Early warning short message pushing method, device, equipment, product and medium

Abstract

The invention relates to the technical field of meteorological disaster early warning, and provides an early warning short message pushing method and device, equipment, a product and a medium. The method comprises the following steps: inputting a detail text in meteorological disaster early warning information and a manually extracted first meteorological disaster feature into an early warning short message classification model to obtain a binary classification of whether an early warning short message is pushed or not; a deep learning network in the early warning short message classification model is used for extracting a second meteorological disaster feature from the meteorological disaster early warning information, and a feature fusion module is used for fusing the first meteorological disaster feature and the second meteorological disaster feature. According to the application, the features are respectively extracted in two modes of manual and deep learning networks, and then the features are fused by using the feature fusion module to obtain the fusion features richer than single-category meteorological disaster features, so that whether to push the early warning short message can be determined more accurately based on the fusion features.

Description

Technical Field

[0001] This application relates to the field of meteorological disaster early warning technology, specifically to a method, device, equipment, product, and medium for sending early warning text messages. Background Technology

[0002] Meteorological disaster early warning information has a profound impact on agriculture. As meteorological forecasting technology becomes increasingly advanced, the types of disasters that can be predicted are also increasing. Existing meteorological early warning services can provide warnings for hundreds of different types of meteorological disasters and push them to users via SMS.

[0003] SMS push has advantages such as timeliness and high message delivery, but it also has the disadvantage of being mandatory to receive messages, which may cause interference to users. Although existing SMS push methods can determine whether to push SMS to users based on extracted meteorological disaster characteristics, the determination of whether to push SMS is not accurate. Summary of the Invention

[0004] This application provides a method, apparatus, device, product, and medium for sending early warning SMS messages, in order to solve the technical problem that existing SMS push methods are inaccurate in determining whether to send an SMS message.

[0005] In a first aspect, the early warning SMS push method of this application includes: Extract detailed text from meteorological disaster early warning information; Input the detailed text and the first meteorological disaster characteristics into the early warning SMS classification model to obtain a binary classification of whether an early warning SMS was pushed, as output by the early warning SMS classification model; The first meteorological disaster feature is extracted manually from the meteorological disaster early warning information; The warning SMS classification model is trained based on any classification model, using historical detailed text, historical first meteorological disaster characteristics and their corresponding historical binary classification labels; Each of the classification models includes a deep learning network and a feature fusion module. The deep learning network is used to extract a second meteorological disaster feature from the meteorological disaster early warning information, and the feature fusion module is used to fuse the first meteorological disaster feature and the second meteorological disaster feature.

[0006] In one embodiment, the deep learning network includes a stacked pointer network and a relational graph convolutional neural network, and the extraction of the second meteorological disaster feature from the meteorological disaster early warning information includes: The detailed text is input into the stacked pointer network to obtain the knowledge graph structure information output by the stacked pointer network; The knowledge graph structure information is input into the relational graph convolutional neural network to obtain the second meteorological disaster feature output by the relational graph convolutional neural network.

[0007] In one embodiment, the warning SMS classification model includes multiple SMS classification models. The step of inputting the detailed text and the first meteorological disaster feature into the warning SMS classification model to obtain a binary classification of whether a warning SMS has been pushed, as output by the warning SMS classification model, includes: The severity levels of the meteorological disaster early warning information are classified to obtain multiple severity level categories; For any of the multiple severity level categories, the detailed text of the corresponding meteorological disaster warning information and the first meteorological disaster feature are input into the SMS classification model that is compatible with it among the multiple SMS classification models, and a binary classification of whether to push a warning SMS is obtained from the output of the compatible SMS classification model. The adapted SMS classification model is trained based on any of the above classification models, using historical detailed text, historical first meteorological disaster features, and historical binary classification labels corresponding to historical meteorological disaster warning information of any severity level category.

[0008] In one embodiment, the meteorological disaster early warning information is obtained based on the following method: The first region within the coverage area of ​​the meteorological early warning service that has already issued meteorological disaster early warning information; The second region to which subscribers who receive weather warning services belong; Determine the overlapping area between the first region and the second region; Obtain meteorological disaster early warning information for the overlapping areas.

[0009] In one embodiment, after obtaining the binary classification of whether or not a warning SMS is pushed, as output by the warning SMS classification model, the process includes: If it is determined that a warning SMS message has been sent, check the cache to see if the warning SMS message has been sent. If the warning SMS has been sent, then query the cache to see if the warning SMS has been sent to every subscriber in the overlapping area. If there are target subscribers who have not received the warning SMS, then send the warning SMS to those target subscribers. If the warning SMS is not sent, then the warning SMS will be sent to every subscriber in the overlapping area.

