Automatic advertisement recommendation method and device based on artificial intelligence

Abstract

The invention provides an automatic advertisement recommendation method and device based on artificial intelligence, and belongs to the technical field of artificial intelligence, the information is comprehensively learned by analyzing user behavior information such as occupations, education backgrounds and language skills of users and attribute information of advertisements, and behaviors of the users in browsing specific advertisements are used as target variables. Through deep learning of interaction characteristics between the user and the advertisement, a personalized advertisement recommendation technology is realized. Based on the recommendation result, the system can automatically recommend the matched advertisement content to the target user. According to the method, significant progress is achieved in the aspects of improving the accuracy and comprehensiveness of advertisement recommendation. And on the aspect of a sorting mechanism, a deep neural network technology is adopted, so that the transparency of an algorithm is enhanced, a recommendation result has higher interpretability, and the user experience is improved. In addition, the long-term interest of the user on the advertisement is enhanced, and the defects in the prior art are effectively overcome.

Description

Technical Field

[0001] The present application relates to the field of automated advertising push, and specifically to an automated advertising recommendation method and device based on artificial intelligence. Background Art

[0002] With the development of artificial intelligence technology and the gradual promotion and application of Internet applications, the method of online advertising promotion combined with Internet applications has the advantages of wide channels, rich content, and low cost compared to traditional offline advertising. Therefore, the Internet-based online advertising method has been greatly promoted and applied.

[0003] When conducting online advertising promotion on the Internet, it is necessary to predict the click-through rate or conversion rate of different Internet advertisements, so as to determine the delivery platform, delivery target, delivery time period, and actual delivery content of the corresponding Internet advertisements based on the prediction results, thereby improving the benefits brought by advertising delivery.

[0004] Traditionally, when promoting online advertising, the click-through rate or conversion rate of an advertisement is usually determined based on the user's historical click and access records of various Internet advertisements, so as to determine the advertisement with a high degree of matching from the candidate advertisement queue for recommendation. However, since the click and access records of the user are relatively messy, and some online advertisements are missed by the user due to the lack of delivery channels, the error of the determined advertisement click-through rate or conversion rate is relatively large. Therefore, the recommended advertisement determined based on the click-through rate of the advertisement still has the problem of low accuracy. Summary of the invention

[0005] The purpose of the present invention is to overcome the deficiencies in the prior art and to provide an automated advertisement recommendation method and device based on artificial intelligence.

[0006] In a first aspect, the present application provides an automatic advertisement recommendation device based on artificial intelligence, comprising:

[0007] A data acquisition module, used to acquire multiple user behavior data and multiple advertisement attribute data;

[0008] A feature processing module, configured to extract at least one user feature by learning the collected multiple user behavior data, and to extract at least one advertisement feature by learning the collected multiple advertisement attribute data; and to learn the at least one user feature and the extracted at least one advertisement feature together to extract at least one interaction feature related to whether the user has clicked and browsed the multiple advertisements;

[0009] An advertisement recommendation module is used to recommend at least one advertisement based on at least one extracted interaction feature. When the feature processing module learns the at least one extracted user feature and the at least one extracted advertisement feature together, it implements flexible sampling of negative instances of single-point functions; when controlling the negative sampling rate of the single-point function according to the deep learning model to extract at least one interaction feature, it reduces the point-by-point error of the deep learning model.

[0010] Preferably, the data acquisition module collects a plurality of user behavior data including a plurality of user data, a plurality of position data, a plurality of certificate data and a plurality of language data;

[0011] The data acquisition module collects a plurality of advertisement attribute data including a plurality of advertisement content data and a plurality of advertisement category data.

[0012] In a possible implementation, the feature processing module uses a deep learning model to first convert multiple user data, multiple position data, multiple certificate data, and multiple language data included in the multiple user behavior data into a first embedding vector, input the converted first embedding vector into a user behavior network layer, and learn the input first embedding vector to extract at least one user feature;

[0013] The feature processing module converts a plurality of main data included in a plurality of advertisement content data and a plurality of advertisement category data into a second embedding vector, and uses the converted second embedding vector as advertisement attribute data; extracts at least one advertisement feature by inputting at least one advertisement attribute data into an advertisement attribute network layer and learning the input second embedding vector;

[0014] The feature processing module inputs the extracted at least one user feature and the extracted at least one advertisement feature into a deep neural network layer, and learns the input at least one user feature and the input at least one advertisement feature together to extract at least one interaction feature.

