Click rate prediction method, system, readable storage medium and computer device

Abstract

The present invention provides a click-through rate prediction method, system, readable storage medium and computer equipment, the method comprising: when a click-through rate prediction task is obtained, the user behavior sequence is divided into a plurality of behavior sequence segments; each behavior sequence segment is converted into an embedding vector through an initial product embedding table; a first interest representation vector is calculated for each embedding vector through a self-attention network; a heterogeneity index and a correlation index are calculated for each first interest representation vector through a heterogeneity learning model and a correlation learning model; the weight of each behavior sequence segment in the click-through rate prediction task and a second interest representation vector are obtained according to each heterogeneity index and each correlation index; and a final prediction result is obtained according to the second interest representation vector and a fully connected neural network. The present invention reduces the negative impact of high heterogeneity and low correlation behavior sequence segments by comprehensively learning the relationship between each behavior in the behavior sequence segment and the relationship between the target and the behavior in the behavior sequence segment.

Description

Technical Field

[0001] The present invention relates to the field of Internet technology, and in particular to a click rate prediction method, system, readable storage medium and computer equipment. Background Art

[0002] With the continuous improvement of computer computing power, online advertising systems have achieved reach efficiency and actual benefits far exceeding traditional advertising models. In the actual e-commerce environment, due to the wide distribution of user groups, diverse needs, and rich types of advertisements, showing similar advertisements to different users through fixed rules greatly limits the revenue capacity of online advertising systems. Click-through rate prediction technology combines the characteristics of different user groups and the advertisements to be distributed, uses neural networks or comprehensive models to perform parameter training on historical data, and then completes the task of estimating the click probability of specific advertisements for different users. It has a wide range of application scenarios in real life.

[0003] In the task of click-through rate prediction, there are often many available contextual features for users and products, which can often reflect the user's potential preferences. In practical applications, user preference information plays an important role in many tasks. For example, in product recommendation, whether a user will buy a product is not only related to his own attributes, but also closely related to the user's recent purchase or browsing history. At present, both industry and academia have proposed many methods for learning user interests, among which the most mainstream is the related method based on deep learning.

[0004] Although many works related to click-through rate prediction based on user behavior have achieved great success in learning user interests, few methods can pay attention to the negative impact of irrelevant products in the behavior sequence on the prediction results. On the one hand, especially in e-commerce applications, a user's interests often change frequently, which makes the behaviors in the user's history records diverse, resulting in heterogeneity in the behavior sequence (i.e., the types of products in a sequence vary greatly). Although the DSIN model attempts to reduce the impact of heterogeneity by segmenting the sequence and has determined a certain effect improvement, the results of such methods depend on the way the sequence is divided. The usual practice is to divide it into two sequence segments when the interval between adjacent behaviors exceeds 30 minutes. However, this approach cannot effectively guarantee the homogeneity of the sequence segments when the user's interests change frequently. Sequence segments with high heterogeneity will have a negative impact on the learning of the overall interest and cause great obstacles to the learning of user interests. Specifically, when the current method of learning interest vectors from sequence segments processes sequences with high heterogeneity, because there are more types of products involved, the extracted interest vectors often cannot effectively represent the real needs of users during this period of time. In the actual estimation process, efforts should be made to avoid the negative impact of irrelevant behaviors on the learned interest vectors. Therefore, after dividing the user behavior sequences, it is necessary to influence behaviors with high heterogeneity and low relevance to the target advertisement. Summary of the invention

[0005] Based on this, the purpose of the present invention is to provide a click rate prediction method, system, readable storage medium and computer device to at least solve the deficiencies in the above-mentioned technology.

[0006] The present invention provides a click rate prediction method, comprising:

[0007] When a click rate prediction task is obtained, a user behavior sequence is obtained, and the user behavior sequence is divided into a sequence according to a preset rule to obtain a plurality of behavior sequence segments;

[0008] Obtaining an initial product embedding table of a target advertisement, and using the initial product embedding table to convert each of the behavior sequence segments into a corresponding embedding vector;

[0009] Inputting each of the embedding vectors into a preset self-attention network to calculate a corresponding first interest representation vector;

[0010] Inputting each of the first interest representation vectors into a preset heterogeneity learning model and a correlation learning model respectively to obtain a corresponding heterogeneity index and a correlation index;

[0011] Obtaining the weight of each of the behavior sequence segments in the click rate prediction task according to each of the heterogeneity indicators and each of the correlation indicators, and obtaining a second interest representation vector using the weight of each of the behavior sequence segments in the click rate prediction task;

[0012] The final estimation result of the click rate estimation task is obtained according to the second interest representation vector and the preset fully connected neural network.

