Business processing method and related device

Abstract

The invention provides a service processing method and a related device, and the method comprises the steps: determining user behavior feature data and user portrait feature data of a service; then, according to a time period corresponding to each feature in the user behavior feature data, the user behavior feature data is divided into first feature data and second feature data, and the time period corresponding to any feature in the first feature data is located before the time period corresponding to any feature in the second feature data; and finally, inputting the first feature data, the second feature data and the user portrait feature data into a service time period model for prediction, and determining a recommendation processing time period of the service. According to the method, the user behavior feature data of the service can be split in time sequence, fusion is performed in combination with the user portrait feature data, the recommendation processing time period of the service can be automatically determined through the service time period model, and the service processing effect is greatly improved.

Description

Technical Field

[0001] The present application relates to the technical field of business processing, and in particular to a business processing method and related devices. Background Art

[0002] In the process of business processing, it is generally necessary to consider when to process the business so as to achieve better processing results. For example, for marketing business, it is necessary to determine when users prefer to receive marketing recommendations, that is, when to make marketing recommendations to achieve a higher conversion rate is a question that needs to be considered. Therefore, how to determine the appropriate recommendation period or processing period for the business has become one of the hot issues in the current business processing field. Summary of the invention

[0003] In view of this, the present application provides a business processing method and related devices, which can split the user behavior feature data of the business in time series, and fuse it with the user portrait feature data, and perform time period prediction on the fusion result through the business time period model, so as to accurately determine the recommended processing time period of the business, thereby improving the effect of business processing.

[0004] In a first aspect, an embodiment of the present application provides a service processing method, the method comprising:

[0005] Determine the user behavior feature data and user portrait feature data of the business;

[0006] According to the time period corresponding to each feature in the user behavior feature data, the user behavior feature data is divided into first feature data and second feature data, wherein the time period corresponding to any feature in the first feature data is before the time period corresponding to any feature in the second feature data;

[0007] The first feature data, the second feature data and the user portrait feature data are input into a business period model for prediction to determine a recommended processing period for the business.

[0008] In a second aspect, an embodiment of the present application provides a service processing device, the device comprising:

[0009] A determination unit, used to determine user behavior feature data and user portrait feature data of a business;

[0010] a dividing unit, configured to divide the user behavior feature data into first feature data and second feature data according to a time period corresponding to each feature in the user behavior feature data, wherein a time period corresponding to any feature in the first feature data is before a time period corresponding to any feature in the second feature data;

[0011] The determination unit is further used to input the first feature data, the second feature data and the user portrait feature data into a business period model for prediction, so as to determine a recommended processing period for the business.

[0012] In a third aspect, an embodiment of the present application provides an electronic device, comprising a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor, and the program includes instructions for executing the steps of any method in the first aspect of the embodiment of the present application.

[0013] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program for electronic data exchange, wherein the computer program enables a computer to execute part or all of the steps described in any method of the first aspect of the embodiment of the present application.

[0014] In a fifth aspect, an embodiment of the present application provides a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute some or all of the steps described in any method of the first aspect of the embodiment of the present application. The computer program product may be a software installation package.

[0015] It can be seen that through the above-mentioned business processing method and related devices, first, the user behavior feature data and user portrait feature data of the business are determined; then, according to the time period corresponding to each feature in the user behavior feature data, the user behavior feature data is divided into the first feature data and the second feature data, and the time period corresponding to any feature in the first feature data is located before the time period corresponding to any feature in the second feature data; finally, the first feature data, the second feature data and the user portrait feature data are input into the business time period model for prediction to determine the recommended processing time period of the business. The user behavior feature data of the business can be split in the time period to obtain the first feature data and the second feature data. Since the first feature data is earlier than the second feature data in time sequence, the user behavior information closer to the current time and the user behavior information farther from the current time can be split, and combined with the user portrait feature data for fusion. Due to the introduction of the user behavior information closer to the current time and the user behavior information farther from the current time, it is possible to dynamically predict what behavior the user is more inclined to perform in the subsequent time period. At the same time, combined with the user portrait information, the recommended processing time period of the business can be automatically determined through the business time period model, which greatly improves the effect of business processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0017] Figure 1 An application scenario diagram of a business processing method provided in an embodiment of the present application;

[0018] Figure 2 A flowchart of a business processing method provided in an embodiment of the present application;

[0019] Figure 3 A schematic diagram of the structure of a business period model provided in an embodiment of the present application;

[0020] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application;

[0021] Figure 5 A block diagram of the functional units of a business processing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0023] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices.

[0024] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. 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 indicates that the associated objects before and after are in an "or" relationship. The "plurality" appearing in the embodiments of the present application refers to two or more.

[0025] In the embodiments of the present application, "at least one item" or similar expressions refer to any combination of these items, including any combination of single items or plural items, and refer to one or more, and multiple refers to two or more. For example, at least one item of a, b, or c can represent the following seven situations: a, b, c, a and b, a and c, b and c, a, b, and c. Among them, each of a, b, and c can be an element or a set containing one or more elements.

[0026] The "connection" that appears in the embodiments of the present application refers to various connection methods such as direct connection or indirect connection to achieve communication between devices, and the embodiments of the present application do not impose any limitations on this.

[0027] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0028] The following is an explanation of the relevant terms involved in the embodiments of the present application:

[0029] 1. Attention mechanism: A deep learning model that is mainly used to process sequence data, such as sentence understanding in natural language processing. In traditional recurrent neural networks (RNNs), the hidden state of each time step is considered evenly, while the attention mechanism model allows the model to focus more on certain specific parts as needed when processing sequence data, rather than considering all parts equally. Specifically, the attention mechanism model not only uses the current input and hidden state at each time step, but also generates an attention weight vector, which represents the attention allocation to the current input. The generation of the attention weight vector usually involves an attention function, which takes the current input and hidden state as input and generates a weight vector.