[0015] In one possible implementation, the feature processing module uses the user behavior network layer and the advertising attribute network layer as parallel sub-networks, which are connected to a deep neural network layer, and uses the deep neural network layer as a prediction layer to implement user-customized advertising recommendations based on deep learning.

[0016] In a possible implementation, the advertisement recommendation module recommends at least one type of advertisement based on implicit feedback, and the implicit feedback indirectly reflects the extracted at least one interactive feature in the recommendation of the at least one type of advertisement;

[0017] The advertisement recommendation module sorts the at least one category of advertisements, lists the sorted at least one category of advertisements, and provides a recommendation list including the listed at least one category of advertisements;

[0018] The advertisement recommendation module randomly selects at least one advertisement with which the user has not interacted from the at least one type of advertisement and sorts the at least one advertisement.

[0019] In a second aspect, the present invention provides an automatic advertisement recommendation method based on artificial intelligence, comprising:

[0020] Obtain multiple user behavior data and multiple advertising attribute data;

[0021] Extracting at least one user feature by learning the collected multiple user behavior data, and extracting at least one advertisement feature by learning the collected multiple advertisement attribute data; learning the at least one user feature and the extracted at least one advertisement feature together to extract at least one interactive feature related to whether the user has clicked and browsed the multiple advertisements; when learning the extracted at least one user feature and the extracted at least one advertisement feature together, implementing flexible sampling of negative instances of single-point functions; when controlling the negative sampling rate of the single-point function according to the deep learning model to extract at least one interactive feature, reducing the point-by-point error of the deep learning model;

[0022] At least one advertisement is recommended based on the extracted at least one interaction feature.

[0023] In a possible implementation, obtaining a plurality of user behavior data and a plurality of advertisement attribute data includes:

[0024] Collect multiple user behavior data including multiple user data, multiple job data, multiple certificate data and multiple language data; collect multiple advertisement attribute data including multiple advertisement content data and multiple advertisement type data.

[0025] In a possible implementation, at least one user feature is extracted by learning the collected multiple user behavior data, multiple user data, multiple position data, multiple certificate data, and multiple language data included in the multiple user behavior data are converted into a first embedding vector, the converted first embedding vector is input into the user behavior network layer, and at least one user feature is extracted by learning the input first embedding vector;

[0026] Extracting at least one advertisement feature by learning the collected multiple advertisement attribute data, converting multiple main data included in the multiple advertisement content data and the multiple advertisement category data into a second embedding vector, and using the converted second embedding vector as the advertisement attribute data; extracting at least one advertisement feature by inputting the at least one advertisement attribute data into the advertisement attribute network layer and learning the input second embedding vector;

[0027] At least one interaction feature related to whether a user participates in multiple advertisements is extracted by jointly learning at least one extracted user feature and at least one extracted advertisement feature, and the at least one extracted user feature and the at least one extracted advertisement feature are input into a deep neural network layer, and at least one interaction feature related to whether a user participates in multiple advertisements is extracted by jointly learning at least one extracted user feature and at least one input advertisement feature.

[0028] In one possible implementation, at least one category of advertisements is recommended based on implicit feedback, and the implicit feedback indirectly reflects the extracted at least one interactive feature in the recommendation of the at least one category of advertisements; the at least one category of advertisements is sorted, the sorted at least one category of advertisements is listed, and a recommendation list including the listed at least one category of advertisements is provided; and at least one advertisement with which the user has not interacted is randomly selected from the at least one category of advertisements and the at least one advertisement is sorted.

[0029] On the other hand, the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of an automatic advertising recommendation method based on artificial intelligence are implemented.

[0030] The artificial intelligence-based automated advertising recommendation provided by the present invention has the following beneficial effects:

[0031] 1. The present invention solves the cold start problem where suitable advertisements cannot be recommended to new users due to insufficient past information or data, and the first start problem where recommendations cannot be provided before the user's preferences are reflected.

[0032] 2. The present invention can solve the scarcity and scalability problems of high-dimensional data in existing collaborative filtering algorithms and the information overload problem in the field of advertising push by simultaneously considering user behavior information and advertising attribute information.

