Advertisement screening method and system based on deep learning

Abstract

The present application relates to an advertisement screening method and system based on deep learning. The method includes: obtaining the time series of user behavior data of the target user; understanding the user behavior pattern of the time series of the user behavior data to obtain the user behavior pattern semantic encoding feature vector; obtaining the text description of the advertisement to be delivered; generating an ideal user behavior pattern inference representation vector based on the text description of the advertisement to be delivered; determining whether to deliver the advertisement to be delivered to the target user based on the semantic distance between the user behavior pattern semantic encoding feature vector and the ideal user behavior pattern inference representation vector, and intelligently determining whether to deliver the advertisement to be delivered to the target user to achieve intelligent screening and precise delivery of advertisements.

Description

Technical Field

[0001] The present application relates to the technical field of advertisement screening, and in particular, to an advertisement screening method and system based on deep learning. Background Art

[0002] With the advent of the digital age, the advertising industry is undergoing unprecedented changes. Traditional advertising methods, such as radio, television, and newspapers, have gradually been replaced by Internet advertising. However, with the massive amount of information on Internet advertising and the growth of user personalized needs, how to accurately place advertisements has become a challenge facing advertisers.

[0003] In daily life, users are exposed to a huge amount of advertising information every day, which leads to the distraction of users' attention to advertisements. Advertisers need to stand out from the crowd and attract users' attention. But more importantly, each user's interests and preferences are unique. Advertisers need to provide personalized advertising content to attract audience users. Traditional advertising often adopts a one-size-fits-all approach, ignoring individual differences among users.

[0004] Therefore, an optimized solution is needed to intelligently determine whether to deliver the advertisements to the target users, so as to achieve intelligent screening and precise delivery of advertisements. Summary of the invention

[0005] This summary is provided to introduce concepts in a brief form that will be described in detail in the detailed description below. This summary is not intended to identify key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0006] In a first aspect, the present application provides an advertisement screening method based on deep learning, the method comprising:

[0007] Obtain the time series of user behavior data of target users;

[0008] Performing user behavior pattern understanding on the time series of the user behavior data to obtain a user behavior pattern semantic encoding feature vector;

[0009] Get the text description of the advertisement to be placed;

[0010] Based on the text description of the advertisement to be placed, generating an ideal user behavior pattern reasoning representation vector;

[0011] Determining whether to deliver the advertisement to the target user based on the semantic distance between the user behavior pattern semantic encoding feature vector and the ideal user behavior pattern inference representation vector;

[0012] Among them, the user behavior pattern is understood on the time series of the user behavior data to obtain the user behavior pattern semantic coding feature vector, including: embedding coding each user behavior data in the time series of the user behavior data to obtain a sequence of user behavior data embedded coding vectors; inputting the sequence of user behavior data embedded coding vectors into a user behavior pattern understander to obtain the user behavior pattern semantic coding feature vector.

[0013] Optionally, the embedded coding includes: word embedding, contextual embedding or one-hot encoding.

[0014] Optionally, the user behavior pattern understander is a user behavior pattern understander based on an RNN model.

[0015] Optionally, based on the text description of the advertisement to be placed, an ideal user behavior pattern inference representation vector is generated, including: performing word segmentation on the text description of the advertisement to be placed and then passing it through an advertisement text semantic encoder including a word embedding layer to obtain a semantic understanding feature vector of the advertisement to be placed; passing the semantic understanding feature vector of the advertisement to be placed through an ideal user behavior simulation generator based on a generative adversarial network model to obtain the ideal user behavior pattern inference representation vector.

[0016] Optionally, after word segmentation processing is performed on the text description of the advertisement to be placed, a semantic understanding feature vector of the advertisement to be placed is obtained by using an advertisement text semantic encoder including a word embedding layer, including: word segmentation processing is performed on the text description of the advertisement to be placed to convert the text description of the advertisement to be placed into a word sequence composed of multiple words; using the word embedding layer of the advertisement text semantic encoder including the word embedding layer to map each word in the word sequence to a word vector to obtain a sequence of word vectors; using the advertisement text semantic encoder including the word embedding layer to perform global contextual semantic encoding on the sequence of word vectors to obtain the semantic understanding feature vector of the advertisement to be placed.

