Reach task processing method, device and server

Abstract

This specification provides a processing method, device and server for a reach task. Based on this method, when it is implemented, the characteristic data, historical reach task sequence, historical reach time slice sequence, and candidate reach task sequence and candidate reach time slice sequence of the target user to be reached can be obtained first; then, based on the characteristic data, historical reach task sequence, historical reach time slice sequence of the target user, and candidate reach task sequence and candidate reach time slice sequence of the target user, the user's interest preference for reach tasks and reach time slices can be considered to accurately determine the target reach time slice and target reach task that match the target user and have a better reach effect, so as to perform a specific reach task on the target user.

Description

Technical Field

[0001] This specification belongs to the field of Internet technology, and in particular, relates to a method, device and server for processing a reach task. Background Art

[0002] In many business scenarios, network platforms often need to perform some tasks involving user contact. For example, a shopping website will push product promotion messages to users on a regular basis every day based on the needs of website users.

[0003] However, when carrying out specific outreach tasks based on existing methods, there are often problems such as low user participation and poor outreach effects.

[0004] Currently, no effective solution has been proposed to the above problems. Summary of the invention

[0005] This specification provides a method, device and server for processing a reach task, which can simultaneously and accurately determine a target reach time slice and a target reach task with a higher probability of acceptance and a higher degree of participation by the target user, so that target reach tasks can be subsequently performed for the target user based on the above-mentioned target reach time slice, thereby improving the user's participation and obtaining a better reach effect.

[0006] The embodiments of this specification provide a method for processing a reach task, including: obtaining characteristic data of a target user, a historical reach task sequence, a historical reach time slice sequence, and a candidate reach task sequence, and a candidate reach time slice sequence; determining a target reach time slice and a target reach task that match the target user based on the characteristic data of the target user, the historical reach task sequence, the historical reach time slice sequence, and the candidate reach task sequence, and the candidate reach time slice sequence.

[0007] An embodiment of the present specification also provides a processing device for a reach task, including: an acquisition module for acquiring characteristic data of a target user, a historical reach task sequence, a historical reach time slice sequence, and a candidate reach task sequence, and a candidate reach time slice sequence; a determination module for determining a target reach time slice and a target reach task that match the target user based on the characteristic data of the target user, the historical reach task sequence, the historical reach time slice sequence, and the candidate reach task sequence, and the candidate reach time slice sequence.

[0008] An embodiment of the present specification also provides a server, including a processor and a memory for storing processor executable instructions, and the processor implements the relevant steps of the processing method of the reach task when executing the instructions.

[0009] The embodiments of this specification also provide a computer-readable storage medium on which computer instructions are stored, and when the instructions are executed by a processor, the relevant steps of the processing method of the touch task are implemented.

[0010] Based on the processing method, device and server of the reach task provided in this specification, before specific implementation, a prediction model that at least includes a parallel task preference evolution structure and a time preference evolution structure can be pre-built and obtained through deep learning training. During specific implementation, the characteristic data, historical reach task sequence, historical reach time slice sequence of the target user to be reached, and the candidate reach task sequence and candidate reach time slice sequence for the target user can be obtained first; then the above-mentioned prediction model is used to process the characteristic data, historical reach task sequence, historical reach time slice sequence of the target user, as well as the candidate reach task sequence and candidate reach time slice sequence, to obtain a prediction result that takes into account the user's interest preference for the reach task and the reach time slice; and then, based on the prediction result, the target reach time slice and target reach task that match the target user and have a better reach effect can be accurately determined. In this way, the target reaching time slices and target reaching tasks with higher probability of acceptance and participation of target users can be accurately determined at the same time, so that the target reaching tasks for target users can be carried out in a targeted manner based on the above target reaching time slices, thereby improving the user's acceptance probability and participation and obtaining better reaching effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the embodiments of this specification, the drawings required for use in the embodiments will be briefly introduced below. The drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0012] Figure 1 is a flowchart of a method for processing a reach task provided by an embodiment of this specification;

[0013] Figure 2 It is a schematic diagram of an embodiment of a method for processing a reach task provided by an embodiment of this specification, in a scenario example;

[0014] Figure 3 It is a schematic diagram of an embodiment of a method for processing a reach task provided by an embodiment of this specification, in a scenario example;

[0015] Figure 4 It is a schematic diagram of an embodiment of a method for processing a reach task provided by an embodiment of this specification, in a scenario example;

[0016] Figure 5It is a schematic diagram of an embodiment of a method for processing a reach task provided by an embodiment of this specification, in a scenario example;

[0017] Figure 6 It is a schematic diagram of an embodiment of a method for processing a reach task provided by an embodiment of this specification, in a scenario example;

[0018] Figure 7 It is a schematic diagram of the structure of a server provided by an embodiment of this specification;

[0019] Figure 8 It is a schematic diagram of the structure of a processing device for a touch task provided by an embodiment of this specification;

[0020] Fig. 9 It is a schematic diagram of an embodiment of a method for processing a reach task provided by an embodiment of this specification, in a scenario example. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0022] See also Figure 1 As shown, the embodiment of this specification provides a method for processing a reach task. Among them, the method can be specifically applied to the server side. When implemented specifically, the method may include the following contents.

[0023] S101: Acquire feature data of the target user, historical reach task sequences, historical reach time slice sequences, and candidate reach task sequences and candidate reach time slice sequences.

[0024] S102: Determine the target reach time slice and target reach task that match the target user based on the target user's feature data, historical reach task sequence, historical reach time slice sequence, and candidate reach task sequence and candidate reach time slice sequence.

[0025] In some embodiments, the above-mentioned determination of the target reach time slice and target reach task matching the target user based on the target user's characteristic data, historical reach task sequence, historical reach time slice sequence, and candidate reach task sequence and candidate reach time slice sequence may include: using a prediction model to process the target user's characteristic data, historical reach task sequence, historical reach time slice sequence, and candidate reach task sequence and candidate reach time slice sequence to obtain corresponding prediction results; wherein the prediction model at least includes a parallel task preference evolution structure and a time preference evolution structure; and according to the prediction results, determining the target reach time slice and target reach task matching the target user.

[0026] In some embodiments, the above-mentioned reach tasks may specifically include pushing promotional messages, for example, pushing promotional information about a merchant's products to users; the above-mentioned reach tasks may also include prompting users to perform related business operations, for example, sending prompt information about the upgrade operation of a certain app application to users; the above-mentioned reach tasks may also include notification reminders to users, for example, reminding users of the day's work schedule, etc. Of course, it should be noted that the above-listed reach tasks are only a schematic description. In specific implementation, according to specific application scenarios and processing requirements, the above-mentioned reach tasks may also include tasks of other types or other contents. This specification does not limit this.

[0027] This specification mainly takes the push of promotional messages as an example of a reach task for specific description. Regarding the processing of other types of reach tasks, you can refer to the embodiment of processing the reach task of pushing promotional information, and this specification will not elaborate on it.

[0028] In some embodiments, the processing method of the above-mentioned reach task can be specifically applied to the server side. Among them, the above-mentioned server can specifically include a background server applied to the side of the business platform (for example, XX promotion platform) that can realize data transmission, data processing and other functions. Specifically, the server can be, for example, an electronic device with data calculation, storage and network interaction functions. Alternatively, the server can also be a software program running in the electronic device to provide support for data processing, storage and network interaction. In this embodiment, the number of the servers is not specifically limited. The server can specifically be one server, or several servers, or a server cluster formed by several servers.