[0033] 3. Improve the personalization of advertising content: Through in-depth analysis and mining of user behavior data and user information data, the user's personality traits and consumption habits can be accurately portrayed, so that the pushed advertising content is more in line with the user's actual needs and preferences.

[0034] 4. The present invention uses the potential features learned in each layer of the user behavior network layer, the advertising attribute network layer and the deep neural network layer to perform deep learning by using individual users to minimize point-by-point errors, adding individual advertisements as target variables, improving the algorithm's recommendation accuracy and comprehensiveness, using deep neural networks in sorting, increasing algorithm transparency, and improving user experience, thereby making up for the shortcomings of the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A schematic diagram of the structure of an automatic advertisement recommendation device based on artificial intelligence provided in an embodiment of the present application;

[0036] Figure 2 A framework of a deep learning model according to an embodiment of the present invention;

[0037] Figure 3 A flowchart of an artificial intelligence-based automated advertising recommendation method provided in an embodiment of the present application;

[0038] Figure 4 A graph showing evaluation results (hit rate (HR)) of recommended results by an automatic advertisement recommendation device based on artificial intelligence according to an embodiment of the present invention;

[0039] Figure 5 A graph showing evaluation results (normalized cumulative gain (NDCG)) of recommendations of an automatic advertisement recommendation device based on artificial intelligence according to an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The present invention will be further described below in conjunction with the embodiments.

[0041] In the description of the present invention, it should be noted that the embodiments described in the present invention are only part of the embodiments of the present invention, rather than all of the embodiments; based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present invention.

[0042] The terms "first", "second", etc. in the specification and claims of this document and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of this document described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, device, product or equipment that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0043] The artificial intelligence-based automated advertising recommendation provided in the embodiments of the present application specifically involves artificial intelligence technology, which can be applied to various scenarios such as cloud technology, artificial intelligence, smart transportation, network media, and assisted driving. With the research and progress of artificial intelligence technology, artificial intelligence technology has been studied and applied in many fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, automatic driving, drones, robots, smart medical care, smart customer service, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0044] Example 1

[0045] This embodiment provides an automatic advertisement recommendation device based on artificial intelligence, such as Figure 1 As shown, including:

[0046] The data acquisition module is used to acquire multiple user behavior data and multiple advertisement attribute data.

[0047] In a possible implementation, the data acquisition module first collects and stores various user data, advertising content data and context data based on a big data or cloud computing server platform; wherein the user data includes but is not limited to: the user's historical behavior data (wherein the main historical behavior data includes: the user's exposure and click content behavior data in the past 30 days, and the content-related vehicle category, vehicle price and vehicle brand data), interest preference data and personal information data; advertising content data includes but is not limited to: advertising feature data, content data and point display data; context data includes but is not limited to: user's environmental information, such as: device type information, geographic location information and time information, etc.;

[0048] In the embodiment of the present application, by embedding code snippets in the front-end code of the application, the user behavior is tracked and recorded through technologies such as JavaScript. For example, the user's page browsing, click events, etc. are recorded. And the user's behavior data is recorded in the back-end server. This includes information recorded when processing user requests, such as the user's purchase behavior, login information, etc.

[0049] After obtaining user behavior data and user information data, the collected user behavior data and user information data can be stored in a database. Commonly used databases include relational databases (such as MySQL, PostgreSQL) and MySQL databases (such as MongoDB, Cassandra). Alternatively, the user behavior data and user information data can be stored in a data warehouse, such as Hadoop, Spark and other big data platforms to support large-scale data processing and analysis.

[0050] Specifically, the data acquisition module collects multiple user behavior data including multiple user data, multiple position data, multiple certificate data and multiple language data; the data acquisition module collects multiple advertisement attribute data including multiple advertisement content data and multiple advertisement type data.

[0051] The data acquisition module may extract at least one extended word for each element included in the user's personal information. The data acquisition module may use the following formula (1) to calculate the trust score of the online behavior data.

[0052]

[0053] In formula (1), t ab is the trust score of online behavior data, T a is the number of words corresponding to the expanded words in the social media activity records in the online behavior data, T b is the number of extended words extracted from user personal information, T' a and T' b It refers to the number of expanded words extracted from user personal information.