[0010] In one embodiment, after obtaining the binary classification of whether or not a warning SMS is pushed, as output by the warning SMS classification model, the process includes: Periodically record and query the cache to see if the meteorological disaster early warning information exists; If it does not exist, the warning SMS will be pushed to every subscriber in the overlapping area, the meteorological disaster warning information will be recorded in the cache, and the validity period of the meteorological disaster warning information will be set to a preset duration. If they exist, then query whether each subscriber in the overlapping area has received the warning SMS message; If there are target subscribers who have not received the alert, the alert SMS will be pushed to the target subscribers, and the identifier of the target subscribers will be recorded in the cache.

[0011] Secondly, embodiments of this application provide a warning SMS push device, comprising: The detailed text extraction module is used to extract detailed text from meteorological disaster early warning information. The warning SMS binary classification module is used to: input the detailed text and the first meteorological disaster characteristics into the warning SMS classification model, and obtain the binary classification of whether a warning SMS has been pushed out by the warning SMS classification model; The first meteorological disaster feature is extracted manually from the meteorological disaster early warning information; The warning SMS classification model is trained based on any classification model, using historical detailed text, historical first meteorological disaster characteristics and their corresponding historical binary classification labels; Each of the classification models includes a deep learning network and a feature fusion module. The deep learning network is used to extract a second meteorological disaster feature from the meteorological disaster early warning information, and the feature fusion module is used to fuse the first meteorological disaster feature and the second meteorological disaster feature.

[0012] Thirdly, embodiments of this application provide an electronic device, including a processor and a memory storing a computer program, wherein the processor executes the program to implement the steps of the early warning SMS push method described in the first aspect.

[0013] Fourthly, embodiments of this application provide a computer program product, including a computer program, which, when executed by a processor, implements the steps of the early warning SMS push method described in the first aspect.

[0014] Fifthly, embodiments of this application provide a non-transitory computer-readable storage medium, including a computer program, which, when executed by a processor, implements the steps of the early warning SMS push method described in the first aspect.

[0015] The method, apparatus, equipment, product, and medium for sending early warning SMS messages provided in this application extract detailed text from meteorological disaster early warning information. The detailed text and a first meteorological disaster feature are input into an early warning SMS classification model, resulting in a binary classification of whether or not an early warning SMS message should be sent. The first meteorological disaster feature is extracted manually from the meteorological disaster early warning information. The early warning SMS classification model is trained based on any existing classification model, using historical detailed text, historical first meteorological disaster features, and their corresponding historical binary classification labels. Each classification model includes a deep learning network and a feature fusion module. The deep learning network extracts a second meteorological disaster feature from the meteorological disaster early warning information, and the feature fusion module fuses the first and second meteorological disaster features. This application uses both manual extraction and deep learning network extraction to extract the first and second meteorological disaster features respectively, and then uses the feature fusion module to fuse the first and second meteorological disaster features to obtain richer fused features than single-category meteorological disaster features. This allows for a more accurate determination of whether or not to send an early warning SMS message based on these fused features. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is one of the flowcharts illustrating the early warning SMS push method provided in the embodiments of this application; Figure 2 This is a schematic diagram of the warning SMS classification model provided in the embodiments of this application; Figure 3 This is a second flowchart illustrating the early warning SMS push method provided in this application embodiment; Figure 4 This is the third flowchart illustrating the early warning SMS push method provided in this application embodiment; Figure 5 This is the fourth flowchart illustrating the early warning SMS push method provided in this application embodiment; Figure 6 This is the fifth flowchart illustrating the early warning SMS push method provided in this application embodiment; Figure 7 This is the sixth flowchart illustrating the early warning SMS push method provided in this application embodiment; Figure 8 This is an interactive diagram of the push notification process for warning text messages provided in an embodiment of this application; Figure 9 This is a schematic diagram of the warning SMS provided in the embodiments of this application; Figure 10 This is a schematic diagram of the structure of the early warning SMS push device provided in the embodiments of this application; Figure 11 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] Figure 1 This is one of the flowcharts illustrating the early warning SMS push method provided in this application embodiment. (Refer to...) Figure 1 This application provides a method for sending early warning SMS messages, which may include: Step 101: Extract detailed text from meteorological disaster early warning information; Step 102: Input the detailed text and the first meteorological disaster characteristics into the warning SMS classification model to obtain the binary classification of whether a warning SMS was pushed, as output by the warning SMS classification model.

