes calculating the advertisement content conversion frame front In the content tag and The specific formula of the corresponding label value.

[0057] Specifically, the above formula ensures that when the advertisement content switches frames, the smoothing process can take into account the correlation between the previous and next frames, thereby more accurately reflecting the characteristic changes of the advertisement content. This smoothing process helps the model capture the dynamic changes of the advertisement content in the time series and improves the model's sensitivity to changes in the advertisement content.

[0058] More specifically, in the process of intra-group smoothing, a weight adjustment factor can be introduced to optimize the smoothing effect. For example, the weight adjustment factor can be dynamically adjusted according to the conversion strength of the advertising content or the relevance of the content. and In addition, machine learning algorithms can be used to predict which frames have active content, or time series analysis methods can be used to determine active indexes and target indexes, thereby further improving the accuracy and effectiveness of intra-group smoothing.

[0059] In some embodiments, based on the confidence level of the advertisement features Ad category label for each content Confidence smoothing is performed, including: for each frame of content, setting the confidence of the main advertisement feature to 1, and calculating the mean of the confidence of all advertisement features; The mean of the confidences and the advertising category label of the frame content Multiply them together to get a smooth label based on the confidence of the frame content.

[0060] Specifically, the confidence smoothing process includes setting the confidence of the main advertising features to 1 for each frame of content, and then calculating the mean of the confidence of all advertising features. This mean is then multiplied by the advertising category label of the frame content to obtain the confidence-based smoothed label of the frame content. The main advertising features can be those that are most critical to the classification of advertising content. For example, in video ads, the main features may include significant visual elements or specific text information.

[0061] Preferably, confidence smoothing can be achieved by the following steps: first, determine the main advertising features of each frame content and set its confidence to 1; then, calculate the mean confidence of all features; finally, multiply the mean by the advertising category label to obtain a smoothed label. This method can ensure that the model can more accurately reflect its classification when processing advertising content with different confidence features.

[0062] Specifically, this embodiment also includes a specific calculation method for smoothing the advertising category label of each frame of content based on confidence. The calculation of the smoothed label of each frame of content based on confidence can be achieved by the following steps: first, for each frame of content, determine its main advertising feature and set its confidence to 1; then, calculate the mean confidence of all advertising features. Next, the mean The ad category label associated with the frame content Multiply to get the smoothed label The calculation formula can be expressed as:

[0063] in, It is the mean confidence value of all advertising features in the frame content.

[0064] More specifically, more confidence adjustment strategies can be introduced during the confidence smoothing process. For example, different weights can be set for different types of advertising features, rather than simply setting the confidence of the main features to 1. In addition, machine learning algorithms can be used to dynamically adjust the confidence of each frame of content to more accurately reflect the importance of advertising features. These methods can further improve the accuracy and robustness of intelligent analysis of advertising content.

[0065] In some embodiments, an advertising content intelligent analysis model is constructed based on a deep learning network. , including: constructing an advertising feature graph with advertising features as nodes and correlations between advertising features as connections ; Construct a deep learning network comprising multiple sequentially connected convolutional layers and fully connected layers ; Each of the convolutional layers Used to perform convolution on the input features to extract the advertising feature map The fully connected layer Used to classify the output of the convolutional layer.

[0066] It should be noted that this embodiment describes in detail the process of building an intelligent analysis model for advertising content based on a deep learning network. This process includes building an advertising feature graph with advertising features as nodes and correlations between advertising features as connection relationships, and building a deep learning network that includes multiple sequentially connected convolutional layers and fully connected layers.

[0067] Specifically, constructing an ad feature graph involves identifying key features in the ad content and treating these features as nodes. The connections between nodes are based on the correlation between features, which can be determined by statistical methods or machine learning algorithms. The construction of a deep learning network involves designing convolutional layers and fully connected layers, which are responsible for extracting features from the input ad feature graph and classifying them.

[0068] Preferably, the construction of a deep learning network can be achieved through the following steps: first, define the number and parameters of convolutional layers, such as the size and number of convolution kernels; then, define the number of fully connected layers and the number of neurons; then, determine the activation function of the network, such as ReLU or Sigmoid; finally, configure the training parameters of the network, such as learning rate and batch size.

[0069] Specifically, this embodiment also includes the specific operation of convolving the input features to extract the features of the advertisement feature map. The operation of the convolution layer can be implemented through the following steps: first, input the input feature map to the convolution layer; then, slide the convolution kernel on the input feature map and calculate the weighted sum of the local area, add the bias term, and get the preliminary representation of the feature map; then, apply the activation function to introduce nonlinearity and enhance the expression ability of the network; finally, use the pooling layer to reduce the spatial size of the feature map and reduce the amount of calculation.

