Platform message pushing method and device, equipment and storage medium

Abstract

The invention relates to the technical field of data processing, in particular to a platform message pushing method and device, equipment and a storage medium, and the method comprises the steps: obtaining historical interaction information, and preprocessing the historical interaction information to obtain preprocessed information; a preference prediction model and an adjustment model are constructed based on the preprocessing information, the preference prediction model is used for obtaining prediction preference information in a certain time period, and the adjustment model is used for integrating and adjusting the generated push elements; when order information related to a user is received, recent historical interaction information is obtained and preprocessed, and recent preprocessed information is obtained; inputting the recent preprocessing information into a preference prediction model to obtain recent preference information; based on the recent preference information, adopting a pre-constructed element generation model to generate push elements; inputting the generated push elements into an adjustment model to obtain a push strategy; according to the method disclosed by the invention, highly personalized information pushing can be realized, and the user experience and the effective reach rate of the message are improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a platform message pushing method, device, equipment and storage medium. Background Art

[0002] For express delivery companies, achieving efficient and accurate delivery of logistics information is a key strategy to enhance customer experience and strengthen user loyalty.

[0003] Traditionally, companies rely on SMS notifications to report the logistics status of packages to users, but this method has gradually shown its limitations in terms of cost control, content customization, and interactivity. Specifically, SMS notifications are expensive, especially for scenarios that require large-scale sending. In addition, SMS content is single in form, making it difficult to carry complex information, and it is impossible to interact with users in real time, making it difficult to meet users' growing personalized needs.

[0004] Compared with traditional SMS notifications, WeChat public account template messages stand out with their unique advantages. First, the sending cost of template messages is relatively low, which helps express delivery companies to effectively control operating costs. Second, its high arrival rate ensures the effective transmission of information and reduces the risk of information omission. More importantly, template messages support multiple media forms such as pictures, texts, and videos, making logistics information more intuitive and vivid. At the same time, it also provides express delivery companies with a broader creative space for customized push content and enhancing the attractiveness and interactivity of information.

[0005] Different users have different needs and preferences for express delivery arrival information. However, the existing message push process is difficult to identify user behavior and preferences, and cannot achieve highly personalized push, resulting in users being unable to obtain the most relevant information, reducing the user experience.

[0006] It can be seen that the existing technology still needs to be improved and enhanced. Summary of the invention

[0007] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a platform message push method that can achieve highly personalized information push, improve user experience and the effective reach of messages.

[0008] The first aspect of the present invention provides a platform message push method, including: obtaining historical interaction information, preprocessing the historical interaction information to obtain preprocessed information; constructing a preference prediction model based on the preprocessed information, the preference prediction model is used to obtain predicted preference information within a certain time period; constructing an adjustment model based on the preprocessed information, the adjustment model is used to integrate and adjust the generated push elements; when order information related to the user is received, extracting recent historical interaction information based on preset extraction rules, and preprocessing the recent historical interactions to obtain recent preprocessed information; inputting the recent preprocessed information into the preference prediction model to obtain recent preference information; based on the recent preference information, using a pre-constructed element generation model to generate a push element; inputting the generated push element into the adjustment model to obtain a push strategy.

[0009] Optionally, in a first implementation method of the first aspect of the present invention, the obtaining of historical interaction information, preprocessing of the historical interaction information, and obtaining preprocessed information include: obtaining historical interaction information, performing data cleaning processing on the historical interaction information to remove invalid information, missing values, or outliers in the historical interaction information, and obtaining cleaned interaction information; using an autoencoder to extract key feature information from the cleaned interaction information; performing time series analysis and push effect analysis based on the key feature information to obtain time analysis data and effect analysis data, and integrating the time analysis data and the effect analysis data to obtain preprocessed information.

[0010] Optionally, in a second implementation method of the first aspect of the present invention, the time series analysis and push effect analysis are performed based on key feature information to obtain time analysis data and effect analysis data, including: the key feature information includes time series data and interaction information; the least squares method is used to capture the long-term trend in the time series data to obtain the long-term trend component; the long-term trend information in the time series data is removed, and the time series of seasonal decomposition is used to identify the seasonal pattern in the key feature information to obtain the seasonal component; the seasonal component in the time series data is removed to obtain the random component; the long-term trend component, the seasonal component and the random component are integrated to obtain the time analysis data; according to the interaction information, a pre-built reward function is used to confirm the push effect to obtain the effect analysis data.