[0054] For example, if T a =10, T b =20, T' a , T' b =50,t ab =0.6, the closer the trust score of the online behavior data is to 1, the higher the reliability and the higher the credibility. It can be inferred that the behavior data is acquired by a single user. Therefore, the data acquisition module acquires online behavior data until the trust score of the online behavior data is calculated to be 0.5 or greater, and controls the further analysis and feature extraction of the online behavior data to be performed only when the trust score is 0.5 or greater, which can improve the accuracy and reliability of extracting personalized advertising content.

[0055] In the data acquisition module, after collecting the aforementioned various user data, advertising content data and context data, the collected data needs to be preprocessed and cleaned; preprocessing and cleaning include but are not limited to: removing noise, processing missing values ​​and processing outliers, etc. The purpose of this is to ensure the quality and availability of the data.

[0056] It should be noted that in the process of obtaining user behavior data and user information data, it is necessary to ensure compliance with relevant privacy policies and laws and regulations to protect user privacy and data security. At the same time, it is also necessary to pay attention to the quality and accuracy of the data to ensure the effectiveness and reliability of data analysis and application.

[0057] A feature processing module, configured to extract at least one user feature by learning the collected multiple user behavior data, and to extract at least one advertisement feature by learning the collected multiple advertisement attribute data; and to learn the at least one user feature and the extracted at least one advertisement feature together to extract at least one interaction feature related to whether the user has clicked and browsed the multiple advertisements;

[0058] According to an embodiment of the present invention, the feature processing module adopts a deep learning model, first converts multiple user data, multiple position data, multiple certificate data, and multiple language data included in the multiple user behavior data into a first embedding vector, inputs the converted first embedding vector into a user behavior network (UBN) layer, and learns the input first embedding vector to extract at least one user feature;

[0059] The feature processing module converts a plurality of main data included in a plurality of advertisement content data and a plurality of advertisement attribute data into a second embedding vector, and converts the converted second embedding vector into advertisement attribute data; extracts at least one advertisement feature by inputting at least one advertisement attribute feature into an advertisement attribute network (GAN) layer and learning the input second embedding vector;

[0060] The feature processing module inputs the extracted at least one user feature and the extracted at least one advertisement feature into a deep neural network (DNN) layer, and learns the input at least one user feature and the input at least one advertisement feature together to extract at least one interaction feature.

[0061] In one possible implementation, the feature processing module uses the user behavior network (UBN) layer and the advertising attribute network (GAN) layer as parallel sub-networks, and the parallel sub-networks are connected to the deep neural network (DNN) layer, and the deep neural network (DNN) layer is used as the prediction layer to achieve user-customized advertising recommendations based on deep learning.

[0062] According to an embodiment of the present invention, the feature processing module uses negative sampling of a single-point function when extracting at least one interactive feature and learning at least one extracted user behavior feature and at least one extracted advertising attribute feature together. By controlling this ratio, the point-by-point error of the deep learning model can be reduced.

[0063] An advertisement recommendation module is used to recommend at least one advertisement based on at least one extracted interaction feature. When the feature processing module learns the at least one extracted user feature and the at least one extracted advertisement feature together, it implements flexible sampling of negative instances of single-point functions; when controlling the negative sampling rate of the single-point function according to the deep learning model to extract at least one interaction feature, it reduces the point-by-point error of the deep learning model.

[0064] In a possible implementation, the advertisement recommendation module recommends at least one type of advertisement based on implicit feedback, and the implicit feedback indirectly reflects the extracted at least one interactive feature in the recommendation of the at least one type of advertisement;

[0065] The advertisement recommendation module sorts the at least one category of advertisements, lists the sorted at least one category of advertisements, and provides a recommendation list including the listed at least one category of advertisements;

[0066] The advertisement recommendation module randomly selects at least one advertisement with which the user has not interacted from the at least one type of advertisement and sorts the at least one advertisement.

[0067] Therefore, the present invention can reduce the information overload of advertising resources and solve the problem of data scarcity based on the deep learning algorithm that models user behavior information and advertising attribute information. In addition, by considering user behavior information and advertising attribute information at the same time, the present invention can solve the high-dimensional data scarcity and scalability problems existing in the existing collaborative filtering algorithm and solve the information overload problem in the field of advertising push.