[0013] Furthermore, the step of dividing the user behavior sequence into a sequence according to a preset rule to obtain a plurality of behavior sequence segments includes:

[0014] Determining whether the interval between two adjacent user behaviors in the user behavior sequence exceeds a preset time;

[0015] If the interval between two adjacent user behaviors in the user behavior sequence exceeds a preset time, the two user behaviors are divided into different behavior sequence segments.

[0016] Furthermore, the steps of inputting each of the first interest representation vectors into a preset heterogeneity learning model and a correlation learning model to obtain corresponding heterogeneity indexes and correlation indexes include:

[0017] Removing one behavior from each of the behavior sequence segments in turn to construct multiple incomplete sequence segments, and calculating a third interest representation vector corresponding to each of the incomplete sequence segments according to the product initial embedding table and the self-attention network;

[0018] A corresponding heterogeneity index is calculated using cosine similarity, the third interest representation vector, and the first interest representation vector.

[0019] Furthermore, the step of inputting each of the first interest representation vectors into a preset heterogeneity learning model and a correlation learning model to obtain corresponding heterogeneity indexes and correlation indexes also includes:

[0020] Adding the target advertisement product to each of the behavior sequence segments to construct a plurality of sequence segments containing the target product, and calculating a fourth interest representation vector corresponding to each of the sequence segments containing the target product according to the product initial embedding table and the self-attention network;

[0021] A corresponding correlation index is calculated using cosine similarity, the fourth interest representation vector, and the first interest representation vector.

[0022] Furthermore, the calculation formula of the embedding vector is:

[0023] B i =b i ×E;

[0024] Where b i represents the product corresponding to the user's ith behavior, E represents the initial embedding table of the product, and B i represents the corresponding embedding vector.

[0025] Furthermore, the calculation formula of the first interest representation vector is:

[0026]

[0027] In the formula, Q t The query matrix representing the embedding vector of the t-th sequence segment of the input is transformed by the self-attention network. The embedding vector of the t-th sequence segment of the input is transformed into the Key matrix by the self-attention network, V t The embedding vector of the t-th sequence segment of the input is transformed into the Value matrix by the self-attention network, and d is the embedding vector dimension.

[0028] The present invention also provides a click rate prediction system, comprising:

[0029] An acquisition module, used for acquiring a user behavior sequence when a click rate estimation task is acquired, and dividing the user behavior sequence into a sequence according to a preset rule to obtain a plurality of behavior sequence segments;

[0030] A conversion module, used to obtain an initial product embedding table of a target advertisement, and convert each of the behavior sequence segments into a corresponding embedding vector using the initial product embedding table;

[0031] A first calculation module, used for inputting each of the embedding vectors into a preset self-attention network to calculate a corresponding first interest representation vector;

[0032] A second calculation module, used for inputting each of the first interest representation vectors into a preset heterogeneity learning model and a correlation learning model respectively, so as to obtain a corresponding heterogeneity index and a correlation index;

[0033] A third calculation module is used to obtain the weight of each of the behavior sequence segments in the click rate prediction task according to each of the heterogeneity indicators and each of the correlation indicators, and obtain a second interest representation vector using the weight of each of the behavior sequence segments in the click rate prediction task;

[0034] An estimation module is used to obtain a final estimation result of the click rate estimation task according to the second interest representation vector and a preset fully connected neural network.

[0035] Furthermore, the acquisition module includes:

[0036] A judging unit, used to judge whether the interval between two adjacent user behaviors in the user behavior sequence exceeds a preset time;

[0037] The dividing unit is used to divide the two adjacent user behaviors in the user behavior sequence into different behavior sequence segments if the interval between the two adjacent user behaviors in the user behavior sequence exceeds a preset time.

