Message push method, device, electronic device and medium based on neural network
By using an LSTM prediction model to predict the message volume within the target push period, the problem of excessive server pressure during peak hours is solved, thus achieving stability and reliability of message push.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- INDUSTRIAL AND COMMERCIAL BANK OF CHINA
- Filing Date
- 2022-09-22
- Publication Date
- 2026-05-26
AI Technical Summary
Existing message push systems experience excessive server load during peak hours due to a large number of concurrent messages, affecting the reliability of scheduled pushes.
A prediction model based on Long Short-Term Memory (LSTM) network is used to predict the amount of other messages within the target push period. Based on the total amount of messages and the server load capacity, it is determined whether to push message content or adjust the push period to avoid peak periods.
This effectively reduced the server load during peak hours, ensuring smooth message delivery and improving system stability and reliability.
Smart Images
Figure CN115563406B_ABST
Abstract
Description
Technical Field
[0001] This application relates to artificial intelligence technology, and more particularly to a message push method, device, electronic device, and medium based on neural networks. Background Technology
[0002] With the rapid development of internet technology, most mobile applications now offer push notification functionality, such as trending news recommendations in news apps, chat message notifications in instant messaging tools, and promotional information for e-commerce products. Push notifications play a crucial role in increasing product activity, improving the usage rate of functional modules, enhancing user stickiness, and increasing user retention. As a key channel in mobile application operations, its proper use can effectively promote the achievement of goals.
[0003] Push notifications specifically refer to proactive message delivery from operations personnel to clients using their own products or third-party tools. After a successful push, users can see the notification on their client's lock screen and in the notification bar. Clicking the notification will launch the client and take them to the corresponding page. Currently, operations personnel sometimes choose scheduled push notifications, where they pre-schedule the push time to deliver the message content to the client within the desired timeframe.
[0004] Due to the growth of various information services, there may be a large number of concurrent pushes within the same time period. Therefore, in the above process, if there are a large number of other messages in addition to the scheduled pushes within the time period scheduled by the operators, it will put a lot of pressure on the message server, which will be detrimental to the scheduled push of messages. Summary of the Invention
[0005] This application provides a message push method, device, electronic device, and medium based on neural networks, which can predict the amount of other messages within a target push time period in advance. When there are too many other messages in the same time period, the operator is prompted to change the target push time period, thereby reducing the risk of server crashes due to excessive message volume in a certain time period.
[0006] On the one hand, this application provides a message push method based on a neural network, the method comprising:
[0007] Obtain the message to be pushed, which includes message content, message volume, and target push time period;
[0008] Based on the Long Short-Term Memory (LSTM) prediction model, the amount of other messages during the target push period is predicted. The amount of other messages includes chat messages, push notifications from accessed applications, and tweets from information accounts.
[0009] Based on the message volume and the other message volumes, a push notification is determined and fed back. The push notification is used to indicate whether the message content can be pushed within the target push time period.
[0010] If the push notification is used to instruct the message content to be pushed within the target push time period, then the message content is pushed within the target push time period.
[0011] In another possible implementation, determining and responding to push notifications based on the message volume and the other message volumes includes:
[0012] The total message volume is obtained by summing the message volume and the other message volumes.
[0013] If the total message volume is less than the preset message volume, then the push notification is determined to be used to instruct the push of the message content within the target push time period; the preset message volume is the maximum message volume that can be handled within the target push time period.
[0014] If the total number of messages is not less than the preset number of messages, then the push notification is determined to indicate that the message content cannot be pushed during the target push time period.
[0015] In another possible implementation, if the push notification is used to indicate that the message content cannot be pushed within the target push time period, the method further includes: performing the following update push process:
[0016] Get the update push time period;
[0017] Based on the LSTM prediction model, predict the amount of new other messages during the update push period;
[0018] Based on the message volume and the new other message volume, determine and provide updated push notifications;
[0019] If the updated push notification is used to instruct the message content to be pushed during the update push period, then the message content will be pushed during the update push period.
[0020] In another possible implementation, if the updated push notification indicates that the message content cannot be pushed within the update push period, the update push process is repeated until the updated push notification indicates that the message content can be pushed within the update push period.
[0021] In another possible implementation, before pushing the message content within the target push time period, the method further includes:
[0022] Get the appointment time slot from the client's feedback;
[0023] The target push time period is determined based on the scheduled time period, and the time period other than the target push time period in the scheduled time period is used as the update push time period.
[0024] In another possible implementation, predicting the amount of other messages within the target push period based on the LSTM prediction model includes:
[0025] Obtain the volume of other historical messages within a preset number of time periods prior to the target push time period;
[0026] The amount of other historical messages within the preset number of time periods is input into the trained LSTM prediction model to obtain the amount of other messages within the target push time period.
[0027] In another possible implementation, before inputting the amount of other historical messages within the preset number of time periods into the trained LSTM prediction model, the amount of other historical messages within the preset number of time periods is preprocessed, the preprocessing including data cleaning, one-hot encoding and data normalization.