[0017] Optionally, the advertisement text semantic encoder including the word embedding layer is used to perform global context semantic encoding on the sequence of word vectors to obtain the semantic understanding feature vector of the advertisement to be placed, including: using the advertisement text semantic encoder including the word embedding layer to perform global context semantic encoding on the sequence of word vectors to obtain multiple feature vectors of the advertisement to be placed; and cascading the multiple feature vectors of the advertisement to be placed to obtain the semantic understanding feature vector of the advertisement to be placed.

[0018] Optionally, based on the semantic distance between the user behavior pattern semantic encoding feature vector and the ideal user behavior pattern inference representation vector, determining whether to deliver the advertisement to be delivered to the target user includes: calculating the Hamming distance between the ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector as the semantic distance; and determining whether to deliver the advertisement to be delivered to the target user based on a comparison between the Hamming distance and a preset threshold.

[0019] Optionally, the Hamming distance between the ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector is calculated as the semantic distance, including: calculating the positional XOR of the ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector to obtain a sequence of user behavior pattern semantic distance quantization values; taking the sum of the sequence of user behavior pattern semantic distance quantization values ​​to obtain the Hamming distance.

[0020] In a second aspect, the present application provides an advertisement screening system based on deep learning, the system comprising:

[0021] A user behavior data acquisition module, used to acquire the time series of user behavior data of target users;

[0022] A user behavior pattern understanding module, used to understand the user behavior pattern of the time series of the user behavior data to obtain a user behavior pattern semantic encoding feature vector;

[0023] A text description module, used to obtain a text description of the advertisement to be placed;

[0024] An ideal user behavior generation module, used to generate an ideal user behavior pattern reasoning representation vector based on the text description of the advertisement to be placed;

[0025] It also includes an advertisement placement judgment module, which is used to determine whether to place the advertisement to be placed to the target user based on the semantic distance between the user behavior pattern semantic encoding feature vector and the ideal user behavior pattern inference representation vector.

[0026] The present application has at least the following technical effects: compared with the prior art, the above technical solution is adopted to obtain the time series of user behavior data of the target user; understand the user behavior pattern of the time series of the user behavior data to obtain the user behavior pattern semantic encoding feature vector; obtain the text description of the advertisement to be delivered; based on the text description of the advertisement to be delivered, generate an ideal user behavior pattern inference representation vector; based on the semantic distance between the user behavior pattern semantic encoding feature vector and the ideal user behavior pattern inference representation vector, determine whether to deliver the advertisement to be delivered to the target user, and it can be intelligently determined whether to deliver the advertisement to be delivered to the target user, so as to realize intelligent screening and precise delivery of advertisements.

[0027] Other features and advantages of the present application will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The above and other features, advantages and aspects of the embodiments of the present application will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale. In the drawings:

[0029] Figure 1 The present invention is a flowchart of an advertisement screening method based on deep learning according to an exemplary embodiment.

[0030] Figure 2 It is a block diagram of an advertisement screening system based on deep learning according to an exemplary embodiment. DETAILED DESCRIPTION

[0031] The embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be construed as being limited to the embodiments described herein. Instead, these embodiments are provided to provide a more thorough and complete understanding of the present application. It should be understood that the drawings and embodiments of the present application are only for exemplary purposes and are not intended to limit the scope of protection of the present application.

[0032] It should be understood that the various steps described in the method implementation of the present application can be performed in different orders and / or performed in parallel. In addition, the method implementation may include additional steps and / or omit the steps shown. The scope of the present application is not limited in this respect.

[0033] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0034] It should be noted that the concepts such as "first" and "second" mentioned in this application are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0035] It should be noted that the modifications of "one" and "plurality" mentioned in the present application are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0036] The names of the messages or information exchanged between multiple devices in the embodiments of the present application are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0037] Although traditional advertising methods are still effective in some cases, they have some obvious flaws in the context of the digital age: traditional advertising usually adopts a unified message and format and cannot be customized to the specific needs and preferences of different user groups; traditional advertising often has difficulty in accurately identifying the target audience, resulting in advertising resources being wasted on non-target users; traditional advertising is usually one-way communication and lacks interaction with users, which limits the attractiveness of advertising and user engagement; the effect of traditional advertising is difficult to quantify, and it is difficult for advertisers to accurately evaluate the ROI (return on investment) of advertising; the production and delivery cycle of traditional advertising is long, and it is difficult to respond quickly Market changes or taking advantage of timely topics; traditional advertising, such as television and radio advertising, is often costly and may not necessarily bring returns that match it; traditional advertising is usually only spread through a single channel, which limits the coverage and influence of advertising; users may become tired of traditional advertising forms, resulting in a decline in advertising effectiveness; traditional advertising often ignores emerging social media and other digital platforms, missing the opportunity to reach young user groups; the format and length of traditional advertising limit the creativity and make it difficult to produce more attractive and memorable advertising content; some traditional advertising forms, such as paper flyers, may have a negative impact on the environment. With the development of technology and changes in consumer behavior, the advertising industry is gradually transforming to more personalized, interactive and cost-effective digital advertising to overcome these shortcomings and improve advertising effectiveness.