[0030] 2. Normalized exponential function Softmax: In deep learning, it is often used in multi-classification prediction problems, and the answers of the classifiers are mutually exclusive; the Softmax function obtains the correlation between different probabilities and maps the variables to [0,1].

[0031] 3. Loss function: In the process of training a deep neural network (DNN), because we hope that the output of the deep neural network is as close as possible to the value we really want to predict, we can compare the predicted value of the current network with the target value we really want, and then update the weight vector of each layer of the neural network according to the difference between the two (of course, there is usually an initialization process before the first update, that is, pre-configuring parameters for each layer in the deep neural network). For example, if the predicted value of the network is high, adjust the weight vector to make it predict a lower value, and keep adjusting until the deep neural network can predict the target value we really want or a value very close to the target value we really want. Therefore, it is necessary to pre-define "how to compare the difference between the predicted value and the target value", which is the loss function or objective function, which are important equations used to measure the difference between the predicted value and the target value. Among them, taking the loss function as an example, the higher the output value (loss) of the loss function, the greater the difference, so the training of the deep neural network becomes a process of minimizing this loss as much as possible.

[0032] In regular time recommendation sequence modeling, most of the time, conventional recurrent neural network modeling is used. This type of network cannot process sequence information in parallel, resulting in the inability to synchronize weight updates during learning. In addition, the long-term and short-term learning methods are more implicit, requiring the network itself to make choices during weight iteration.

[0033] For example, in common marketing campaign scenarios, it is often necessary to consider personalized marketing timing that suits users (i.e., user preferred timing) in order to improve the effectiveness of the campaign and stimulate greater benefit value; accurately hitting the user's timing preferences and doing marketing at the right time can bring positive feedback from users, thereby promoting marketing goals.

[0034] Among the existing timing prediction technologies, most use conventional sequence modeling and perform sequence feature mining from a universal perspective. This method can often only obtain the general preferences of users over a period of time and cannot capture the user's interest preferences that change dynamically over time.

[0035] To solve the above problems, the embodiments of the present application provide a business processing method and related devices, which can split the user behavior feature data of the business in time series, and integrate it with the user portrait feature data, and automatically determine the recommended processing time period of the business through the business period model, which greatly improves the effect of business processing. In the marketing scenario, the advantage of integrating the user's long-term and short-term preference modeling is that it can simultaneously consider the user's recent preferences and interest adjustments that change according to long-term data dynamics, effectively learn different short-term and long-term weights and interest deviations, and thus more accurately obtain the user's possible personalized preference time period in the future.

[0036] The embodiments of the present application can be applied to the following application scenarios, including but not limited to: information recommendation systems based on deep learning, neural network-based processing systems deployed on electronic devices, and scenarios for natural language processing. The embodiments of the present application can be adjusted and improved according to the specific application environment, and are not specifically limited here.

[0037] The business processing method provided in the embodiment of the present application can be executed by an electronic device, and the electronic device can be a portable electronic device that also includes other functions such as a personal digital assistant and / or a music player function, such as a mobile phone, a tablet computer, a wearable electronic device with a wireless communication function (such as a smart watch), etc. Exemplary embodiments of portable electronic devices include but are not limited to portable electronic devices equipped with an IOS system, an Android system, a Microsoft system or other operating systems. The above-mentioned portable electronic device may also be other portable electronic devices, such as a laptop computer (Laptop), etc. It should also be understood that in some other embodiments, the above-mentioned electronic device may not be a portable electronic device, but a desktop computer, a server, etc.

[0038] Based on the service processing method provided in the embodiment of this application, this application provides an application scenario diagram of the service processing method. Figure 1 , Figure 1 This is an application scenario diagram of a business processing method provided in an embodiment of the present application. Figure 1 As shown, the device 100 can obtain a business and determine the user behavior feature data and user portrait feature data of the business. The device 100 can be equipped with a trained business period model 110. By inputting the user portrait feature data and the divided user behavior feature data into the business period model 110, the recommended processing period of the business can be determined through the output of the business period model 110.

[0039] Among them, since the determined user behavior feature data can be sorted in time series, the device 100 can split the user behavior feature data into first feature data that is farther away from the current moment and second feature data that is closer to the current moment. In this way, the global and local characteristics of the business can be obtained, thereby improving the accuracy of the subsequently determined recommended processing time period.

[0040] The recommended processing time period may be displayed to the personnel who handle the business, so that the personnel can handle the business according to the recommended processing time period and improve the processing effect of the business as much as possible.

[0041] For example, if the business is a marketing recommendation business, the user behavior feature data at this time is the historical behavior features of the user who needs to be recommended, such as shopping behavior features, behavior features of answering telephone sales, behavior features of logging in to view shopping pages, etc. The user portrait feature data at this time can be the portrait features of the user who needs to be recommended, such as the age characteristics, gender characteristics, occupational characteristics, income characteristics, expenditure characteristics, shopping preference characteristics, etc. of the user who needs to be recommended, which are not specifically limited here. Specifically, the user behavior feature data at this time can be sorted with 1 day as a preset cycle and 1 hour as a time period, and arranged in chronological order from the past to the present, that is, all the behavior features of 24 time periods every day in the past N consecutive days can be obtained, and the user's historical behavior features can be split into long-term behavior features and recent behavior features to obtain the user's general preferences and recent preference fluctuations in different periods, so that the subsequent determined recommendation processing period is more accurate, and adding the user's portrait features for fusion processing can enhance the group expression ability and further improve the accuracy of the determined recommendation processing period.

[0042] See also Figure 2 , which is a flow chart of a business processing method provided in an embodiment of the present application. Figure 2 The business processing method shown may include the following steps:

[0043] Step 201, determine user behavior feature data and user portrait feature data of the business.