[0038] Furthermore, the second calculation module includes:

[0039] A first construction unit is used to remove one behavior from each of the behavior sequence segments in turn to construct multiple incomplete sequence segments, and calculate a third interest representation vector corresponding to each of the incomplete sequence segments according to the initial product embedding table and the self-attention network;

[0040] The first calculation unit is used to calculate a corresponding heterogeneity index by using cosine similarity, the third interest representation vector and the first interest representation vector.

[0041] Furthermore, the second calculation module also includes:

[0042] A second construction unit is used to add the target advertisement product to each of the behavior sequence segments to construct a plurality of sequence segments containing the target product, and calculate a fourth interest representation vector corresponding to each of the sequence segments containing the target product according to the product initial embedding table and the self-attention network;

[0043] The second calculation unit is used to calculate a corresponding correlation index by using the cosine similarity, the fourth interest representation vector and the first interest representation vector.

[0044] The present invention also provides a readable storage medium on which a computer program is stored. When the program is executed by a processor, the above-mentioned click rate prediction method is implemented.

[0045] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned click rate prediction method when executing the computer program.

[0046] The click-through rate prediction method, system, readable storage medium and computer device of the present invention divide the user behavior sequence into multiple behavior sequence segments, and use the initial embedding table of the product in the target advertisement to convert each behavior sequence segment into a corresponding embedding vector, calculate the first interest representation vector corresponding to each embedding vector through the self-attention network, and use the heterogeneity learning model and the correlation learning model to calculate the heterogeneity index and correlation index of each first interest representation vector, and then obtain the weight of each behavior sequence segment in the click-through rate prediction task and the second interest representation vector through each heterogeneity index and each correlation index, and obtain the final prediction result of the click-through rate prediction task according to the second interest representation vector. The present invention reduces the negative impact of high heterogeneity and low correlation behavior sequence segments on the prediction task by comprehensively learning the relationship between each behavior in the behavior sequence segment and the relationship between the target and the behavior sequence segment, and measures the heterogeneity of the sequence segments by calculating the similarity between the interest representation vectors of different sequence segments. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a flow chart of a click rate prediction method in a first embodiment of the present invention;

[0048] Figure 2 for Figure 1 Detailed flow chart of step S101;

[0049] Figure 3 is a framework diagram of a click rate prediction model in the first embodiment of the present invention;

[0050] Figure 4 for Figure 1 Detailed flow chart of step S104;

[0051] Figure 5 is a structural block diagram of a click rate prediction system in a second embodiment of the present invention;

[0052] Figure 6 It is a structural block diagram of a computer device in the third embodiment of the present invention.

[0053] Description of main component symbols:

[0054] Memory 10 Creating a Module 11 processor 20 Calling Modules 12 Computer Programs 30 Execution Module 13

[0055] The following specific implementation manner will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION

[0056] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Several embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0057] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0058] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0059] First of all, it should be noted that embedding technology is widely used in the field of deep learning. This technology can transform high-dimensional, sparse features into low-dimensional dense vectors, provide initial input for deep neural networks, and enable deep models to learn and converge better. Among them, the most common approach in the field of click-through rate prediction is to create a trainable nx d-dimensional embedding table for each feature, where n represents all types of the feature, and transform the feature into a d-dimensional vector by looking up the table.

[0060] The self-attention mechanism is to construct three different linear transformation matrices to transform the same input into Query, Key, and Value. For each input in a series of inputs, its matching degree with other inputs (i.e., Attention score) is calculated through its corresponding Query-Key pair, and each input vector is given contextual information in a weighted summation manner. An additional feedforward neural network structure will be used to integrate all input vectors in the sequence and obtain the output of the entire sequence.

[0061] Embodiment 1

[0062] See also Figure 1 , which shows a click rate estimation method in a first embodiment of the present invention, and the method specifically includes steps S101 to S106:

[0063] S101, when a click rate prediction task is obtained, a user behavior sequence is obtained, and the user behavior sequence is divided into a sequence according to a preset rule to obtain a plurality of behavior sequence segments;

[0064] For further information, see Figure 2 , the step S101 specifically includes steps S1011 to S1012:

[0065] S1011, determining whether the interval between two adjacent user behaviors in the user behavior sequence exceeds a preset time;

[0066] S1012: If the interval between two adjacent user behaviors in the user behavior sequence exceeds a preset time, the two user behaviors are divided into different behavior sequence segments.