[0038] In order to solve the above problems, the present application provides an advertisement screening method and system based on deep learning, which obtains the time series of user behavior data of the target user; understands the user behavior pattern of the time series of the user behavior data to obtain a user behavior pattern semantic encoding feature vector; obtains a text description of the advertisement to be placed; based on the text description of the advertisement to be placed, generates an ideal user behavior pattern inference representation vector; based on the semantic distance between the user behavior pattern semantic encoding feature vector and the ideal user behavior pattern inference representation vector, determines whether to place the advertisement to be placed to the target user, and can intelligently judge whether to place the advertisement to be placed to the target user, so as to realize intelligent screening and precise delivery of advertisements.

[0039] The specific implementation of the present application is described in detail below with reference to the accompanying drawings.

[0040] Figure 1 is a flowchart of an advertisement screening method based on deep learning according to an exemplary embodiment. Figure 1 As shown, the method includes:

[0041] Step 101: Obtain the time series of user behavior data of the target user;

[0042] Step 102: understanding the user behavior pattern of the time series of the user behavior data to obtain a user behavior pattern semantic encoding feature vector;

[0043] Step 103: Obtain a text description of the advertisement to be placed;

[0044] Step 104: Generate an ideal user behavior pattern reasoning representation vector based on the text description of the advertisement to be placed;

[0045] Step 105: Determine whether to deliver the advertisement to the target user based on the semantic distance between the user behavior pattern semantic encoding feature vector and the ideal user behavior pattern inference representation vector.

[0046] In response to the above technical problems, the technical concept of the present application is: to use deep learning algorithms to intelligently deduce the behavior patterns and character portraits of ideal users from the text descriptions of the advertisements to be placed, and at the same time analyze the complex behavior patterns of the target users, and to determine whether to place the advertisements to be placed to the target users by quantitatively measuring the behavioral semantic distance between the behavior patterns of the ideal users and the behavior patterns of the target users, thereby achieving intelligent screening and precise placement of advertisements.

[0047] Based on this, in the technical solution of the present application, first, the time series of user behavior data of the target user is obtained. In actual application scenarios, the user behavior data may include but is not limited to search records, browsing history, click behavior, purchase records, etc. These data are the key to understanding user needs and preferences. Specifically, search records reflect the user's interest preferences or current needs; browsing history can reveal the user's attention to specific content; the user's click behavior on advertisements or links is a direct indicator of their interest in specific content, reflecting the reaction, willingness and participation in advertisements; and the purchase history is the most direct manifestation of user preferences and needs, from which the user's brand preferences, purchase frequency and consumption capacity can be understood. By obtaining these data, it is helpful to understand the user's complex behavior patterns and real consumption needs, thereby guiding the precise delivery of advertisements. It should be noted that in the process of obtaining and collecting the above data, it is based on respecting the user's wishes and right to know, and the user's privacy will be protected in the process of using the data.

[0048] Then, each user behavior data in the time series of the user behavior data is embedded and encoded to obtain a sequence of user behavior data embedded coding vectors; and the sequence of user behavior data embedded coding vectors is input into a user behavior pattern understander based on the RNN model to obtain a user behavior pattern semantic coding feature vector. Specifically, each user behavior data is converted into a unified vector representation through embedding coding methods such as word embedding to represent the semantic information of user behavior. The recurrent neural network (RNN) model is then used to construct a user behavior pattern understander to capture the time dependency and temporal dynamic characteristics of user behavior in the sequence of user behavior data embedded coding vectors, thereby modeling the user's complex behavior pattern.