[0028] In another possible implementation, the training process of the LSTM prediction model includes:
[0029] Training sample data is constructed based on historical sample data; the training sample data includes an input set of other message volumes arranged in chronological order and a target output set of other message volumes arranged in chronological order.
[0030] The initial LSTM prediction model is trained based on the training sample data to obtain the trained LSTM model.
[0031] In another possible implementation, training the initial LSTM prediction model based on the training sample data to obtain the trained LSTM model includes:
[0032] The amount of other messages in the input set before the current time period is taken as the sample features of the current time period, and the amount of other messages in the target output set during the current time period is taken as the sample labels of the current time period.
[0033] Obtain the predicted value output by the initial LSTM prediction model based on the sample features;
[0034] Based on the predicted value of the current time period and the sample label of the current time period, determine the loss function;
[0035] Based on the loss function, the initial LSTM prediction model is adjusted until the loss function converges, resulting in a trained LSTM prediction model.
[0036] On the other hand, this application provides a message push device based on a neural network, including a receiving module, a prediction module, a feedback module, and a push module, wherein...
[0037] The receiving module is used to acquire the message to be pushed, which includes message content, message volume and target push time period;
[0038] The prediction module is used to predict the volume of other messages within the target push time period based on the LSTM prediction model. The volume of other messages includes chat messages, push notifications from access applications, and tweets from information accounts.
[0039] The feedback module is used to determine and provide a push notification based on the message volume and the other message volumes. The push notification is used to indicate whether the message content can be pushed within the target push time period.
[0040] The push module is used to push the message content within the target push time period when the push notification indicates that the message content should be pushed within the target push time period.
[0041] In another possible implementation, the feedback module is specifically used for:
[0042] The total message volume is obtained by summing the message volume and the other message volumes.
[0043] If the total message volume is less than the preset message volume, then the push notification is determined to be used to instruct the push of the message content within the target push time period; the preset message volume is the maximum message volume that can be handled within the target push time period.
[0044] If the total number of messages is not less than the preset number of messages, then the push notification is determined to indicate that the message content cannot be pushed during the target push time period.
[0045] In another possible implementation, the apparatus further includes an update push module, specifically configured to execute an update push process, the update push process including:
[0046] Get the update push time period;
[0047] Based on the LSTM prediction model, predict the amount of new other messages during the update push period;
[0048] Based on the message volume and the new other message volume, determine and provide updated push notifications;
[0049] If the updated push notification is used to instruct the message content to be pushed during the update push period, then the message content will be pushed during the update push period.
[0050] In another possible implementation, the update push module is also used for:
[0051] If the updated push notification indicates that the message content cannot be pushed within the updated push period, the update push process is repeated until the updated push notification indicates that the message content can be pushed within the updated push period.
[0052] In another possible implementation, the device further includes a sending module, an acquisition module, and a determining module, wherein,
[0053] The acquisition module is used to obtain the appointment time slot returned by the client.
[0054] The determination module is used to determine the target push time period based on the scheduled time period, and to use the time period other than the target push time period in the scheduled time period as the update push time period.
[0055] In another possible implementation, the prediction module is specifically used for:
[0056] Obtain the volume of other historical messages within a preset number of time periods prior to the target push time period;
[0057] The amount of other historical messages within the preset number of time periods is input into the trained LSTM prediction model to obtain the amount of other messages within the target push time period.
[0058] In another possible implementation, the prediction module is also used for:
[0059] Before inputting the amount of other historical messages within the preset number of time periods into the trained LSTM prediction model, the amount of other historical messages within the preset number of time periods is preprocessed, including data cleaning, one-hot encoding, and data normalization.
[0060] In another possible implementation, the training process of the LSTM prediction model includes:
[0061] Training sample data is constructed based on historical sample data; the training sample data includes an input set of other message volumes arranged in chronological order and a target output set of other message volumes arranged in chronological order.
[0062] The initial LSTM prediction model is trained based on the training sample data to obtain the trained LSTM model.
[0063] In another possible implementation, training the initial LSTM prediction model based on the training sample data to obtain the trained LSTM model includes:
[0064] The amount of other messages in the input set before the current time period is taken as the sample features of the current time period, and the amount of other messages in the target output set during the current time period is taken as the sample labels of the current time period.
[0065] Obtain the predicted value output by the initial LSTM prediction model based on the sample features;
[0066] Based on the predicted value of the current time period and the sample label of the current time period, determine the loss function;
[0067] Based on the loss function, the initial LSTM prediction model is adjusted until the loss function converges, resulting in a trained LSTM prediction model.
[0068] Thirdly, the present invention provides an electronic device, comprising:
[0069] At least one processor and memory;
[0070] The memory stores computer-executed instructions;
[0071] The at least one processor executes computer execution instructions stored in the memory, causing the at least one processor to perform the neural network-based message push method as described in any of the first aspects above.