[0067] In the specific implementation, this embodiment proposes a click rate prediction model, please refer to Figure 3 , segment the user's behavior sequence. Specifically, if the interval between two adjacent behaviors in the sequence exceeds 30 minutes, the two behaviors will be regarded as being in different sequence segments. Different users will obtain different numbers of sequence segments. In this embodiment, the number of sequence segments is set to T.

[0068] S102, obtaining an initial product embedding table of a target advertisement, and using the initial product embedding table to convert each of the behavior sequence segments into a corresponding embedding vector;

[0069] In the specific implementation, embedding technology is used to convert the products in the user behavior sequence and the products in the target advertisement into low-dimensional vectors that can be input into the neural network.

[0070] Obtain the initial embedding table E of the product in the target advertisement, and obtain the embedding vector of the product according to the following formula (1). i is the product corresponding to the user’s i-th behavior, B i is the corresponding embedding vector.

[0071] Formula (1): B i =b i ×E;

[0072] S103, inputting each of the embedding vectors into a preset self-attention network to calculate a corresponding first interest representation vector;

[0073] In the specific implementation, the self-attention mechanism is used to input the embedding vector of the product in each sequence segment into the self-attention network based on formula (2) to obtain the corresponding T sequence segment interest representation vectors. t The query matrix representing the embedding vector of the t-th sequence segment of the input is transformed by the self-attention network. The embedding vector of the t-th sequence segment of the input is transformed into the Key matrix by the self-attention network, V t The embedding vector of the t-th sequence segment of the input is transformed into the Value matrix by the self-attention network, and d is the embedding vector dimension.

[0074] Formula (2):

[0075] S104, inputting each of the first interest representation vectors into a preset heterogeneity learning model and a correlation learning model respectively to obtain a corresponding heterogeneity index and a correlation index;

[0076] For further information, see Figure 4 , the step S104 specifically includes steps S1041 to S1044:

[0077] S1041, removing one behavior from each of the behavior sequence segments in turn to construct multiple incomplete sequence segments, and calculating a third interest representation vector corresponding to each of the incomplete sequence segments according to the product initial embedding table and the self-attention network;

[0078] S1042, calculating a corresponding heterogeneity index using cosine similarity, the third interest representation vector, and the first interest representation vector;

[0079] S1043, adding the target advertisement product to each of the behavior sequence segments to construct a plurality of sequence segments containing the target product, and calculating a fourth interest representation vector corresponding to each of the sequence segments containing the target product according to the product initial embedding table and the self-attention network;

[0080] S1044: Calculate a corresponding correlation index using the cosine similarity, the fourth interest representation vector, and the first interest representation vector.

[0081] In the specific implementation, in the heterogeneous learning model, this embodiment designs a heterogeneous learning mechanism. For a segmented user behavior sequence segment, one of the behaviors is removed in turn to construct a new sequence segment. For a sequence segment with n behaviors, n derived incomplete sequence segments will be obtained. Similarly, according to formula (2), the interest representation vector is calculated for each incomplete sequence segment. According to formula (3), the cosine similarity between the n interest representation vectors and the original sequence segment representation vector is calculated and the average value is taken as the heterogeneity index of the sequence segment;

[0082] Formula (3):

[0083] In the correlation learning module, in order to mine the correlation between the target and the sequence segment behavior, this embodiment designs a variant based on the heterogeneous learning mechanism to perform correlation learning. For each sequence segment, the product in the target advertisement is added to the end of the sequence segment to construct a new sequence segment containing the target product. According to formula (2), the interest representation vector I′ of the new sequence segment is calculated. t According to formula (4), the cosine similarity between the vector and the original sequence segment vector is calculated as the relevance index of the sequence segment to the target advertisement;

[0084] Formula (4):

[0085] S105, obtaining the weight of each of the behavior sequence segments in the click rate prediction task according to each of the heterogeneity indicators and each of the correlation indicators, and obtaining a second interest representation vector using the weight of each of the behavior sequence segments in the click rate prediction task;

[0086] In the specific implementation, according to formula (5), the heterogeneity index and correlation index of the T sequence segments obtained by the above steps are combined to obtain the weight of each sequence segment with respect to the above target advertisement in the above click-through rate prediction task; according to formula (6), the interest representation vectors of all sequence segments are integrated in a weighted summation manner to obtain the overall user interest representation vector.