[0020] Among them, the first meteorological disaster characteristic is extracted manually from meteorological disaster early warning information; The warning SMS classification model is trained based on any classification model, using historical detailed text, historical first meteorological disaster characteristics and their corresponding historical binary classification labels; Any classification model includes a deep learning network and a feature fusion module. The deep learning network is used to extract the second meteorological disaster feature from the meteorological disaster early warning information, and the feature fusion module is used to fuse the first meteorological disaster feature and the second meteorological disaster feature.

[0021] Meteorological disaster warning information can be obtained from meteorological disaster warnings already issued by meteorological warning service providers. Based on their own needs, individuals can selectively obtain meteorological disaster warning information for specific regions, time periods, severity levels, and disaster types. In this embodiment, in order to ensure the completeness of meteorological disaster early warning information, it is possible to obtain meteorological disaster early warning information for all regions, all time periods, all severity levels, and all disaster types that have been released.

[0022] Furthermore, due to the frequent changes in meteorological data, meteorological disaster early warning information can be obtained periodically to update previously acquired meteorological disaster early warning information in a timely manner.

[0023] In addition, to ensure data quality, the acquired meteorological disaster early warning information can be preprocessed by data cleaning, data integration, and data transformation.

[0024] Deep learning networks extract features from meteorological disaster warning information differently than human intervention. Typically, meteorological disaster warning information can include information such as disaster severity level, time of occurrence, disaster area, disaster type, and disaster details. Humans can extract information from meteorological disaster warning information and further extract explicit meteorological disaster features that are easy for humans to identify, such as disaster severity level, disaster time, disaster area, and disaster type. Deep learning networks, on the other hand, can further extract implicit meteorological disaster features that are not easy for humans to identify from the extracted disaster details text information. By fusing explicit and implicit meteorological disaster features, both the dimensionality and depth of the features are increased, resulting in richer fused features in both dimensionality and depth.

[0025] Furthermore, the feature fusion module can be an attention module, which uses explicit features as query Q and implicit features as key-value pairs K and V, and fuses them according to the following formula: ; in, This indicates the attention module. Let K be the dimension of the key.

[0026] In the process of feature fusion, attention mechanisms can uncover the interaction relationships between different features, thereby obtaining key features between explicit and implicit features and improving the representational ability of fused features.

[0027] Will The explicit and implicit features are further input into a binary classification layer to obtain a binary classification result indicating whether a warning SMS was pushed. This binary classification layer can be a fully connected layer, such as... Figure 2 As shown.

[0028] In this embodiment, the loss function of the warning SMS classification model is... It can be defined as: ; in, To decide whether to push the first The actual classification of weather disaster warning text messages. The output of the warning SMS classification model is whether or not the message is pushed. The predictive classification of weather disaster warning text messages. This represents the total number of meteorological disaster early warning messages entered into the early warning SMS classification model.

[0029] The early warning SMS push method provided in this embodiment extracts the detailed text from meteorological disaster early warning information, inputs the detailed text and a first meteorological disaster feature into an early warning SMS classification model, and obtains a binary classification of whether to push an early warning SMS based on the output of the early warning SMS classification model. The first meteorological disaster feature is extracted manually from the meteorological disaster early warning information. The early warning SMS classification model is trained based on any classification model, using historical detailed text, historical first meteorological disaster features, and their corresponding historical binary classification labels. Each classification model includes a deep learning network and a feature fusion module. The deep learning network is used to extract the second meteorological disaster feature from the meteorological disaster early warning information, and the feature fusion module is used to fuse the first and second meteorological disaster features. This embodiment uses both manual extraction and deep learning network extraction to extract the first and second meteorological disaster features respectively, and then uses the feature fusion module to fuse the first and second meteorological disaster features to obtain richer fused features than single-category meteorological disaster features, thereby enabling a more accurate determination of whether to push an early warning SMS based on these fused features.

[0030] Figure 3 This is the second flowchart illustrating the early warning SMS push method provided in this application embodiment. (Refer to...) Figure 3 In one embodiment, the deep learning network includes a stacked pointer network and a relational graph convolutional neural network. Extracting second meteorological disaster features from meteorological disaster early warning information may include: Step 301: Input the detailed text into the stacked pointer network to obtain the knowledge graph structure information output by the stacked pointer network; Step 302: Input the knowledge graph structure information into the relational graph convolutional neural network to obtain the second meteorological disaster feature output by the relational graph convolutional neural network.

[0031] Reference Figure 2 : In step 301, the text detailing the disaster is first obtained from the meteorological disaster warning information. This text is then input into a cascaded pointer network, which is used to transform the entities and relationships in the text into knowledge graph structure information and output it.