[0044] Among them, a business is a business that needs to be processed, such as the need to conduct telephone marketing to a target user, which is a business. The embodiment of the present application can determine the recommended processing time period for processing this business, and improve the effect of business processing, which is equivalent to determining in which time period to conduct telephone marketing to the target user to maximize the conversion rate. The business includes first business data and second business data, and the first business data includes multiple first businesses, and these multiple first businesses can come from multiple time periods of multiple business processing cycles.

[0045] The second business data may be basic data of the business.

[0046] For example, different business processing cycle lengths and business processing cycle numbers can be set according to different scenarios. For example, 1 day can be set as a business processing cycle, 30 can be set as the number of business processing cycles, and 1 hour can be set as a time period. In this way, the first business of each hour of each day in the past 30 days can be obtained. In the marketing recommendation scenario, the first business can be the data of the user's telemarketing, login, shopping and other behaviors. It should be noted that the first business data can be text data.

[0047] Among them, the second business data may include basic information dimensions: age, gender, education level, occupation, income, etc., which are helpful to identify the basic socioeconomic characteristics of users; geographical location, the user's place of residence can reflect their living environment and cultural background, which is instructive for market segmentation and regional product development; family structure: the user's marital status, number of children and other information, which is helpful to understand the user's living conditions and consumption needs.

[0048] In a possible embodiment, the second business data may include purchasing behavior, namely, information such as the user's purchase history, frequency, and preferred brands, which can reveal their consumption habits and brand loyalty; usage habits, namely, the user's frequency of use, usage time, and functional preferences for the product, which can help optimize product functions and enhance user stickiness; psychological characteristics, namely, the user's personality tendencies, lifestyles, and values, which can be used to design product features that are more in line with the user's psychological expectations.

[0049] In a possible embodiment, the second business data may include social network activities, i.e., the user's activity, interaction methods, social circles, etc. on social platforms, which helps to understand their social influence and information acquisition channels; technology acceptance, i.e., the user's acceptance of new technologies, common applications, operating system preferences, etc., which can guide the design and marketing strategies of technology products; information acquisition channels, i.e., the main channels for users to obtain information, such as search engines, news applications, professional forums, etc., which affect the formulation of content marketing and information dissemination strategies. By obtaining basic business data, data support can be provided for the subsequent determination of user portrait feature data.

[0050] Each first service may be embedded and sorted in time sequence to obtain the user behavior feature data.

[0051] Embedding plays a vital role in natural language processing (NLP). It converts text data into vector form so that the model can understand and process the data. Embedding mainly solves problems such as input limitation, context coherence, efficiency and accuracy. Specifically:

[0052] Breaking through input limitations: By encoding long text into compact high-dimensional vectors, Embedding enables large models to process text beyond their original input limits.

[0053] Maintaining contextual coherence: Preserving the contextual information of the text during encoding ensures that large models can still generate coherent output when processing segmented text.

[0054] Improve efficiency and accuracy: Pre-trained embeddings accelerate model training, improve the accuracy of various natural language processing tasks, and achieve cross-task knowledge transfer.

[0055] In addition, Embedding technology can capture the semantic information of data, making similar data closer in the embedding space, which helps the model better understand the relationship between data. In a typical Transformer architecture, the Embedding layer converts the current sequence into a vector, builds an embedding_table and finds the word vector corresponding to each word, and then returns the result to the output, so that the input word becomes a word vector. This conversion not only considers the input word sequence, but also needs to consider other additional information and position information to ensure that the embedding vector contains the input word sequence information, additional information and position information, so as to fully reflect the characteristics of the text. Application cases show that Embedding solves the input and coherence problems when large models process long texts, and optimizes the answer quality through vector retrieval and prompt engineering. The main operation methods of hybrid Embedding include stacking different pre-trained expected word vectors or using pre-trained word vector linear fusion to increase the effective information of Embedding. These methods perform well in improving the evaluation indicators F1-score and AUC values ​​of the model, further proving the effectiveness of Embedding operations in improving model performance.

[0056] Specifically, a sliding window with a length of N can be set to obtain all the behavioral features of users in t time periods over the past N consecutive days. An embedding operation can be performed on each first service to obtain the behavioral features of each user. For example, embedding the i-th first service x i Perform embedding operation to obtain where g i is the length of the encoding query list of the i-th first service. Finally, we get X = {x1, x2…, x f}, where i∈f represents the user behavior feature data containing f user behavior features.

[0057] The second business data of the business may be obtained, and then the second business data may be embedded to obtain the user portrait feature data. For example, for a marketing recommendation business, the second business data may be the personal portrait data of the user who needs marketing recommendation.

[0058] Specifically, the business basic data may be embedded by embedding the Embedding layer to obtain the second feature data.

[0059] It can be seen that in this way, accurate user behavior feature data and user portrait feature data of the business can be obtained, providing reliable data support for subsequent business processing.

[0060] Step 202: dividing the user behavior characteristic data into first characteristic data and second characteristic data according to a time period corresponding to each characteristic in the user behavior characteristic data.

[0061] The time period corresponding to any feature in the first feature data is before the time period corresponding to any feature in the second feature data.

[0062] Among them, the division time can be determined according to the time period corresponding to each feature in the user behavior feature data; then, the characteristics of the time period before the division time are determined as the first feature data; finally, the characteristics of the time period after the division time are determined as the second feature data.

[0063] For example, the length of user behavior feature data is N, and it is arranged in time sequence. The user behavior feature data can be divided into the first feature data composed of the first NK time lengths and the second feature data composed of the last K time lengths. The first feature data {R(1), R(2), ..., R(N-K+1)} composed of the first NK time slices are more distant time slice features, and the second feature data {R(N-K+1), ..., R(N)} composed of the last K time slices closer to the current moment. Among them, the N-K+1th features are global and local shared features.

