Daily active user number prediction method and device, electronic equipment and storage medium

Abstract

The invention discloses a daily active user number prediction method and device, electronic equipment and a storage medium, and the method comprises the steps: responding to a prediction request for the daily active user number of a target product on the Nth day after a preset reference day, determining an active user retention number of the initial active user group on the preset reference day after N days of attenuation, a daily newly-added user number from the preset reference day to the Nth day and a retention rate of the daily newly-added user on the Nth day; determining influence coefficients of the external influence factors on the retention number of active users, the number of newly-added users every day and the retention rate of the newly-added users every day in the Nth day; according to the respective corresponding influence coefficients, respectively adjusting the retention number of active users, the number of newly-added users every day and the retention rate of the newly-added users every day on the Nth day; and estimating the daily active user number of the Nth day according to the adjusted active user retention number, the daily new user number and the retention rate of the daily new user in the Nth day. By adopting the scheme of the invention, the accuracy of predicting the daily active user number can be improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly to a method, device, electronic device and storage medium for predicting the number of daily active users. Background Art

[0002] Driven by the digital wave, the number of daily active users (DAU) has become a key indicator to measure the degree of attention and user participation of a product or service in the market. Accurate prediction of the number of daily active users is crucial for enterprises to formulate strategic plans, develop product strategies, grasp market dynamics, optimize resource allocation, and reasonably arrange advertising placement time and budget, thereby improving operational efficiency and return on investment. Even for Internet products with a large number of users, accurate prediction of the number of daily active users in real-time bidding for advertising can play a key role in increasing advertising revenue.

[0003] Currently, traditional methods for estimating the number of daily active users mainly rely on predicting based on the trend of the DAU curve. However, this method has certain deficiencies: it only relies on the time-series characteristics of historical data for prediction, resulting in relatively low accuracy of the estimated number of daily active users using this prediction method. Therefore, how to improve the prediction accuracy of the number of daily active users has become a technical problem that urgently needs to be solved. Summary of the Invention

[0004] The present invention provides a method, device, electronic device and storage medium for predicting the number of daily active users.

[0005] According to one aspect of the present invention, there is provided a method for predicting the number of daily active users, including:

[0006] In response to a prediction request for the number of daily active users on the Nth day after a preset reference day for a target product, determining the number of active user retentions after N days of decay of the initial active user group on the preset reference day, the number of new users added each day between the preset reference day and the Nth day, and the retention rate of new users added each day on the Nth day;

[0007] Determining the influence coefficients of external influencing factors on the number of active user retentions, the number of new users added each day, and the retention rate of new users added each day on the Nth day;

[0008] According to the influence coefficients corresponding to the number of active user retentions, the number of new users added each day, and the retention rate of new users added each day on the Nth day, respectively adjusting the number of active user retentions, the number of new users added each day, and the retention rate of new users added each day on the Nth day;

[0009] Estimating the number of daily active users on the Nth day based on the adjusted number of active user retentions, the number of new users added each day, and the retention rate of new users added each day on the Nth day.

[0010] According to another aspect of the present invention, there is provided a daily active user number prediction device, including:

[0011] A first estimation module, configured to, in response to a prediction request for the daily active user number on the Nth day after a preset reference day for a target product, determine the active user retention number after N days of decay of the initial active user group on the preset reference day, the number of new users added each day between the preset reference day and the Nth day, and the daily new user retention rate;

[0012] An influence coefficient determination module, configured to determine the influence coefficients of external influence factors on the active user retention number, the number of new users added each day, and the daily new user retention rate;

[0013] An adjustment module, configured to adjust the active user retention number, the number of new users added each day, and the daily new user retention rate respectively according to the influence coefficients corresponding to the active user retention number, the number of new users added each day, and the daily new user retention rate;

[0014] A second estimation module, configured to estimate the daily active user number on the Nth day according to the adjusted active user retention number, the number of new users added each day, and the daily new user retention rate.

[0015] According to another aspect of the present invention, there is provided an electronic device, including:

[0016] At least one processor; and

[0017] A memory communicatively connected to the at least one processor; wherein,

[0018] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the daily active user number prediction method of the embodiment of the present invention.

[0019] According to another aspect of the present invention, there is provided a computer-readable storage medium storing computer instructions for causing a processor to implement the daily active user number prediction method of the embodiment of the present invention when executed.

[0020] The technical solution of the embodiment of the present invention comprehensively considers the influence degree of external influence factors on the estimation of the daily active user number, so that the accuracy of the predicted daily active user number is higher.