[0029] In some embodiments, the above-mentioned target users can be specifically understood as users to be reached.

[0030] In some embodiments, the characteristic data of the target user may specifically include the portrait characteristics of the target user and / or the behavioral characteristics of the target user.

[0031] Specifically, the target user's portrait features may include at least one of the following: the target user's gender, the target user's city, the target user's occupation, the target user's education level, etc.

[0032] The above-mentioned target user's behavioral characteristics may specifically include at least one of the following: the target user's behavioral operation characteristics for historical contact tasks, the target user's behavioral operation characteristics when using the terminal device, the target user's behavioral operation characteristics for historical push information (for example, redemption vouchers, coupons, etc.), etc.

[0033] The terminal device may specifically include a front-end device applied to the user side and capable of realizing functions such as data collection and data transmission. Specifically, the terminal device may be, for example, an electronic device such as a desktop computer, a tablet computer, a laptop computer, a smart phone, a smart wearable device, etc. Alternatively, the terminal device may also be a software application that can be run in the above electronic devices. For example, it may be a client of an APP application installed and run on a smart phone.

[0034] Specifically, the above-mentioned target user's behavioral operation characteristics for historical reach tasks may include: the target user's click to accept operation for historical reach tasks, the target user's swipe to ignore operation for historical reach tasks, or the target user's postponement operation for historical reach tasks, etc. The above-mentioned target user's behavioral operation characteristics for using terminal devices may include: the time when the target user turns on the terminal device, the length of time the target user keeps the terminal device turned on after turning on the terminal device, the frequency of the target user turning on the terminal device every day, etc. The above-mentioned target user's behavioral operation characteristics for historical push information may specifically include: the target user uses coupons, the target user deletes coupons, the target user blocks push information, etc.

[0035] It should be noted that the characteristic data of the target users listed above are all collected and used in accordance with compliant means with the knowledge and consent of the users. In addition, in the process of collecting and using the above data, corresponding data encryption means will be used to prevent the leakage of relevant user data and protect the user's data security and privacy.

[0036] In some embodiments, see Fig. 9As shown, the above historical reach task sequence (which can be recorded as user tasksequence) can specifically include multiple historical reach tasks, which can be specifically expressed as: [tk(1), tk(2), ... tk(i), ... tk(N)]. Among them, tk(i) represents any historical reach task numbered i. The above historical reach tasks can specifically be understood as the reach tasks that have been executed for the target user in the past. For example, the reach tasks executed for the target user in the past week.

[0037] In some embodiments, the above historical reach task sequence may specifically include the task features of each historical reach task (e.g., the task type, task content, reach time, etc. of the historical reach task). In addition, the above historical reach task sequence may also include the reach performance of the historical reach tasks, for example, the overall reach success rate of each historical reach task in history, the overall user acceptance probability, etc.

[0038] In some embodiments, the above-mentioned historical reach task sequence may also include the target user's interaction characteristics for each historical reach task (for example, the target user's operation type, operation time, operation frequency, etc. for the historical reach task).

[0039] In some embodiments, see Fig. 9 As shown, the above historical reach time slice sequence (which can be recorded as user time-slot sequence) can specifically include multiple historical reach time slices, which can be specifically expressed as: [ts(1), ts(2), ... ts(j), ... ts(M)]. Among them, ts(i) represents any historical reach time slice numbered j. The above historical reach time slice can be specifically understood as the time point or time period when a certain reach task was performed in history. For example, 8:00, or 8:00-9:30, etc. Specifically, in each historical reach time slice, one or more historical reach tasks can be performed on the target user.

[0040] In some embodiments, the above historical reach task sequence and historical reach time slice sequence can be generated specifically based on the interaction record of the target user's historical reach task. The above target user's feature data can be generated specifically based on the target user's user database, combined with the target user's interaction record of starting the reach task.

[0041] In some embodiments, the above-mentioned candidate reach task sequence may specifically include one or more candidate reach tasks to be processed, which can be specifically recorded as: Target task.

[0042] In some embodiments, during specific implementation, the server may select from a candidate reach task list (e.g., <Task i,Group i >) to generate a candidate reach task sequence for the target user. In addition, the server can also select one or more candidate reach tasks randomly from the target user's historical reach task interaction records (for example, <User i ,List <Task j >>), filter out the reach tasks that the target user is relatively willing to accept (for example, the reach tasks that match the user type of the target user) from the candidate reach task list to generate a candidate reach task sequence for the target user.

[0043] In some embodiments, the above-mentioned candidate touch time slot sequence may specifically include one or more candidate touch time slots, which can be specifically recorded as: Target time-slot.

[0044] In some embodiments, during specific implementation, the server can select the time slices in which the target user frequently uses the terminal device from multiple historical contact time slices based on the historical contact task records of the target user to generate a candidate contact time slice sequence. In addition, the server can also generate a candidate contact time slice sequence for full-score merchants and platforms based on the customized requirements of merchants and platforms.

[0045] In some embodiments, the above-mentioned prediction model can be specifically understood as a neural network model that is pre-trained based on deep learning and can simultaneously predict the user's acceptance probability for different reach time slices and different reach tasks. Specifically, the above-mentioned prediction module can be a model based on the DTIEN network. Among them, the above-mentioned DTIEN network can be specifically understood as a deep time interest evolution network, which can use the user's historical reach tasks, historical interaction time, user portraits and other information data; by extracting interest preferences and time preferences, and adding an evolution module of the attention mechanism to finally estimate the user's click rate on the target task and target reach time, so as to avoid the loss of the effect of offline optimization after the previous reach tasks and reach times are modeled separately, and obtain better application effects.

[0046] In some embodiments, the above prediction model may include at least a parallel task preference evolution structure and a time preference evolution structure.

[0047] For details, please refer to Figure 2 As shown, the prediction model may further specifically include an embedding layer, a connection layer and a classification layer; wherein the embedding layer is respectively connected to the task preference evolution structure and the time preference evolution structure; and the task preference evolution structure and the time preference evolution structure are respectively connected to the connection layer.

[0048] The embedding layer (also referred to as an embedding layer) can be used to convert input data into a corresponding fixed-size vector (which can be recorded as an embedding vector or an embedding vector).

[0049] The above-mentioned task preference evolution structure can be specifically used to jointly process the target user's historical reach tasks and candidate reach tasks, and output an evolved task preference hidden state sequence related to the candidate reach tasks that meets the target user's interest preference expectations based on the historical reach tasks.

[0050] The above-mentioned time preference evolution structure can be specifically used to jointly process the target user's historical contact time slices and candidate contact time slices to output an evolved time preference hidden state sequence related to the candidate contact time slices that meets the target user's interest preference expectations based on the historical contact time slices.

[0051] The above connection layer can be specifically used to splice different vectors and sequences to obtain an aggregated vector that integrates information of multiple different dimensions.

[0052] The above classification layer can be specifically used to calculate and output the acceptance probability of the target user for different candidate reach tasks for different candidate reach time slices according to the input aggregation vector as a processing result.