[0072] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions, wherein when a processor executes the computer-executable instructions, the message push method based on a neural network as described in any of the first aspects above is implemented.
[0073] This application provides a message push method, apparatus, electronic device, and medium based on neural networks. Through this application, when operators set a target push time period, the electronic device predicts the amount of other messages within that target push time period based on a trained LSTM prediction model. Only when the sum of the amount of other messages within the target push time period and the amount of the message to be pushed is less than the server's maximum load, a push notification prompts the operator that the message content can be pushed within the target push time period, and the message content is pushed within that target push time period. Conversely, when the sum of the amount of other messages within the target push time period and the amount of the message to be pushed is not less than the server's maximum load, a push notification prompts the operator that the message content cannot be pushed within the target push time period, allowing operators to avoid peak message periods, reducing the server's pressure at the same time, and thus facilitating timely message pushes. Attached Figure Description
[0074] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0075] Figure 1 A schematic diagram illustrating a scenario for periodically pushing messages using an information number, provided as an embodiment of this application;
[0076] Figure 2 A flowchart illustrating a message push method based on a neural network provided in this application embodiment. Figure 1 ;
[0077] Figure 3 A flowchart illustrating a message push method based on a neural network provided in this application embodiment. Figure 2 ;
[0078] Figure 4a A schematic diagram illustrating a method for training an LSTM prediction model provided in an embodiment of this application;
[0079] Figure 4b A topology diagram of an LSTM prediction model provided in an embodiment of this application;
[0080] Figure 4c This is a schematic diagram of the structure of an LSTM prediction model provided in an embodiment of this application;
[0081] Figure 5 A schematic diagram of the structure of a message push device based on a neural network provided in this application embodiment;
[0082] Figure 6 An electronic device provided in an embodiment of this application.
[0083] The accompanying drawings have illustrated specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation
[0084] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.
[0085] Figure 1 This is a schematic diagram illustrating a scenario where an information signal is periodically pushed out in an embodiment of this application. Figure 1 As shown, enterprise operators can input messages to be pushed in advance via terminal device 101. The messages include the message content and the target push time. The cloud server 102 will push the message content to the client user's terminal device 103 at the target push time. The client user's terminal device 103 can be any terminal device that can install the client, such as a mobile phone, tablet, or computer.
[0086] In practical applications, due to the growth of various information services, within the target push time selected by enterprise operators, there may be additional messages that need to be pushed in addition to the messages that the enterprise needs to push. These include instant push messages generated within the target push time and other scheduled push messages from other enterprises.
[0087] Therefore, if the above issues are not considered when selecting the target push time, and the target push time is blindly selected based on one's own wishes, it may result in an excessive amount of messages at a certain time, which will put great pressure on the message server and thus hinder the smooth push of message content from various enterprises.
[0088] Figure 2 A flowchart illustrating a message push method based on a neural network provided in this application embodiment. Figure 1 .like Figure 2 As shown, the method in this embodiment may include steps S201, S202, S203, and S204, wherein,
[0089] Step S201: Obtain the message to be pushed.
[0090] The messages to be pushed include message content, message volume, and target push time period.
[0091] Specifically, the messages to be pushed are entered into the electronic device by relevant staff of the enterprise. The message content and message volume are manually entered by relevant staff of the enterprise, and the target push time period can be manually entered by relevant staff of the enterprise, or it can be automatically and randomly selected by the electronic device in combination with the client user's wishes.
[0092] Step S202: Based on the Long Short-Term Memory (LSTM) prediction model, predict the amount of other messages within the target push period.
[0093] Specifically, the LSTM prediction model predicts the amount of other messages within the target push period based on the amount of other messages before the target push period. In other words, it predicts the amount of messages that should exist within the target push period based on the historical message volume, namely the amount of chat messages, application pushes, and information account tweets that should occur within the target push period.
[0094] Step S203: Based on the message volume and other message volumes, determine and provide push notifications.
[0095] Push notifications are used to indicate whether message content can be pushed within the target push time period.
[0096] Specifically, if the target push time period is manually entered by relevant company staff into the electronic device, the push notification may be a pop-up displayed on the screen of the electronic device. The pop-up message may indicate whether the message to be pushed can / cannot be pushed within the specified target push time period, thus informing the relevant company staff. If the target push time period is automatically or randomly selected by the electronic device in conjunction with the client user's wishes, the electronic device may or may not display a pop-up notification.
[0097] Step S204: If the push notification is used to indicate that message content is to be pushed within the target push time period, then the message content is pushed within the target push time period.
[0098] Specifically, when an electronic device learns that it can push message content within a target push time period, it pushes the message content to be pushed to the client within the target push time period.