[0064] The first characteristic data may reflect the user's longer-term preferences, and the second characteristic data may reflect the user's recent preferences. It is understandable that the choice of dividing the moments may be flexibly switched according to the scenario and actual needs.

[0065] It can be seen that in this way, the feature data can be divided at the time, which can improve the accuracy of the output results of the subsequent business period model.

[0066] Step 203: Input the first feature data, the second feature data and the user portrait feature data into the business period model for prediction to determine the recommended processing period for the business.

[0067] Among them, the business period model includes an extraction network and a fusion network. The first feature data and the second feature data can be processed by the extraction network to obtain first characterization data and second characterization data; the first characterization data, the second characterization data and the user portrait feature data are fused by the fusion network to obtain fused feature data; and the time period prediction is performed based on the fused feature data to obtain the recommended processing time period for the business.

[0068] Specifically, the extraction network may include a first extraction module and a second extraction module. The first extraction module may be used to process the first feature data and the second feature data to obtain a first vector set; and the first vector set may be processed to obtain a first weight matrix; the first weight matrix may be processed by the first extraction module to obtain the first representation data; the second feature data may be processed by the second extraction module to obtain a second vector set; and the second vector set may be processed to obtain a second weight matrix; and the second weight matrix may be processed by the second extraction module to obtain the second representation data.

[0069] Specifically, the first characterization data and the second characterization data can be first fused through the fusion network to obtain characterization fusion data; and the characterization fusion data and the user portrait feature data can be second fused through the fusion network to obtain the fused feature data.

[0070] For ease of understanding, an exemplary description is given below.

[0071] Among them, the first extraction module may include a first linear layer, a first attention layer and a first activation layer.

[0072] Among them, the first feature data and the second feature data can be input into the first linear layer for processing to obtain a first vector set, so that each vector in the first vector set obtained can correspond one-to-one to each feature in the first feature data and each feature in the second feature data; the first vector set is input into the first attention layer for processing to obtain the first weight matrix; the first weight matrix is ​​input into the first activation layer for processing to obtain the first representation data.

[0073] Among them, the linear layer is a basic layer in the neural network, also known as the fully connected layer or dense layer. Each node in this layer is connected to each node in the previous layer. Each connection has a weight to adjust the influence of the input data, and a bias to increase the flexibility of the network. The linear layer is usually located in the hidden layer or output layer of the neural network. In the linear layer, the input data is flattened into a vector, each input feature is multiplied by the corresponding weight, and then all the results are added together and the bias is added. This operation is actually a linear transformation that maps the input data to the output space. The output of the linear layer can be used as the input of the next layer of the neural network, or it can be the final output result, depending on the structure of the network and the task requirements.

[0074] For example, each first vector in the first vector set may be obtained by using the following formula:

[0075] h(i)=W h ·R(i)+b h ,i∈{1,2,…,N}

[0076] Among them, W h 、b h are the learnable parameters in the first linear layer, with dimensions N and K respectively.

[0077] For example, the first weight matrix can be obtained by using the following formula:

[0078]

[0079] Among them, W Q , W K , W V and b Q 、b K 、b V are the weight and bias learnable parameter matrices in the attention mechanism, K d is the feature dimension of input Y; e W is the differentiated representation weight after personalized selection, e A It is the weight factor for feature selection, Q represents query, K represents key, and V represents value.

[0080] For example, the first characterization data may be determined using the following formula:

[0081] In order to increase the nonlinear expression ability of time series feature learning, the first activation layer is introduced here, including the activation function (Gated Linear Unit, GLU). The GLU activation function is an activation function used in neural networks. It has a gating mechanism that can help the network better capture the long-term dependencies in sequence data. For the GLU activation module, the output of the attention mechanism network neuron needs to be twice the feature dimension of Y; the final output is:

[0082] [e A t ,e A s ]=e A

[0083] e Y =e A t ⊙σ(e A s )

[0084] Here A s and e A t They are all intermediate states in the GLU module, have no practical significance, and have a dimension of e A One half of Y is the output of the GLU module. The above formula is a general calculation formula, which is applicable to the first characterization data and the second characterization data. For the sake of distinction, in the embodiment of the present application, the first characterization data is represented by e Y h express.

[0085] It should be noted that the GLU activation function is defined as: Where x is the input vector, represents element-by-element multiplication, σ represents the Sigmoid function, and g(x) is the intermediate vector obtained by the fully connected layer or the convolutional layer.

[0086] In the formula, the intermediate vector g(x) is first gated and the Sigmoid function σ is used to map it to a range between 0 and 1, indicating the probability of each element being retained. Then, the input vector x is multiplied element by element with the gated vector (i.e. operation) to obtain the final output vector. This gating mechanism allows the GLU activation function to selectively filter certain parts of the input vector and adjust the output according to the context of the input. The function of the gating part is to classify the input into two categories and decide which parts should be retained and which parts should be suppressed. In this way, the GLU activation function can selectively activate or suppress different input information according to task requirements. It should be noted that the calculation of the intermediate vector g(x) is usually obtained through a fully connected layer or a convolutional layer, and the specific form and parameters will vary depending on the specific network structure and task.

[0087] It can be seen that the first representation data can be obtained through the first extraction module, which is used to represent the global representation of the user's longer-term preference.

[0088] Among them, the second extraction module may include a second linear layer, a second attention layer and a second activation layer.

[0089] Among them, the second feature data can be input into the second linear layer for processing to obtain a second vector set, so that each vector in the second vector set obtained corresponds one-to-one to each feature in the second feature data; the second vector set is input into the second attention layer for processing to obtain a second weight matrix; the second weight matrix is ​​input into the second activation layer for processing to obtain the second representation data.