[0011] Optionally, in a third implementation of the first aspect of the present invention, the preference prediction model is constructed based on preprocessing information, and the preference prediction model is used to obtain predicted preference information within a certain time period, including: dividing the collected preprocessing information into a first training set, a first validation set and a first test set according to a preset division ratio; inputting the first training set and the first validation set into a first model constructed based on a recurrent neural network, training the first model, and evaluating the performance of the first model during the training process based on a cross-validation method; inputting the first test set into the trained first model, and according to the test results of the first model, using a Bayesian optimization method to adjust the hyperparameters of the first model to obtain a preference prediction model.

[0012] Optionally, in a fourth implementation method of the first aspect of the present invention, the adjustment model is constructed based on the preprocessing information, and the adjustment model is used to integrate and adjust the generated push elements, including: dividing the collected preprocessing information into a second training set, a second verification set, and a second test set according to a preset division ratio; inputting the first training set and the first verification set into the second model constructed based on the deep Q network, respectively, and setting the action space of the second model; training the second model, and evaluating the performance of the second model during the training process based on the cross-validation method; inputting the second test set into the trained second model, and according to the test results of the second model, using the Bayesian optimization method to adjust the hyperparameters of the second model to obtain the adjustment model.

[0013] Optionally, in a fifth implementation of the first aspect of the present invention, when order information related to the user is received, recent historical interaction information is extracted based on preset extraction rules, and the recent historical interactions are pre-processed to obtain recent pre-processed information, including: when order information related to the user is received, historical interaction information of the past month is obtained as recent historical interaction information; data cleaning processing and feature extraction processing are performed on the recent historical interaction information in sequence to obtain recent key feature information; time series analysis and push effect analysis are performed based on the recent key feature information to obtain recent time analysis data and recent effect analysis data; and recent time analysis data and recent effect analysis data are integrated to obtain recent pre-processed information.

[0014] Optionally, in a sixth implementation method of the first aspect of the present invention, the push element is generated based on the recent preference information using a pre-built element generation model, including: pre-building an element generation model based on a generative adversarial network, the element generation model including a generator and a discriminator; using natural language processing technology to perform keyword recognition on the recent preference information to obtain key information; inputting the key information and the order information into the generator to obtain push content information; using the discriminator to judge the authenticity of the push content information, and outputting the push content information that passes the authenticity judgment as a push element.

[0015] The second aspect of the present invention provides a platform message push device, including: a first processing module, used to obtain historical interaction information, pre-process the historical interaction information, and obtain pre-processed information; a first construction module, used to construct a preference prediction model based on the pre-processed information, and the preference prediction model is used to obtain predicted preference information within a certain time period; a second construction module, used to construct an adjustment model based on the pre-processed information, and the adjustment model is used to integrate and adjust the generated push elements; a second processing module, used to extract recent historical interaction information based on preset extraction rules when receiving order information related to the user, and pre-process the recent historical interactions to obtain recent pre-processed information; a first input module, used to input the recent pre-processed information into the preference prediction model to obtain recent preference information; a generation module, used to generate a push element based on the recent preference information and using a pre-constructed element generation model; a second input module, used to input the generated push element into the adjustment model to obtain a push strategy.

[0016] Optionally, in a first implementation method of the second aspect of the present invention, the first processing module includes: a first acquisition unit, used to acquire historical interaction information, and perform data cleaning on the historical interaction information to remove invalid information, missing values ​​or abnormal values ​​in the historical interaction information to obtain cleaned interaction information; a first extraction unit, used to use an automatic encoder to extract key feature information from the cleaned interaction information; a first analysis unit, used to perform time series analysis and push effect analysis based on the key feature information to obtain time analysis data and effect analysis data, and integrate the time analysis data and effect analysis data to obtain preprocessing information.

[0017] Optionally, in a second implementation of the second aspect of the present invention, the first processing module also includes: a capture unit, used to capture long-term trends in time series data using the least squares method to obtain long-term trend components; a first identification unit, used to remove long-term trend information in time series data, and use seasonally decomposed time series to identify seasonal patterns in key feature information to obtain seasonal components; a removal unit, used to remove seasonal components in time series data to obtain random components; a first integration unit, used to integrate long-term trend components, seasonal components and random components to obtain time analysis data; a confirmation unit, used to confirm the push effect based on the interaction information using a pre-built reward function to obtain effect analysis data.