[0099] The method provided in this embodiment involves an electronic device manually or automatically selecting a target push time period. Then, an LSTM prediction model predicts the volume of other messages within that target push time period. Based on the volume of other messages and the volume of the message to be pushed within the target push time period, it is determined whether the message content of the message to be pushed can be pushed within the target push time period. By predicting the volume of other messages within the target push time period in advance, the situation where a large volume of messages pushes within a certain time period could lead to server overload can be effectively avoided, thus facilitating the smooth progress of message pushes.
[0100] Optionally, electronic devices can use an LSTM prediction model to predict the message volume for each time period in the next few days, filter out the time periods for pushing message content, and display these time periods to relevant enterprise staff so that they can select the target push time period based on these time periods, thereby reducing the probability of incorrectly selecting the target push time period.
[0101] Figure 3 This is a flowchart illustrating a message push method based on a neural network provided in an embodiment of this application. Figure 2 The following is combined with Figure 3 The specific implementation process of the embodiments of this application will be described in detail. Specifically, based on the above embodiments, this embodiment specifies in detail the method for determining push notifications.
[0102] like Figure 3 As shown, the method includes steps S301 to S312, wherein,
[0103] Step S301: Obtain the appointment time slot from the client.
[0104] Specifically, in this embodiment, the reservation time period is the time period during which the client user can receive push messages based on their own wishes, and the electronic device obtains the reservation time periods fed back by each client.
[0105] Step S302: Determine the target push time period based on the scheduled time period, and use the time period other than the target push time period in the scheduled time period as the update push time period.
[0106] Both the target push time period and the update push time period are scheduled time periods. Furthermore, the target push time period or the update push time period can be selected by relevant staff of the enterprise based on the scheduled time period, or the electronic device can automatically select the target push time period or the update push time period based on the scheduled time period.
[0107] In this embodiment, the electronic device automatically selects the target push time period for each client. Specifically, for each client's requested time period, the electronic device automatically divides the time period into N time periods according to preset rules. The start time of the first time period is the same as the start time of the requested time period, and the end time of the Nth time period is the same as the end time of the requested time period. The electronic device automatically selects the first time period within the requested time period as the target push time period, and the time periods within the requested time period other than the first time period can be used as update push time periods.
[0108] Optionally, when selecting an update push time period, the electronic device automatically selects the second time period after the first time period as the update push time period, that is, the electronic device automatically selects the next time period after the current time period as the update push time period.
[0109] Step S303: Obtain the message to be pushed.
[0110] In this embodiment, the target push time period in the message to be pushed is automatically selected by the electronic device in combination with the client user's wishes.
[0111] Step S304: Obtain the historical message volume of a preset number of time periods before the target push time period.
[0112] The target push time period is an integer multiple of the time period duration. In this embodiment, the target push time period duration is consistent with the duration of each time period within a preset number of time periods. The electronic device stores the amount of other messages in each time period and the predicted message volume for the next time period corresponding to each time period, and updates this information in real time. When making predictions, the electronic device automatically obtains the historical amount of other messages within the preset number of time periods preceding the target push time period.
[0113] Step S305: Input the historical message volume of a preset number of time periods into the trained LSTM prediction model to obtain the message volume of other messages within the target push time period.
[0114] Specifically, before inputting the other messages from a preset number of time periods into the trained LSTM prediction model, the other messages from the preset number of time periods are first preprocessed. The preprocessing may include data cleaning, one-hot encoding, and data normalization to improve the quality of the other messages from the preset number of time periods input into the LSTM prediction model.
[0115] Optionally, the amount of other historical messages within a preset number of time periods before the target push time period can be either the amount of other historical messages within a preset number of time periods before the start time of the target push time period, or the time periods corresponding to the target push time period each day within a preset number of days before the date of the target push time period. For example, if the target push time period is 07:00-07:20 on July 20th, then the amount of other messages from 07:00-07:20 within the preset number of days before July 20th can be input into the trained LSTM prediction model.
[0116] Step S306: Sum the message volume and other message volumes to obtain the total message volume.
[0117] Specifically, after obtaining the amount of other messages in the target delivery time period through the LSTM prediction model, the electronic device sums the amount of other messages in the target push time period and the amount of messages to be pushed to obtain the total amount of messages in the target push time period.
[0118] In step S307, when the total message volume is obtained, determine whether the total message volume is less than the preset message volume; if yes, proceed to step S308; otherwise, proceed to step S309.
[0119] The preset message volume is the maximum message volume that the server can handle. Furthermore, to ensure smooth message pushing at different times, the preset message volume can be set to 90% of the server's maximum message volume.
[0120] Specifically, after obtaining the total number of messages during the target push period, the electronic device automatically compares the total number of messages with the preset number of messages. If the total number of messages is less than the preset number of messages, the electronic device continues to execute step S308. If the total number of messages is greater than or equal to the preset number of messages, the electronic device executes step S309.
[0121] Step S308: Push the message content within the target push time period.
[0122] Specifically, electronic devices push message content to the client through third-party tools during the target push period.
[0123] Step S309: Obtain the update push time period.