[0090] For example, each second vector in the second vector set may be obtained by using the following formula:

[0091] h′(i)=W h′ ·R(i)+b h′ ,i∈{N-K+1,…,N}

[0092] Among them, W h′ 、b h′ are the learnable parameters in the second linear layer, with dimensions N and K respectively.

[0093] For example, the second weight matrix can be obtained by using the following formula:

[0094]

[0095] Among them, W Q , W K , W V and b Q 、b k 、b v are the weight and bias learnable parameter matrices in the attention mechanism, K d is the feature dimension of input Y; e W is the differentiated representation weight after personalized selection, eA It is the weight factor for feature selection, Q represents query, K represents key, and V represents value.

[0096] For example, the second characterization data may be determined using the following formula:

[0097] In order to increase the nonlinear expression ability of time series feature learning, the first activation layer is introduced here, including the activation function (Gated Linear Unit, GLU). The GLU activation function is an activation function used in neural networks. It has a gating mechanism that can help the network better capture the long-term dependencies in sequence data. For the GLU activation module, the output of the attention mechanism network neuron needs to be twice the feature dimension of Y; the final output is:

[0098] [e A t ,e A s ]=e A

[0099] e Y =e A t ⊙σ(e A s )

[0100] Here A s and e A t They are all intermediate states in the GLU module, have no practical significance, and have a dimension of e A One half of 11 is the output of the GLU module. The above formula is a general calculation formula, which is applicable to the first characterization data and the second characterization data. For the sake of distinction, in the embodiment of the present application, the second characterization data is represented by e Y h′ express.

[0101] It can be seen that the second representation data can be obtained through the second representation extraction network, which is used to represent the local representation of the user's recent preferences.

[0102] Among them, the fusion network can adopt a gating mechanism, and the first characterization data and the second characterization data can be subjected to a first fusion processing through the fusion network to obtain characterization fusion data; the characterization fusion data and the second feature data can be subjected to a second fusion processing through the fusion network to obtain the fused feature data.

[0103] For example, the fusion feature data can be determined using the following formula:

[0104]

[0105] O=O Y *X u

[0106] in, is the weight factor of long-term and short-term information, which represents the degree to which user preferences are affected by long-term and short-term information; O∈R 1*o , o is the output feature number, the first characterization data is e Y h , the second characterization data is e Y h′ , the second feature data is X u , characterizing the fused data as O Y , the fused feature data is O.

[0107] Finally, based on the fused feature data, a time period prediction is performed to obtain a predicted recommended processing time period. The fused feature data can be input into the prediction layer to obtain the recommended processing time period.

[0108] It can be seen that based on the feature representation learning of long-term and short-term modeling, and through sequence splitting modeling and attention mechanism, the global interests of the general preferences of users in the long term and short term and the similar interest information of dynamic fluctuations are effectively extracted respectively, which makes the features under different sequence meanings fully mined. Based on the obtained sequence interest representation and the enhanced fusion of user-side information, the gating mechanism makes the selection of weights of users affected by long-term and short-term more flexible, and the attribute information enhancement can further bring effective feature gain. The long-term and short-term fusion interest modeling involved in the present invention can mine different user interests from different sequence expression angles in effective sequence information, and help the model obtain the user's final personalized timing preference through reasonable representation learning and information selection. That is, the long-term and short-term sequences are split from the input modeling form, and then a more complex series of learning modules are designed in the feature learning, so that the extraction process of the entire sequence information is more parallelized, and the weight update between different time points is carried out synchronously, and a gating mechanism and information enhancement for selecting long-term and short-term features are designed, so that user personalization is more fully learned. For example, for marketing recommendation scenarios, users’ personalized interests are first explicitly modeled as the sum of long-term and short-term interest factors, which can reflect the general preferences and recent fluctuations of users in the telemarketing system at different times, and a gating mechanism is used to strengthen both, rather than simply modeling them as sequence models or recommendation tasks. Finally, static side information is added to enhance its group expression ability.

[0109] For ease of understanding, the model training process is illustrated below.

[0110] For example, in a telephone marketing scenario, the average call connection rate of each time period can be used as the connection rate representation of each time period, and the time period with a connection rate greater than a preset value can be selected as the label of the recommended processing time period.

[0111] In a possible embodiment, the attention mechanism in the embodiment of the present application can adopt a self-attention architecture, the number of neurons is 256, the learning rate is 0.003, and the Adam learner can be used for training.

[0112] The loss function in the embodiment of the present application may be a cross entropy loss function, which calculates the index between the predicted score output by the model and the actual preference of the user, thereby performing positive feedback training of the model. The specific mathematical expression is as follows:

[0113]

[0114] Among them, y i and Represent the user's real preference and predicted preference respectively. The business period model can be obtained by iteration until the model converges, which will not be described here.

[0115] For easier understanding, see Figure 3 , the following is an exemplary description of a business period model in an embodiment of the present application, Figure 3 A structural diagram of a business period model provided for an embodiment of the present application. First, the first business data is input into a shared embedding layer to obtain user behavior feature data. The first business data includes {R(1), R(2), ..., R(N)}, and the user behavior feature data includes {r(1), r(2), ..., r(N)}. It should be noted that the second business data also obtains user portrait feature data through the shared embedding layer (not shown in the figure).

[0116] Next, the user behavior feature data is divided into first feature data {r(1), r(2), …, r(N-K+1)} and second feature data {r(N-K+1), …, r(N)}, the first feature data {r(1), r(2), …, r(N)} are input into the first linear layer to obtain a first vector set {h(1), h(2), …, h(N)}, and the second feature data {r(N-K+1), …, r(N)} are input into the second linear layer to obtain a second vector set {h'(N-K+1), …, h'(N)}.