[0032] The stacked pointer network includes an encoding layer, a head entity recognition layer, and a relation and tail entity joint recognition layer. The encoding layer is used to obtain the contextual semantics of the detailed text to encode the words in the text and obtain their latent representations. The head entity recognition layer is used to decode these latent representations to identify the start and end positions of all possible head entities. The relation and tail entity joint recognition layer is used to identify the tail entities related to the head entities given the relations, so as to finally obtain the triple information composed of all head entities-relationships-tail entities, that is, the knowledge graph structure information.

[0033] Specifically, the start and end positions of the head entity can be identified based on the following formula: ; ; in, For the first A potential representation, and All of these are trainable weight matrices. and All are bias terms, but , , This represents the activation function. for It is the probability of the starting position of the head entity. for It is the probability of the end position of the head entity. If it exceeds a certain threshold, then... Mark as 1, confirm This is the starting position of the head entity; otherwise, it will be... Mark as 0, confirm Not the starting position of the head entity; for Similarly, determine Is it the end position of the head entity? According to the nearest matching principle, each start position of the head entity is matched with its nearest end position to obtain the head entity set.

[0034] Furthermore, the start and end positions of the tail entity associated with the head entity can be identified based on the following formula: ; ; in, For the head entity set, the first The encoding vector of each head entity and All are given relations The trainable weight matrix below, and All are given relations The bias term below, but , , This represents the activation function. for Is with The probability of the starting position of the relevant tail entity. for Is with The probability of the ending position of the relevant tail entity, if If it exceeds a certain threshold, then... Mark as 1, confirm Is with The starting position of the relevant tail entity, otherwise... Mark as 0, confirm Not with The starting position of the relevant tail entity; for Similarly, determine Is it with The ending position of the relevant tail entity; according to the nearest matching principle, it will be related to... Each start position of the relevant tail entity is matched with its nearest end position to obtain the corresponding... The relevant tail entity set can be obtained, and similarly, the tail entity set related to all head entities can be obtained. Combining the given relationship between each head entity and its related tail entities, the triple information composed of all head entities-relationships-tail entities is obtained, which is the knowledge graph structure information.

[0035] In step 302, the propagation model of the relational graph convolutional neural network can be expressed by the following formula: ; in, For the knowledge graph structure information, the first The nth entity in the relational graph convolutional neural network Layer feature representation, For the first The entity in the first Layer self-embedding weight matrix, For the first In the knowledge graph structure information of each entity, the first... The first entity The neighboring entities in the th Layer feature representation, For the first The neighboring entities in the th Layer and First The relationship between the entities is: The weight matrix at time, In order to be with the first Entities and Relationships The relevant regularization parameters, For a set of relations, In order to be with the first The relationship between the entities is: The set of indices of neighboring entities, For activation function, For the first The nth entity in the relational graph convolutional neural network Layer feature representation.

[0036] To address the problem of relation overfitting, cardinality decomposition can be used for different types of relations. ,Will Decompose according to the following formula: ; in, For the first The first layer The nth basis matrix, whose dimension is the nth basis matrix. Layer feature dimensions No. The feature dimensions of the layer For the first Layer Relationship In the Combination coefficients on each basis matrix For the first The total number of basis matrices in each layer. After decomposition. With Relationship related, It can be shared across all relation types, thus avoiding relation overfitting.

[0037] This step can employ a two-layer graph convolutional structure. After obtaining the feature representations of all entities from the last layer output according to the above formula, average pooling is performed on these feature representations to obtain the global feature representation of the knowledge graph structure information, which is also the second meteorological disaster feature.

[0038] This embodiment utilizes a stacked pointer network to extract the knowledge graph structure information of the detailed text of meteorological disaster early warning information, and simultaneously mines the various entities in the detailed text and the diverse relationships between them, improving the accuracy of entity and relationship mining. This avoids the traditional pipeline method of identifying entities first and then relationships, which leads to the error of entity identification being passed on to subsequent relationship identification, resulting in error accumulation. Furthermore, a relation graph convolutional neural network is used to extract global features from the knowledge graph structure information. Through hierarchical analysis of the knowledge graph structure information, the global feature extraction can be made more accurate.

[0039] In one embodiment, the warning SMS classification model includes multiple SMS classification models. The detailed text and the first meteorological disaster characteristic are input into the warning SMS classification model to obtain a binary classification of whether a warning SMS was sent. This classification may include: The severity levels of meteorological disaster warning information are classified to obtain multiple severity level categories. For any one of the multiple severity level categories, the detailed text of the corresponding meteorological disaster warning information and the first meteorological disaster feature are input into the SMS classification model that is adapted to it among multiple SMS classification models. The result is a binary classification of whether a warning SMS should be sent, output by the adapted SMS classification model.