[0068] Figure 2 The framework of the deep learning model used by the automatic advertisement recommendation device based on artificial intelligence according to an embodiment of the present invention is shown, and the framework of the deep learning model according to an embodiment of the present invention can be used with Figure 1 Related to the feature processing module described in .

[0069] Reference Figure 2 , the framework of the deep learning model according to the embodiment of the present invention includes an input network layer 210 for inputting an embedding vector, a user behavior network layer 220 for extracting user behavior features, an advertisement attribute network layer 230, and a deep neural network layer 240. For example, the deep learning model can be referred to as an advertisement recommendation method model based on deep learning.

[0070] According to one embodiment of the present invention, the input network layer 210 is a hidden layer, and user behavior data related to users, occupations, certificates, and languages ​​are converted into embedding vectors and input into the input network layer 210 , and attribute data related to advertisements and professions can be converted into embedding vectors and input into the input network layer 210 .

[0071] According to an embodiment of the present invention, data input to the input network layer 210 may be sent to the user behavior network layer 220 and the advertisement attribute network layer 230 .

[0072] For example, the user behavior network layer 220 may be referred to as a user behavior network (UBN) layer, and the advertisement attribute network layer 230 may be referred to as an advertisement attribute network (GAN) layer.

[0073] For example, the user behavior network layer 220 can use a feed-forward neural network (FFNN) to extract user behavior features. The first embedding vector input to the user behavior network layer 220 is a vector with low-dimensional dense real values ​​because it includes highly sparse and high-dimensional categorical and continuous values. The first embedding vector can be fed to the hidden layer of the neural network in a forward pass.

[0074] The user behavior network layer 220 according to an embodiment of the present invention may provide user features as output, as shown in the following formula (2).

[0075] a (0) =[e user ,e job , e cortificate , e language ] (2)

[0076] In formula (2), e user , e job , e certificate , e language can represent users, jobs, certificates, and languages, and a (0) Can be associated with the process of feeding into the deep neural network layer 240.

[0077] The process provided to the deep neural network layer 240 can be defined by the following formula (3).

[0078] a (l+1) =σ (l) a (l) +b (l) ) (3)

[0079] In formula (3), l represents the depth of the layer, σ represents the activation function, and W (l) 、a (l) 、b (l) They can represent the output, model weight and bias of the lth layer respectively.

[0080] In a specific embodiment, the advertisement attribute network layer 230 may use a feed-forward neural network (FFNN) to extract user behavior features.

[0081] The second embedding vector input to the advertisement attribute network layer 230 is a vector with low-dimensional dense real values ​​because it includes highly sparse and high-dimensional categorical and continuous values.

[0082] The second embedding vector may be fed to the hidden layer of the neural network in a forward pass. The advertisement attribute network layer 230 according to an embodiment of the present invention may provide advertisement features as output as shown in the following formula (4).

[0083] a(0) =[e coursc ,e major ] (4)

[0084] In formula (4), e course , e major can represent the advertising content characteristics and category characteristics respectively, and a (0) It may be related to the process fed into the deep neural network layer 240. The process provided to the deep neural network layer 240 may be defined by the above formula (3).

[0085] According to another embodiment of the present invention, the deep neural network layer 240 can also aggregate the output data of the user behavior network layer 220 and the advertising attribute network layer 230. The deep neural network layer 240 can extract the user features of the user behavior network layer 220 and the advertising features of the advertising attribute network layer 230 in an end-to-end process.

[0086] The deep neural network layer 240 can jointly learn all parameters and network parameters of the deep learning model based on the following formula (5).

[0087]

[0088] In formula (5), can represent a binary class label, σ can represent a (sigmoid) function, and y UBN can represent the output of the user behavior network layer 220, and y GAN The advertisement attributes network layer 230 may represent the display output.

[0089] Therefore, the present invention uses a multi-layer perceptron to construct an advanced nonlinear deep learning algorithm, reduces the dimensionality of the sparse data of advertising push, and generates user behavior information and advertising attribute information by finely extracting interactive feature information, which can fully predict the user's interest preferences. In addition, the present invention solves the problem of being unable to recommend suitable advertisements to new users due to insufficient past information or data, and the problem of being unable to provide recommendations before reflecting the user's preferences.