[0087] Formula (5):

[0088] Formula (6):

[0089] S106: Obtain a final estimation result of the click rate estimation task according to the second interest representation vector and a preset fully connected neural network.

[0090] In the specific implementation, a fully connected neural network is used to construct a click rate prediction layer. The input is the concatenation of the overall user interest representation vector obtained by the above step and the embedding vector obtained in the above step. The output is the probability of the user clicking on the target advertisement. Formula (7) is the loss function used for model training. Then the model is trained using the stochastic gradient descent method to obtain the corresponding final prediction result.

[0091] Formula (7):

[0092]

[0093] Compared with other methods, the present application can effectively detect heterogeneous sequence segments in behaviors and reduce their negative impact, while increasing the proportion of sequence segments with high correlation in the prediction task. Table 1 shows the four public data sets used by the present application in comparative experiments. Table 2 shows the comparison of the evaluation indicators AUC and logloss of the present application and other models in the click-through rate prediction task on four different data sets. It can be seen from Table 2 that the present application has achieved better results in the click-through rate prediction task compared with other models. Table 3 shows the results of the ablation experiment. RM-R, RM-H, and RM-HR represent the model variants of removing correlation learning, heterogeneous learning, and both removed. It can be seen from Table 3 that the heterogeneous learning mechanism of the model and the variants of its correlation learning mechanism can improve the model effect.

[0094] Table 1

[0095]

[0096] Table 2

[0097]

[0098] Table 3

[0099]

[0100] In summary, the click-through rate prediction method in the above embodiment of the present invention divides the user behavior sequence into multiple behavior sequence segments, and uses the initial embedding table of the product in the target advertisement to convert each behavior sequence segment into a corresponding embedding vector, calculates the first interest representation vector corresponding to each embedding vector through the self-attention network, and uses the heterogeneity learning model and the correlation learning model to calculate the heterogeneity index and correlation index of each first interest representation vector, and then obtains the weight of each behavior sequence segment in the click-through rate prediction task and the second interest representation vector through each heterogeneity index and each correlation index, and obtains the final prediction result of the click-through rate prediction task according to the second interest representation vector. The present invention reduces the negative impact of high heterogeneity and low correlation behavior sequence segments on the prediction task by comprehensively learning the relationship between each behavior in the behavior sequence segment and the relationship between the target and the behavior sequence segment, and measures the heterogeneity of the sequence segments by calculating the similarity between the interest representation vectors of different sequence segments.

[0101] Embodiment 2

[0102] Another aspect of the present invention is to provide a click rate prediction system. Figure 5 , which shows a click rate prediction system in a second embodiment of the present invention, and the system includes:

[0103] An acquisition module 11 is used to acquire a user behavior sequence when a click rate estimation task is acquired, and to divide the user behavior sequence into a sequence according to a preset rule to obtain a plurality of behavior sequence segments;

[0104] Furthermore, the acquisition module 11 includes:

[0105] A judging unit, used to judge whether the interval between two adjacent user behaviors in the user behavior sequence exceeds a preset time;

[0106] The dividing unit is used to divide the two adjacent user behaviors in the user behavior sequence into different behavior sequence segments if the interval between the two adjacent user behaviors in the user behavior sequence exceeds a preset time.

[0107] A conversion module 12, configured to obtain an initial product embedding table of a target advertisement, and convert each of the behavior sequence segments into a corresponding embedding vector using the initial product embedding table;

[0108] A first calculation module 13, used for inputting each of the embedding vectors into a preset self-attention network to calculate a corresponding first interest representation vector;

[0109] A second calculation module 14 is used to input each of the first interest representation vectors into a preset heterogeneity learning model and a correlation learning model to obtain a corresponding heterogeneity index and a correlation index;

[0110] Furthermore, the second calculation module 14 includes:

[0111] A first construction unit is used to remove one behavior from each of the behavior sequence segments in turn to construct multiple incomplete sequence segments, and calculate a third interest representation vector corresponding to each of the incomplete sequence segments according to the initial product embedding table and the self-attention network;

[0112] The first calculation unit is used to calculate a corresponding heterogeneity index by using cosine similarity, the third interest representation vector and the first interest representation vector.