[0040] The adapted SMS classification model is trained based on any classification model, using historical detailed text, historical first meteorological disaster features, and historical binary classification labels corresponding to historical meteorological disaster warning information of any severity level category.

[0041] Specifically, refer to Figure 4 Meteorological disaster warning information severity levels can be divided into "Extreme", "Severe", "Major", "Moderate", "Minor", "Standard", "Unknown", "None" and "Cancel", which correspond to "extreme", "serious", "major", "moderate", "minor", "standard", "unknown", "none" and "cancel" respectively.

[0042] In this embodiment, "Severe", "Major", and "Moderate" are classified as the first severity level, and "Minor", "Standard", "Unknown", "None", and "Cancel" are classified as the second severity level. The first SMS classification model adapted to the first severity level is trained based on any classification model, using historical detailed text, historical first meteorological disaster features, and historical binary classification labels corresponding to historical meteorological disaster warning information of the first severity level. The second SMS classification model adapted to the second severity level is trained based on any classification model, using historical detailed text, historical first meteorological disaster features, and historical binary classification labels corresponding to historical meteorological disaster warning information of the second severity level.

[0043] Furthermore, for meteorological disaster warnings with a severity level of "Extreme", there is no need to perform SMS classification model prediction; the warning SMS can be directly sent.

[0044] Warning SMS push has unique business requirements. It needs to be sensitive to warnings of extremely severe meteorological disasters, such as flash floods and mudslides, and must promptly send warning SMS messages. It also needs to differentiate warning SMS messages for meteorological disasters of different severity levels. For example, in some areas where low temperature warnings are frequently issued in winter, users do not need to receive warning SMS messages frequently. Ideally, warning SMS messages should only be sent to users when the low temperature warning reaches a certain level of severity.

[0045] This embodiment pre-classifies meteorological disaster warning information based on its severity level and adopts different adaptive processing procedures for different categories of meteorological disaster warning information. For meteorological disaster warning information of extreme severity level, the impact of not sending warning SMS messages far outweighs the impact of differentiated push notifications. Therefore, there is no need to reclassify whether to push the warning; the warning SMS message is directly sent to the user, avoiding the occurrence of not sending severe disaster warning information. For meteorological disaster warning information of "severe," "major," and "moderate," as well as "minor," "standard," "unknown," "none," and "cancelled," since the former has a higher severity level and the latter has a lower severity level, their respective adapted SMS classification models are used for reclassification to determine whether to push the warning, in order to improve prediction accuracy.

[0046] Figure 5 This is the fourth flowchart illustrating the early warning SMS push method provided in this application embodiment. (Refer to...) Figure 5 In one embodiment, meteorological disaster early warning information can be obtained based on the following method: Step 501: Obtain the first area within the coverage of the meteorological early warning service that has issued meteorological disaster early warning information; Step 502: Obtain the second region to which the subscribers of the weather warning service belong; Step 503: Determine the overlapping area of ​​the first region and the second region; Step 504: Obtain meteorological disaster early warning information for overlapping areas.

[0047] In this embodiment, since the area where the user who subscribes to the weather warning service is located may differ from the area where the weather disaster warning information has been issued, and the subscribing user only cares about the weather disaster warning information in their own area, it is necessary to first obtain the overlapping area of ​​the two areas. This area is both the area where the subscribing user is located and the area where the weather disaster warning information has been issued. Obtaining the weather disaster warning information in this area helps to accurately determine whether to send warning SMS messages to the subscribing user in this area.

[0048] Reference Figure 6 In one embodiment, after obtaining the binary classification of whether or not a warning SMS message is pushed from the warning SMS classification model, the following can be included: Step 601: If it is determined that a warning SMS has been sent, check the cache to see if the warning SMS has been sent. Step 602: If the warning SMS has been sent, check the cache to see if the warning SMS has been sent to every subscriber in the overlapping area. If there are target subscribers who have not been sent the warning SMS, send the warning SMS to the target subscribers. Step 603: If the warning SMS is not sent, then send the warning SMS to every subscriber in the overlapping area.

[0049] In step 601, when determining to push a warning SMS, the corresponding meteorological disaster warning information can be encoded, and the encoded information can be used as a key to query the cache whether the target meteorological disaster warning information has been pushed in the form of a warning SMS.