[0018] Optionally, in a third implementation of the second aspect of the present invention, the first construction module includes: a first division unit, used to divide the collected preprocessing information into a first training set, a first verification set and a first test set according to a preset division ratio; a first training unit, used to input the first training set and the first verification set into a first model constructed based on a recurrent neural network, train the first model, and evaluate the performance of the first model during the training process based on a cross-validation method; a first testing unit, used to input the first test set into the trained first model, and according to the test results of the first model, use a Bayesian optimization method to adjust the hyperparameters of the first model to obtain a preference prediction model.

[0019] Optionally, in a fourth implementation of the second aspect of the present invention, the second construction module includes: a second division unit, used to divide the collected preprocessing information into a second training set, a second verification set, and a second test set according to a preset division ratio; a first input unit, used to input the first training set and the first verification set into the second model constructed based on the deep Q network, respectively, and set the action space of the second model; a second training unit, used to train the second model, and evaluate the performance of the second model during the training process based on a cross-validation method; a second testing unit, used to input the second test set into the trained second model, and according to the test results of the second model, use a Bayesian optimization method to adjust the hyperparameters of the second model to obtain an adjusted model.

[0020] Optionally, in a fifth implementation of the second aspect of the present invention, the second processing module includes: a second acquisition unit, used to obtain historical interaction information of the past month as recent historical interaction information when order information related to the user is received; a second extraction unit, used to perform data cleaning processing and feature extraction processing on the recent historical interaction information in sequence to obtain recent key feature information; a second analysis unit, used to perform time series analysis and push effect analysis based on the recent key feature information to obtain recent time analysis data and recent effect analysis data; a second integration unit, used to integrate recent time analysis data and recent effect analysis data to obtain recent preprocessing information.

[0021] Optionally, in a sixth implementation of the second aspect of the present invention, the generation module includes: a construction unit, used to pre-construct an element generation model based on a generative adversarial network, the element generation model including a generator and a discriminator; a second identification unit, used to use natural language processing technology to perform keyword recognition on recent preference information to obtain key information; a second input unit, used to input the key information and the order information into the generator to obtain push content information; a discriminator, used to use the discriminator to judge the authenticity of the push content information, and output the push content information that passes the authenticity judgment as a push element.

[0022] The third aspect of the present invention provides a platform message push device, which includes: a memory and at least one processor, wherein instructions are stored in the memory; at least one of the processors calls the instructions in the memory so that the platform message push device executes each step of the platform message push method described in any one of the above items.

[0023] A fourth aspect of the present invention provides a computer-readable storage medium, on which instructions are stored, and when the instructions are executed by a processor, each step of any of the above-mentioned platform message push methods is implemented.

[0024] In the technical solution of the present invention, preprocessing information is obtained by obtaining historical interaction information and performing preprocessing; a preference prediction model and an adjustment model are constructed based on the preprocessing information, the preference prediction model is used to obtain predicted preference information within a certain time period, and the adjustment model is used to integrate and adjust the generated push elements; when order information related to the user is received, recent historical interaction information is obtained and preprocessed to obtain recent preprocessing information; the recent preprocessing information is input into the preference prediction model to obtain recent preference information; based on the recent preference information, a push element is generated using a pre-constructed element generation model; the generated push element is input into the adjustment model to obtain a push strategy; the method disclosed in the present application can realize highly personalized information push, improve user experience and the effective reach rate of messages. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A first flow chart of a data processing method provided by an embodiment of the present invention;

[0026] Figure 2 A second flow chart of the data processing method provided by an embodiment of the present invention;

[0027] Figure 3 A third flow chart of the data processing method provided by an embodiment of the present invention;

[0028] Figure 4 A fourth flow chart of the data processing method provided by an embodiment of the present invention;

[0029] Figure 5 A fifth flow chart of the data processing method provided by the embodiment of the present invention;

[0030] Figure 6 A sixth flow chart of the data processing method provided by an embodiment of the present invention;

[0031] Figure 7 A seventh flow chart of the data processing method provided by the embodiment of the present invention;