[0124] As defined in step S302 above, the update push time period can be any time period within the scheduled time period other than the target push time period.
[0125] Specifically, in this embodiment, when the electronic device determines through the above steps that it cannot push message content within the target push time period, it automatically selects a time period after the target push time period as the update push time period, that is, the start time of the update push time period is the same as the end time of the target push time period.
[0126] Step S310: Obtain the amount of other historical messages within a preset number of time periods before the update push time period.
[0127] The duration of the update push period is an integer multiple of the duration of each of the preset number of time periods. In this embodiment, the duration of the update push period is the same as the duration of each of the preset number of time periods.
[0128] Step S311: Input the historical message volume within a preset number of time periods into the trained LSTM prediction model to obtain the message volume within the update push time period.
[0129] Specifically, before inputting the amount of other historical messages within a preset number of time periods into the trained LSTM prediction model, the amount of other historical messages within a preset number of time periods before the update push time period is also preprocessed as described in step S305 above.
[0130] Step S312: Sum the message volume and the other message volume within the update push period to obtain the total message volume.
[0131] Specifically, the electronic device sums the message volume and the other message volume within the update push period to obtain the total message volume within the update push period. At this time, the electronic device automatically triggers step S307 to compare the total message volume within the update push period with the preset message volume. If the total message volume within the update push period is less than the preset message volume, the electronic device continues to execute step S308; otherwise, the electronic device continues to execute steps S309-S312.
[0132] For example, on July 19th, relevant staff members of the enterprise input the messages to be pushed on July 20th into an electronic device. The input includes the message content, the number of messages, and the push time (July 20th). If a client user selects a scheduled time slot of 07:00-09:00 every day, after obtaining the client's scheduled push time slot, the electronic device divides the scheduled time slot into N time slots according to pre-set rules. Specifically, it divides the scheduled time slot into 6 time slots, each lasting 20 minutes. When the electronic device receives the push message, it automatically selects 07:00-07:20 as the target push time slot. At the same time, it obtains the number of other messages within the 07:00-07:20 time slot in the preset number of days before July 20th, preprocesses this data, and inputs it into a trained LSTM prediction model to obtain the number of other messages between 07:00 and 07:20 on July 20th.
[0133] After determining the volume of other messages between 07:00 and 07:20 on July 20th, the electronic device sums this volume with the volume of messages to be pushed, and then compares this sum with a preset message volume. If the total message volume is less than the preset message volume, the electronic device will push the message content between 07:00 and 07:20 on July 20th. If the total message volume is not less than the preset message volume, the electronic device automatically selects 07:20-07:40 as the update push time period. Simultaneously, it retrieves the volume of other messages between 07:20 and 07:40 from the preset number of days prior to July 20th, preprocesses this data, and inputs it into a trained LSTM prediction model to obtain the volume of other messages between 07:20 and 07:40 on July 20th.
[0134] After determining the volume of other messages between 07:20 and 07:40 on July 20th, the electronic device sums this volume with the volume of messages to be pushed, and then compares this sum with a preset message volume. If the total message volume is less than the preset message volume, the electronic device will push the message content between 07:20 and 07:40 on July 20th. Otherwise, the electronic device automatically selects 07:40 to 08:00 as the next update push time period.
[0135] Optionally, in this embodiment, the preset message volume is set to 90% of the maximum message volume that the server can handle. If the electronic device ultimately determines that there is no time period within the scheduled time period where the total message volume is less than the preset message volume, then the message content should be pushed during the time period with the smallest total message volume.
[0136] The method provided in this embodiment, when pushing messages, first considers the client user's wishes and determines the scheduled time period. Secondly, based on the scheduled time period and the total message volume of each time period within that period, it determines whether to push messages within that time period. Messages are only pushed when the total message volume in a certain time period is less than a preset message volume, reducing the risk of server overload and ensuring the smooth progress of message pushing for all parties.
[0137] Figure 4a This is a schematic diagram illustrating a method for training an LSTM prediction model provided in an embodiment of this application. Figure 4b This is a topology diagram of an LSTM prediction model provided in an embodiment of this application. Figure 4c This is a schematic diagram of the structure of an LSTM prediction model provided in an embodiment of this application. The following is in conjunction with... Figure 4a , Figure 4b and Figure 4c The specific implementation process of the embodiments of this application will be further described in detail below. Specifically, this embodiment specifies how to train the LSTM prediction model.
[0138] like Figure 4a As shown, the method may include steps S401, S402, S403, S404, and S405, wherein,
[0139] Step S401: Construct training sample data based on historical sample data; the training sample data includes an input set of other message volumes arranged in chronological order and a target output set of other message volumes arranged in chronological order.
[0140] The sample data includes the actual volume of other messages in each time period.
[0141] Specifically, the electronic device first determines the duration of each time period, and then records the amount of other messages every other time period in real time as the amount of other messages for the corresponding time period.