[0090] Example 2

[0091] This embodiment provides an automatic advertisement recommendation method based on artificial intelligence. Figure 3 As shown, including:

[0092] Obtain multiple user behavior data and multiple advertising attribute data;

[0093] Extracting at least one user feature by learning the collected multiple user behavior data, and extracting at least one advertisement feature by learning the collected multiple advertisement attribute data; learning the at least one user feature and the extracted at least one advertisement feature together to extract at least one interaction feature related to whether the user has clicked and browsed the multiple advertisements;

[0094] At least one advertisement is recommended based on the at least one extracted interaction feature. When the feature processing module learns the at least one extracted user feature and the at least one extracted advertisement feature together, the negative instances of the single-point function are flexibly sampled. When the negative sampling rate of the single-point function is controlled according to the deep learning model to extract at least one interaction feature, the point-by-point error of the deep learning model is reduced.

[0095] In a possible implementation, obtaining a plurality of user behavior data and a plurality of advertisement attribute data includes:

[0096] Collect multiple user behavior data including multiple user data, multiple job data, multiple certificate data and multiple language data; collect multiple advertisement attribute data including multiple advertisement content data and multiple advertisement type data.

[0097] In a possible implementation, at least one user feature is extracted by learning the collected multiple user behavior data, multiple user data, multiple position data, multiple certificate data, and multiple language data included in the multiple user behavior data are converted into a first embedding vector, the converted first embedding vector is input into a user behavior network (UBN) layer, and at least one user feature is extracted by learning the input first embedding vector;

[0098] Extracting at least one advertisement feature by learning the collected multiple advertisement attribute data, converting the multiple advertisement content data and the multiple main data included in the multiple advertisement attribute data into a second embedding vector, connecting the converted second embedding vector to an advertisement attribute network (GAN) layer, and extracting at least one advertisement feature by learning the input second embedding vector;

[0099] At least one interaction feature related to whether a user participates in multiple advertisements is extracted by jointly learning at least one extracted user feature and at least one extracted advertisement feature, and the at least one extracted user feature and the at least one extracted advertisement feature are input into a deep neural network (DNN) layer, and at least one interaction feature related to whether a user participates in multiple advertisements is extracted by jointly learning at least one extracted user feature and at least one input advertisement feature.

[0100] In one possible implementation, at least one category of advertisements is recommended based on implicit feedback, and the implicit feedback indirectly reflects the extracted at least one interactive feature in the recommendation of the at least one category of advertisements; the at least one category of advertisements is sorted, the sorted at least one category of advertisements is listed, and a recommendation list including the listed at least one category of advertisements is provided; and at least one advertisement with which the user has not interacted is randomly selected from the at least one category of advertisements and the at least one advertisement is sorted.

[0101] That is, the artificial intelligence-based automated advertising recommendation method provided by the embodiment of the present invention ranks at least one category of recommended advertisements according to interactive features, lists at least one category of ranked advertisements, and includes at least one category of listed advertisement recommendations, thereby providing targeted personalized advertisements.

[0102] Example 3

[0103] This embodiment provides a computer-readable storage medium, including:

[0104] The storage medium is used to store computer software instructions used to implement the artificial intelligence-based automated advertising recommendation method described in the above-mentioned embodiment 2, which includes a program for executing the above-mentioned program set for the artificial intelligence-based automated advertising recommendation method; specifically, the executable program can be built into the artificial intelligence-based automated advertising recommendation device described in embodiment 1, so that the neural network-based advertising recommendation system can implement the neural network-based advertising recommendation method described in embodiment 2 by executing the built-in executable program.

[0105] In addition, the computer-readable storage medium of this embodiment may adopt any combination of one or more computer-readable storage media, wherein the computer-readable storage medium includes electrical, optical, electromagnetic, infrared or semiconductor systems, devices or components, or any combination thereof.

[0106] Example 4

[0107] This embodiment provides an electronic device, which may include: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus.

[0108] Memory, used to store computer programs;

[0109] The processor is used to implement the steps of the automatic advertising recommendation method based on artificial intelligence described in the above embodiment 2 when executing the computer program stored in the memory.

[0110] As an embodiment of the present invention, the communication bus mentioned in the above terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc.

[0111] As an implementation mode of the present invention, the communication interface is used for communication between the above-mentioned terminal and other devices.