[0049] In one embodiment of the present application, user behavior pattern understanding is performed on the time series of the user behavior data to obtain a user behavior pattern semantic coding feature vector, including: embedding and encoding each user behavior data in the time series of the user behavior data to obtain a sequence of user behavior data embedded coding vectors; inputting the sequence of user behavior data embedded coding vectors into a user behavior pattern understander to obtain the user behavior pattern semantic coding feature vector.

[0050] The embedded coding includes: word embedding, contextual embedding or one-hot encoding.

[0051] Among them, the user behavior pattern understander is a user behavior pattern understander based on the RNN model.

[0052] At the same time, a text description of the advertisement to be placed is obtained. The text description of the advertisement to be placed is a textual representation of the content of the advertisement to be placed, and is the core of the advertisement content. Specifically, the text description may include the core content, main information, promotion purpose, emotional tendency and shooting style of the advertisement to be placed. Next, the text description of the advertisement to be placed is processed by word segmentation and then passed through an advertisement text semantic encoder including a word embedding layer to obtain a semantic understanding feature vector of the advertisement to be placed. Specifically, continuous text characters are divided into meaningful units, namely words or phrases, through word segmentation; then, the words or phrases are converted into structured vector representations using the word embedding layer, so that the model can capture the semantic information and contextual associations contained therein; then, the advertisement text semantic encoder is used to perform contextual semantic understanding and analysis on the vectorized representation of the words or phrases, thereby learning the text expression and semantic meaning of the text description of the advertisement to be placed.

[0053] Furthermore, the semantic understanding feature vector of the advertisement to be delivered is passed through an ideal user behavior simulation generator based on a generative adversarial network model to obtain an ideal user behavior pattern inference representation vector. Here, the generative adversarial network (GAN) model consists of a generator and a discriminator. The discriminator can learn the complex distribution patterns of real data samples during the training process, and the generator can generate samples close to the real data distribution in the inference stage. In the technical solution of the present application, the ideal user behavior simulation generator constructed using a generative adversarial network can convert the high-dimensional implicit expression of the ideal user behavior pattern in the semantic understanding feature vector of the advertisement to be delivered into a user portrait close to the real one, so as to simulate the audience users of the advertisement to be delivered.

[0054] In one embodiment of the present application, based on the text description of the advertisement to be placed, an ideal user behavior pattern inference representation vector is generated, including: performing word segmentation on the text description of the advertisement to be placed and then passing it through an advertisement text semantic encoder including a word embedding layer to obtain a semantic understanding feature vector of the advertisement to be placed; passing the semantic understanding feature vector of the advertisement to be placed through an ideal user behavior simulation generator based on a generative adversarial network model to obtain the ideal user behavior pattern inference representation vector.

[0055] Among them, after the text description of the advertisement to be placed is processed by word segmentation, the advertisement text semantic encoder including a word embedding layer is used to obtain the semantic understanding feature vector of the advertisement to be placed, including: performing word segmentation on the text description of the advertisement to be placed to convert the text description of the advertisement to be placed into a word sequence composed of multiple words; using the word embedding layer of the advertisement text semantic encoder including the word embedding layer to map each word in the word sequence to a word vector to obtain a sequence of word vectors; using the advertisement text semantic encoder including the word embedding layer to perform global contextual semantic encoding on the sequence of word vectors to obtain the semantic understanding feature vector of the advertisement to be placed.

[0056] Furthermore, the advertisement text semantic encoder including the word embedding layer is used to perform global context semantic encoding on the sequence of word vectors to obtain the semantic understanding feature vector of the advertisement to be placed, including: using the advertisement text semantic encoder including the word embedding layer to perform global context semantic encoding on the sequence of word vectors to obtain multiple feature vectors of the advertisement to be placed; and cascading the multiple feature vectors of the advertisement to be placed to obtain the semantic understanding feature vector of the advertisement to be placed.

[0057] Subsequently, the Hamming distance between the inference representation vector of the ideal user behavior pattern and the semantic encoding feature vector of the user behavior pattern is calculated. Here, the Hamming distance is usually used to measure the semantic difference between two vectors. In the technical solution of the present application, the calculation of the Hamming distance can be used to evaluate the semantic distance and semantic similarity between the ideal user behavior pattern and the actual behavior pattern of the target user, thereby achieving a quick comparison. Then, the Hamming distance is used as a decision basis to determine whether to deliver the advertisement to the target user based on the comparison between the Hamming distance and the preset threshold, thereby achieving accurate delivery of advertisements.