[0053] In some embodiments, see Figure 3 As shown, the above prediction model is used to process the historical reach task sequence, the historical reach time slice sequence, the candidate reach task sequence, and the candidate reach time slice sequence of the target user to obtain the corresponding prediction result. When implemented specifically, the following contents may be included:

[0054] S1: using the embedding layer to process the target user's feature data, historical reach task sequence, historical reach time slice sequence, and candidate reach task sequence and candidate reach time slice sequence respectively, and obtain the corresponding target user's feature data embedding vector, historical reach task embedding vector, historical reach time slice embedding vector, and candidate reach task embedding vector and candidate reach time slice embedding vector;

[0055] S2: Use the task preference evolution structure to process the historical reach task embedding vector and the candidate reach task embedding vector to obtain the evolved task preference hidden state sequence; use the time preference evolution structure to process the historical reach time slice embedding vector and the candidate reach time slice embedding vector to obtain the evolved time preference hidden state sequence;

[0056] S3: Using a connection layer, concatenate the evolved task preference hidden state sequence, the evolved time preference hidden state sequence, the target user's feature data embedding vector, the candidate reach task embedding vector, and the candidate reach time slice embedding vector to obtain an aggregate vector;

[0057] S4: Process the aggregation vector using a classification layer to obtain a corresponding processing result.

[0058] In some embodiments, see Fig. 9 As shown, in specific implementation, the embedding layer can be used to process the input historical reach task sequence [tk(1), tk(2), ... tk(i), ... tk(N)] using the one-hot encoding method to obtain and output the corresponding historical reach task embedding vector, which can be expressed as: [e(1), e(2), ... e(i), ... te(N)]. Among them, e(i) can be expressed as the embedding vector corresponding to the historical reach task tk(i).

[0059] Similarly, the embedding layer can be used to process the input historical reach time slice sequence [ts(1), ts(2), … ts(j), … ts(j)] using the one-hot encoding method to obtain and output the corresponding historical reach time slice embedding vector, which can be expressed as: [e(1), e(2), … e(j), … e(M)]. Among them, e(j) can be expressed as the embedding vector corresponding to the historical reach time slice ts(j).

[0060] In addition, the embedding layer can be used to process the input feature data of the target user (for example, it can be recorded as userprofile feature) to obtain and output the corresponding target user feature data embedding vector (for example, it can be recorded as e(userp-feature)); the embedding layer can be used to process the input candidate reach task sequence (for example, Target task) to obtain and output the corresponding candidate reach task embedding vector (which can be recorded as e(tgt-tk)); the embedding layer can be used to process the input candidate reach time slice sequence (for example, Target time-slot) to obtain and output the corresponding candidate reach time slice embedding vector (which can be recorded as e(tgt-ts)).

[0061] In some embodiments, see Figure 4 As shown, the task preference evolution structure may specifically include: a task preference extraction layer and a task preference evolution layer.

[0062] Among them, the input end of the task preference extraction layer is connected to the output end of the embedding layer, the output end of the task preference extraction layer is connected to the input end of the task preference evolution layer, and the output end of the task preference evolution layer is connected to the input end of the connection layer.

[0063] See also Figure 5 As shown, the time preference evolution structure may specifically include: a time preference extraction layer and a time preference evolution layer.

[0064] Among them, the input end of the time preference extraction layer is connected to the output end of the embedding layer, the output end of the time preference extraction layer is connected to the input end of the time preference evolution layer, and the output end of the time preference evolution layer is connected to the input end of the connection layer.

[0065] In some embodiments, see Figure 4 As shown, based on the above task preference evolution structure, the above task preference evolution structure is used to process the historical touch task embedding vector and the candidate touch task embedding vector to obtain the evolved task preference hidden state sequence. When implemented specifically, the following contents may be included:

[0066] S1: Use the task preference extraction layer to process the historical access task embedding vector and extract the hidden state sequence of task preference;

[0067] S2: Use the task preference evolution layer to process the hidden state sequence of the task preference and the candidate reach task embedding vector to obtain the evolved task preference hidden state sequence.

[0068] In some embodiments, the task preference extraction layer may specifically include a plurality of GRU modules, wherein each GRU module is used to process a vector segment corresponding to a historical reach task in the input historical reach task embedding vector, and output a hidden state of a task preference corresponding to the historical reach task.

[0069] The above-mentioned GRU (Gate Recurrent Unit) module can be specifically understood as a type of recurrent neural network (RNN).

[0070] Specifically, take the GRU module that processes the vector segment e(i) corresponding to the historical reach task tk(i) numbered i as an example. The GRU module can receive the intermediate data (which can be recorded as z(i-1)) generated and transmitted when the adjacent previous GRU module processes the vector segment e(i-1) corresponding to the previous historical reach task tk(1-1); and then process it according to the input vector segment e(i) and the intermediate data z(i-1) transmitted by the previous GRU module at the same time to extract and output the hidden state of the task preference for tk(i), which can be recorded as: h(i).

[0071] In the above manner, by using the task preference extraction layer to process the historical reach task embedding vector, a hidden state sequence of task preferences containing multiple vector segments corresponding to multiple historical reach tasks can be extracted and output, which can be expressed in the following form: [h(1), h(2), ... h(i), ... h(N)].

[0072] In some embodiments, the task preference evolution layer may further include multiple AUGRU modules. Similarly, the time preference evolution layer may further include multiple AUGRU modules.

[0073] Each AUGRU module is used to process a hidden state of an input task preference, generate and output a sequence segment in the evolved task preference hidden state sequence; and pass the sequence segment to the next adjacent AUGRU module.

[0074] The above-mentioned AUGRU module can be specifically understood as an update and improvement based on the GUR module, introducing an attention mechanism, which can pay more attention to learning and updating the hidden state of user interest preferences that are more relevant to the candidate reach tasks, and then more accurately use historical reach tasks to predict users' preference expectations for candidate reach tasks.

[0075] According to the above method, each GUR module in the task preference extraction layer can be processed in turn to obtain multiple hidden state sequences corresponding to a single access task; then the above multiple hidden state sequences are spliced ​​to obtain and output a complete hidden state sequence of task preference.

[0076] In some embodiments, the specific structure of the AUGRU module can be found in Figure 6 shown.

[0077] In some embodiments, the above-mentioned processing of the hidden state sequence of the task preference and the candidate reach task embedding vector by the task preference evolution layer to obtain the evolved task preference hidden state sequence may include: obtaining the current sequence segment in the evolved task preference hidden state sequence in the following manner:

[0078] S1: Using the current sequence segment in the hidden state sequence of the task preference and the embedding vector of the candidate reach task, an attention operation is performed to obtain the current correlation parameter between the current task preference and the candidate reach task;

[0079] S2: The AUGRU module is used to process the previous sequence segment and the current correlation parameter in the evolved task preference hidden state sequence to obtain the current sequence segment in the evolved task preference hidden state sequence.

[0080] In some embodiments, taking the current sequence segment in the hidden state sequence of processing task preference (e.g., the task preference hidden state h(i) corresponding to the historical reach task tk(i)) as an example, the current relevance parameter can be calculated as follows:

[0081]

[0082] Among them, a i is the current correlation parameter, h i is the current sequence segment in the hidden state sequence preferred by the task, W e(tgt-tk) is the matrix parameter corresponding to e(tgt-tk), e(tgt-tk) is the embedding vector of the candidate reach task, h j is the task preference hidden state corresponding to the historical reach task numbered j, and N is the total number of historical reach tasks.