[0142] After the electronic device performs preprocessing on the real other message volume for each time period as described in step S305 of the aforementioned embodiment, it stores the processed data with the time period as the unique index to obtain the input set and the target output set.
[0143] Step S402: Use the amount of other messages in the input set that are a preset number of time periods before the current time period as the sample features of the current time period, and use the amount of other messages in the target output set that are the current time period as the sample labels of the current time period.
[0144] Specifically, for any given time period, the amount of other messages in a preset number of time periods preceding the current time period is obtained as the sample features of the current time period, and the amount of other messages in the target output set corresponding to the current time period is obtained as the sample label of the current time period.
[0145] Step S403: Obtain the predicted value output by the LSTM prediction model based on the sample features.
[0146] Specifically, the sample features of any current time period are input into the LSTM prediction model to obtain the predicted value for that current time period.
[0147] Step S404: Determine the loss function based on the predicted value and sample label of the current time period.
[0148] Specifically, in this embodiment, the mean absolute value (MAE) error calculated from the predicted value and the sample label is used as the loss function. Therefore, the mathematical model of the loss function is: In this mathematical model, This represents the predicted value for any given time period. This represents the sample label for any given time period.
[0149] Step S405: Based on the loss function, adjust the LSTM prediction model until the loss function converges to obtain the trained LSTM prediction model.
[0150] Specifically, the electronic device backpropagates the gradients calculated from the loss function to adjust all values in the LSTM prediction model until the loss function converges.
[0151] like Figure 4b As shown, in this embodiment, the LSTM prediction model includes an input layer, a hidden layer, and an output layer. The input layer is used to convert the preprocessed input sequence and the target output sequence into data suitable for supervised learning. For example, as shown... Figure 4b As shown, the input X_1 for the current time period consists of the amount of other messages corresponding to a preset number of time periods (3 periods, ts01, ts02, ts03) before the current time period, and the sample label for the current time period. These time periods are used as unique indices to input the data into the hidden layer. The number of hidden layers is flexibly adjusted based on the input data features and model training performance. In this embodiment, a genetic algorithm (GA) is incorporated into the hidden layer to optimize the number of units and feature selection, addressing the problems of imbalanced predictions and getting trapped in local maxima, which lead to poor performance. Correspondingly, when X_1 is input, the output prediction result is O_1. Based on the prediction results O_1, O_2, O_3, etc., and their corresponding sample labels, the loss function MAE is determined.
[0152] like Figure 4cAs shown, the LSTM prediction model adds a new state unit C compared to the RNN model. The LSTM prediction model's gate mechanism includes an input gate, a forget gate, and an output gate.
[0153] The input gate is used to filter new information, and its mathematical model is as follows: In this expression, This represents the input to the gate at time t. This indicates that the sigmoid function retains a certain value after passing through the input gate. .
[0154] Forget gates are used to determine discarded information, and their mathematical model is as follows: In this expression, It is the threshold of the Forgetting Gate; what is retained after passing through the Forgetting Gate is... .
[0155] The output gate is used to determine the final output and the information retained. Its mathematical model is as follows: In this expression, This represents the state C after passing through the input gate and the forget gate. This indicates that the output layer's state O at time t is ultimately retained in the hidden layer. (i.e., state s).
[0156] Correspondingly, in this embodiment, the sample feature corresponding to each current time period is the amount of other messages in the corresponding time period of each day within a preset number of days prior to the current time period. For example, if the current time period is 07:00-07:20 on July 20th, then the sample feature corresponding to this current time period is: the amount of other messages in the 07:00-07:20 time period of each day within the preset number of days prior to July 20th. The sample label for this current time period is the actual amount of other messages in the 07:00-07:20 time period of July 20th. Training the LSTM prediction model using the above sample features and sample labels can effectively increase the accuracy of the prediction results.
[0157] The above embodiments describe a message push method based on a neural network from the perspective of the method flow. The following embodiments describe a message push device based on a neural network from the perspective of a virtual module or virtual unit. For details, please refer to the following embodiments.
[0158] This application provides a message push device based on a neural network, such as... Figure 5 As shown, the device includes a receiving module 51, a prediction module 52, a feedback module 53, and a push module 54, wherein...
[0159] The receiving module 51 is used to obtain the message to be pushed, which includes the message content, message volume and target push time period;
[0160] Prediction module 52 is used to predict the amount of other messages within the target push period based on the LSTM prediction model;
[0161] Feedback module 53 is used to determine and provide feedback on push notifications based on message volume and other message volumes. The push notifications are used to indicate whether message content can be pushed within the target push time period.
[0162] The push module 54 is used to push message content within the target push time period when the push notification indicates that the message content should be pushed within the target push time period.
[0163] In another possible implementation of this embodiment, the feedback module 53 is specifically used for:
[0164] Sum the message volume and other message volumes to get the total message volume;
[0165] If the total message volume is less than the preset message volume, then the push notification is determined to indicate the message content to be pushed within the target push time period; the preset message volume is the maximum message volume that can be handled within the target push time period.