[0058] In one embodiment of the present application, based on the semantic distance between the user behavior pattern semantic encoding feature vector and the ideal user behavior pattern inference representation vector, determining whether to deliver the advertisement to be delivered to the target user includes: calculating the Hamming distance between the ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector as the semantic distance; and determining whether to deliver the advertisement to be delivered to the target user based on a comparison between the Hamming distance and a preset threshold.

[0059] Among them, calculating the Hamming distance between the ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector as the semantic distance includes: calculating the positional XOR of the ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector to obtain a sequence of user behavior pattern semantic distance quantization values; taking the sum of the sequence of user behavior pattern semantic distance quantization values ​​to obtain the Hamming distance.

[0060] Specifically, the Hamming distance between the ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector is calculated as the semantic distance using the following distance formula; wherein the distance formula is:

[0061] in, The first inference vector for the ideal user behavior pattern is The eigenvalue at position, The semantic encoding feature vector of the user behavior pattern is The eigenvalue at position, Indicates XOR processing, Infer the length of the representation vector for the ideal user behavior pattern, is the Hamming distance.

[0062] In the technical solution of the present application, the semantic understanding feature vector of the advertisement to be placed is used to represent the text semantic features of the advertisement to be placed, and the semantic understanding feature vector of the advertisement to be placed is passed through an ideal user behavior simulation generator based on an adversarial generative network model, which utilizes the adversarial generation mechanism of the ideal user behavior simulation generator based on the adversarial generation network model to infer the ideal user behavior pattern features that match the advertisement to be placed. However, considering that the ideal user behavior pattern inference representation vector is simulated and generated by an adversarial generation mechanism, there is a difference in feature order between it and the real user behavior pattern semantic encoding feature vector, and considering that the data noise implicit in the training data will also be introduced in the adversarial generation process, this makes the ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector have a more significant feature structure heterogeneity, which leads to the influence of the mapping consistency of the ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector to the common semantic Hamming distance representation space when calculating the Hamming distance between the ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector, thereby affecting the calculation accuracy of the Hamming distance between the ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector.

[0063] In a preferred example, calculating the Hamming distance between the ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector includes the following steps: performing feature structure modulation on the ideal user behavior pattern inference representation vector to obtain a corrected ideal user behavior pattern inference representation vector; and calculating the Hamming distance between the corrected ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector.

[0064] The step of performing feature structure modulation on the ideal user behavior pattern inference representation vector to obtain a corrected ideal user behavior pattern inference representation vector includes:

[0065] Performing L random sampling on the ideal user behavior pattern inference representation vector to obtain L user behavior pattern randomly disturbed feature values;

[0066] Arrange the L user behavior pattern random perturbation eigenvalues ​​in order from large to small to obtain a user behavior pattern random perturbation ordered vector;

[0067] Calculating the inner product of the ideal user behavior pattern inference representation vector to obtain the energy modulation feature value of the user behavior pattern;

[0068] Performing semantic intensity scaling on the randomly perturbed ordered vector of the user behavior pattern based on the energy modulation characteristic value of the user behavior pattern to obtain the randomly perturbed ordered vector of the semantically enhanced user behavior pattern;

[0069] Calculating the product of the randomly disturbed ordered vector of the semantically enhanced user behavior pattern and the transposed vector of the ideal user behavior pattern reasoning representation vector to obtain an orthogonal modulation matrix of the best sample prompt of the user behavior pattern;

[0070] The matrix product between the user behavior pattern best example prompt orthogonal modulation matrix and the ideal user behavior pattern inference representation vector is calculated to obtain the corrected ideal user behavior pattern inference representation vector.