[0050] In steps 602 to 603: If the meteorological disaster warning information has been pushed in the form of a warning SMS, then check the cache to see if the warning SMS has been pushed to every subscriber in the overlapping area. If there are target subscribers who have not been pushed, then push the meteorological disaster warning information to the mobile phone number of the target subscriber in the form of a warning SMS. If there are no target subscribers who have not been pushed, then no further processing will be carried out. If the meteorological disaster warning information has not been sent in the form of a warning SMS, then the meteorological disaster warning information will be sent in the form of a warning SMS to the mobile phone number of each subscriber in the overlapping area.

[0051] Furthermore, referring to Figure 7 After determining whether to send a warning SMS, it may also include: Step 701: Periodically record and query the cache to see if meteorological disaster early warning information exists; Step 702: If not, push the warning SMS to every subscriber in the overlapping area, record the meteorological disaster warning information in the cache, and set the validity period of the meteorological disaster warning information to the preset duration; Step 703: If it exists, check whether each subscriber in the overlapping area has received a warning SMS. Step 704: If there are target subscribers who have not received the message, push the warning SMS to the target subscribers and record the identifier of the target subscribers in the cache.

[0052] In step 701, the encoded meteorological disaster early warning information is periodically recorded in the cache and its existence is checked.

[0053] In steps 702 to 703: If it does not exist, the meteorological disaster warning information will be pushed to the mobile phone number of each subscriber in the overlapping area in the form of a warning SMS, and the encoded meteorological disaster warning information will be recorded in the cache, which can be set to a validity period of 3 days; If it exists, check whether the mobile phone number of each subscriber in the overlapping area has received the warning SMS. This can be confirmed by checking the receipt sent by the mobile phone number.

[0054] In step 704: If there are target subscribers who have not received the warning SMS, the meteorological disaster warning information will be pushed to the target subscriber's mobile phone number in the form of a warning SMS, and the target subscriber's mobile phone number will be recorded in the cache; If there are no target subscribers who have not received the warning SMS, no further processing will be carried out.

[0055] In this embodiment, Redis can be used as a cache and base64 encoding can be used.

[0056] This embodiment encodes meteorological disaster warning information and records the encoded information along with the subscriber's mobile phone number in a cache. This facilitates subsequent repeated queries and updates of relevant information, ensuring the timeliness of warning SMS pushes and effectively avoiding the problem of duplicate warning SMS pushes. Furthermore, by adopting a subscription-based push model, warning SMS pushes are not sent to users who have not subscribed to the service, reducing the push burden and minimizing the number of calls to third-party meteorological warning services.

[0057] Reference Figure 8 In one embodiment, the process of sending alert SMS messages is briefly described; 1. User clients subscribe to weather warning services from the system server; 2. The system server sends a request to a third-party meteorological early warning service to obtain the first area within its coverage that has issued meteorological disaster early warning information; 3. The third-party weather warning service will send the first regional warning to the system server; 4. The system server obtains the second region to which the subscribers of the weather warning service within its internal system belong; 5. After the system server determines the overlapping area between the first region and the second region, it sends a request to a third-party meteorological early warning service to obtain meteorological disaster early warning information that has been issued in the overlapping area; 6. Third-party meteorological early warning services will send meteorological disaster early warning information already issued in overlapping areas to the system server; 7. The system server obtains the mobile phone numbers of subscribers in overlapping areas; 8. The system server selects the appropriate process based on the severity level of meteorological disaster warning information in overlapping areas to determine whether to push meteorological disaster warning information of different severity levels in the form of warning SMS messages; 9. If the system server determines that a warning SMS has been sent, it queries Redis to see if the warning SMS has been sent. 10. Once the system server confirms that the warning SMS has been sent, it queries Redis to see if the warning SMS has been sent to the mobile phone number of each subscriber in the overlapping area. 11. If the system server determines that there are target subscribers who have not received the notification, it will send a warning SMS to those target subscribers.

[0058] The alert text messages received by subscribers can be like... Figure 9 As shown.

[0059] The warning SMS push device provided in the embodiments of this application is described below. The warning SMS push device described below can be referred to in correspondence with the warning SMS push method described above.

[0060] Figure 10 This is a schematic diagram of the structure of the early warning SMS push device provided in an embodiment of this application. (Refer to...) Figure 10 This application provides an early warning SMS push device, which may include: The detailed text extraction module 1001 is used to extract detailed text from meteorological disaster early warning information; The warning SMS binary classification module 1002 is used to: input the details text and the first meteorological disaster characteristics into the warning SMS classification model, and obtain the binary classification of whether a warning SMS is pushed out by the warning SMS classification model; The first meteorological disaster feature is extracted manually from the meteorological disaster early warning information; The warning SMS classification model is trained based on any classification model, using historical detailed text, historical first meteorological disaster characteristics and their corresponding historical binary classification labels; Each of the classification models includes a deep learning network and a feature fusion module. The deep learning network is used to extract a second meteorological disaster feature from the meteorological disaster early warning information, and the feature fusion module is used to fuse the first meteorological disaster feature and the second meteorological disaster feature.