[0070] More specifically, the design of the convolutional layer can be further refined. For example, different sizes of convolution kernels can be used to capture features of different scales, or dilated convolutions can be used to increase the receptive field without increasing the number of parameters. In addition, batch normalization layers can be introduced to speed up the training process, or dropout layers can be used to reduce overfitting. In the design of the fully connected layer, the complexity of the model can be controlled by adjusting the number of neurons and the number of layers to achieve the best classification performance.

[0071] In some embodiments, the convolutional layer Used to perform convolution on the input features to extract the advertising feature map The invention relates to a method for partitioning the neighborhood of each node in the advertisement feature graph into subsets based on correlation; and constructing an adjacency matrix of each node based on the subsets after partitioning. ; Convolutional Layer According to the formula Perform convolution operation to extract features; in, represents the input features of the convolutional layer, represents the output features of the convolutional layer, represents the convolution kernel weight, represents the bias term, represents the activation function, Represents a convolution operation.

[0072] It should be noted that this embodiment describes in detail the process of using the convolution layer in the deep learning network to convolve the input features to extract the features of the advertising feature graph. This includes using a correlation-based partitioning method to subset the neighborhood of each node in the advertising feature graph, and constructing an adjacency matrix for each node based on the partitioned subsets.

[0073] Specifically, the correlation-based partitioning method involves determining the neighborhoods of nodes based on the strength of correlation between ad features and partitioning these neighborhoods into subsets. An adjacency matrix is ​​a matrix that represents the relationships between nodes in a graph, where the elements of the matrix represent the strength of the connection or correlation between nodes. These steps help the network better understand the relationships between ad features and extract useful features.

[0074] Preferably, the partitioning method can be implemented by calculating the similarity or distance between features, for example, using cosine similarity or Euclidean distance. The construction of the adjacency matrix can be implemented by setting the connection between the features with strong correlation to a higher weight.

[0075] Specifically, this embodiment also includes the specific step of performing a convolution operation on the convolution layer according to a specific formula to extract features. The convolution operation can be implemented by the following formula:

[0076] in, represents the input features of the convolutional layer, represents the output features of the convolutional layer, represents the convolution kernel weight, represents the bias term, represents the activation function, Represents a convolution operation.

[0077] Preferably, the convolution kernel weight and the bias term It can be learned through the back-propagation algorithm during training. The activation function can choose the ReLU function because it is widely used in deep learning networks and can effectively introduce nonlinearity.

[0078] More specifically, the design of the convolutional layer can be further refined. For example, different sizes of convolution kernels can be used to capture features of different scales, or dilated convolutions can be used to increase the receptive field without increasing the number of parameters. In addition, batch normalization layers can be introduced to speed up the training process, or dropout layers can be used to reduce overfitting. These design choices help improve the performance and generalization ability of the network.

[0079] In some embodiments, the convolutional layer It includes a feature enhancement unit for adaptively adjusting the feature importance of each node to other adjacent nodes; The feature enhancement unit includes batch normalization layers connected in sequence , Relu layer , dropout layer , convolutional layer and Sigmoid layer ; The normalization layer Used to make the feature enhancement matrix asymmetric; Layers are used for nonlinear transformations; The convolutional layer is used to prevent overfitting. The convolution kernel is 1×1, which is used to make the enhancement matrix consistent with the dimension of the corresponding convolution layer; The layer is used to map the output to the range [0,1].

[0080] It should be noted that this embodiment describes in detail the implementation of the feature enhancement units contained in the convolution layer, which are used to adaptively adjust the feature importance of each node to other adjacent nodes. The feature enhancement unit is implemented by a series of sequentially connected layers, including a batch normalization layer, a ReLU layer, a dropout layer, a 1x1 convolution layer, and a Sigmoid layer.

[0081] Specifically, the implementation of the feature enhancement unit involves the following steps: 1. Batch normalization layers are used to make the feature augmentation matrix asymmetric, which helps the network learn more complex features.

[0082] 2. The ReLU layer is used for nonlinear transformation to increase the expressive power of the network.

[0083] 3. The dropout layer is used to prevent overfitting and improve the generalization performance of the model by randomly discarding some neurons.