[0071] The above optimization process is expressed as follows:

[0072]

[0073]

[0074]

[0075]

[0076]

[0077]

[0078]

[0079]

[0080] in, Indicates The user behavior pattern randomly perturbs the eigenvalue, represents L random samplings, represents the Gaussian density probability function, represents the random perturbation eigenvalue of the user behavior pattern, represents the ideal user behavior pattern inference representation vector, represents the ideal user behavior pattern inference representation vector eigenvalues, It means to calculate the average value of the vector. represents the length of the ideal user behavior pattern inference representation vector, and They represent the mean and variance of the ideal user behavior pattern inference representation vector, respectively, It indicates that the random perturbation eigenvalues ​​of user behavior patterns are arranged in order from large to small. represents the randomly disturbed ordered vector of user behavior patterns, Represents the energy modulation characteristic value of the user behavior pattern, represents the square of the vector's two-norm, It means point multiplication by position. represents matrix multiplication, represents the transpose of a vector, The best example of user behavior pattern prompts the orthogonal modulation matrix, The ideal user behavior pattern inference representation vector representing the correction.

[0081] Accordingly, by randomly sampling the ideal user behavior pattern inference representation vector L times to obtain a set of new feature values ​​based on the original features but containing slight changes, randomness is introduced to explore different aspects of the feature space, and by simulating the noise or variation in the actual data, the model can be better generalized to unseen data. Then, these user behavior pattern random perturbation feature values ​​are arranged in order from large to small to form a user behavior pattern random perturbation ordered vector to identify which perturbation directions may be the most important for the current problem. Such an ordering not only helps the algorithm focus on the feature differences that can best distinguish different categories, but also helps to reveal hidden data structures or patterns. Subsequently, the inner product of the ideal user behavior pattern inference representation vector is calculated to obtain the user behavior pattern energy modulation feature value, which reflects the overall strength or "energy" of the original feature vector and provides a benchmark for measuring the importance of the feature. Using this user behavior pattern energy modulation feature value as a reference point, the previously formed user behavior pattern random perturbation ordered vector is semantically scaled to construct a semantically enhanced user behavior pattern random perturbation ordered vector. In this way, those perturbations that appear to be more important in the original feature space are amplified, enhancing the model's sensitivity to key feature changes while retaining relatively small but potentially still useful perturbation information. Further, the product of the semantically enhanced user behavior pattern random perturbation ordered vector and the transpose of the ideal user behavior pattern inference representation vector is calculated to obtain the user behavior pattern best example prompt orthogonal modulation matrix, which aims to find the way in which features interact from two different perspectives. The resulting matrix can be regarded as a conversion tool for capturing and representing complex linear relationships between features. Finally, the final feature structure modulation is completed by applying the above-constructed user behavior pattern best example prompt orthogonal modulation matrix to the original ideal user behavior pattern inference representation vector, so that the features are more in line with the requirements of the objective function, reducing redundant information while maximizing the retention of useful information, so as to improve learning efficiency and prediction performance.

[0082] In summary, the feature manifold optimization process of the ideal user behavior pattern inference representation vector is clarified, which extracts the best prompt examples based on fine-grained target retrieval with random perturbations, and then guides adversarial robustness learning based on the best prompt examples to complete the structure of the high-dimensional manifold of the feature vector in the feature space, so as to improve the detail and structure of the feature expression. In this way, the feature distribution heterogeneity of the ideal user behavior pattern inference representation vector relative to the user behavior pattern semantic encoding feature vector is improved, so as to improve the mapping consistency of the ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector to the common semantic Hamming distance representation space under cross-sample domains, and improve the calculation accuracy of the Hamming distance between the ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector.

[0083] In summary, the above scheme uses a deep learning algorithm to intelligently deduce the behavior pattern and character portrait of an ideal user from the text description of the advertisement to be placed, and analyzes the complex behavior pattern of the target user at the same time, and determines whether to place the advertisement to be placed to the target user by quantitatively measuring the behavioral semantic distance between the behavior pattern of the ideal user and the behavior pattern of the target user. In this way, intelligent screening and precise placement of advertisements are achieved.

[0084] Figure 2 FIG. 1 is a block diagram of an advertisement screening system based on deep learning according to an exemplary embodiment. Figure 2 As shown, the advertisement screening system 200 includes:

[0085] A user behavior data acquisition module 201 is used to acquire the time series of user behavior data of a target user;

[0086] A user behavior pattern understanding module 202, configured to understand the user behavior pattern of the time series of the user behavior data to obtain a user behavior pattern semantic encoding feature vector;

[0087] A text description module 203 is used to obtain a text description of the advertisement to be placed;

[0088] An ideal user behavior generating module 204, configured to generate an ideal user behavior pattern reasoning representation vector based on the text description of the advertisement to be delivered;

[0089] The advertisement placement determination module 205 is used to determine whether to place the advertisement to be placed to the target user based on the semantic distance between the semantic encoding feature vector of the user behavior pattern and the inference representation vector of the ideal user behavior pattern.