[0112] As an embodiment of the present invention, the memory may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0113] As an embodiment of the present invention, the above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.

[0114] The recommendation effect of the present invention is measured using index data related to hit rate (HR) and normalized difference cumulative gain (NDCG). Figure 4 and Figure 5 The results of negative sampling changes in the artificial intelligence-based automatic advertisement recommendation method and apparatus according to an embodiment of the present invention are compared and explained with the advertisement recommendation system based on the recommendation algorithm according to the prior art.

[0115] Prior art 701 and prior art 711 may represent a matrix factor (MF) recommendation algorithm, and prior art 702 and prior art 712 may represent a neural collaborative filtering (NCF) recommendation algorithm. Figure 4 To compare the prior art 701 and the prior art 702 with the present invention 703, Figure 5The following is a comparison between the prior art 711 and the prior art 712 and the present invention 713 .

[0116] according to Figure 4 The graph shown and Figure 5 As shown in the graph, the advantage of the single-point function compared to the paired function is that it has flexible sampling for negative instances. For example, it can be confirmed that the paired function can be paired with only one negative case, while the single-point function has the advantage that the sampling rate can be set to be suitable for the model.

[0117] When Figure 4 The graphs 700 and Figure 5 When checking the performance for changes in the negative sampling rate in the graph 710 shown, it can be confirmed that the optimal sampling rate is about 3 to 6. That is, the artificial intelligence-based automated advertisement recommendation method and apparatus according to one embodiment of the present invention has the advantage of reducing the single-point error of the deep learning model when extracting at least one interactive feature by controlling the negative sampling rate of the single-point function.

[0118] Therefore, different from the prior art, the present application adopts an artificial intelligence-based automatic advertising recommendation method, device and medium, which learns user behavior information and advertising attribute information including user work, education, language ability and other related information, and learns the learned user information and advertising attribute information together, takes the situation of each user browsing individual advertisements as the target variable, and learns interactive features to achieve user-customized advertising recommendation technology, improve the algorithm's recommendation accuracy and comprehensiveness, use deep neural networks for sorting, increase algorithm transparency, make recommendation results more explainable, and improve user experience; enhance users' long-term interest in advertisements, and make up for the shortcomings of the prior art.

[0119] It should be understood that in the various embodiments of this document, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this document.

[0120] It should also be understood that in the embodiments of this article, the term "and / or" is only a description of the association relationship of the associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0121] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this article.

[0122] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0123] In the several embodiments provided herein, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, or it can be an electrical, mechanical or other form of connection.

[0124] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of this article.

[0125] In addition, each functional unit in each embodiment of this invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware or in the form of software functional unit.

[0126] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this article is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of this article. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0127] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An automatic advertisement recommendation device based on artificial intelligence, characterized in that: include: A data acquisition module, used to acquire multiple user behavior data and multiple advertisement attribute data; A feature processing module, configured to extract at least one user feature by learning the collected multiple user behavior data, and to extract at least one advertisement feature by learning the collected multiple advertisement attribute data; Learning the at least one user feature and the extracted at least one advertisement feature together to extract at least one interactive feature related to whether the user has clicked and browsed the advertisement; An advertisement recommendation module is used to recommend at least one advertisement based on at least one extracted interaction feature; when the feature processing module learns the at least one extracted user feature and the at least one extracted advertisement feature together, it implements flexible sampling of negative instances of single-point functions; when extracting at least one interaction feature by controlling the negative sampling rate of the single-point function according to the deep learning model, it reduces the point-by-point error of the deep learning model.

2. The automatic advertisement recommendation device based on artificial intelligence according to claim 1, characterized in that: The data acquisition module collects a plurality of user behavior data including a plurality of user data, a plurality of position data, a plurality of certificate data and a plurality of language data; The data acquisition module collects a plurality of advertisement attribute data including a plurality of advertisement content data and a plurality of advertisement category data.