[0117] Next, the first vector set is input into the first attention layer, and the first representation data F(g) is obtained through the first activation layer. The second vector set is input into the second attention layer, and the second representation data F(r) is obtained through the second activation layer. It should be noted that the first attention layer and the second attention layer can be the same attention layer, processing the first vector set and the second vector set respectively; the first attention layer and the second attention layer can be different attention layers, processing the first vector set and the second vector set in parallel.

[0118] Next, the user portrait feature data, the first representation data and the second representation data are input into the fusion network, and the fused feature data is obtained by using the gating mechanism. Finally, the fused feature data is input into the prediction layer, and the recommended processing time period is obtained according to the output of the prediction layer.

[0119] It can be seen that based on the feature representation learning of long-term and short-term modeling, and through sequence splitting modeling and attention mechanism, the global interests of the general preferences of users in the long term and short term and the similar interest information of dynamic fluctuations are effectively extracted respectively, which makes the features under different sequence meanings fully mined. Based on the obtained sequence interest representation and the enhanced fusion of user-side information, the gating mechanism makes the selection of weights of users affected by long-term and short-term more flexible, and the attribute information enhancement can further bring effective feature gain. The long-term and short-term fusion interest modeling involved in the present invention can mine different user interests from different sequence expression angles in effective sequence information, and help the model obtain the user's final personalized timing preference through reasonable representation learning and information selection. That is, the long-term and short-term sequences are split from the input modeling form, and then a more complex series of learning modules are designed in the feature learning, so that the extraction process of the entire sequence information is more parallelized, and the weight update between different time points is carried out synchronously, and a gating mechanism and information enhancement for selecting long-term and short-term features are designed, so that user personalization is more fully learned. For example, for marketing recommendation scenarios, users’ personalized interests are first explicitly modeled as the sum of long-term and short-term interest factors, which can reflect the general preferences and recent fluctuations of users in the telemarketing system at different times, and a gating mechanism is used to strengthen both, rather than simply modeling them as sequence models or recommendation tasks. Finally, static side information is added to enhance its group expression ability.

[0120] Combine the following Figure 4 An electronic device in an embodiment of the present application is described. Figure 4 4 is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. The electronic device 400 includes a processor 401 , a memory 402 , and a communication bus 403 for connecting the processor 401 and the memory 402 .

[0121] In some possible implementations, the memory 402 includes but is not limited to random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or portable read-only memory (CD-ROM), and the memory 402 is used to store the program code executed by the electronic device 400 and the transmitted data.

[0122] In some possible implementations, the electronic device 400 further includes a communication interface for receiving and sending data.

[0123] In some possible implementations, the processor 401 may be one or more central processing units (CPUs). When the processor 401 is a central processing unit (CPU), the central processing unit (CPU) may be a single-core central processing unit (CPU) or a multi-core central processing unit (CPU).

[0124] In some possible implementations, the processor 401 may be a baseband chip, a chip, a central processing unit (CPU), a general-purpose processor, a DSP, an ASIC, an FPGA or other programmable logic device, a transistor logic device, a hardware component or any combination thereof.

[0125] In a specific implementation, the processor 401 in the electronic device 400 is used to execute the program instructions 421 stored in the memory 402 to perform the following operations:

[0126] Determine first feature data and second feature data of a service, wherein features in the first feature data are arranged in time sequence;

[0127] Dividing the first feature data into first feature data and second feature data, any feature in the first feature data is located before any feature in the second feature data;

[0128] The first feature data, the second feature data and the second feature data are input into a business period model to determine a recommended processing period for the business.

[0129] It can be seen that through the above-mentioned business processing method and related devices, the user behavior feature data and user portrait feature data of the business are determined; then, according to the time period corresponding to each feature in the user behavior feature data, the user behavior feature data is divided into the first feature data and the second feature data, and the time period corresponding to any feature in the first feature data is located before the time period corresponding to any feature in the second feature data; finally, the first feature data, the second feature data and the user portrait feature data are input into the business time period model for prediction to determine the recommended processing time period of the business. The user behavior feature data of the business can be split in the time period to obtain the first feature data and the second feature data. Since the first feature data is earlier than the second feature data in time sequence, the user behavior information closer to the current time and the user behavior information farther from the current time can be split, and combined with the user portrait feature data for fusion. Due to the introduction of the user behavior information closer to the current time and the user behavior information farther from the current time, it is possible to dynamically predict what behavior the user is more inclined to perform in the subsequent time period. At the same time, combined with the user portrait information, the recommended processing time period of the business can be automatically determined through the business time period model, which greatly improves the effect of business processing.

[0130] It should be noted that the specific implementation of each operation can adopt the corresponding description of the method embodiment shown above, and the electronic device 400 can be used to execute the above method embodiment of the present application, which will not be described in detail.

[0131] The above mainly introduces the scheme of the embodiment of the present application from the perspective of the execution process on the method side. It is understandable that, in order to realize the above functions, the electronic device includes a hardware structure and / or software module corresponding to the execution of each function. It should be easily appreciated by those skilled in the art that, in combination with the units and algorithm steps of each example described in the embodiments provided herein, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware 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 the present application.

[0132] The embodiment of the present application can divide the electronic device into functional units according to the above method example. For example, each functional unit can be divided according to each function, or two or more functions can be integrated into one processing unit. The above integrated unit can be implemented in the form of hardware or in the form of software functional units. It should be noted that the division of units in the embodiment of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation.

[0133] In the case of dividing each functional module into corresponding functional modules, Figure 5 This is a block diagram of the functional units of a service processing device provided in an embodiment of the present application. The service processing device 500 includes:

[0134] A determination unit 510, configured to determine user behavior feature data and user portrait feature data of a service;

[0135] A division unit 520, configured to divide the user behavior feature data into first feature data and second feature data according to a time period corresponding to each feature in the user behavior feature data, wherein a time period corresponding to any feature in the first feature data is before a time period corresponding to any feature in the second feature data;

[0136] The determination unit 510 is further used to input the first feature data, the second feature data and the user portrait feature data into a business period model for prediction, so as to determine a recommended processing period for the business.