[0090] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present application is not limited to the technical solution 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 disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in this application (but not limited to) by each other to form a technical solution.

[0091] In addition, although each operation is described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or to be performed in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be construed as limiting the scope of the application. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0092] Although the subject matter has been described in language specific to structural features and / or method logic actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing the claims. Regarding the device in the above embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.

Claims

1. The advertisement screening method based on deep learning is characterized by: include: Obtain the time series of user behavior data of target users; Performing user behavior pattern understanding on the time series of the user behavior data to obtain a user behavior pattern semantic encoding feature vector; Get the text description of the advertisement to be delivered; Based on the text description of the advertisement to be placed, generating an ideal user behavior pattern reasoning representation vector; Performing feature structure modulation on the ideal user behavior pattern inference representation vector to obtain a corrected ideal user behavior pattern inference representation vector; Determining whether to deliver the advertisement to the target user based on the semantic distance between the user behavior pattern semantic encoding feature vector and the corrected ideal user behavior pattern reasoning representation vector; The step of understanding the user behavior pattern of the time series of the user behavior data to obtain a user behavior pattern semantic coding feature vector includes: Embedding and encoding each user behavior data in the time series of the user behavior data to obtain a sequence of user behavior data embedded coding vectors; Inputting the sequence of the user behavior data embedded in the coding vector into a user behavior pattern understander to obtain the user behavior pattern semantic coding feature vector; Determining whether to deliver the advertisement to be delivered to the target user based on the semantic distance between the user behavior pattern semantic encoding feature vector and the corrected ideal user behavior pattern inference representation vector includes: Calculating the Hamming distance between the corrected ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector as the semantic distance; Based on the comparison between the Hamming distance and a preset threshold, determining whether to deliver the advertisement to be delivered to the target user; Calculating the Hamming distance between the corrected ideal user behavior pattern reasoning representation vector and the user behavior pattern semantic encoding feature vector as the semantic distance includes: Calculating the positional XOR of the corrected ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector to obtain a sequence of user behavior pattern semantic distance quantization values; Taking the sum of the sequence of quantized values ​​of the semantic distance of the user behavior pattern to obtain the Hamming distance; The step of performing feature structure modulation on the ideal user behavior pattern inference representation vector to obtain a corrected ideal user behavior pattern inference representation vector includes: Performing L random sampling on the ideal user behavior pattern inference representation vector to obtain L user behavior pattern randomly disturbed feature values; Arrange the L user behavior pattern random perturbation eigenvalues ​​in order from large to small to obtain a user behavior pattern random perturbation ordered vector; Calculating the inner product of the ideal user behavior pattern inference representation vector to obtain the energy modulation feature value of the user behavior pattern; Based on the energy modulation characteristic value of the user behavior pattern, the randomly perturbed ordered vector of the user behavior pattern is semantically scaled to obtain the randomly perturbed ordered vector of the semantically enhanced user behavior pattern; Calculating the product of the randomly disturbed ordered vector of the semantically enhanced user behavior pattern and the transposed vector of the ideal user behavior pattern reasoning representation vector to obtain an orthogonal modulation matrix of the best sample prompt of the user behavior pattern; The matrix product between the user behavior pattern best example prompt orthogonal modulation matrix and the ideal user behavior pattern inference representation vector is calculated to obtain the corrected ideal user behavior pattern inference representation vector.

2. The advertisement screening method based on deep learning according to claim 1, characterized in that: The embedded coding includes: word embedding, contextual embedding or one-hot encoding.

3. The advertisement screening method based on deep learning according to claim 2, characterized in that: The user behavior pattern understander is a user behavior pattern understander based on an RNN model.

4. The advertisement screening method based on deep learning according to claim 3, characterized in that: Based on the text description of the advertisement to be placed, generating an ideal user behavior pattern reasoning representation vector, including: After word segmentation processing is performed on the text description of the advertisement to be placed, a semantic understanding feature vector of the advertisement to be placed is obtained by passing the advertisement text semantic encoder including a word embedding layer; The semantic understanding feature vector of the advertisement to be placed is passed through an ideal user behavior simulation generator based on a generative adversarial network model to obtain the ideal user behavior pattern reasoning representation vector.