[0083] The above relevance parameter can also be called attention score. In the future, the value of the next sequence segment can be limited by using the above relevance parameter to focus on updating and evolving the interest preference hidden state that is highly relevant to the candidate reach task, so as to achieve the effect of updating the hidden state vector more for the behavior that is more relevant to the candidate reach task.

[0084] In some embodiments, the above attention operation can be specifically understood as an operation based on the attention mechanism, which can guide the automatic update and evolution in the direction of attention. Specifically, the current sequence segment in the above evolved task preference hidden state sequence can be specifically recorded as u t '. In specific implementation, the AURUG module can be used according to the following formula to focus on updating and evolving the hidden states of interest preferences that are highly relevant to the candidate reach tasks:

[0085] u t ′=u t *a t

[0086]

[0087] Among them, u t ′ is the current sequence segment in the evolved task preference hidden state sequence, u t is the previous sequence segment before the current sequence segment in the evolved task preference hidden state sequence, a t is the current correlation parameter between the current task preference and the candidate reachable tasks, For u t ′ corresponds to the hidden state of the task preference in vector form, h t-1 For ut The corresponding hidden state of task preference, Represents the inner product operation.

[0088] According to the above method, each AUGRU module in the task preference evolution layer can process a sequence segment in turn, and the last AUGRU module in the task preference evolution layer outputs the final evolved task preference hidden state sequence, which can be recorded as Utk(N)'.

[0089] Through the above-mentioned embodiments, based on the introduction and attention operation of the AUGRU module, a soft correlation weight sequence can be obtained as the evolved task preference hidden state sequence. Based on the evolved task preference hidden state sequence, it can effectively reflect the characteristics that the interest preference state with high relevance to the candidate reach task target-task will have a relatively large attention score, and vice versa. In order to achieve the effect of only connecting those interest preference states related to the target-task into a subsequence for modeling, the present application designs a sequence modeling method that can effectively use the size of the attention score to determine the intensity of the hidden state being updated during the sequence modeling process, that is, those interest preference states that are more relevant to the candidate reach task target-task can obtain relatively greater updates and evolutions, while interest preference states that are not related to the candidate reach task target-task can only obtain relatively smaller updates and evolutions, or even no updates or evolutions.

[0090] Similarly, the above-mentioned use of the time preference evolution structure to process the historical reach time slice embedding vector and the candidate reach time slice embedding vector to obtain the evolved time preference hidden state sequence may include: using the time preference extraction layer to process the historical reach time slice embedding vector to extract the time preference hidden state sequence; using the time preference evolution layer to process the time preference hidden state sequence and the candidate reach time slice embedding vector to obtain the evolved time preference hidden state sequence.

[0091] The above-mentioned use of the time preference evolution layer to process the hidden state sequence of the time preference and the embedding vector of the candidate touch time slice to obtain the evolved time preference hidden state sequence may include: obtaining the current sequence segment in the evolved time preference hidden state sequence in the following manner:

[0092] The current sequence segment in the hidden state sequence of the time preference and the embedding vector of the candidate reach time slice are used to perform an attention operation to obtain the current correlation parameter between the current time preference and the candidate reach time slice; the AUGRU module is used to process the previous sequence segment and the current correlation parameter in the evolved time preference hidden state sequence to obtain the current sequence segment in the evolved time preference hidden state sequence.

[0093] According to the above method, each AUGRU module in the task preference evolution layer can process a sequence segment in turn, and the last AUGRU module in the task preference evolution layer outputs the final evolved task preference hidden state sequence, which can be recorded as Uts(M)'.

[0094] In some embodiments, see Fig. 9 As shown, the above connection layer may specifically include: a concat&flatten structure, and two PRelu layers.

[0095] In specific implementation, the concat&flatten structure in the above-mentioned connection layer can be used to concatenate the input evolved task preference hidden state sequence, evolved time preference hidden state sequence, target user's feature data embedding vector, candidate reach task embedding vector, and candidate reach time slice embedding vector, and then process them through a fully connected layer with 256 neurons (the first PRelu layer), and then through another fully connected layer with 128 neurons (the second PRelu layer) to aggregate data information of multiple different dimensions, and finally output a 128-dimensional vector as the required aggregated vector.

[0096] In some embodiments, the above classification layer may specifically be a Softmax binary classification layer.

[0097] In specific implementation, the classification layer can be used to process the input 128-dimensional aggregation vector, calculate the user's acceptance probability values ​​for different candidate reach tasks under different candidate time slices as the corresponding processing results, and output the processing results as the model output of the prediction model.

[0098] In some embodiments, the prediction result may specifically include multiple acceptance probability parameter groups, wherein one acceptance probability parameter group corresponds to one candidate reach time slice; the acceptance probability parameter group includes multiple acceptance probabilities, wherein one acceptance probability corresponds to one candidate reach task;

[0099] Correspondingly, according to the prediction results, the above-mentioned determination of the target reaching time slice and the target reaching task matching the target user may include the following contents during specific implementation:

[0100] S1: sorting the acceptance probabilities contained in each acceptance probability parameter group according to the probability value to obtain a plurality of sorted acceptance probability parameter groups;

[0101] S2: comparing the acceptance probabilities ranked first among the plurality of ranked acceptance probability parameter groups, and determining the acceptance probability parameter group where the acceptance probability ranked first with the largest value is located as the target acceptance probability parameter group;

[0102] S3: Determine the candidate touch time slice corresponding to the target acceptance probability parameter group as the target touch time slice; determine the candidate touch task corresponding to the acceptance probability ranked first in the target acceptance probability parameter group as the target touch task.

[0103] In some embodiments, specifically, for example, for the target user User1, when the candidate reach task sequence includes N candidate reach tasks and the candidate reach time slice sequence includes M candidate reach time slices, by sorting the processing results, the following multiple sorted acceptance probability parameter groups can be obtained:

[0104] User1,1,[(task1,0.73),(task2,0.68),(task3,0.66),...(taski,0.65),...(taskN,0.61)],

[0105] User1,2,[(task1,0.70),(task2,0.70),(task4,0.69),...(taski,0.66),...(taskN,0.63)],

[0106] User1,3,[(task3,0.81),(task4,0.76),(task5,0.67),...(taski,0.66),...(taskN,0.56)],

[0107] …

[0108] User1,j,[(task3,0.80),(task4,0.77),(task6,0.69),...(taski,0.67),...(taskN,0.63)],

[0109] …

[0110] User1,M,[(task4,0.69),(task5,0.66),(task7,0.63),...(taski,0.61),...(taskN,0.52)].

[0111] Among them, User1,j,[(task3,0.80),(task4,0.77),(task6,0.69),...(taski,0.67),...(taskN,0.63)] can be expressed as a probability parameter group corresponding to the candidate reach time slice numbered j. (taski,0.67) in this probability parameter group can be expressed as the acceptance probability of User1 is 0.67 when performing the candidate reach task numbered i on user User1 in the candidate reach time slice numbered j.