[0021] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0023] Figure 1 It is a schematic flowchart of a daily active user number prediction method provided by an embodiment of the present invention;

[0024] Figure 2 It is a schematic flowchart of another daily active user number prediction method provided by an embodiment of the present invention;

[0025] Figure 3 It is a schematic structural diagram of a daily active user number prediction device provided by an embodiment of the present invention;

[0026] Figure 4 It is a schematic structural diagram of an electronic device for implementing the daily active user number prediction method of the embodiment of the present invention. Detailed implementation manners

[0027] In order to enable those skilled in the art to better understand the solution of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0028] When predicting the daily active user count, traditional prediction methods based on the trend of the DAU (Daily Active User) curve can be adopted. However, this approach has certain deficiencies. That is, this prediction method only relies on the temporal characteristics of historical data and seriously ignores the rich internal and external factors that affect user behavior. From the internal dimension, the diverse attributes of individual users, including age, gender, regional cultural background, consumption psychology and habits, and the dynamic evolution of interests and hobbies, etc., will all cause significant differences and changes in their usage patterns of products or services. From the perspective of the external environment, the demand fluctuations caused by seasonal changes, the diversion of user attention due to social hot events, the changes in the market pattern caused by the strategic adjustments of competitors, etc., will all make the DAU data show highly complex non-linear change characteristics. Although in a controlled experimental data environment, such curve fitting methods may achieve good fitting effects with limited data, once migrated to a real and variable production scenario, due to the inability to effectively integrate the above complex factors, the prediction results often deviate far from the actual DAU value. Therefore, the present invention proposes a new method for predicting the daily active user count, and the specific implementation process can be seen in the following embodiments.

[0029] Embodiment 1

[0030] Figure 1 The following is a flowchart of a method for predicting the daily active user count provided by an embodiment of the present invention. This embodiment is applicable to scenarios where the daily active user count of a product needs to be predicted. This method can be executed by a device for predicting the daily active user count, and the device for predicting the daily active user count can be implemented in the form of hardware and / or software, and the device for predicting the daily active user count can be configured in an electronic device.

[0031] As Figure 1 shown, the method for predicting the daily active user count includes:

[0032] S101. In response to a prediction request for the daily active user count of the target product on the Nth day after the preset reference day, determine the active user retention count of the initial active user group of the preset reference day after N days of decay, the number of new users per day between the preset reference day and the Nth day, and the retention rate of new users on the Nth day.

[0033] In the embodiments of the present invention, the target product can be any one of various websites, Internet applications, and online games. The preset reference day can be optionally the first day when the target product is launched, or the current day when the daily active user count needs to be predicted can be used as the preset reference day, and no specific limitation is made here. N can be an integer value set according to the prediction requirements. The initial active user group of the preset reference day refers to the total number of active users of the target product on the preset reference day.

[0034] In the embodiments of the present invention, through analysis, it is determined that the daily active user number of the target product on the Nth day after the preset reference day consists of two parts. The first part is the part remaining after the initial active user group on the preset reference day has naturally decayed and fluctuated for N days, and the second part is the sum of the newly added users every day between the preset reference day and the Nth day and the retention contribution on the Nth day. Therefore, in order to estimate the daily active user number on the Nth day, it is necessary to calculate the active user retention number of the initial active user group on the preset reference day after N days of decay, as well as the number of newly added users every day between the preset reference day and the Nth day and the retention rate of the newly added users on the Nth day. Optionally, a fixed calculation formula can be used, or other calculation formulas can be adopted, which are not specifically limited herein.

[0035] S102. Determine the influence coefficients of external influencing factors on the active user retention number, the number of newly added users every day, and the retention rate of the newly added users on the Nth day.

[0036] The applicant has found that during special periods or when major adjustments are made to the product business strategy, it will have a great impact on the prediction of the daily active user number of the target product. Taking Ramadan as an example, during Ramadan, some users will have unique changes in aspects such as the time distribution, frequency, and preferred functions of using the product or service. It is difficult for ordinary exponential decay models to depict such changes; at the same time, during Ramadan, market promotion is restricted and user consumption preferences change, which also makes the scale and retention pattern of newly added users every day different from usual. Another example is that during the summer and winter vacations, the significant adjustment of the work and rest rules of the student group deeply affects their usage activity of various products or services, thus interfering with the calculation logic of the overall daily active user number. In addition, strategy adjustments such as internal product function updates, interface design revisions, and charging mode reforms within an enterprise will all trigger deep changes in the user inflow and outflow patterns and usage behaviors, resulting in the failure of the calculation method of the daily active user number based on a fixed formula, leading to a significant increase in the prediction deviation in some months and inaccurate estimation of the daily active user number.

[0037] The applicant has found that the external influencing factors that have a greater impact on the estimation of the daily active user number mainly include Ramadan time, summer and winter vacation time, the product strategy of the target product, uncontrollable factors, etc. On this basis, in order to accurately estimate the daily active user number of the target product on a certain day, the present invention focuses on analyzing the influence coefficient k of external influencing factors on the active user retention number o , the influence coefficient k of external influencing factors on the number of newly added users every day n and the influence coefficient k of external influencing factors on the retention rate of the newly added users on the Nth day. r . Among them, the influence coefficient k o is used to measure the influence degree of external influencing factors on the active user retention number; similarly, the influence coefficient k n is used to measure the influence degree of external influencing factors on the number of newly added users every day; the influence coefficient kr Used to measure the influence degree of external influencing factors on the retention rate of newly added users on the Nth day.

[0038] In an alternative implementation, an agent based on the Deep Deterministic Policy Gradient (DDPG) algorithm is pre-trained; wherein, the structure of the agent includes an actor network and a critic network. On this basis, determine the influence coefficients of external influencing factors on the number of retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day, including: inputting the Ramadan time, the time of summer and winter vacations, the product strategy of the target product, uncontrollable factors, the number of retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day into the pre-trained agent based on the deep deterministic policy gradient algorithm; through the actor network in the agent structure, predict the influence coefficients of the Ramadan time, the time of summer and winter vacations, the product strategy of the target product, and uncontrollable factors on the number of retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day.