[0137] In a possible embodiment, in determining the user behavior feature data and the user portrait feature data of the service, the determining unit 510 is specifically used to:

[0138] Acquire first business data of the business, where the first business data includes a plurality of first businesses, and the plurality of first businesses are from a plurality of time periods of a plurality of business processing cycles;

[0139] Embedding each first service and sorting them in time sequence to obtain the user behavior feature data;

[0140] Acquire second business data of the business;

[0141] The second business data is embedded to obtain the user portrait feature data.

[0142] In a possible embodiment, in terms of dividing the user behavior feature data into first feature data and second feature data according to a time period corresponding to each feature in the user behavior feature data, the dividing unit 520 is specifically configured to:

[0143] Determine the division time according to the time period corresponding to each feature in the user behavior feature data;

[0144] Determine the feature of the time period before the division time as the first feature data;

[0145] The characteristics of the time period after the division time are determined as the second characteristic data.

[0146] In a possible embodiment, the business period model includes an extraction network and a fusion network; in terms of inputting the first feature data, the second feature data, and the user portrait feature data into the business period model for prediction and determining the recommended processing period of the business, the determination unit 510 is specifically used to:

[0147] Processing the first feature data and the second feature data through the extraction network to obtain first characterization data and second characterization data;

[0148] The first characterization data, the second characterization data, and the user portrait feature data are fused through the fusion network to obtain fused feature data;

[0149] A time period prediction is performed based on the fused feature data to obtain a recommended processing time period for the service.

[0150] In a possible embodiment, the extraction network includes a first extraction module and a second extraction module; in terms of processing the first feature data and the second feature data through the extraction network to obtain the first characterization data and the second characterization data, the determination unit 510 is specifically used to:

[0151] Processing the first feature data and the second feature data by the first extraction module to obtain a first vector set; and processing the first vector set to obtain a first weight matrix;

[0152] Processing the first weight matrix by the first extraction module to obtain the first representation data;

[0153] Processing the second feature data by the second extraction module to obtain a second vector set; and processing the second vector set to obtain a second weight matrix;

[0154] The second weight matrix is ​​processed by the second extraction module to obtain the second characterization data.

[0155] In a possible embodiment, in terms of fusing the first characterization data, the second characterization data, and the user portrait feature data through the fusion network to obtain fused feature data, the determining unit 510 is specifically used to:

[0156] Performing a first fusion on the first representation data and the second representation data through the fusion network to obtain representation fusion data;

[0157] The representation fusion data and the user portrait feature data are secondly fused through the fusion network to obtain the fused feature data.

[0158] It can be seen that through the above-mentioned business processing method and related devices, first, the user behavior feature data and user portrait feature data of the business are determined; then, according to the time period corresponding to each feature in the user behavior feature data, the user behavior feature data is divided into the first feature data and the second feature data, and the time period corresponding to any feature in the first feature data is located before the time period corresponding to any feature in the second feature data; finally, the first feature data, the second feature data and the user portrait feature data are input into the business time period model for prediction to determine the recommended processing time period of the business. The user behavior feature data of the business can be split in the time period to obtain the first feature data and the second feature data. Since the first feature data is earlier than the second feature data in time sequence, the user behavior information closer to the current time and the user behavior information farther from the current time can be split, and combined with the user portrait feature data for fusion. Due to the introduction of the user behavior information closer to the current time and the user behavior information farther from the current time, it is possible to dynamically predict what behavior the user is more inclined to perform in the subsequent time period. At the same time, combined with the user portrait information, the recommended processing time period of the business can be automatically determined through the business time period model, which greatly improves the effect of business processing.

[0159] It should be noted that the specific implementation of each operation can adopt the corresponding description of the method embodiment shown above, and the business processing device 500 can be used to execute the above method embodiment of the present application, which will not be described in detail.

[0160] 6. Other examples

[0161] An embodiment of the present application also provides a chip, including a processor, a memory, and a computer program or instructions stored in the memory, wherein the processor executes the computer program or instructions to implement the steps described in the above method embodiment.

[0162] An embodiment of the present application also provides a chip module, including a transceiver component and a chip, the chip including a processor, a memory and a computer program or instructions stored in the memory, wherein the processor executes the computer program or instructions to implement the steps described in the above method embodiment.

[0163] An embodiment of the present application also provides a computer storage medium, wherein the computer storage medium stores a computer program for electronic data exchange, wherein the computer program enables a computer to execute part or all of the steps of any method recorded in the above method embodiment, and the above computer includes an electronic device.

[0164] The embodiment of the present application also provides a computer program product, the computer program product includes a non-transitory computer-readable storage medium storing a computer program, the computer program is operable to cause a computer to execute some or all of the steps of any method described in the method embodiment. The computer program product may be a software installation package, and the computer includes an electronic device.

[0165] It should be noted that, for the above-mentioned various embodiments, for the sake of simple description, they are all expressed as a series of action combinations. Those skilled in the art should be aware that the present application is not limited by the described order of actions, because some steps in the embodiments of the present application can be performed in other orders or simultaneously. In addition, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions, steps, modules or units involved are not necessarily required by the embodiments of the present application.

[0166] In the above embodiments, the embodiments of the present application have different focuses on the description of each embodiment. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0167] The steps of the method or algorithm described in the embodiments of the present application can be implemented in hardware or by executing software instructions by a processor. The software instructions can be composed of corresponding software modules, and the software modules can be stored in RAM, flash memory, ROM, EPROM, electrically erasable programmable read-only memory (electrically EPROM, EEPROM), registers, hard disks, mobile hard disks, read-only compact disks (CD-ROMs) or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an ASIC. In addition, the ASIC can be located in a terminal device or a management device. Of course, the processor and the storage medium can also be present in a terminal device or a management device as discrete components.