[0166] If the total number of messages is not less than the preset number of messages, then the push notification is determined to indicate that no message content can be pushed during the target push time period.
[0167] In another possible implementation of this application embodiment, the apparatus further includes an update push module, which is specifically used to execute an update push process, the update push process including:
[0168] Get the update push time period;
[0169] Based on the LSTM prediction model, predict the amount of new and other messages within the update push period;
[0170] Based on the message volume and other new message volumes, determine and provide updated push notifications;
[0171] If the updated push notification is used to indicate that message content will be pushed during the update push period, then the message content will be pushed during the update push period.
[0172] In another possible implementation of this application embodiment, the update push module is further used for:
[0173] If the updated push notification indicates that the message content cannot be pushed within the update push period, the update push process is repeated until the updated push notification indicates that the message content can be pushed within the update push period.
[0174] In another possible implementation of this application embodiment, the apparatus further includes a sending module, an acquiring module, and a determining module, wherein...
[0175] The acquisition module is used to obtain the appointment time slot returned by the client.
[0176] The determination module is used to determine the target push time period based on the scheduled time period, and to use the time period other than the target push time period in the scheduled time period as the update push time period.
[0177] In another possible implementation of this application embodiment, the prediction module 52 is specifically used for:
[0178] Get the historical message volume of a preset number of time periods before the target push time period;
[0179] Input the historical message volume within a preset number of time periods into the trained LSTM prediction model to obtain the message volume within the target push time period.
[0180] In another possible implementation of this application embodiment, the prediction module 52 is further used for:
[0181] Before inputting the historical other messages within a preset number of time periods into the trained LSTM prediction model, the historical other messages within the preset number of time periods are preprocessed. The preprocessing includes data cleaning, one-hot encoding, and data normalization.
[0182] Another possible implementation of this application embodiment, the training process of the LSTM prediction model includes:
[0183] Training sample data is constructed based on historical sample data; the training sample data includes the input sequence of other message volumes arranged in chronological order and the target output sequence of other message volumes arranged in chronological order.
[0184] The training sample data is input into the LSTM prediction model for training, resulting in a trained LSTM model.
[0185] In another possible implementation of this application embodiment, training sample data is input into the LSTM prediction model for training to obtain the trained LSTM model, including:
[0186] The amount of other messages in the input sequence that precedes the current time period is used as the sample features of the current time period, and the amount of other messages in the target output sequence that precedes the current time period is used as the sample label of the current time period.
[0187] Obtain the predicted values output by the LSTM prediction model based on sample features;
[0188] Determine the loss function based on the predicted value and sample label for the current time period;
[0189] Based on the loss function, adjust the LSTM prediction model until the loss function converges to obtain the trained LSTM prediction model.
[0190] This application provides a message push device based on a neural network, which is applicable to the above method embodiments and will not be described again here.
[0191] This application provides an electronic device, such as... Figure 6 As shown, Figure 6 The illustrated electronic device includes a processor 61 and a memory 62. The processor 61 and the memory 62 are connected, for example, via a bus 63. Optionally, the electronic device may also include a transceiver 64. It should be noted that in practical applications, the transceiver 64 is not limited to one type, and the structure of this electronic device does not constitute a limitation on the embodiments of this application.
[0192] Processor 61 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), a FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 61 may also be a combination that implements computational functions, such as a combination of one or more microprocessors 61, a combination of a DSP and a microprocessor 61, etc.
[0193] Bus 63 may include a pathway for transmitting information between the aforementioned components. Bus 631002 may be a PCI (Peripheral Component Interconnect) bus 63 or an EISA (Extended Industry Standard Architecture) bus 63, etc. Bus 63 can be divided into address bus 63, data bus 63, control bus 63, etc. For ease of representation, Figure 3 The bus 63 is represented by only one thick line, but this does not mean that there is only one bus 63 or one type of bus 63.
[0194] The memory 62 may be a ROM (Read Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.
[0195] The memory 62 is used to store application code that executes the solution of this application, and its execution is controlled by the processor 61. The processor 61 is used to execute the application code stored in the memory 62 to implement the content shown in the foregoing method embodiments.
[0196] Electronic devices include, but are not limited to: mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (such as in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Servers can also be included. Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.
[0197] This application provides a computer-readable storage medium storing a computer program that, when run on a computer, enables the computer to execute the corresponding content in the aforementioned method embodiments.
[0198] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and / or data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.
[0199] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.
[0200] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.