[0039] It should be noted that before the external influencing factors are input into the agent, the Ramadan time, the time of summer and winter vacations, the product strategy of the target product, and uncontrollable factors need to be quantified and encoded respectively to obtain their corresponding vector data. Exemplarily, for the Ramadan time vector data, when encoding, if it is during the Ramadan period, the corresponding element in the vector is 1, otherwise it is 0; the summer vacation vector data can set the vector element value between 0 and 1 according to the summer vacation time range and influence in different regions, indicating the degree of influence of the summer vacation on user behavior; the product strategy vector data can be quantified and encoded according to the type and intensity of product function updates, promotional activities, etc.; other uncontrollable factor vectors can set the vector element value between 0 and 1 according to the time corresponding to some abnormal traffic. For example, a major function update can set the corresponding element in the vector to 0.8, and a small promotional activity to 0.3, etc.

[0040] S103. According to the respective influence coefficients of the number of retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day, adjust the number of retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day respectively.

[0041] In the embodiments of the present invention, in addition to determining the influence coefficients of external influencing factors on the number of retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day, the pre-trained agent can also execute steps S103 - S104 to realize the prediction of the number of daily active users. Specifically, through the pre-trained agent, adjust the number of retained active users, the number of newly added users per day, and the retention rate of newly added users according to the following formula:

[0042] ODAU adjusted (N)=ODAU(N)×(1 + k o );

[0043] NDAU adjusted (i) = NDAU(i) × (1 + k n );

[0044] R adjusted (i) = R(i) × (1 + k r );

[0045] Among them, ODAU adjusted (N) represents the number of active user retentions on the Nth day after adjustment; ODAU(N) represents the number of active user retentions on the Nth day before adjustment; NDAU adjusted (i) represents the number of new users on the ith day after adjustment, and the ith day is any day between the preset reference day and the Nth day; NDAU(i) represents the data of new users on the ith day before adjustment; R adjusted (i) represents the retention rate of new users on the ith day after adjustment on the Nth day; R(i) represents the retention rate of new users on the ith day before adjustment on the Nth day; k o represents the influence coefficient of external influencing factors on the number of active user retentions; k n represents the influence coefficient of external influencing factors on the number of new users per day; k r represents the influence coefficient of external influencing factors on the retention rate of new users on the Nth day per day.

[0046] S104. Estimate the daily active user number on the Nth day according to the adjusted number of active user retentions, the number of new users per day, and the retention rate of new users on the Nth day per day.

[0047] In an optional implementation manner, the intelligent body determines the daily active user number on the Nth day according to the following formula through pre-trained intelligence:

[0048]

[0049] Among them, DAU predict (N) represents the estimated daily active user number on the Nth day.

[0050] To illustrate the solution of the present invention in detail, the intelligent body involved in the present invention is described. The intelligent body based on the deep deterministic policy gradient algorithm includes an actor network and a critic network. Let the parameter of the actor network be θ, and the input state be The output action A t , then the actor network can be expressed as: Among them, μ θ represents the prediction function of the trained actor network based on the input state vector. The loss function of the actor network aims to maximize the cumulative reward and is updated through the policy gradient method: Among them, m is the number of samples, For taking action A in state The obtained reward. The reward function can be defined according to the error between the predicted number of daily active users and the actual number of daily active users. Update the actor network parameters through the gradient ascent algorithm: t where α is the learning rate. When calculating , according to the policy gradient theorem: where is the output of the critic network, that is, the value estimate of taking action A in state . For taking action A t in state

[0051] Specifically, in each iteration: ① Randomly sample a batch of experience tuples from the experience replay buffer ② According to the current actor network parameters θ, calculate the action prediction A in state, calculate the predicted value of the number of daily active users DAU and the reward t ③ Use the critic network to calculate the Q value of taking action A in state. ④ According to the feedback of the critic network, calculate the loss function J(θ) of the actor network, and use a suitable optimization algorithm (such as the Adam optimizer) to calculate the gradient of the loss function with respect to the actor network parameters θ For taking action A t in state. ⑤ Update the actor network parameters according to the calculated gradient Critic network: Used to evaluate the value of the actions generated by the actor network. It takes the state and action as inputs and outputs an estimated value function Let the parameters of the critic network be Then the critic network can be expressed as The target value of the critic network is: The loss function of the critic network adopts the mean squared error (MSE): Update the parameters of the critic network by minimizing this loss function. Calculate the gradient of the loss function with respect to , at the same time, use the backpropagation algorithm for calculation, and then use an optimization algorithm (such as Adam) to update

[0052]

[0052] Specifically, in each iteration: ① Randomly sample a batch of experience tuples from the experience replay buffer ② Use the actor network μ θ to generate an action for the current state ③ The environment gives a reward according to the real data and the next state Calculate the target value of the critic network: ​ Among them, γ is the discount factor, representing the importance of future rewards. ④ Calculate the loss of the critic network And update the parameter gradient of the critic network ⑤ According to the calculated gradient, update the parameters of the critic network where β is the learning rate of the critic network.