[0032] Figure 8 A schematic diagram of the structure of a data processing device provided by an embodiment of the present invention;

[0033] Fig. 9 A schematic diagram of the structure of a data processing device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The present invention provides a platform message push method, apparatus, device and storage medium. In the present invention, the terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0035] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of the platform message push method in the embodiment of the present invention includes:

[0036] 101. Obtain historical interaction information, and pre-process the historical interaction information to obtain pre-processed information;

[0037] 102. Constructing a preference prediction model based on the preprocessing information, wherein the preference prediction model is used to obtain predicted preference information within a certain time period;

[0038] 103. Building an adjustment model based on the preprocessing information, wherein the adjustment model is used to integrate and adjust the generated push elements to customize a more accurate and effective push strategy, thereby improving satisfaction and participation;

[0039] 104. When order information related to the user is received, recent historical interaction information is extracted based on a preset extraction rule, and recent historical interactions are preprocessed to obtain recent preprocessed information;

[0040] In this embodiment, when the order information is received and then the recent historical interaction information is extracted, the timeliness and relevance of the recent preference information generated subsequently can be ensured.

[0041] 105. Input the recent preprocessing information into the preference prediction model to obtain recent preference information;

[0042] 106. Generate push elements using a pre-built element generation model based on recent preference information;

[0043] 107. Input the generated push elements into the adjustment model to obtain the push strategy.

[0044] The present application discloses a platform message push method, which obtains preprocessed information by acquiring historical interaction information and performing preprocessing; constructs a preference prediction model and an adjustment model based on the preprocessed information, wherein the preference prediction model is used to obtain predicted preference information within a certain period of time, and the adjustment model is used to integrate and adjust the generated push elements; when order information related to a user is received, recent historical interaction information is acquired and preprocessed to obtain recent preprocessed information; the recent preprocessed information is input into the preference prediction model to obtain recent preference information; based on the recent preference information, a push element is generated using a pre-constructed element generation model; the generated push element is input into the adjustment model to obtain a push strategy; the method disclosed in the present application can realize highly personalized information push, effectively improve user experience and the effective reach rate of messages, enhance the interactivity and stickiness between users and logistics companies, and promote the growth of logistics business.

[0045] See also Figure 2 , a second embodiment of the platform message push method in an embodiment of the present invention includes:

[0046] 201. Obtain historical interaction information, and perform data cleaning on the historical interaction information to remove invalid information, missing values, or abnormal values ​​in the historical interaction information to obtain cleaned interaction information;

[0047] In this embodiment, a regular expression is used to remove invalid information and missing values ​​in the historical interaction information, and a statistical analysis algorithm is used to process abnormal values ​​in the historical interaction information.

[0048] 202. Use an automatic encoder to extract key feature information from the cleaned interaction information;

[0049] In this embodiment, an autoencoder is used to extract key feature information. The autoencoder is often used in feature learning tasks. The autoencoder is a neural network that compresses input data into a smaller representation through learning, and then reconstructs the original data from the representation. The autoencoder includes an encoder and a decoder. The encoder is responsible for compressing the input data into a low-dimensional representation, while the decoder restores the low-dimensional representation to the original data. The mean square error loss function is selected as the loss function of the autoencoder.

[0050] 203. Perform time series analysis and push effect analysis based on key feature information to obtain time analysis data and effect analysis data, and integrate the time analysis data and effect analysis data to obtain preprocessing information.

[0051] See also Figure 3 , a third embodiment of the platform message push method in the embodiment of the present invention includes:

[0052] The key feature information includes time series data and interaction information;

[0053] 301. Use the least squares method to capture the long-term trend in time series data and obtain the long-term trend component;

[0054] In this embodiment, the least square method obtains the long-term trend from the time series data by fitting a straight line or a curve.

[0055] 302. Remove the long-term trend information in the time series data, use the seasonal decomposition of the time series to identify the seasonal pattern in the key feature information, and obtain the seasonal component;

[0056] 303. Remove the seasonal component from the time series data to obtain the random component;

[0057] In this embodiment, after removing the long-term trend component and the seasonal component from the time series data, the random component is separated and obtained. The random component includes irregular fluctuations and may be caused by sudden events or noise.