[0084] 4. The convolution kernel of the 1x1 convolutional layer is used to make the augmentation matrix consistent with the corresponding convolutional layer dimension while keeping the number of parameters small.

[0085] 5. The Sigmoid layer is used to map the output results to the range of [0,1], providing a normalized measure of feature importance.

[0086] Preferably, the parameter setting of the feature enhancement unit may include: Parameters of the batch normalization layer, such as the decay rate momentum of the moving average.

[0087] The dropout rate of the dropout layer is usually set between 0.2 and 0.5.

[0088] The number of convolution kernels in the 1x1 convolution layer can be determined based on the dimension of the input features and the required output feature dimension.

[0089] Specifically, the feature enhancement unit can be implemented by sequentially connecting the above layers, and the implementation steps of the feature enhancement unit can be: 1. The input features are first normalized through a batch normalization layer.

[0090] 2. Then the nonlinear activation is performed through the ReLU layer.

[0091] 3. Then randomly discard through the dropout layer.

[0092] 4. Then the dimension is adjusted through a 1x1 convolutional layer.

[0093] 5. Finally, the output is normalized through the Sigmoid layer.

[0094] More specifically, the design of the feature enhancement unit can be further refined or provide alternatives: You can explore different activation functions, such as Leaky ReLU or ELU, to improve the training effect of the network.

[0095] The dropout rate of the dropout layer can be adjusted according to the performance of the model on the validation set to find the optimal anti-overfitting strategy.

[0096] The 1x1 convolutional layers can be replaced by other types of convolutional layers or fully connected layers to explore different feature fusion strategies.

[0097] The Sigmoid layer can be replaced by other normalization techniques, such as Softmax or Tanh, to adapt to different output range requirements.

[0098] The above-mentioned embodiments of the present invention have the following beneficial effects: The present invention can effectively perform category analysis on advertising content by constructing an advertising content intelligent analysis system based on machine learning. The system obtains advertising content data and constructs an initial training sample set, and then performs label smoothing on the advertising category labels to obtain a final training sample set. The advertising content intelligent analysis model constructed using a deep learning network can conduct in-depth learning and understanding of advertising feature data, thereby achieving accurate classification of advertising content. In addition, the system can automatically complete intelligent analysis of advertising content by extracting feature data of the advertising content to be analyzed and inputting it into the trained model, thereby improving the efficiency and accuracy of advertising analysis.

[0099] Furthermore, the present invention also provides a method for intelligent analysis of advertising content, which includes label smoothing of advertising category labels in an initial sample set, and building an advertising content intelligent analysis model based on a deep learning network. This method can effectively handle the dynamic changes of advertising content, especially when the advertising content switches frames, through intra-group advertising category label smoothing and confidence smoothing, it can better adapt to the complexity and diversity of advertising content. Through the application of these technologies, the accuracy and efficiency of intelligent analysis of advertising content can be improved, providing strong technical support for advertising delivery and effect evaluation.

[0100] like Figure 2 As shown, in some embodiments, a device 200 for intelligently analyzing advertisement content based on machine learning includes: The training set construction module 201 is used to obtain advertising content data and construct an initial training sample set for the advertising content intelligent analysis system based on the advertising content data; the initial sample set includes advertising feature data and ad category labels ; Label the advertising categories in the initial sample set Perform label smoothing to obtain the final training sample set; Model training module 202, used to build an advertising content intelligent analysis model based on a deep learning network ; Based on the advertising content intelligent analysis training sample set, the advertising content intelligent analysis model Conduct training to obtain a trained advertising content intelligent analysis model ; Advertisement content intelligent analysis module 203, used to extract feature data of the advertisement content to be analyzed ; The advertising feature data Input the trained advertising content intelligent analysis model In the process of advertising content classification analysis,

[0101] It is understandable that the modules recorded in the machine learning-based advertising content intelligent analysis device 200 and the reference Figure 1 Therefore, the operations, features and beneficial effects described above for the advertising content intelligent analysis system based on machine learning are also applicable to the advertising content intelligent analysis device 200 based on machine learning and the modules contained therein, and will not be repeated here.

[0102] Furthermore, the training set construction module includes: The overall smoothing module is used to smooth all the advertising category labels in the sample set. Perform overall smoothing; An intra-group smoothing module is used to determine an advertisement content conversion frame in a sample set and perform intra-group advertisement category label smoothing on a group of contents before the advertisement content conversion frame; Confidence smoothing module, used for the advertisement feature data Includes confidence in ad features ; Confidence based on ad features Ad category label for each content Perform confidence smoothing.