[0053] During the training process, the agent continuously interacts with the environment (constituted by historical data and external factors), generates actions according to the current state, receives the rewards feedback by the environment (based on the error between the predicted daily active user number DAU value and the real daily active user number DAU value), and uses this information to update the parameters of the actor network and the critic network, gradually optimizing the prediction strategy to improve the accuracy of DAU prediction. The trained agent can implement steps S102 - S104.

[0054] In the embodiments of the present invention, the influence degree of external influencing factors on the prediction of the daily active user number is comprehensively considered, making the prediction of the daily active user number more accurate.

[0055] Embodiment 2

[0056] Figure 2 It is a flowchart of a method for predicting the daily active user number provided by the embodiments of the present invention.

[0057] In this embodiment, it is creatively proposed to perform fitting processing on the number of retained active users, the number of newly added users per day, and the retention rate of newly added users on the t-th day respectively to obtain the best fitting expression or model.

[0058] Specifically, the fitting process of the number of retained active users is as follows:

[0059] According to the historical user data of the target product, construct the first fitting sample data set; among them, each sample in the first fitting sample data set represents whether a user retains on the t-th day; based on the first fitting sample data set, through the combination of the exponential function and the sine function, perform fitting processing on the change of the number of retained active users over time to obtain the fitting expression of the number of retained active users. Specifically, construct the expression of the retention data on the t-th day through the combination of the exponential function and the sine function. For example, the constructed expression is as follows: where ODAU(t) represents the retention data of users on the t-th day after a certain day in history; t iDenote the retention on the i-th day t; a, b, c, d, e are parameters to be determined; where the initial amplitude of the attenuation function. When a > 0, the larger its value, the higher the starting value of the function; when a < 0, the starting value of the function is in the negative y-axis direction and the larger the absolute value, the lower the starting value; b represents the parameter affecting the attenuation rate in the exponential decay function, the larger the b value, the faster the function decays; c is the coefficient of the sine function part, used to control the amplitude of the sine function, the larger |c|, the greater the fluctuation amplitude of the sine function in the axis direction; d determines the period of the sine function, the larger the d value, the shorter the period, meaning the sine function fluctuates more frequently; e is the initial phase of the sine function. The fitting process is to solve the parameters to be determined, construct an error function, and take the partial derivative of the error function; then use an iterative optimization algorithm (such as the gradient descent method) to solve the parameters to be determined. After the solution is completed, the fitting expression of the active user retention number can be obtained.

[0060] It should be noted that for the fitting expression of the active user retention number, the power function part can be used to capture the long-term attenuation trend of the user retention number. As time increases, gradually decreases, which is in line with the actual situation that users gradually lose over time. The sine function part c×sin(dt i +e) is used to characterize the possible periodic fluctuations. For example, the user activity of some products may show regular changes within a weekly or monthly cycle, and this periodic factor can be simulated by the sine function. In this way, the fitting expression of the active user retention number determined based on the combination of the exponential function and the sine function can calculate the active user retention number on any day more accurately. In addition, it should be noted that because the retention numbers of the historical active users on each subsequent day when counting the historical active users on a certain day will be affected by some abnormal factors and inaccurate rules will be obtained (for example, Day1 is a certain day determined as the benchmark, and there is a natural disaster factor on Day1 + 10. It is possible that the 1-day retention number of the historical active users on Day1 + 9 is affected, while the 1-day retention number of the historical active users on Day1 + 8 is normal. It is also possible that the 100-day retention number of the historical active users on Day1 - 90 is affected, while the 100-day retention number of the historical active users on Day1 - 89 is normal, etc.). Therefore, when constructing the sample data set, if abnormal data is found in the historical data, the abnormal data is cleared to avoid negative impacts on the fitting process.

[0061] Specifically, the fitting process for the daily new user count includes: constructing a second fitting sample data set based on the historical user data of the target product; where each sample in the second fitting sample data set is a historical new user sequence; that is, a sample is a sequence of daily new user counts before a certain day; the true value of the sample is the sequence of daily new user counts after a certain day. Based on the second fitting sample data set, the long short-term memory network (i.e., LSTM network) is used to fit the change of the daily new user count over time, and the target long short-term memory network model is obtained. During the fitting process, after the sample is input into the long short-term memory network, the loss error is calculated according to the sequence of daily new user counts predicted by the long short-term memory network and the true value of the sample; furthermore, the gradients of the loss error with respect to the weight matrix and the bias vector are calculated; the weight and bias parameters are adjusted by the gradient descent method to minimize the calculated loss, thereby achieving effective fitting.

[0062] It should be noted that the long short-term memory network selectively retains or forgets historical information through the gating mechanism, so it can learn the long-term dependence relationships and dynamic change patterns in the data. For example, a marketing campaign may have a greater impact on the new user count in the short term, but as time goes by, its influence may gradually weaken. The long short-term memory network can capture this complex time dependence relationship, thus more accurately predicting the future new user count.

[0063] The fitting process for the retention rate of daily new users on the t-th day includes: constructing a third fitting sample data set according to the historical user data of the target product; where each sample in the third fitting sample data set represents the retention rate of the new users on the i-th day on the t-th day; and the retention rate of the new users on the i-th day on the t-th day is equal to the quotient of the number of users who are still active on the t-th day among the new users on the i-th day and the number of new users on the i-th day; based on the third fitting sample data set, a power function is used to fit the change trend of the retention rate of daily new users on the t-th day over time, and the fitting expression of the retention rate of daily new users on the t-th day is obtained.