[0058] In this embodiment, by decomposing the time series data into three parts: trend, seasonality and residual, the time series data can be comprehensively analyzed, and the obtained time analysis data can improve the accuracy and effectiveness of subsequent model training.

[0059] 304. Integrate long-term trend components, seasonal components and random components to obtain time analysis data;

[0060] 305. According to the interaction information, a pre-built reward function is used to confirm the push effect and obtain effect analysis data;

[0061] In this embodiment, the interaction information includes click-through rate, reading time and conversion rate. The reward function can quantify the effect of user interaction with the pushed content, that is, the reward value of each push can be calculated to obtain effect analysis data; through time analysis data and effect analysis data, the user's browsing time preference and browsing content preference can be confirmed.

[0062] See also Figure 4 The fourth embodiment of the platform message push method in the embodiment of the present invention includes:

[0063] 401. Divide the collected preprocessing information into a first training set, a first validation set, and a first test set according to a preset division ratio;

[0064] In this embodiment, the predicted division ratio can be training set: validation set: test set

[0065] =6:2:2.

[0066] 402. Input the first training set and the first validation set into a first model constructed based on a recurrent neural network, train the first model, and evaluate the performance of the first model during the training process based on a cross-validation method;

[0067] In this embodiment, the performance of the first model in the training process is evaluated based on the cross-validation method. Cross-validation usually involves dividing the data set into k subsets, taking one of the subsets as a validation set and the rest as training sets in turn, so as to evaluate the generalization ability of the model. Through the training and cross-validation of the recurrent neural network, it can be ensured that the model has a good fit on the first training set and has good generalization ability.

[0068] 403. Input the first test set into the first model that has completed training, and adjust the hyperparameters of the first model using a Bayesian optimization method according to the test result of the first model to obtain a preference prediction model;

[0069] In this embodiment, the Bayesian optimization method is used to adjust the hyperparameters of the first model. Bayesian optimization is a global optimization algorithm that predicts the location of the optimal hyperparameters by establishing a probability model, and uses this model to guide the search process to find the hyperparameter combination that optimizes the model performance. Compared with traditional grid search or random search, Bayesian optimization can converge to the optimal solution more quickly, thereby reducing unnecessary consumption of computing resources. By optimizing the hyperparameters, the preference prediction model can be better adapted to the preference prediction of different types of user groups, thereby improving the flexibility and applicability of the preference prediction model.

[0070] See also Figure 5 The fifth embodiment of the platform message push method in the embodiment of the present invention includes:

[0071] 501. Divide the collected preprocessed information into a second training set, a second verification set, and a second test set according to a preset division ratio;

[0072] 502. Input the first training set and the first validation set into the second model constructed based on the deep Q network respectively, and set the action space of the second model;

[0073] In this embodiment, the action space refers to actions that can be performed by adjusting the model, and the actions include changing the push time, adjusting the push frequency, selecting different types of push content, etc.

[0074] In this embodiment, a second model is constructed based on a deep Q network, which is an algorithm that combines deep learning and reinforcement learning. It solves reinforcement learning problems with high-dimensional state spaces by using a deep neural network to approximate the Q function. DQN uses a convolutional neural network to process image inputs and stabilizes the learning process through experience replay and target network techniques. The experience replay mechanism allows the network to randomly extract samples from past experiences to learn, breaking the temporal correlation between data and improving learning efficiency.

[0075] 503. Train the second model, and evaluate the performance of the second model during the training process based on a cross-validation method;

[0076] 504. Input the second test set into the trained second model, and adjust the hyperparameters of the second model using a Bayesian optimization method according to the test result of the second model to obtain an adjusted model;

[0077] In this embodiment, the Bayesian optimization method is used to adjust the hyperparameters of the second model; Bayesian optimization is a global optimization algorithm that predicts the position of the optimal hyperparameters by establishing a probability model, and uses this model to guide the search process to find the hyperparameter combination that optimizes the model performance; compared with traditional grid search or random search, Bayesian optimization can converge to the optimal solution more quickly, thereby reducing unnecessary consumption of computing resources; by optimizing hyperparameters, the adjustment model can be better adapted to adjustment tasks at different times, thereby improving the flexibility and applicability of the adjustment model.