[0168] Those skilled in the art should be aware that in one or more of the above examples, the functions described in the embodiments of the present application can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiments of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server, or data center to another website site, computer, server, or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrations. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a digital video disc (DVD)), or a semiconductor medium (eg, a solid state disk (SSD)).

[0169] The modules / units included in the devices and products described in the above embodiments may be software modules / units or hardware modules / units, or may be partially software modules / units and partially hardware modules / units. For example, for the devices and products applied to or integrated in the chip, the modules / units included therein may all be implemented in the form of hardware such as circuits, or at least some of the modules / units may be implemented in the form of software programs, which run on the processor integrated inside the chip, and the remaining (if any) modules / units may be implemented in the form of hardware such as circuits; for the devices and products applied to or integrated in the chip module, the modules / units included therein may all be implemented in the form of hardware such as circuits, and different modules / units may be located in the same component (such as a chip, circuit module, etc.) or in different components of the chip module, or at least some of the modules / units may be implemented in the form of software programs. The software programs run on the processor integrated inside the chip, and the remaining (if any) modules / units may be implemented in the form of hardware such as circuits. It is implemented in the form of a software program, which runs on a processor integrated inside the chip module, and the remaining (if any) modules / units can be implemented in hardware such as circuits; for various devices and products applied to or integrated in the terminal equipment, the various modules / units contained therein can be implemented in hardware such as circuits, and different modules / units can be located in the same component (for example, chip, circuit module, etc.) or in different components in the terminal equipment, or, at least some modules / units can be implemented in the form of a software program, which runs on a processor integrated inside the terminal equipment, and the remaining (if any) modules / units can be implemented in hardware such as circuits.

[0170] The specific implementation methods described above further illustrate the purpose, technical solutions and beneficial effects of the embodiments of the present application. It should be understood that the above description is only the specific implementation method of the embodiments of the present application and is not intended to limit the protection scope of the embodiments of the present application. Any modifications, equivalent substitutions, improvements, etc. made on the basis of the technical solutions of the embodiments of the present application should be included in the protection scope of the embodiments of the present application.

Claims

1. A business processing method, characterized in that: The method comprises: Determine the user behavior feature data and user portrait feature data of the business; According to the time period corresponding to each feature in the user behavior feature data, the user behavior feature data is divided into first feature data and second feature data, wherein the time period corresponding to any feature in the first feature data is before the time period corresponding to any feature in the second feature data; The first feature data, the second feature data and the user portrait feature data are input into a business period model for prediction to determine a recommended processing period for the business.

2. The method according to claim 1, characterized in that The user behavior feature data and user portrait feature data for determining the service include: Acquire first business data of the business, where the first business data includes a plurality of first businesses, and the plurality of first businesses are from a plurality of time periods of a plurality of business processing cycles; Embedding each first service and sorting them in time sequence to obtain the user behavior feature data; Acquire second business data of the business; The second business data is embedded to obtain the user portrait feature data.

3. The method according to claim 1, characterized in that The dividing the user behavior feature data into first feature data and second feature data according to a time period corresponding to each feature in the user behavior feature data includes: Determine the division time according to the time period corresponding to each feature in the user behavior feature data; Determine the feature of the time period before the division time as the first feature data; The characteristics of the time period after the division time are determined as the second characteristic data.

4. The method according to claim 1, characterized in that The business period model includes an extraction network and a fusion network; the inputting the first feature data, the second feature data and the user portrait feature data into the business period model for prediction to determine the recommended processing period of the business includes: Processing the first feature data and the second feature data through the extraction network to obtain first characterization data and second characterization data; The first characterization data, the second characterization data, and the user portrait feature data are fused through the fusion network to obtain fused feature data; A time period prediction is performed based on the fused feature data to obtain a recommended processing time period for the service.

5. The method according to claim 4, characterized in that The extraction network includes a first extraction module and a second extraction module; the first feature data and the second feature data are processed by the extraction network to obtain first characterization data and second characterization data, including: Processing the first feature data and the second feature data by the first extraction module to obtain a first vector set; and processing the first vector set to obtain a first weight matrix; Processing the first weight matrix by the first extraction module to obtain the first representation data; Processing the second feature data by the second extraction module to obtain a second vector set; and processing the second vector set to obtain a second weight matrix; The second weight matrix is ​​processed by the second extraction module to obtain the second characterization data.

6. The method according to claim 4, characterized in that The fusing the first characterization data, the second characterization data, and the user portrait feature data through the fusion network to obtain fused feature data includes: Performing a first fusion on the first representation data and the second representation data through the fusion network to obtain representation fusion data; The representation fusion data and the user portrait feature data are secondly fused through the fusion network to obtain the fused feature data.

7. A service processing device, characterized in that: The device comprises: A determination unit, used to determine user behavior feature data and user portrait feature data of a business; a dividing unit, configured to divide the user behavior feature data into first feature data and second feature data according to a time period corresponding to each feature in the user behavior feature data, wherein a time period corresponding to any feature in the first feature data is before a time period corresponding to any feature in the second feature data; The determination unit is further used to input the first feature data, the second feature data and the user portrait feature data into a business period model for prediction, so as to determine a recommended processing period for the business.

8. An electronic device, characterized in that: include: A processor, a memory, and one or more programs; the one or more programs are stored in the memory and are configured to be executed by the processor, the programs comprising instructions for executing the steps in the method according to any one of claims 1-6.

9. A computer storage medium, characterized in that: The computer storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor is caused to perform the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The computer program product comprises a non-transitory computer-readable storage medium storing a computer program, wherein the computer program is operable to cause a computer to execute the method according to any one of claims 1-6.

Images