[0064] In this embodiment, the constructed power function is as follows: where m is the number of samples; t i is the retention rate of the i-th t-th day; a, b, and c are parameters to be determined; where a is the coefficient of the power function part, which determines the amplitude of the power function part; b is the parameter that affects the change trend of the function in the power function; c can be a constant term. During specific fitting, an error function is constructed, and partial derivatives are taken with respect to a, b, and c of the error function respectively; furthermore, an iterative optimization algorithm is used to solve the values of a, b, and c.

[0065] It should be noted that the power function a×t i -bIt can better describe the decreasing trend of the retention rate of new users over time. Usually, the retention rate of new users is relatively high in the initial stage. However, as time goes by, due to various factors (such as the decrease in product freshness, the attraction of competing products, etc.), the retention rate will gradually decline, and the decline rate may gradually slow down. The characteristics of the power function conform to this law.

[0066] Based on the above, the daily active user count can be predicted according to the steps of S201 - S206.

[0067] S201. In response to a prediction request for the daily active user count of the target product on the Nth day after the preset reference day, according to the fitting expression of the active user retention count, determine the active user retention count after N days of decay for the initial active user group on the preset reference day.

[0068] For example, directly substitute the value of N into the fitting expression of the active user retention count for solution to obtain the active user retention count after N days of decay for the initial active user group on the preset reference day.

[0069] S202. In response to a prediction request for the daily active user count of the target product on the Nth day after the preset reference day, according to the target long - short - term memory network model, determine the number of new users per day between the preset reference day and the Nth day.

[0070] The sequence of the number of new users per day in part before the preset reference day can be input into the fitted target long - short - term memory network model, and according to the model output, the number of new users per day between the preset reference day and the Nth day can be obtained.

[0071] S203. In response to a prediction request for the daily active user count of the target product on the Nth day after the preset reference day, according to the fitting expression of the retention rate of new users on the t - th day, determine the retention rate of new users on the Nth day for each day between the preset reference day and the Nth day.

[0072] The values between the reference day and the Nth day can be respectively input into the fitting expression of the retention rate of new users on the t - th day. According to the calculation results of the fitting expression, the retention rate of new users on the Nth day for each day between the preset reference day and the Nth day can be obtained.

[0073] S204. Determine the influence coefficients of external influencing factors on the active user retention count, the number of new users per day, and the retention rate of new users on the Nth day.

[0074] In an alternative implementation, an agent based on the Deep Deterministic Policy Gradient (DDPG) algorithm is pre-trained; wherein, the structure of the agent includes an actor network and a critic network. On this basis, the influence coefficients of external influencing factors on the number of retained active users, the number of new users per day, and the retention rate of new users on the Nth day are determined, including: inputting the Ramadan time, the time of winter and summer vacations, the product strategy of the target product, uncontrollable factors, the number of retained active users, the number of new users per day, and the retention rate of new users on the Nth day into the pre-trained agent based on the Deep Deterministic Policy Gradient algorithm; through the actor network in the agent structure, predicting the influence coefficients of the Ramadan time, the time of winter and summer vacations, the product strategy of the target product, and uncontrollable factors on the number of retained active users, the number of new users per day, and the retention rate of new users on the Nth day. It can be understood that the above implementation process is essentially to directly use the Ramadan time, the time of winter and summer vacations, the product strategy of the target product, uncontrollable factors, the number of retained active users, the number of new users per day, and the retention rate of new users on the Nth day, etc. as static inputs and directly input them into the agent for processing. In another alternative implementation, the external influencing factors (Ramadan time, the time of winter and summer vacations, the product strategy of the target product, different seasonal times, data related to the characteristics of each season, uncontrollable factors) can be concatenated with the number of retained active users, the number of new users per day, and the retention rate of new users on the Nth day to obtain an input sequence, and then the input sequence is input into the Transformer Encoder for feature extraction to obtain a dynamic feature vector with a fixed dimension. This vector combines the external influencing factors and user behavior data (i.e., the number of retained active users, the number of new users per day, and the retention rate of new users on the Nth day), and each dimension represents a certain latent feature learned by the model. The structure of the above Transformer Encoder may include a multi-head attention mechanism, positional encoding, a feed-forward neural network, etc.; wherein, the multi-head attention mechanism allows the model to simultaneously focus on the input sequence from multiple perspectives. For example, for the seasonal factor part in the input sequence, one head may focus on the relationship between the activity arrangements in each season and the historical user retention data of ODAU, and find the change rules of user retention during seasonal activities; for the product strategy part, another head may focus on the immediate impact of the launch of new features on the new user data of NDAU. Through the parallel calculation of multiple heads, the model can mine the complex interactions between different factors. Positional Encoding: Since the input sequence contains time-related information (such as the alternation of seasons, the order of product strategy updates, etc.), positional encoding adds position information to the input at each position. This helps the model understand the time series characteristics of the data. For example, through positional encoding, the model can distinguish the differences in user behavior before and after the change of the product strategy, or the changing trend of the impact of different seasons on user behavior.Feed-Forward Network: After the multi-head attention mechanism, the feed-forward neural network performs independent non-linear transformations on the features at each position. This step further enhances the model's ability to represent features and enables it to learn more complex patterns in the input sequence. For example, it can learn the non-linear relationship between market environment changes and competitor dynamics changes, and how this relationship interacts with user behavior data. After being processed by the above model structure, a dynamic feature vector that synthesizes various information can be output, and then this dynamic feature vector is input into the agent. The agent calculates the influence coefficient based on the received dynamic feature vector and executes subsequent steps S205 - S206 to complete the prediction of the daily active user count.