[0112] In some embodiments, during specific implementation, the acceptance probabilities ranked first in each of the sorted acceptance probability parameter groups can be compared based on multiple sorted acceptance probability parameter groups, and the acceptance probability ranked first with the largest acceptance probability value (for example, the acceptance probability 0.83 ranked first in the third acceptance probability group) is found to be the target acceptance probability parameter group; and the candidate reach time slice corresponding to the target acceptance probability parameter group (for example, the candidate reach time slice numbered 3) is determined as the target reach time slice matching the target user User1; at the same time, the candidate reach task corresponding to the acceptance probability (0.83) ranked first in the target acceptance probability parameter group (for example, the candidate reach task numbered 3) is determined as the target reach task matching the target user User1.

[0113] In some embodiments, after determining the target reach time slice and target reach task matching the target user based on the prediction results, the method may also include the following contents when implemented: performing the target reach task for the target user in the target reach time slice.

[0114] Specifically, for example, the server can perform candidate reach task No. 3 for target user User1 in candidate reach time slice No. 3. At this time, target user User1 will be relatively more willing and have a greater probability of accepting the reach task, thereby improving the user experience and achieving a better reach effect.

[0115] In some embodiments, during specific implementation, according to specific reach requirements, multiple sorted acceptance probability parameter groups can also be sorted according to the numerical value of the acceptance probability ranked first in the multiple sorted acceptance probability parameter groups, and a preset number (for example, 5) of acceptance probability parameter groups with the highest ranking are selected as target probability parameter groups to determine a preset number of target reach time slices; and then, the candidate reach tasks corresponding to the acceptance probability with the largest numerical value are determined from the preset number of target probability parameter groups as the target reach tasks under the target reach time slice corresponding to the target probability parameter group.

[0116] In some embodiments, according to the reach requirements, while allowing multiple target reach tasks to be performed on the target user simultaneously in the same target reach time slice, multiple candidate reach tasks corresponding to multiple (for example, 3) acceptance probabilities with the highest rankings can be screened out in the target probability parameter group according to corresponding reach rules as the multiple target reach tasks in the target reach time slice corresponding to the target probability parameter group.

[0117] For example, the three candidate reach tasks, namely, candidate reach task 3, candidate reach task 4 and candidate reach task 5, which correspond to the top three acceptance probabilities of 0.81, 0.76 and 0.67 in the third probability parameter group, can be screened out as the three target reach tasks for the target user User1 in the target reach time slice.

[0118] In some embodiments, according to the reach requirements, when the same target reach task is not allowed to be performed in different target reach time slices, when specifically determining the target reach task, it is also possible to prioritize, according to the corresponding reach rules, the candidate reach task (for example, candidate reach task q) corresponding to the first-ranked acceptance probability in the relatively forward-ranked target probability parameter group (for example, the first target probability parameter group) as the target reach task under the corresponding target reach time slice (for example, the first target reach time slice).

[0119] When determining the target touch task under the target touch time slice (for example, the second target touch time slice) corresponding to the target probability parameter group that is ranked relatively low (for example, the second target probability parameter group that is ranked behind the first target probability parameter group), if the candidate touch task corresponding to the acceptance probability ranked first in the second target probability parameter group is the same as the target touch task under the first target touch time slice, and is also the candidate touch task q, then the candidate touch task corresponding to the acceptance probability ranked second in the second target probability parameter group (for example, candidate touch task p) can be determined as the target touch task under the second target touch time slice. If the above-mentioned candidate touch task p is also the same as the target touch task under other target touch time slices that are determined in priority, then the candidate touch task corresponding to the acceptance probability ranked third in the second target probability parameter group can be selected in order as the target touch task under the second target touch time slice, and so on.

[0120] In some embodiments, when the target user includes multiple target users, the target reach time slice for each target user and the target reach task under the corresponding target reach time slice can be determined for each target user among the multiple target users in the above manner. Furthermore, different target users can be distinguished, and targeted reach task processing strategies can be formulated for different targets. For different target users, based on the corresponding reach task processing strategies, corresponding target reach tasks are performed under different target reach time slices.

[0121] In some embodiments, the method may further include the following when implemented:

[0122] S1: Obtain the interaction records of the sample users’ historical reach tasks;

[0123] S2: Create sample training data based on the sample user's historical interaction records of the task.

[0124] S3: constructing an initial prediction model; wherein the initial prediction model at least includes an initial task preference evolution structure and an initial time preference evolution structure in parallel;

[0125] S4: Using the sample training data to train the initial prediction model to obtain a prediction model that meets the requirements.

[0126] In some embodiments, the interaction records of the historical reach tasks of the above-mentioned sample users may specifically include one or more of the following content information: historical reach tasks that have been performed on sample users in history, behavioral operations of sample users on historical reach tasks (for example, whether the sample user issues an operation representing acceptance of the historical reach tasks), historical reach time slices corresponding to historical reach tasks, historical delivery performance of historical reach tasks, portrait features of sample users, historical behavioral features of sample users, etc.

[0127] In some embodiments, the sample training data is established based on the interaction records of the sample user's historical contact tasks. In specific implementation, for a sample user, the content information of interest can be extracted from the historical contact task interaction records of the sample user and used to establish the sample training data for the sample user. The sample data can be specifically expressed in the following form: (List <f k >,Timeslot,Label).

[0128] Among them, f kIt can be the characteristic data of sample users, such as portrait features, behavior sequences or behavior features, historical task delivery performance, etc.). Among them, the behavior sequence can specifically refer to: the sample user's verification sequence and click sequence of the coupons pushed by the historical reach task, the sample user's click on the information to be reached by the historical reach task or the business line sequence used, etc. The task historical delivery performance can specifically refer to the sample user's feedback statistics on the historical reach task, such as the click-through rate in the last 1 day, the click-through rate in the last 3 days, the click-through rate in the last 7 days, the verification rate in the last 1 day, the verification rate in the last 3 days, the verification rate in the last 7 days, etc.

[0129] Timeslot can be a historical reach time slice, for example, 7:00, 8:00, 12:00, 16:00, 20:00, 21:00, etc. The above historical reach time slice can be customized, and the value of Timeslot can be a value based on one-hot encoding, for example, 1 to 6.

[0130] Label can specifically indicate whether the user accepts the historical reach task, for example, 1 indicates acceptance and 0 indicates rejection.

[0131] In some embodiments, the constructed initial prediction model may further specifically include an initial embedding layer, an initial connection layer, and an initial classification layer.

[0132] In some embodiments, when training a model, sample data can be divided into a training set and a test set. The test set can be used to train the initial prediction model through deep learning; and the test set can be used to test the accuracy of the trained prediction model. After multiple training and testing, when it is determined that the accuracy of the prediction model meets the preset accuracy requirements, the training and testing are stopped, and the prediction model at this time is determined to be a prediction model that meets the requirements.

[0133] In some embodiments, after obtaining the interaction records of the sample user's historical reach tasks, the method may also include: based on the interaction records of the sample user's historical reach tasks, screening out historical reach time slices in which the sample user accepts historical reach tasks with a frequency greater than a preset frequency threshold as candidate reach time slices, to generate a candidate reach time slice sequence for the target user.

[0134] Through the above embodiments, it is possible to screen out the reach time slices that the target user is relatively more likely to accept based on the interaction records of the historical reach tasks of the sample users, and obtain a candidate reach time slice sequence with relatively better reach effect.