[0061] The early warning SMS push device provided in this embodiment extracts the detailed text from meteorological disaster early warning information, inputs the detailed text and a first meteorological disaster feature into an early warning SMS classification model, and obtains a binary classification of whether or not to push an early warning SMS. The first meteorological disaster feature is extracted manually from the meteorological disaster early warning information. The early warning SMS classification model is trained based on any classification model, using historical detailed text, historical first meteorological disaster features, and their corresponding historical binary classification labels. Each classification model includes a deep learning network and a feature fusion module. The deep learning network is used to extract a second meteorological disaster feature from the meteorological disaster early warning information, and the feature fusion module is used to fuse the first and second meteorological disaster features. This embodiment uses both manual extraction and deep learning network extraction to extract the first and second meteorological disaster features respectively, and then uses the feature fusion module to fuse the first and second meteorological disaster features to obtain richer fused features than single-category meteorological disaster features, thereby enabling a more accurate determination of whether to push an early warning SMS based on these fused features.

[0062] In one embodiment, the deep learning network includes a stacked pointer network and a relational graph convolutional neural network, and the warning SMS binary classification module 1002 is specifically used for: The detailed text is input into the stacked pointer network to obtain the knowledge graph structure information output by the stacked pointer network; The knowledge graph structure information is input into the relational graph convolutional neural network to obtain the second meteorological disaster feature output by the relational graph convolutional neural network.

[0063] In one embodiment, the warning SMS classification model includes multiple SMS classification models, with the warning SMS binary classification module 1002 specifically used for: The severity levels of the meteorological disaster early warning information are classified to obtain multiple severity level categories; For any of the multiple severity level categories, the detailed text of the corresponding meteorological disaster warning information and the first meteorological disaster feature are input into the SMS classification model that is compatible with it among the multiple SMS classification models, and a binary classification of whether to push a warning SMS is obtained from the output of the compatible SMS classification model. The adapted SMS classification model is trained based on any of the above classification models, using historical detailed text, historical first meteorological disaster features, and historical binary classification labels corresponding to historical meteorological disaster warning information of any severity level category.

[0064] In one embodiment, a meteorological disaster early warning information acquisition module (not shown in the figure) is further included, for: The first region within the coverage area of ​​the meteorological early warning service that has already issued meteorological disaster early warning information; The second region to which subscribers who receive weather warning services belong; Determine the overlapping area between the first region and the second region; Obtain meteorological disaster early warning information for the overlapping areas.

[0065] In one embodiment, a warning SMS push module (not shown in the figure) is also included, for: If it is determined that a warning SMS message has been sent, check the cache to see if the warning SMS message has been sent. If the warning SMS has been sent, then query the cache to see if the warning SMS has been sent to every subscriber in the overlapping area. If there are target subscribers who have not received the warning SMS, then send the warning SMS to those target subscribers. If the warning SMS is not sent, then the warning SMS will be sent to every subscriber in the overlapping area.

[0066] In one embodiment, the warning SMS push module is further used for: Periodically record and query the cache to see if the meteorological disaster early warning information exists; If it does not exist, the warning SMS will be pushed to every subscriber in the overlapping area, the meteorological disaster warning information will be recorded in the cache, and the validity period of the meteorological disaster warning information will be set to a preset duration. If they exist, then query whether each subscriber in the overlapping area has received the warning SMS message; If there are target subscribers who have not received the alert, the alert SMS will be pushed to the target subscribers, and the identifier of the target subscribers will be recorded in the cache.

[0067] Figure 11 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application, such as... Figure 11 As shown, the electronic device may include: a processor 1110, a communication interface 1120, a memory 1130, and a communication bus 1140, wherein the processor 1110, the communication interface 1120, and the memory 1130 communicate with each other via the communication bus 1140. The processor 1110 can call a computer program in the memory 1130 to execute the steps of the early warning SMS push method, such as including: Extract detailed text from meteorological disaster early warning information; Input the detailed text and the first meteorological disaster characteristics into the early warning SMS classification model to obtain a binary classification of whether an early warning SMS was pushed, as output by the early warning SMS classification model; The first meteorological disaster feature is extracted manually from the meteorological disaster early warning information; The warning SMS classification model is trained based on any classification model, using historical detailed text, historical first meteorological disaster characteristics and their corresponding historical binary classification labels; Each of the classification models includes a deep learning network and a feature fusion module. The deep learning network is used to extract a second meteorological disaster feature from the meteorological disaster early warning information, and the feature fusion module is used to fuse the first meteorological disaster feature and the second meteorological disaster feature.