[0075] S205. Adjust the active user retention count, the daily new user count, and the retention rate of new users on the Nth day respectively according to the corresponding influence coefficients of the active user retention count, the daily new user count, and the retention rate of new users on the Nth day.

[0076] Optionally, through a pre-trained agent, adjust the active user retention count, the daily new user count, and the new user retention rate according to the following formula:

[0077] ODAU adjusted (N) = ODAU(N) × (1 + k o );

[0078] NDAU adjusted (i) = NDAU(i) × (1 + k n );

[0079] R adjusted (i) = R(i) × (1 + k r );

[0080] Where, ODAU adjusted (N) represents the adjusted active user retention count on the Nth day; ODAU(N) represents the active user retention count on the Nth day before adjustment; NDAU adjusted (i) represents the adjusted new user count on the ith day, where the ith day is any day between the preset reference day and the Nth day; NDAU(i) represents the new user data on the ith day before adjustment; R adjusted (i) represents the retention rate of new users on the Nth day for the adjusted new users on the ith day; R(i) represents the retention rate of new users on the Nth day for the new users on the ith day before adjustment; k o represents the influence coefficient of external influencing factors on the active user retention count; k n represents the influence coefficient of external influencing factors on the daily new user count; k r represents the influence coefficient of external influencing factors on the retention rate of new users on the Nth day for daily new users.

[0081] S206. Estimate the daily active user count on the Nth day based on the adjusted retained active user count, the daily new user count, and the retention rate of daily new users on the Nth day.

[0082] Optionally, determine the daily active user count on the Nth day through a pre-trained agent according to the following formula:

[0083]

[0084] where DAU predict (N) represents the estimated daily active user count on the Nth day.

[0085] In this embodiment, by using different functions or formulas, fit the retained active user count, the daily new user count, and the t-day retention rate of the daily new user count. Compared with the fixed formula, a more accurate fitting expression or model can be fitted, providing a guarantee for accurately estimating the daily active user count subsequently.

[0086] Embodiment III

[0087] Figure 3 The following is a schematic structural diagram of a daily active user count prediction device provided by an embodiment of the present invention. This embodiment is applicable to scenarios where the daily active user count of a product needs to be predicted. For example Figure 3 As shown, the device includes:

[0088] A first estimation module 301, configured to, in response to a prediction request for the daily active user count on the Nth day after a preset reference day for a target product, determine the retained active user count after N days of decay of the initial active user group on the preset reference day, the daily new user count between the preset reference day and the Nth day, and the retention rate of daily new users on the Nth day;

[0089] An influence coefficient determination module 302, configured to determine the influence coefficients of external influence factors on the retained active user count, the daily new user count, and the retention rate of daily new users on the Nth day;

[0090] An adjustment module 303, configured to respectively adjust the retained active user count, the daily new user count, and the retention rate of daily new users on the Nth day according to the respective influence coefficients corresponding to the retained active user count, the daily new user count, and the retention rate of daily new users on the Nth day;

[0091] A second estimation module 304, configured to estimate the daily active user count on the Nth day based on the adjusted retained active user count, the daily new user count, and the retention rate of daily new users on the Nth day.

[0092] In some embodiments, it further includes a first fitting module, configured to:

[0093] Construct a first fitting sample data set based on the historical user data of the target product; wherein, each sample in the first fitting sample data set represents whether a user retains on the t-th day.

[0094] Based on the first fitting sample data set, perform a fitting process on the change of the number of retained active users over time through a combination of an exponential function and a sine function to obtain a fitting expression for the number of retained active users.

[0095] Correspondingly, the first estimation module 301 is further configured to:

[0096] In response to a prediction request for the daily active user number of the target product on the N-th day after the preset reference day, determine the number of retained active users after N days of decay of the initial active user group on the preset reference day according to the fitting expression of the number of retained active users.

[0097] In some embodiments, it further includes a second fitting module, which is configured to:

[0098] Construct a second fitting sample data set according to the historical user data of the target product; wherein, each sample in the second fitting sample data set is a historical new user sequence.

[0099] Based on the second fitting sample data set, perform a fitting process on the change of the daily new user number over time through a long short-term memory network to obtain a target long short-term memory network model.

[0100] Correspondingly, the first estimation module 301 is further configured to:

[0101] In response to a prediction request for the daily active user number of the target product on the N-th day after the preset reference day, determine the daily new user number between the preset reference day and the N-th day according to the target long short-term memory network model.

[0102] In some embodiments, it further includes a third fitting module, which is configured to:

[0103] Construct a third fitting sample data set according to the historical user data of the target product; wherein, each sample in the third fitting sample data set represents the retention rate of new users added on the i-th day on the t-th day.

[0104] Based on the third fitting sample data set, perform a fitting on the change trend of the retention rate of new users added each day on the t-th day over time through a power function to obtain a fitting expression for the retention rate of new users added each day on the t-th day.