[0135] In some embodiments, after obtaining the interaction records of the historical reach tasks of the sample users, the method may further include: determining different sample user types and historical reach tasks with higher acceptance corresponding to different sample user types through clustering processing according to the interaction records of the historical reach tasks of the sample users. Accordingly, in the specific implementation, the sample user type corresponding to the target user may be determined first; then, referring to the historical reach tasks with higher acceptance corresponding to the sample user type, the reach tasks with relatively higher acceptance of the target user may be screened out from multiple candidate reach tasks to generate a candidate reach task sequence for the target user.

[0136] Through the above-mentioned embodiments, it is possible to screen out the reach tasks that the target users are relatively more likely to accept based on the interaction records of the historical reach tasks of the sample users, and obtain a candidate reach task sequence with relatively better reach effects.

[0137] As can be seen from the above, based on the processing method of the reach task provided in the embodiment of this specification, before specific implementation, a prediction model that at least includes a parallel task preference evolution structure and a time preference evolution structure can be pre-constructed and obtained through deep learning training. During specific implementation, the characteristic data, historical reach task sequence, historical reach time slice sequence of the target user to be reached, and the candidate reach task sequence and candidate reach time slice sequence for the target user can be obtained first; then the above prediction model is used to process the characteristic data, historical reach task sequence, historical reach time slice sequence of the target user, and the candidate reach task sequence and candidate reach time slice sequence, to obtain a prediction result that simultaneously considers the user's interest preference for the reach task and the reach time slice; and then, according to the prediction result, the target reach time slice and target reach task that match the target user can be accurately determined. Thereby, the target reach time slice and target reach task with a high probability of acceptance and a high degree of participation of the target user can be accurately determined at the same time, so that the target reach task for the target user can be carried out in a targeted manner according to the above target reach time slice in the future, and a better reach effect can be obtained.

[0138] An embodiment of the present specification also provides a server, including a processor and a memory for storing processor executable instructions. When the processor is implemented, it can perform the following steps according to the instructions: obtain the characteristic data of the target user, the historical contact task sequence, the historical contact time slice sequence, and the candidate contact task sequence, and the candidate contact time slice sequence; determine the target contact time slice and target contact task that match the target user based on the characteristic data of the target user, the historical contact task sequence, the historical contact time slice sequence, and the candidate contact task sequence, and the candidate contact time slice sequence.

[0139] In order to complete the above instructions more accurately, refer to Figure 7As shown, the embodiment of this specification also provides another specific server, wherein the server includes a network communication port 701, a processor 702 and a memory 703, and the above structures are connected through internal cables so that each structure can perform specific data interaction.

[0140] Among them, the network communication port 701 can be specifically used to obtain characteristic data of the target user, historical reach task sequence, historical reach time slice sequence, as well as candidate reach task sequence and candidate reach time slice sequence.

[0141] The processor 702 can be specifically used to determine the target contact time slice and target contact task that match the target user based on the target user's feature data, historical contact task sequence, historical contact time slice sequence, and candidate contact task sequence and candidate contact time slice sequence.

[0142] The memory 703 may be specifically used to store corresponding instruction programs.

[0143] In this embodiment, the network communication port 701 can be a virtual port that is bound to different communication protocols so that different data can be sent or received. For example, the network communication port can be a port responsible for web data communication, a port responsible for FTP data communication, or a port responsible for email data communication. In addition, the network communication port can also be a physical communication interface or communication chip. For example, it can be a wireless mobile network communication chip, such as GSM, CDMA, etc.; it can also be a Wifi chip; it can also be a Bluetooth chip.

[0144] In this embodiment, the processor 702 may be implemented in any appropriate manner. For example, the processor may take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) executable by the (micro)processor, a logic gate, a switch, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, etc. This specification does not limit this.

[0145] In this embodiment, the memory 703 may include multiple levels. In a digital system, anything that can store binary data can be a memory; in an integrated circuit, a circuit with a storage function that has no physical form is also called a memory, such as RAM, FIFO, etc.; in a system, a storage device with a physical form is also called a memory, such as a memory stick, TF card, etc.

[0146] The embodiments of this specification also provide a computer storage medium based on the above-mentioned method for processing the reach task, wherein the computer storage medium stores computer program instructions, which, when executed, implement: obtaining the characteristic data of the target user, the historical reach task sequence, the historical reach time slice sequence, and the candidate reach task sequence, and the candidate reach time slice sequence; determining the target reach time slice and target reach task that match the target user based on the characteristic data of the target user, the historical reach task sequence, the historical reach time slice sequence, and the candidate reach task sequence, and the candidate reach time slice sequence.

[0147] In this embodiment, the storage medium includes, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a cache, a hard disk (HDD), or a memory card. The memory may be used to store computer program instructions. The network communication unit may be an interface for network connection communication set in accordance with the standard specified by the communication protocol.

[0148] In this embodiment, the functions and effects specifically implemented by the program instructions stored in the computer storage medium can be explained in comparison with other implementations and will not be described in detail here.

[0149] See also Figure 8 As shown, at the software level, the embodiment of this specification also provides a processing device for reaching tasks, which may specifically include the following structural modules:

[0150] The acquisition module 801 may be specifically used to acquire characteristic data of the target user, historical reach task sequences, historical reach time slice sequences, and candidate reach task sequences and candidate reach time slice sequences;

[0151] Determination module 802 can be specifically used to determine the target contact time slice and target contact task that match the target user based on the target user's characteristic data, historical contact task sequence, historical contact time slice sequence, and candidate contact task sequence and candidate contact time slice sequence.

[0152] In some embodiments, when the above-mentioned determination module 802 is implemented, the target reach time slice and target reach task matching the target user can be determined in the following manner: the characteristic data of the target user, the historical reach task sequence, the historical reach time slice sequence, and the candidate reach task sequence, the candidate reach time slice sequence are processed by a prediction model to obtain corresponding prediction results; wherein the prediction model at least includes a parallel task preference evolution structure and a time preference evolution structure; based on the prediction results, the target reach time slice and target reach task matching the target user are determined.

[0153] In some embodiments, the characteristic data of the target user may specifically include the portrait characteristics and / or behavioral characteristics of the target user.

[0154] In some embodiments, the prediction model also includes an embedding layer, a connection layer and a classification layer; wherein the embedding layer is connected to the task preference evolution structure and the time preference evolution structure respectively; and the task preference evolution structure and the time preference evolution structure are connected to the connection layer respectively.

[0155] In some embodiments, when the processing module 802 is implemented, the prediction model can be used to process the target user's historical reach task sequence, historical reach time slice sequence, and candidate reach task sequence, candidate reach time slice sequence in the following manner to obtain the corresponding prediction result: the embedding layer is used to process the target user's feature data, historical reach task sequence, historical reach time slice sequence, and candidate reach task sequence, candidate reach time slice sequence, respectively, to obtain the corresponding target user's feature data embedding vector, historical reach task embedding vector, historical reach time slice embedding vector, and candidate reach task embedding vector, candidate reach time slice sequence. time slice embedding vector; using the task preference evolution structure to process the historical reach task embedding vector and the candidate reach task embedding vector to obtain the evolved task preference hidden state sequence; using the time preference evolution structure to process the historical reach time slice embedding vector and the candidate reach time slice embedding vector to obtain the evolved time preference hidden state sequence; using the connection layer to splice the evolved task preference hidden state sequence, the evolved time preference hidden state sequence, the target user's feature data embedding vector, the candidate reach task embedding vector, and the candidate reach time slice embedding vector to obtain an aggregate vector; using the classification layer to process the aggregate vector to obtain the corresponding processing result.