[0078] See also Figure 6 The sixth embodiment of the platform message push method in the embodiment of the present invention includes:

[0079] 601. When order information related to the user is received, historical interaction information of the past month is obtained as recent historical interaction information;

[0080] In this embodiment, for example, when the time of receiving the order information related to the user is August 15, the historical interaction information from July 15 to August 15 is obtained as the recent historical interaction information.

[0081] 602. Perform data cleaning and feature extraction on the recent historical interaction information in sequence to obtain recent key feature information;

[0082] In this embodiment, regular expressions are used to remove invalid information and missing values ​​in recent historical interaction information, and a statistical analysis algorithm is used to process outliers in recent historical interaction information to perform data cleaning on the recent historical interaction information; further, an automatic encoder is used to extract recent key feature information from the cleaned recent interaction information.

[0083] 603. Perform time series analysis and push effect analysis based on recent key feature information to obtain recent time analysis data and recent effect analysis data;

[0084] 604. Integrate the recent time analysis data and the recent effect analysis data to obtain recent preprocessing information. Figure 7 The seventh embodiment of the platform message push method in the embodiment of the present invention includes:

[0085] 701. Pre-constructing an element generation model based on a generative adversarial network, wherein the element generation model includes a generator and a discriminator;

[0086] 702. Use natural language processing technology to identify keywords in recent preference information to obtain key information;

[0087] 703. Input the key information and the order information into the generator to obtain push content information;

[0088] 704. Using the discriminator to judge the authenticity of the pushed content information, and outputting the pushed content information that passes the authenticity judgment as a push element;

[0089] In this embodiment, by generating push elements through the adversarial network pre-constructed element generation model, it can be ensured that the pushed content is in line with user preferences and authentic, and that the pushed elements are both personalized and reliable, thereby improving user experience and satisfaction.

[0090] The platform message push method in the embodiment of the present invention is described above. The platform message push device in the embodiment of the present invention is described below. Figure 8 In one embodiment of the present invention, a platform message push device includes:

[0091] The first processing module 801 is used to obtain historical interaction information, pre-process the historical interaction information, and obtain pre-processed information; the first construction module 802 is used to construct a preference prediction model based on the pre-processed information, and the preference prediction model is used to obtain predicted preference information within a certain time period; the second construction module 803 is used to construct an adjustment model based on the pre-processed information, and the adjustment model is used to integrate and adjust the generated push elements; the second processing module 804 is used to extract recent historical interaction information based on preset extraction rules when receiving order information related to the user, and pre-process the recent historical interactions to obtain recent pre-processed information; the first input module 805 is used to input the recent pre-processed information into the preference prediction model to obtain recent preference information; the generation module 806 is used to generate push elements based on the recent preference information using a pre-constructed element generation model; the second input module 807 is used to input the generated push elements into the adjustment model to obtain a push strategy.

[0092] In this embodiment, the first processing module 801 includes: a first acquisition unit 8011, which is used to acquire historical interaction information, perform data cleaning on the historical interaction information to remove invalid information, missing values ​​or abnormal values ​​in the historical interaction information, and obtain cleaned interaction information; a first extraction unit 8012, which is used to use an automatic encoder to extract key feature information from the cleaned interaction information; a first analysis unit 8013, which is used to perform time series analysis and push effect analysis based on the key feature information to obtain time analysis data and effect analysis data, and integrate the time analysis data and the effect analysis data to obtain preprocessing information.

[0093] In this embodiment, the first processing module 801 also includes: a capture unit 8014, which is used to capture the long-term trend in the time series data using the least squares method to obtain the long-term trend component; a first identification unit 8015, which is used to remove the long-term trend information in the time series data, and use the seasonally decomposed time series to identify the seasonal pattern in the key feature information to obtain the seasonal component; a removal unit 8016, which is used to remove the seasonal component in the time series data to obtain the random component; a first integration unit 8017, which is used to integrate the long-term trend component, the seasonal component and the random component to obtain the time analysis data; a confirmation unit 8018, which is used to confirm the push effect according to the interaction information using a pre-built reward function to obtain the effect analysis data.

[0094] In this embodiment, the first construction module 802 includes: a first division unit 8021, which is used to divide the collected preprocessing information into a first training set, a first validation set and a first test set according to a preset division ratio; a first training unit 8022, which is used to input the first training set and the first validation set into the first model constructed based on the recurrent neural network, train the first model, and evaluate the performance of the first model during the training process based on the cross-validation method; a first testing unit 8023, which is used to input the first test set into the first model that has completed the training, and according to the test result of the first model, use the Bayesian optimization method to adjust the hyperparameters of the first model to obtain a preference prediction model.