[0105] Correspondingly, the first estimation module 301 is further configured to:

[0106] In response to a prediction request for the number of daily active users of a target product on the Nth day after a preset reference day, according to the fitting expression of the retention rate of newly added users on the tth day, determine the retention rate of newly added users on the Nth day between the preset reference day and the Nth day.

[0107] In some embodiments, the external influencing factors include the Ramadan time, the winter and summer vacation time, the product strategy of the target product, and uncontrollable factors;

[0108] The influence coefficient determination module 302 is further configured to:

[0109] Input the Ramadan time, the winter and summer vacation time, the product strategy of the target product, uncontrollable factors, the number of retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day into a pre-trained agent based on the deep deterministic policy gradient algorithm; wherein, the structure of the agent includes an actor network and a critic network;

[0110] Through the actor network in the agent structure, predict the influence coefficients of the Ramadan time, the winter and summer vacation time, the product strategy of the target product, and uncontrollable factors on the number of retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day.

[0111] In some embodiments, the adjustment module 303 is further configured to:

[0112] Adjust the number of retained active users, the number of newly added users per day, and the new user retention rate through the agent according to the following formula:

[0113] ODAU adjusted (N)=ODAU(N)×(1 + k o );

[0114] NDAU adjusted (i)=NDAU(i)×(1 + k n );

[0115] R adjusted (i)=R(i)×(1 + k r );

[0116] Wherein, ODAU adjusted (N) represents the number of retained active users on the Nth day after adjustment; ODAU(N) represents the number of retained active users on the Nth day before adjustment; NDAU adjusted (i) represents the number of newly added users on the ith day after adjustment, and the ith day is any day between the preset reference day and the Nth day; NDAU(i) represents the data of newly added users on the ith day before adjustment; R adjusted (i) represents the retention rate of newly added users on the Nth day on the ith day after adjustment; R(i) represents the retention rate of newly added users on the Nth day on the ith day before adjustment; ko represents the influence coefficient of external influencing factors on the number of retained active users; k n represents the influence coefficient of external influencing factors on the number of newly added users per day; k r represents the influence coefficient of external influencing factors on the retention rate of newly added users on the Nth day.

[0117] In some embodiments, the second prediction module 304 is further configured to:

[0118] The agent determines the daily active user number on the Nth day according to the following formula:

[0119]

[0120] where DAU predict (N) represents the predicted daily active user number on the Nth day.

[0121] The daily active user number prediction device provided by the embodiments of the present invention can execute the daily active user number prediction method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0122] Embodiment 4

[0123] Figure 4 FIG. shows a schematic structural diagram of an electronic device 10 that can be used to implement the embodiments of the present invention. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0124] As Figure 4 shown, the electronic device 10 includes at least one processor 11, and a memory communicatively connected to at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc. Among them, the memory stores a computer program executable by at least one processor. The processor 11 can execute various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. The input / output (I / O) interface 15 is also connected to the bus 14.

[0125] Multiple components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disc, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0126] The processor 11 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as executing the daily active user number prediction method.

[0127] In some embodiments, the daily active user number prediction method can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the daily active user number prediction method described above can be executed. Alternatively, in other embodiments, the processor 11 can be configured to execute the daily active user number prediction method in any other suitable manner (e.g., by means of firmware).

[0128] The various embodiments of the systems and technologies described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a dedicated or general-purpose programmable processor, and can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0129] A computer program for implementing the method of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable daily active user number prediction devices, such that when the computer programs are executed by the processor, the functions / operations specified in the flowchart and / or block diagram are implemented. The computer programs can be executed entirely on the machine, partially on the machine, executed partially on the machine as an independent software package and partially on a remote machine, or executed entirely on a remote machine or server.

[0130] In the context of the present invention, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, the computer-readable storage medium can be a machine-readable signal medium. More specific examples of the machine-readable storage medium would include electrical connections based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0131] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) through which the user can provide input to the electronic device. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0132] The systems and techniques described herein can be implemented in a computing system including backend components (e.g., as a data server), or a computing system including middleware components (e.g., an application server), or a computing system including frontend components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described herein), or a computing system including any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected with each other by digital data communication in any form or medium (e.g., a communication network). Examples of communication networks include: local area network (LAN), wide area network (WAN), blockchain network, and the Internet.

[0133] A computing system can include a client and a server. The client and the server are generally remote from each other and typically interact via a communication network. The client-server relationship is created by computer programs running on respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system and solves the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services.

[0134] It should be understood that various forms of the processes shown above can be used, with steps reordered, added, or deleted. For example, the steps recited in the present invention can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved, and no limitation is imposed herein.

[0135] The above specific embodiments do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for predicting the number of daily active users, characterized in that, Including: In response to a prediction request for the number of daily active users on the Nth day after the preset benchmark day of the target product, determining the number of retained active users after N days of decay of the initial active user group on the preset benchmark day, the number of newly added users per day between the preset benchmark day and the Nth day, and the retention rate of newly added users on the Nth day; Determining the influence coefficients of external influencing factors on the number of retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day; According to the influence coefficients corresponding to the number of retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day respectively, adjusting the number of retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day; Estimating the number of daily active users on the Nth day according to the adjusted number of retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day.