5. The advertisement screening method based on deep learning according to claim 4, characterized in that: After word segmentation processing is performed on the text description of the advertisement to be placed, a semantic understanding feature vector of the advertisement to be placed is obtained through an advertisement text semantic encoder including a word embedding layer, including: Performing word segmentation processing on the text description of the advertisement to be placed to convert the text description of the advertisement to be placed into a word sequence consisting of multiple words; Using the word embedding layer of the advertisement text semantic encoder including the word embedding layer, mapping each word in the word sequence to a word vector to obtain a sequence of word vectors; The advertisement text semantic encoder including the word embedding layer is used to perform global context semantic encoding on the sequence of word vectors to obtain the semantic understanding feature vector of the advertisement to be delivered.

6. The advertisement screening method based on deep learning according to claim 5, characterized in that: Using the advertisement text semantic encoder including the word embedding layer to perform global context semantic encoding on the sequence of word vectors to obtain the semantic understanding feature vector of the advertisement to be delivered, including: Using the advertisement text semantic encoder including the word embedding layer to perform global context semantic encoding on the sequence of word vectors to obtain a plurality of feature vectors of advertisements to be delivered; The multiple feature vectors of the advertisements to be placed are cascaded to obtain the semantic understanding feature vector of the advertisements to be placed.

7. The advertisement screening system based on deep learning is characterized by: include: A user behavior data acquisition module, used to acquire the time series of user behavior data of target users; A user behavior pattern understanding module, used to understand the user behavior pattern of the time series of the user behavior data to obtain a user behavior pattern semantic encoding feature vector; The step of understanding the user behavior pattern of the time series of the user behavior data to obtain a user behavior pattern semantic coding feature vector includes: Embedding and encoding each user behavior data in the time series of the user behavior data to obtain a sequence of user behavior data embedded coding vectors; Inputting the sequence of the user behavior data embedded in the coding vector into a user behavior pattern understander to obtain the user behavior pattern semantic coding feature vector; A text description module, used to obtain a text description of the advertisement to be placed; An ideal user behavior generation module, used to generate an ideal user behavior pattern reasoning representation vector based on the text description of the advertisement to be placed; Performing feature structure modulation on the ideal user behavior pattern inference representation vector to obtain a corrected ideal user behavior pattern inference representation vector; The system further comprises: An advertisement placement judgment module, configured to determine whether to place the advertisement to be placed to the target user based on the semantic distance between the user behavior pattern semantic encoding feature vector and the corrected ideal user behavior pattern reasoning representation vector; The advertisement placement judgment module is used to: Calculating the Hamming distance between the corrected ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector as the semantic distance; Based on the comparison between the Hamming distance and a preset threshold, determining whether to deliver the advertisement to be delivered to the target user; The advertisement placement judgment module is used to: Calculating the positional XOR of the corrected ideal user behavior pattern inference representation vector and the user behavior pattern semantic encoding feature vector to obtain a sequence of user behavior pattern semantic distance quantization values; Taking the sum of the sequence of quantized values ​​of the semantic distance of the user behavior pattern to obtain the Hamming distance; The step of performing feature structure modulation on the ideal user behavior pattern inference representation vector to obtain a corrected ideal user behavior pattern inference representation vector includes: Performing L random sampling on the ideal user behavior pattern inference representation vector to obtain L user behavior pattern randomly disturbed feature values; Arrange the L user behavior pattern random perturbation eigenvalues ​​in order from large to small to obtain a user behavior pattern random perturbation ordered vector; Calculating the inner product of the ideal user behavior pattern inference representation vector to obtain the energy modulation feature value of the user behavior pattern; Based on the energy modulation characteristic value of the user behavior pattern, the randomly perturbed ordered vector of the user behavior pattern is semantically scaled to obtain the randomly perturbed ordered vector of the semantically enhanced user behavior pattern; Calculating the product of the randomly disturbed ordered vector of the semantically enhanced user behavior pattern and the transposed vector of the ideal user behavior pattern reasoning representation vector to obtain an orthogonal modulation matrix of the best sample prompt of the user behavior pattern; The matrix product between the user behavior pattern best example prompt orthogonal modulation matrix and the ideal user behavior pattern inference representation vector is calculated to obtain the corrected ideal user behavior pattern inference representation vector.

Images