3. The automatic advertisement recommendation device based on artificial intelligence according to claim 1, characterized in that: The feature processing module adopts a deep learning model, first converts multiple user data, multiple position data, multiple certificate data, and multiple language data included in the multiple user behavior data into a first embedding vector, inputs the converted first embedding vector into the user behavior network layer, and learns the input first embedding vector to extract at least one user feature; The feature processing module converts a plurality of main data included in a plurality of advertisement content data and a plurality of advertisement category data into a second embedding vector, and uses the converted second embedding vector as advertisement attribute data; extracts at least one advertisement feature by inputting at least one advertisement attribute data into an advertisement attribute network layer and learning the input second embedding vector; The feature processing module inputs the extracted at least one user feature and the extracted at least one advertisement feature into a deep neural network layer, and learns the input at least one user feature and the input at least one advertisement feature together to extract at least one interaction feature.

4. The automatic advertisement recommendation device based on artificial intelligence according to claim 3, characterized in that: The feature processing module uses the user behavior network layer and the advertising attribute network layer as parallel sub-networks, and the parallel sub-networks are connected to the deep neural network layer, and the deep neural network layer is used as the prediction layer to achieve user-customized advertising recommendations based on deep learning.

5. The automatic advertisement recommendation device based on artificial intelligence according to claim 1, characterized in that: The advertisement recommendation module recommends at least one type of advertisement based on implicit feedback, where the implicit feedback indirectly reflects the extracted at least one interactive feature in the recommendation of the at least one type of advertisement; The advertisement recommendation module sorts the at least one category of advertisements, lists the sorted at least one category of advertisements, and provides a recommendation list including the listed at least one category of advertisements; The advertisement recommendation module randomly selects at least one advertisement with which the user has not interacted from the at least one type of advertisement and sorts the at least one advertisement.

6. An automated advertising recommendation method based on artificial intelligence, characterized in that: include: Obtain multiple user behavior data and multiple advertising attribute data; Extracting at least one user feature by learning the collected multiple user behavior data, and extracting at least one advertisement feature by learning the collected multiple advertisement attribute data; The at least one user feature and the at least one extracted advertisement feature are learned together to extract at least one interactive feature related to whether the user has clicked on and browsed the advertisement; when the at least one extracted user feature and the at least one extracted advertisement feature are learned together, a flexible sampling of negative instances of a single-point function is achieved; when the negative sampling rate of the single-point function is controlled according to a deep learning model to extract at least one interactive feature, a point-by-point error of the deep learning model is reduced; At least one advertisement is recommended based on the extracted at least one interaction feature.

7. The method for automatic advertisement recommendation based on artificial intelligence according to claim 6, characterized in that: The obtaining of multiple user behavior data and multiple advertisement attribute data includes: Collect multiple user behavior data including multiple user data, multiple job data, multiple certificate data and multiple language data; collect multiple advertisement attribute data including multiple advertisement content data and multiple advertisement type data.

8. The method for automatic advertisement recommendation based on artificial intelligence according to claim 7, characterized in that: Extracting at least one user feature by learning the collected multiple user behavior data, converting multiple user data, multiple position data, multiple certificate data, and multiple language data included in the multiple user behavior data into a first embedding vector, inputting the converted first embedding vector into a user behavior network layer, and extracting at least one user feature by learning the input first embedding vector; Extracting at least one advertisement feature by learning the collected multiple advertisement attribute data, converting multiple main data included in the multiple advertisement content data and the multiple advertisement category data into a second embedding vector, and using the converted second embedding vector as the advertisement attribute data; extracting at least one advertisement feature by inputting the at least one advertisement attribute data into the advertisement attribute network layer and learning the input second embedding vector; At least one interaction feature related to whether a user participates in multiple advertisements is extracted by jointly learning at least one extracted user feature and at least one extracted advertisement feature, and the at least one extracted user feature and the at least one extracted advertisement feature are input into a deep neural network layer, and at least one interaction feature related to whether a user participates in multiple advertisements is extracted by jointly learning at least one extracted user feature and at least one input advertisement feature.

9. The method for automatic advertisement recommendation based on artificial intelligence according to claim 6, characterized in that: recommending at least one category of advertisements based on implicit feedback, the implicit feedback indirectly reflecting the extracted at least one interactive feature in the recommendation of at least one category of advertisements; sorting the at least one category of advertisements, listing the sorted at least one category of advertisements, and providing a recommendation list including the listed at least one category of advertisements; At least one advertisement with which the user has not interacted is randomly selected from the at least one type of advertisements and the at least one advertisement is sorted.

Images