[0103] Specifically, the training set construction module is responsible for obtaining advertising content data and constructing an initial training sample set based on this data, including advertising feature data and advertising category labels. The model training module constructs an advertising content intelligent analysis model based on a deep learning network and trains the model. The advertising content intelligent analysis module is responsible for extracting the feature data of the advertising content to be analyzed and inputting this data into the trained model for category analysis.

[0104] Preferably, the training set construction module may include the following submodules: a data collection submodule, a feature extraction submodule, a label generation submodule, and a label smoothing submodule. The model training module may include a network architecture design submodule, a training strategy submodule, and a performance evaluation submodule. The advertising content intelligent analysis module may include a feature extraction submodule and a classification prediction submodule.

[0105] The specific implementation of the present embodiment also includes specific implementation steps of the training set construction module, and the implementation steps of the training set construction module may include: 1. The data collection submodule is responsible for collecting advertising content data from different sources.

[0106] 2. The feature extraction submodule is responsible for extracting useful features from the advertising content.

[0107] 3. The label generation submodule is responsible for assigning the correct category label to each advertising content.

[0108] 4. The label smoothing submodule is responsible for smoothing the category labels to improve the generalization ability of the model.

[0109] Preferably, the implementation of the model training module may include the following steps: 1. The network architecture design submodule is responsible for selecting the appropriate deep learning model architecture.

[0110] 2. The training strategy submodule is responsible for setting training parameters such as learning rate, batch size, and number of iterations.

[0111] 3. The performance evaluation submodule is responsible for evaluating the performance of the model using the validation set and making corresponding adjustments.

[0112] More specifically, the implementation of the advertising content intelligent analysis module can be further refined: 1. The feature extraction submodule can use a pre-trained deep learning model to extract high-level features.

[0113] 2. The classification prediction submodule is responsible for inputting the extracted features into the trained model and outputting the prediction results.

[0114] 3. Post-processing steps such as threshold adjustment or ensemble learning can be introduced to improve the final classification accuracy.

[0115] In addition, different feature extraction techniques and model training strategies can be explored to further improve the performance of the advertising content intelligent analysis device. For example, autoencoders or generative adversarial networks (GANs) can be used to enhance the expressiveness of features, or transfer learning strategies can be adopted to utilize the knowledge of pre-trained models.

[0116] Reference below Figure 3 , which shows a schematic diagram of a structure 300 of an electronic device suitable for implementing some embodiments of the present invention. The electronic devices in some embodiments of the present invention may include, but are not limited to, mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 3 The terminal device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0117] like Figure 3 As shown, the electronic device 300 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 to a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic device 300 are also stored. The processing device 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0118] Typically, the following devices may be connected to the I / O interface 305: an input device 306 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 308 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 309. The communication device 309 may allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. Figure 3 The electronic device 300 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 3 Each block shown in the figure may represent one device, or may represent multiple devices as required.

[0119] Furthermore, the storage medium of the embodiment of the present application stores program instructions that can implement all the above methods, wherein the program instructions can be stored in the above storage medium in the form of a software product, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or terminal devices such as a computer, a server, a mobile phone, and a tablet.

[0120] The above descriptions are only some preferred embodiments of the present invention and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present invention is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the embodiments of the present invention.

Claims

1. An advertising content intelligent analysis system based on machine learning, characterized in that: The system performs the following steps: Acquire advertisement content data, and construct an initial training sample set for an advertisement content intelligent analysis system based on the advertisement content data; the initial sample set includes advertisement feature data and ad category labels ; The advertisement category labels in the initial sample set Perform label smoothing to obtain the final training sample set; Building an intelligent advertising content analysis model based on deep learning network ; The advertising content intelligent analysis model is trained based on the advertising content intelligent analysis training sample set. Conduct training to obtain a trained advertising content intelligent analysis model ; Extract feature data of the advertising content to be analyzed ; The advertising feature data Input the trained advertising content intelligent analysis model In the process of advertising content classification analysis, 2. The machine learning-based advertising content intelligent analysis system according to claim 1, characterized in that: The advertisement category labels in the initial sample set Perform label smoothing to obtain the final training sample set, including: all advertising category labels in the sample set Perform overall smoothing; Determine an advertisement content conversion frame in the sample set, and perform intra-group advertisement category label smoothing on a group of contents before the advertisement content conversion frame; The advertisement characteristic data Includes confidence in ad features ; Confidence based on ad features Ad category label for each content Perform confidence smoothing.