[0156] In some embodiments, the task preference evolution structure may specifically include: a task preference extraction layer and a task preference evolution layer; the time preference evolution structure includes: a time preference extraction layer and a time preference evolution layer.

[0157] In some embodiments, when the processing module 802 is implemented, the task preference extraction layer can be used to process the historical reach task embedding vector to extract the hidden state sequence of task preference; the task preference evolution layer can be used to process the hidden state sequence of task preference and the candidate reach task embedding vector to obtain the evolved task preference hidden state sequence.

[0158] In some embodiments, the task preference evolution layer further includes an AUGRU module; the time preference evolution layer further includes an AUGRU module.

[0159] In some embodiments, when the processing module 802 is implemented, the current sequence segment in the evolved task preference hidden state sequence can be obtained in the following manner: using the current sequence segment in the hidden state sequence of the task preference and the candidate reach task embedding vector, an attention operation is performed to obtain the current correlation parameter between the current task preference and the candidate reach task; using the AUGRU module to process the previous sequence segment and the current correlation parameter in the evolved task preference hidden state sequence to obtain the current sequence segment in the evolved task preference hidden state sequence.

[0160] In some embodiments, the prediction result may specifically include multiple acceptance probability parameter groups, wherein one acceptance probability parameter group corresponds to one candidate reach time slice; the acceptance probability parameter group includes multiple acceptance probabilities, wherein one acceptance probability corresponds to one candidate reach task;

[0161] Accordingly, when the above-mentioned determination module 803 is implemented, the target reach time slice and target reach task matching the target user can be determined according to the prediction results in the following manner: sort the acceptance probabilities contained in each acceptance probability parameter group according to the probability value to obtain multiple sorted acceptance probability parameter groups; compare the acceptance probabilities ranked first in the multiple sorted acceptance probability parameter groups, and determine the acceptance probability parameter group with the largest value and ranked first as the target acceptance probability parameter group; determine the candidate reach time slice corresponding to the target acceptance probability parameter group as the target reach time slice; determine the candidate reach task corresponding to the acceptance probability ranked first in the target acceptance probability parameter group as the target reach task.

[0162] In some embodiments, the device may also include a reach module, which, when implemented, can be used to perform a target reach task for the target user in the target reach time slice after determining the target reach time slice and target reach task matching the target user based on the prediction results.

[0163] In some embodiments, the device may also include a training module, which can be used to obtain the interaction records of the sample user's historical reach tasks; establish sample training data based on the interaction records of the sample user's historical reach tasks; construct an initial prediction model; wherein the initial prediction model at least includes an initial task preference evolution structure and an initial time preference evolution structure in parallel; and use the sample training data to train the initial prediction model to obtain a prediction model that meets the requirements.

[0164] In some embodiments, when the device is implemented, after obtaining the interaction records of the sample user's historical reach tasks, it can also filter out historical reach time slices in which the frequency of the sample user accepting historical reach tasks is greater than a preset frequency threshold as candidate reach time slices based on the interaction records of the sample user's historical reach tasks.

[0165] It should be noted that the units, devices or modules described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. For the convenience of description, the above devices are described separately by functions divided into various modules. Of course, when implementing this specification, the functions of each module can be implemented in the same or more software and / or hardware, or the modules that implement the same function can be implemented by a combination of multiple sub-modules or sub-units. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0166] As can be seen from the above, the processing device for the reach task provided in the embodiment of this specification can be pre-built and trained through deep learning to obtain a prediction model that at least includes a parallel task preference evolution structure and a time preference evolution structure before specific implementation. During specific implementation, the characteristic data, historical reach task sequence, historical reach time slice sequence of the target user to be reached, and the candidate reach task sequence and candidate reach time slice sequence for the target user can be obtained first; then the above prediction model is used to process the characteristic data, historical reach task sequence, historical reach time slice sequence of the target user, and the candidate reach task sequence and candidate reach time slice sequence, to obtain a prediction result that simultaneously considers the user's interest preference for the reach task and the reach time slice; and then, according to the prediction result, the target reach time slice and target reach task that match the target user can be accurately determined. Thereby, the target reach time slice and target reach task that the target user has a high probability of acceptance and a high degree of participation can be accurately determined at the same time, so that the target reach task for the target user can be carried out in a targeted manner according to the above target reach time slice in the future, and a better reach effect can be obtained.

[0167] Although the present specification provides method operation steps as described in the embodiments or flow charts, more or less operation steps may be included based on conventional or non-creative means. The order of steps listed in the embodiments is only one way of executing the order of many steps, and does not represent a unique execution order. When the device or client product in practice is executed, it can be executed in sequence or in parallel according to the method shown in the embodiments or the drawings (for example, a parallel processor or a multi-threaded processing environment, or even a distributed data processing environment). The term "include", "include" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, product or device including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such a process, method, product or device. In the absence of more restrictions, it is not excluded that there are other identical or equivalent elements in the process, method, product or device including the elements. The first, second, etc. words are used to represent the name, and do not represent any particular order.

[0168] Those skilled in the art also know that, in addition to implementing the controller in a purely computer-readable program code, the controller can be made to implement the same function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered as a hardware component, and the devices for implementing various functions included therein can also be considered as structures within the hardware component. Or even, the devices for implementing various functions can be considered as both software modules for implementing the method and structures within the hardware component.

[0169] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, classes, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media including storage devices.

[0170] Through the description of the above embodiments, it can be known that those skilled in the art can clearly understand that the present specification can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present specification can essentially be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a mobile terminal, a server, or a network device, etc.) to execute the methods described in each embodiment of the present specification or some parts of the embodiments.

[0171] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. This specification can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc.

[0172] Although the present specification is described through embodiments, those skilled in the art will appreciate that there are many modifications and changes to the present specification without departing from the spirit of the present specification, and it is intended that the appended claims include these modifications and changes without departing from the spirit of the present specification.

Claims

1. A method for processing a reach task, characterized in that: include: Acquire the target user's characteristic data, historical reach task sequence, historical reach time slice sequence, as well as candidate reach task sequence and candidate reach time slice sequence; wherein, the candidate reach time slice sequence is determined according to the target user's historical reach task record; the target user's characteristic data includes the target user's portrait characteristics and / or the target user's behavioral characteristics, the target user's portrait characteristics include at least one of the following: the target user's gender, the target user's city, the target user's occupation, the target user's education level, the target user's behavioral characteristics include at least one of the following: the target user's behavioral operation characteristics for historical reach tasks, the target user's behavioral operation characteristics when using a terminal device, the target user's behavioral operation characteristics for historical push information; the historical reach task sequence includes at least one of the following: the task characteristics of historical reach tasks, the reach performance of historical reach tasks, the target user's interaction characteristics for historical reach tasks; according to the target user's characteristic data, historical reach task sequence, historical reach time The method comprises: using a prediction model to process the characteristic data of the target user, the historical reach task sequence, the historical reach time slice sequence, the candidate reach task sequence, and the candidate reach time slice sequence, and obtaining corresponding prediction results; wherein the prediction model at least includes a parallel task preference evolution structure and a time preference evolution structure; according to the prediction results, determining the target reach time slice and the target reach task that match the target user; the task preference evolution structure includes: a task preference extraction layer and a task preference evolution layer; the time preference evolution structure includes: a time preference extraction layer and a time preference evolution layer; the prediction results include multiple acceptance probability parameter groups, wherein one acceptance probability parameter group corresponds to one candidate reach time slice; the acceptance probability parameter group includes multiple acceptance probabilities, wherein one acceptance probability corresponds to one candidate reach task; Among them, according to the prediction result, the target reach time slice and target reach task matching the target user are determined, including: sorting the acceptance probabilities contained in each acceptance probability parameter group according to the probability value to obtain multiple sorted acceptance probability parameter groups; comparing the acceptance probabilities ranked first in the multiple sorted acceptance probability parameter groups, and determining the acceptance probability parameter group with the largest value and ranked first as the target acceptance probability parameter group; determining the candidate reach time slice corresponding to the target acceptance probability parameter group as the target reach time slice; determining the candidate reach task corresponding to the acceptance probability ranked first in the target acceptance probability parameter group as the target reach task.