2. The method according to claim 1, wherein Also including: Constructing a first fitting sample data set according to the historical user data of the target product; wherein each sample in the first fitting sample data set represents whether a user retains on the t-th day; Based on the first fitting sample data set, performing a fitting process on the change of the number of retained active users over time through a combination of an exponential function and a sine function to obtain a fitting expression for the number of retained active users; Correspondingly, in response to a prediction request for the number of daily active users on the Nth day after the preset benchmark day of the target product, determining the number of retained active users after N days of decay of the initial active user group on the preset benchmark day includes: In response to a prediction request for the number of daily active users on the Nth day after the preset benchmark day of the target product, determining the number of retained active users after N days of decay of the initial active user group on the preset benchmark day according to the fitting expression of the number of retained active users.

3. The method according to claim 1, characterized in that, Also including: Constructing a second fitting sample data set according to the historical user data of the target product; wherein each sample in the second fitting sample data set is a historical new user sequence; Based on the second fitting sample data set, performing a fitting process on the change of the number of newly added users per day over time through a long short-term memory network to obtain a target long short-term memory network model; Correspondingly, in response to a prediction request for the number of daily active users on the Nth day after the preset benchmark day of the target product, determining the number of newly added users per day between the preset benchmark day and the Nth day includes: In response to a prediction request for the number of daily active users on the Nth day after the preset benchmark day of the target product, determining the number of newly added users per day between the preset benchmark day and the Nth day according to the target long short-term memory network model.

4. The method according to claim 1, wherein Also including: Constructing a third fitting sample data set according to the historical user data of the target product; wherein each sample in the third fitting sample data set represents the retention rate of newly added users on the i-th day on the t-th day; Based on the third fitting sample data set, performing a fitting on the change trend of the retention rate of newly added users on the t-th day over time through a power function to obtain a fitting expression for the retention rate of newly added users on the t-th day. Correspondingly, in response to a prediction request for the daily active user count of the target product on the Nth day after the preset reference day, determining the retention rate of newly added users on the Nth day for each day between the preset reference day and the Nth day includes: In response to a prediction request for the daily active user count of the target product on the Nth day after the preset reference day, determining the retention rate of newly added users on the Nth day for each day between the preset reference day and the Nth day according to the fitting expression of the retention rate of newly added users on the tth day.

5. The method according to claim 1, wherein The external influencing factors include Ramadan time, summer and winter vacation time, the product strategy of the target product, and uncontrollable factors; Determining the influence coefficients of external influencing factors on the retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day includes: Inputting the Ramadan time, the summer and winter vacation time, the product strategy of the target product, the uncontrollable factors, the retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day into an intelligent agent based on the deep deterministic policy gradient algorithm that has been pre-trained; wherein, the structure of the intelligent agent includes an actor network and a critic network; Through the actor network in the intelligent agent structure, predicting the influence coefficients of the Ramadan time, the summer and winter vacation time, the product strategy of the target product, and the uncontrollable factors on the retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day.

6. The method according to claim 5, wherein The adjusting the retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day according to the respective influence coefficients corresponding to the retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day includes: The intelligent agent adjusts the retained active users, the number of newly added users per day, and the retention rate of newly added users according to the following formula: ODAU adjusted (N) = ODAU(N) × (1 + k o ); NDAU adjusted (i) = NDAU(i) × (1 + k n ); R adjusted (i) = R(i) × (1 + k r ); Among them, ODAU adjusted (N) represents the number of active user retentions on the Nth day after adjustment; ODAU(N) represents the number of active user retentions on the Nth day before adjustment; NDAU adjusted (i) represents the number of newly added users on the ith day after adjustment, where the ith day is any day between the preset reference day and the Nth day; NDAU(i) represents the data of newly added users on the ith day before adjustment; R adjusted (i) represents the retention rate of newly added users on the ith day after adjustment on the Nth day; R(i) represents the retention rate of newly added users on the ith day before adjustment on the Nth day; k o represents the influence coefficient of external influencing factors on the number of active user retentions; k n represents the influence coefficient of external influencing factors on the number of newly added users per day; k r represents the influence coefficient of external influencing factors on the retention rate of newly added users on the Nth day per day.

7. The method according to claim 6, wherein Estimating the daily active user count on the Nth day according to the adjusted retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day includes: The intelligent agent determines the daily active user count on the Nth day according to the following formula: wherein, DAUpredict(N) represents the estimated daily active user count on the Nth day.

8. A daily active user number prediction device, characterized in that including: A first estimation module, configured to, in response to a prediction request for the daily active user count of the target product on the Nth day after the preset reference day, determine the number of retained active users after N days of decay of the initial active user group on the preset reference day, the number of newly added users per day between the preset reference day and the Nth day, and the retention rate of newly added users on the Nth day; An influence coefficient determination module, configured to determine the influence coefficients of external influencing factors on the retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day; An adjustment module, configured to respectively adjust the retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day according to the respective influence coefficients corresponding to the retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day; A second estimation module, configured to estimate the daily active user number on the Nth day according to the adjusted number of retained active users, the number of newly added users per day, and the retention rate of newly added users on the Nth day.

9. An electronic device, characterized in that, Comprising: At least one processor; And A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and when the computer program is executed by the at least one processor, the at least one processor is enabled to execute the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the method according to any one of claims 1-7 is implemented.

Images