[0113] Furthermore, the second calculation module 14 also includes:

[0114] A second construction unit is used to add the target advertisement product to each of the behavior sequence segments to construct a plurality of sequence segments containing the target product, and calculate a fourth interest representation vector corresponding to each of the sequence segments containing the target product according to the product initial embedding table and the self-attention network;

[0115] The second calculation unit is used to calculate a corresponding correlation index by using the cosine similarity, the fourth interest representation vector and the first interest representation vector.

[0116] A third calculation module 15 is used to obtain the weight of each of the behavior sequence segments in the click rate prediction task according to each of the heterogeneity indicators and each of the correlation indicators, and obtain a second interest representation vector using the weight of each of the behavior sequence segments in the click rate prediction task;

[0117] The estimation module 16 is used to obtain the final estimation result of the click rate estimation task according to the second interest representation vector and a preset fully connected neural network.

[0118] The functions or operation steps implemented when the above modules and units are executed are generally the same as those in the above method embodiments, and will not be repeated here.

[0119] The click rate prediction system provided in the embodiment of the present invention has the same implementation principle and technical effects as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the system embodiment, reference may be made to the corresponding contents in the aforementioned method embodiment.

[0120] Embodiment 3

[0121] The present invention also provides a computer device, see Figure 6 , shown is a computer device in the third embodiment of the present invention, including a memory 10, a processor 20, and a computer program 30 stored in the memory 10 and executable on the processor 20. When the processor 20 executes the computer program 30, the above-mentioned click rate prediction method is implemented.

[0122] The memory 10 includes at least one type of readable storage medium, and the readable storage medium includes a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., an SD or DX memory, etc.), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the memory 10 may be an internal storage unit of a computer device, such as a hard disk of the computer device. In other embodiments, the memory 10 may also be an external storage device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card, etc. Further, the memory 10 may also include both an internal storage unit of a computer device and an external storage device. The memory 10 may be used not only to store application software and various types of data installed in the computer device, but also to temporarily store data that has been output or is to be output.

[0123] Among them, in some embodiments, the processor 20 can be an electronic control unit (Electronic Control Unit, abbreviated as ECU, also known as a vehicle computer), a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, a microprocessor or other data processing chip, used to run the program code stored in the memory 10 or process data, such as executing access restriction programs, etc.

[0124] It should be pointed out that Figure 6 The structure shown does not constitute a limitation on the computer device. In other embodiments, the computer device may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0125] The embodiment of the present invention further provides a readable storage medium on which a computer program is stored. When the program is executed by a processor, the click rate prediction method as described above is implemented.

[0126] Those skilled in the art will appreciate that the logic and / or steps represented in the flowchart or otherwise described herein, for example, may be considered as an ordered list of executable instructions for implementing logical functions, and may be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For purposes of this specification, a "computer-readable medium" may be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0127] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0128] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or a combination thereof: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0129] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0130] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A click rate prediction method, characterized in that: include: When a click rate prediction task is obtained, a user behavior sequence is obtained, and the user behavior sequence is divided into a sequence according to a preset rule to obtain a plurality of behavior sequence segments; Obtaining an initial product embedding table of a target advertisement, and using the initial product embedding table to convert each of the behavior sequence segments into a corresponding embedding vector; Inputting each of the embedding vectors into a preset self-attention network to calculate a corresponding first interest representation vector; Inputting each of the first interest representation vectors into a preset heterogeneity learning model and a correlation learning model respectively to obtain a corresponding heterogeneity index and a correlation index, wherein the step of inputting each of the first interest representation vectors into a preset heterogeneity learning model and a correlation learning model respectively to obtain a corresponding heterogeneity index and a correlation index comprises: Removing one behavior from each of the behavior sequence segments in turn to construct multiple incomplete sequence segments, and calculating a third interest representation vector corresponding to each of the incomplete sequence segments according to the product initial embedding table and the self-attention network; Calculating a corresponding heterogeneity index using cosine similarity, the third interest representation vector, and the first interest representation vector; Adding the target advertisement product to each of the behavior sequence segments to construct a plurality of sequence segments containing the target product, and calculating a fourth interest representation vector corresponding to each of the sequence segments containing the target product according to the product initial embedding table and the self-attention network; Calculating a corresponding correlation index using cosine similarity, the fourth interest representation vector, and the first interest representation vector; Obtaining the weight of each of the behavior sequence segments in the click rate prediction task according to each of the heterogeneity indicators and each of the correlation indicators, and obtaining a second interest representation vector using the weight of each of the behavior sequence segments in the click rate prediction task; The final estimation result of the click rate estimation task is obtained according to the second interest representation vector and the preset fully connected neural network.