3. The advertising content intelligent analysis system based on machine learning according to claim 2 is characterized in that: Use the following formula to classify all advertisements in the sample set: Perform overall smoothing: in, represents the sample ad category label, represents the number of categories, represents the smoothing parameter.

4. The machine learning-based advertising content intelligent analysis system according to claim 2, characterized in that: Performing intra-group advertisement category label smoothing on a group of content before the advertisement content conversion frame, including: for each advertisement content conversion frame, determining an active index of the content group before the conversion according to a label value of a frame of content before the advertisement content conversion frame and a label value of the advertisement content conversion frame and the target index ; According to the formula , calculate the advertising content before switching frames In the content tag The corresponding label value; According to the formula , calculate the advertising content before switching frames In the content tag The corresponding label value; Among them, The frame is the advertisement content conversion frame. Indicates In the ad category label of the frame content The corresponding label value; Indicates In the ad category label of the frame content The corresponding label value, It is The index of the maximum value of the label in the ad category label of the frame content. It is The index of the maximum value of the ad category labels in the frame content.

5. The machine learning-based advertising content intelligent analysis system according to claim 2, characterized in that: Confidence based on ad features Ad category label for each content Confidence smoothing is performed, including: for each frame of content, setting the confidence of the main advertisement feature to 1, and calculating the mean of the confidence of all advertisement features; The mean of the confidences and the advertising category label of the frame content Multiply them together to get a smooth label based on the confidence of the frame content.

6. The advertising content intelligent analysis system based on machine learning according to claim 1, characterized in that: Building an intelligent advertising content analysis model based on deep learning network , including: constructing an advertising feature graph with advertising features as nodes and correlations between advertising features as connections ; Construct a deep learning network comprising multiple sequentially connected convolutional layers and fully connected layers ; Each of the convolutional layers Used to perform convolution on the input features to extract the advertising feature map The fully connected layer Used to classify the output of the convolutional layer.

7. The machine learning-based advertising content intelligent analysis system according to claim 6, characterized in that: The convolutional layer Used to perform convolution on the input features to extract the advertising feature map The invention relates to a method for partitioning the neighborhood of each node in the advertisement feature graph into subsets based on correlation; and constructing an adjacency matrix of each node based on the subsets after the partitioning. ; Convolutional Layer According to the formula Perform convolution operation to extract features; in, represents the input features of the convolutional layer, represents the output features of the convolutional layer, represents the convolution kernel weight, represents the bias term, represents the activation function, Represents a convolution operation.

8. The machine learning-based advertising content intelligent analysis system according to claim 7, characterized in that: The convolutional layer It includes a feature enhancement unit for adaptively adjusting the feature importance of each node to other adjacent nodes; The feature enhancement unit includes batch normalization layers connected in sequence , Relu layer , dropout layer , convolutional layer and Sigmoid layer ; The normalization layer Used to make the feature enhancement matrix asymmetric; Layers are used for nonlinear transformations; The convolutional layer is used to prevent overfitting. The convolution kernel is 1×1, which is used to make the enhancement matrix consistent with the dimension of the corresponding convolution layer; The layer is used to map the output to the range [0,1].

9. An intelligent advertising content analysis device based on machine learning, characterized in that: Includes the following modules: The training set construction module is used to obtain advertising content data and construct an initial training sample set for the advertising content intelligent analysis system based on the advertising content data; the initial sample set includes advertising feature data and ad category labels ; The advertisement category labels in the initial sample set Perform label smoothing to obtain the final training sample set; Model training module, used to build an intelligent analysis model for advertising content based on deep learning network ; Based on the advertising content intelligent analysis training sample set, the advertising content intelligent analysis model Conduct training to obtain a trained advertising content intelligent analysis model ; Advertisement content intelligent analysis module, used to extract characteristic data of the advertisement content to be analyzed ; The advertising feature data Input the trained advertising content intelligent analysis model In the process of advertising content classification analysis, 10. The device for intelligently analyzing advertisement content based on machine learning according to claim 9, characterized in that: The training set construction module includes: an overall smoothing module for smoothing all advertisement category labels in the sample set. Perform overall smoothing; An intra-group smoothing module is used to determine an advertisement content conversion frame in a sample set and perform intra-group advertisement category label smoothing on a group of contents before the advertisement content conversion frame; Confidence smoothing module, used for the advertisement feature data Includes confidence in ad features ; Confidence based on ad features Ad category label for each content Perform confidence smoothing.

Images