[0095] In this embodiment, the second construction module 803 includes: a second division unit 8031, which is used to divide the collected preprocessing information into a second training set, a second verification set, and a second test set according to a preset division ratio; a first input unit 8032, which is used to input the first training set and the first verification set into the second model constructed based on the deep Q network, and set the action space of the second model; a second training unit 8033, which is used to train the second model and evaluate the performance of the second model during the training process based on the cross-validation method; a second testing unit 8034, which is used to input the second test set into the trained second model, and according to the test results of the second model, use the Bayesian optimization method to adjust the hyperparameters of the second model to obtain the adjusted model.

[0096] In this embodiment, the second processing module 804 includes: a second acquisition unit 8041, which is used to obtain historical interaction information of the past month as recent historical interaction information when order information related to the user is received; a second extraction unit, which is used to perform data cleaning processing and feature extraction processing on the recent historical interaction information in sequence to obtain recent key feature information; a second analysis unit 8042, which is used to perform time series analysis and push effect analysis based on the recent key feature information to obtain recent time analysis data and recent effect analysis data; a second integration unit 8043, which is used to integrate the recent time analysis data and the recent effect analysis data to obtain recent preprocessing information.

[0097] In this embodiment, the generation module 806 includes: a construction unit 8061, which is used to pre-construct an element generation model based on a generative adversarial network, and the element generation model includes a generator and a discriminator; a second recognition unit, which is used to use natural language processing technology to perform keyword recognition on recent preference information to obtain key information; a second input unit 8062, which is used to input the key information and the order information into the generator to obtain push content information; a discriminator unit 8063, which is used to use the discriminator to judge the authenticity of the push content information, and output the push content information that passes the authenticity judgment as a push element.

[0098] above Figure 8 The platform message push device in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The platform message push device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0099] Fig. 9 1 is a schematic diagram of the structure of a platform message push device provided by an embodiment of the present invention. The platform message push device 900 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 910 (for example, one or more processors) and a memory 920, and one or more storage media 930 (for example, one or more mass storage devices) storing application programs 933 or data 932. Among them, the memory 920 and the storage medium 930 can be short-term storage or permanent storage. The program stored in the storage medium 930 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the platform message push device 900. Furthermore, the processor 910 can be configured to communicate with the storage medium 930, and execute a series of instruction operations in the storage medium 930 on the platform message push device 900 to implement the steps of the platform message push method provided in the above-mentioned method embodiments.

[0100] The platform message push device 900 may also include one or more power supplies 940, one or more wired or wireless network interfaces 950, one or more input and output interfaces 960, and / or one or more operating systems 931, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Fig. 9 The illustrated structure of the platform message push device does not constitute a limitation on the platform message push device, and may include more or fewer components than illustrated, or a combination of certain components, or a different arrangement of components.

[0101] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the platform message push method.

[0102] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0103] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0104] Finally, it should be noted that the above description is only a preferred example of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A platform message push method, characterized in that: include: Acquire historical interaction information, pre-process the historical interaction information, and obtain pre-processed information; Building a preference prediction model based on the preprocessed information, wherein the preference prediction model is used to obtain predicted preference information within a certain period of time; Building an adjustment model based on the preprocessing information, wherein the adjustment model is used to integrate and adjust the generated push elements; When order information related to the user is received, recent historical interaction information is extracted based on a preset extraction rule, and recent historical interactions are preprocessed to obtain recent preprocessed information; Inputting recent preprocessed information into the preference prediction model to obtain recent preference information; Based on recent preference information, a pre-built element generation model is used to generate push elements; The generated push elements are input into the adjustment model to obtain the push strategy.

2. The platform message push method according to claim 1, characterized in that: The acquiring of historical interaction information and preprocessing of the historical interaction information to obtain preprocessed information includes: Acquire historical interaction information, and perform data cleaning on the historical interaction information to remove invalid information, missing values ​​or abnormal values ​​in the historical interaction information to obtain cleaned interaction information; Use autoencoder to extract key feature information from the cleaned interaction information; Based on the key feature information, time series analysis and push effect analysis are performed to obtain time analysis data and effect analysis data, and the time analysis data and effect analysis data are integrated to obtain preprocessing information.