2. The click rate estimation method according to claim 1, characterized in that: The step of dividing the user behavior sequence into a sequence according to a preset rule to obtain a plurality of behavior sequence segments comprises: Determining whether the interval between two adjacent user behaviors in the user behavior sequence exceeds a preset time; If the interval between two adjacent user behaviors in the user behavior sequence exceeds a preset time, the two user behaviors are divided into different behavior sequence segments.

3. The click rate prediction method according to claim 1, characterized in that: The calculation formula of the embedding vector is: ； In the formula, represents the product corresponding to the user's ith behavior, E represents the initial embedding table of the product, represents the corresponding embedding vector.

4. The click rate estimation method according to claim 3, characterized in that: The calculation formula of the first interest representation vector is: ； In the formula, The query matrix representing the embedding vector of the t-th sequence segment of the input is transformed by the self-attention network. The embedding vector of the t-th sequence segment of the input is transformed into the Key matrix by the self-attention network, The embedding vector of the t-th sequence segment of the input is transformed into the Value matrix by the self-attention network, and d is the embedding vector dimension.

5. A click rate prediction system, characterized in that: include: An acquisition module, used for acquiring a user behavior sequence when a click rate estimation task is acquired, and dividing the user behavior sequence into a sequence according to a preset rule to obtain a plurality of behavior sequence segments; A conversion module, used to obtain an initial product embedding table of a target advertisement, and convert each of the behavior sequence segments into a corresponding embedding vector using the initial product embedding table; A first calculation module, used for inputting each of the embedding vectors into a preset self-attention network to calculate a corresponding first interest representation vector; A second calculation module is used to input each of the first interest representation vectors into a preset heterogeneity learning model and a correlation learning model to obtain a corresponding heterogeneity index and a correlation index, wherein further, the second calculation module includes: A first construction unit is used to remove one behavior from each of the behavior sequence segments in turn to construct multiple incomplete sequence segments, and calculate a third interest representation vector corresponding to each of the incomplete sequence segments according to the initial product embedding table and the self-attention network; A first calculation unit, configured to calculate a corresponding heterogeneity index using cosine similarity, the third interest representation vector, and the first interest representation vector; A second construction unit is used to add the target advertisement product to each of the behavior sequence segments to construct a plurality of sequence segments containing the target product, and calculate a fourth interest representation vector corresponding to each of the sequence segments containing the target product according to the product initial embedding table and the self-attention network; A second calculation unit, configured to calculate a corresponding correlation index using cosine similarity, the fourth interest representation vector and the first interest representation vector; A third calculation module is used to obtain the weight of each of the behavior sequence segments in the click rate prediction task according to each of the heterogeneity indicators and each of the correlation indicators, and obtain a second interest representation vector using the weight of each of the behavior sequence segments in the click rate prediction task; An estimation module is used to obtain a final estimation result of the click rate estimation task according to the second interest representation vector and a preset fully connected neural network.

6. The click rate prediction system according to claim 5, characterized in that: The acquisition module comprises: A judging unit, used to judge whether the interval between two adjacent user behaviors in the user behavior sequence exceeds a preset time; The dividing unit is used to divide two adjacent user behaviors in the user behavior sequence into different behavior sequence segments if the occurrence interval between the two adjacent user behaviors in the user behavior sequence exceeds a preset time.

7. A readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the click rate prediction method as described in any one of claims 1 to 4 is implemented.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the click rate prediction method as described in any one of claims 1 to 4 is implemented.

Images