Claims
1. A message push method based on neural networks, characterized in that, include: Obtain the message to be pushed, which includes message content, message volume, and target push time period; Based on the Long Short-Term Memory (LSTM) prediction model, the amount of other messages during the target push period is predicted. The amount of other messages includes chat messages, push notifications from accessed applications, and tweets from information accounts. Based on the message volume and the other message volumes, a push notification is determined and fed back. The push notification is used to indicate whether the message content can be pushed within the target push time period. If the push notification is used to instruct the message content to be pushed within the target push time period, then the message content is pushed within the target push time period; Before pushing the message content within the target push time period, the method further includes: Obtain the appointment time period from the client; the appointment time period is the time period during which the client user can receive push messages, set by the client user based on their own wishes; The scheduled time period is divided into N time periods according to a preset rule, wherein the start time of the first time period is the same as the start time of the scheduled time period, and the end time of the Nth time period is the same as the end time of the scheduled time period. Select the first time period from the N time periods as the target push time period, and use the time periods other than the target push time period from the N time periods as update push time periods; If the push notification is used to indicate that the message content cannot be pushed within the target push time period, obtain the update push time period; Based on the message volume and the volume of other new messages during the update push period, determine and provide feedback on the updated push notification; If the updated push notification is used to instruct the message content to be pushed during the update push period, then the message content will be pushed during the update push period.
2. The method according to claim 1, characterized in that, The step of determining and responding to push notifications based on the message volume and the other message volumes includes: The total message volume is obtained by summing the message volume and the other message volumes. If the total message volume is less than the preset message volume, then the push notification is determined to be used to instruct the push of the message content within the target push time period; the preset message volume is the maximum message volume that can be handled within the target push time period. If the total number of messages is not less than the preset number of messages, then the push notification is determined to indicate that the message content cannot be pushed during the target push time period.
3. The method according to claim 1, characterized in that, If the push notification is used to indicate that the message content cannot be pushed within the target push time period, the method further includes: predicting the amount of new other messages within the update push time period based on the LSTM prediction model.
4. The method according to claim 3, characterized in that, If the updated push notification indicates that the message content cannot be pushed within the update push period, the update push process is repeated until the updated push notification indicates that the message content can be pushed within the update push period.
5. The method according to any one of claims 1-4, characterized in that, The prediction of other message volume within the target push time period based on the LSTM prediction model includes: Obtain the volume of other historical messages within a preset number of time periods prior to the target push time period; The amount of other historical messages within the preset number of time periods is input into the trained LSTM prediction model to obtain the amount of other messages within the target push time period.
6. The method according to claim 5, characterized in that, Before inputting the amount of other historical messages within the preset number of time periods into the trained LSTM prediction model, the amount of other historical messages within the preset number of time periods is preprocessed, including data cleaning, one-hot encoding, and data normalization.
7. The method according to claim 5, characterized in that, The training process of the LSTM prediction model includes: Training sample data is constructed based on historical sample data; the training sample data includes an input set of other message volumes arranged in chronological order and a target output set of other message volumes arranged in chronological order. The initial LSTM prediction model is trained based on the training sample data to obtain the trained LSTM model.
8. The method according to claim 7, characterized in that, The step of training the initial LSTM prediction model based on the training sample data to obtain the trained LSTM model includes: The amount of other messages in the input set before the current time period is taken as the sample features of the current time period, and the amount of other messages in the target output set during the current time period is taken as the sample labels of the current time period. Obtain the predicted value output by the initial LSTM prediction model based on the sample features; Based on the predicted value of the current time period and the sample label of the current time period, determine the loss function; Based on the loss function, the initial LSTM prediction model is adjusted until the loss function converges, resulting in a trained LSTM prediction model.
9. A message push device based on a neural network, characterized in that, include: The receiving module is used to acquire the message to be pushed, which includes message content, message volume and target push time period; The prediction module is used to predict the volume of other messages within the target push time period based on the LSTM prediction model. The volume of other messages includes chat messages, push notifications from access applications, and tweets from information accounts. The feedback module is used to determine and provide a push notification based on the message volume and the other message volumes. The push notification is used to indicate whether the message content can be pushed within the target push time period. The push module is used to push the message content within the target push time period when the push notification indicates that the message content should be pushed within the target push time period. The device further includes: The acquisition module is used to acquire the scheduled time period fed back by the client before pushing the message content within the target push time period; the scheduled time period is the time period that the client user can receive push messages based on their own wishes; The determination module is used to divide the reservation time period into N time periods according to a preset rule, wherein the start time of the first time period is the same as the start time of the reservation time period, and the end time of the Nth time period is the same as the end time of the reservation time period. Select the first time period from the N time periods as the target push time period, and use the time periods other than the target push time period from the N time periods as update push time periods; If the push notification is used to indicate that the message content cannot be pushed within the target push time period, obtain the update push time period; The feedback module is also used to determine and provide feedback on the updated push notification based on the message volume and the volume of other new messages during the update push period. The push module is also configured to push the message content within the update push time period when the updated push notification indicates that the message content should be pushed within the update push time period.
10. An electronic device, characterized in that, It includes: One or more processors; Memory; One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, the one or more applications being configured to: execute the neural network-based push notification method according to any one of claims 1-8.
11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the neural network-based message push method as described in any one of claims 1-8.