2. The method according to claim 1, characterized in that After determining the target reach time slice and the target reach task matching the target user, the method further includes: During the target reach time slice, carry out target reach tasks for target users.

3. The method according to claim 1, characterized in that The prediction model also includes an embedding layer, a connection layer and a classification layer; wherein the embedding layer is connected to the task preference evolution structure and the time preference evolution structure respectively; and the task preference evolution structure and the time preference evolution structure are connected to the connection layer respectively.

4. The method according to claim 3, characterized in that The prediction model is used to process the historical reach task sequence, the historical reach time slice sequence, the candidate reach task sequence, and the candidate reach time slice sequence of the target user to obtain corresponding prediction results, including: Using the embedding layer to process the target user's feature data, historical reach task sequence, historical reach time slice sequence, and candidate reach task sequence, candidate reach time slice sequence, respectively, to obtain the corresponding target user's feature data embedding vector, historical reach task embedding vector, historical reach time slice embedding vector, and candidate reach task embedding vector, candidate reach time slice embedding vector; The task preference evolution structure is used to process the historical touch task embedding vector and the candidate touch task embedding vector to obtain the evolved task preference hidden state sequence; the time preference evolution structure is used to process the historical touch time slice embedding vector and the candidate touch time slice embedding vector to obtain the evolved time preference hidden state sequence; Using a connection layer, concatenate the evolved task preference hidden state sequence, the evolved time preference hidden state sequence, the target user's feature data embedding vector, the candidate reach task embedding vector, and the candidate reach time slice embedding vector to obtain an aggregate vector; The aggregation vector is processed by using a classification layer to obtain a corresponding processing result.

5. The method according to claim 4, characterized in that The task preference evolution structure is used to process the historical touch task embedding vectors and the candidate touch task embedding vectors to obtain the evolved task preference hidden state sequence, including: Use the task preference extraction layer to process the historical access task embedding vector and extract the hidden state sequence of task preference; The task preference evolution layer is used to process the hidden state sequence of the task preference and the candidate reach task embedding vector to obtain the evolved task preference hidden state sequence.

6. The method according to claim 5, characterized in that The task preference evolution layer also includes an AUGRU module; the time preference evolution layer also includes an AUGRU module.

7. The method according to claim 6, characterized in that The task preference evolution layer is used to process the hidden state sequence of the task preference and the candidate reach task embedding vector to obtain the evolved task preference hidden state sequence, including: The current sequence segment in the evolved task preference hidden state sequence is obtained in the following way: Using the current sequence segment in the hidden state sequence of the task preference and the candidate reach task embedding vector, an attention operation is performed to obtain a current correlation parameter between the current task preference and the candidate reach task; The AUGRU module is used to process the previous sequence segment and the current correlation parameter in the evolved task preference hidden state sequence to obtain the current sequence segment in the evolved task preference hidden state sequence.

8. The method according to claim 1, characterized in that The method further comprises: Obtain the interaction records of historical reach tasks of sample users; Create sample training data based on the sample users' historical interaction records of the task. Constructing an initial prediction model; wherein the initial prediction model at least includes an initial task preference evolution structure and an initial time preference evolution structure in parallel; The sample training data is used to train the initial prediction model to obtain a prediction model that meets the requirements.

9. The method according to claim 8, characterized in that After obtaining the interaction records of the historical reach tasks of the sample users, the method further includes: According to the interaction records of the sample users' historical reach tasks, the historical reach time slices in which the frequency of the sample users accepting the historical reach tasks is greater than a preset frequency threshold are screened out as candidate reach time slices.

10. A processing device for a touch task, characterized in that: include: An acquisition module is used to acquire characteristic data of a target user, a historical reach task sequence, a historical reach time slice sequence, and a candidate reach task sequence and a candidate reach time slice sequence; wherein the candidate reach time slice sequence is determined according to the historical reach task record of the target user; the characteristic data of the target user includes the portrait characteristics of the target user and / or the behavioral characteristics of the target user, the portrait characteristics of the target user include at least one of the following: the gender of the target user, the city where the target user is located, the occupation of the target user, and the educational background of the target user; the behavioral characteristics of the target user include at least one of the following: the behavioral operation characteristics of the target user for historical reach tasks, the behavioral operation characteristics of the target user when using a terminal device, and the behavioral operation characteristics of the target user for historical push information; the historical reach task sequence includes at least one of the following: the task characteristics of historical reach tasks, the reach performance of historical reach tasks, and the interactive characteristics of the target user for historical reach tasks; A determination module is used to determine a target reach time slice and a target reach task that match the target user based on the target user's feature data, historical reach task sequences, historical reach time slice sequences, and candidate reach task sequences and candidate reach time slice sequences; Wherein, the determination module is specifically used to: use the prediction model to process the characteristic data of the target user, the historical reach task sequence, the historical reach time slice sequence, and the candidate reach task sequence, and the candidate reach time slice sequence to obtain the corresponding prediction result; wherein, the prediction model at least includes a parallel task preference evolution structure and a time preference evolution structure; according to the prediction result, determine the target reach time slice and target reach task that match the target user; the task preference evolution structure includes: a task preference extraction layer and a task preference evolution layer; the time preference evolution structure includes: a time preference extraction layer and a time preference evolution layer; the prediction result includes multiple acceptance probability parameter groups, wherein one acceptance probability parameter group corresponds to one candidate reach time slice; the acceptance probability parameter group includes multiple acceptance probabilities, wherein one acceptance probability corresponds to one candidate reach task; Among them, according to the prediction result, the target reach time slice and target reach task matching the target user are determined, including: sorting the acceptance probabilities contained in each acceptance probability parameter group according to the probability value to obtain multiple sorted acceptance probability parameter groups; comparing the acceptance probabilities ranked first in the multiple sorted acceptance probability parameter groups, and determining the acceptance probability parameter group with the largest value and ranked first as the target acceptance probability parameter group; determining the candidate reach time slice corresponding to the target acceptance probability parameter group as the target reach time slice; determining the candidate reach task corresponding to the acceptance probability ranked first in the target acceptance probability parameter group as the target reach task.

11. A server, characterized in that: The method comprises a processor and a memory for storing processor-executable instructions, wherein the processor implements the steps of the method according to any one of claims 1 to 9 when executing the instructions.

12. A computer-readable storage medium, characterized in that: Computer instructions are stored thereon, and when the instructions are executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

Images