3. The platform message push method according to claim 2, characterized in that: The time series analysis and push effect analysis based on the key feature information are performed to obtain time analysis data and effect analysis data, including: The key feature information includes time series data and interaction information; The least square method is used to capture the long-term trend in time series data and obtain the long-term trend component; Remove the long-term trend information in the time series data, use the seasonal decomposition of the time series to identify the seasonal pattern in the key feature information, and obtain the seasonal component; Remove the seasonal component from the time series data to obtain the random component; Integrate long-term trend components, seasonal components and random components to obtain time analysis data; Based on the interaction information, a pre-built reward function is used to confirm the push effect and obtain effect analysis data.

4. The platform message push method according to claim 1, characterized in that: The preference prediction model is constructed based on the preprocessing information, and the preference prediction model is used to obtain predicted preference information within a certain period of time, including: According to a preset division ratio, the collected preprocessed information is divided into a first training set, a first validation set and a first test set; Inputting the first training set and the first validation set into a first model constructed based on a recurrent neural network, respectively, training the first model, and evaluating the performance of the first model during the training process based on a cross-validation method; The first test set is input into the first model that has completed training. According to the test result of the first model, the hyperparameters of the first model are adjusted by using the Bayesian optimization method to obtain a preference prediction model.

5. The platform message push method according to claim 1, characterized in that: The step of constructing an adjustment model based on the preprocessing information, wherein the adjustment model is used to integrate and adjust the generated push elements, includes: According to a preset division ratio, the collected preprocessed information is divided into a second training set, a second verification set, and a second test set; Inputting the first training set and the first validation set into the second model constructed based on the deep Q network respectively, and setting the action space of the second model; Training the second model, and evaluating the performance of the second model during the training process based on a cross-validation method; The second test set is input into the trained second model, and according to the test result of the second model, the hyperparameters of the second model are adjusted by using the Bayesian optimization method to obtain the adjusted model.

6. The platform message push method according to claim 1, characterized in that: When order information related to the user is received, recent historical interaction information is extracted based on a preset extraction rule, and recent historical interactions are preprocessed to obtain recent preprocessed information, including: When order information related to the user is received, historical interaction information of the past month is obtained as recent historical interaction information; Perform data cleaning and feature extraction on recent historical interaction information to obtain recent key feature information; Perform time series analysis and push effect analysis based on recent key feature information to obtain recent time analysis data and recent effect analysis data; Integrate recent time analysis data and recent effect analysis data to obtain recent preprocessing information.

7. The platform message push method according to claim 1, characterized in that: The method of generating a push element based on recent preference information by using a pre-built element generation model includes: Pre-constructing an element generation model based on a generative adversarial network, wherein the element generation model includes a generator and a discriminator; Use natural language processing technology to identify keywords in recent preference information and obtain key information; Input the key information and the order information into the generator to obtain push content information; The discriminator is used to judge the authenticity of the pushed content information, and the pushed content information that passes the authenticity judgment is output as a push element.

8. A platform message push device, characterized in that: include: A first processing module is used to obtain historical interaction information, pre-process the historical interaction information, and obtain pre-processed information; A first building module is used to build a preference prediction model based on preprocessing information, wherein the preference prediction model is used to obtain predicted preference information within a certain time period; A second building module is used to build an adjustment model based on the preprocessing information, wherein the adjustment model is used to integrate and adjust the generated push elements; The second processing module is used to extract recent historical interaction information based on a preset extraction rule when receiving order information related to the user, and pre-process the recent historical interaction to obtain recent pre-processed information; A first input module is used to input recent preprocessing information into the preference prediction model to obtain recent preference information; A generation module, used to generate push elements based on recent preference information using a pre-built element generation model; The second input module is used to input the generated push elements into the adjustment model to obtain the push strategy.

9. A platform message push device, characterized in that: The platform message push device comprises: a memory and at least one processor, wherein instructions are stored in the memory; At least one of the processors calls the instructions in the memory to enable the platform message pushing device to execute each step of the platform message pushing method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the steps of the platform message pushing method as described in any one of claims 1 to 7 are implemented.

Images