[0068] Furthermore, the logical instructions in the aforementioned memory 1130 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0069] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the steps of the early warning SMS push method provided in the above embodiments, such as including: Extract detailed text from meteorological disaster early warning information; Input the detailed text and the first meteorological disaster characteristics into the early warning SMS classification model to obtain a binary classification of whether an early warning SMS was pushed, as output by the early warning SMS classification model; The first meteorological disaster feature is extracted manually from the meteorological disaster early warning information; The warning SMS classification model is trained based on any classification model, using historical detailed text, historical first meteorological disaster characteristics and their corresponding historical binary classification labels; Each of the classification models includes a deep learning network and a feature fusion module. The deep learning network is used to extract a second meteorological disaster feature from the meteorological disaster early warning information, and the feature fusion module is used to fuse the first meteorological disaster feature and the second meteorological disaster feature.

[0070] On the other hand, embodiments of this application also provide a non-transitory computer-readable storage medium storing a computer program thereon, the computer program being used to cause a processor to execute the steps of the early warning SMS push method provided in the above embodiments, for example including: Extract detailed text from meteorological disaster early warning information; Input the detailed text and the first meteorological disaster characteristics into the early warning SMS classification model to obtain a binary classification of whether an early warning SMS was pushed, as output by the early warning SMS classification model; The first meteorological disaster feature is extracted manually from the meteorological disaster early warning information; The warning SMS classification model is trained based on any classification model, using historical detailed text, historical first meteorological disaster characteristics and their corresponding historical binary classification labels; Each of the classification models includes a deep learning network and a feature fusion module. The deep learning network is used to extract a second meteorological disaster feature from the meteorological disaster early warning information, and the feature fusion module is used to fuse the first meteorological disaster feature and the second meteorological disaster feature.

[0071] The non-transitory computer-readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical memory (e.g., CD, DVD, BD, HVD), and semiconductor memory (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)).

[0072] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0073] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for sending early warning text messages, characterized in that, include: Extract detailed text from meteorological disaster early warning information; Input the detailed text and the first meteorological disaster characteristics into the early warning SMS classification model to obtain a binary classification of whether an early warning SMS was pushed, as output by the early warning SMS classification model; The first meteorological disaster feature is extracted manually from the meteorological disaster early warning information; The warning SMS classification model is trained based on any classification model, using historical detailed text, historical first meteorological disaster characteristics and their corresponding historical binary classification labels; Each of the classification models includes a deep learning network and a feature fusion module. The deep learning network is used to extract a second meteorological disaster feature from the meteorological disaster early warning information, and the feature fusion module is used to fuse the first meteorological disaster feature and the second meteorological disaster feature.

2. The method for sending early warning SMS messages according to claim 1, characterized in that, The deep learning network includes a stacked pointer network and a relational graph convolutional neural network. Extracting the second meteorological disaster feature from the meteorological disaster early warning information includes: The detailed text is input into the stacked pointer network to obtain the knowledge graph structure information output by the stacked pointer network; The knowledge graph structure information is input into the relational graph convolutional neural network to obtain the second meteorological disaster feature output by the relational graph convolutional neural network.

3. The method for sending early warning text messages according to claim 1, characterized in that, The warning SMS classification model includes multiple SMS classification models. The step of inputting the detailed text and the first meteorological disaster feature into the warning SMS classification model yields a binary classification of whether a warning SMS has been sent, including: The severity levels of the meteorological disaster early warning information are classified to obtain multiple severity level categories; For any of the multiple severity level categories, the detailed text of the corresponding meteorological disaster warning information and the first meteorological disaster feature are input into the SMS classification model that is compatible with it among the multiple SMS classification models, and a binary classification of whether to push a warning SMS is obtained from the output of the compatible SMS classification model. The adapted SMS classification model is trained based on any of the above classification models, using historical detailed text, historical first meteorological disaster features, and historical binary classification labels corresponding to historical meteorological disaster warning information of any severity level category.

4. The method for sending early warning text messages according to claim 3, characterized in that, The meteorological disaster early warning information is obtained based on the following methods: The first region within the coverage area of ​​the meteorological early warning service that has already issued meteorological disaster early warning information; The second region to which subscribers who receive weather warning services belong; Determine the overlapping area between the first region and the second region; Obtain meteorological disaster early warning information for the overlapping areas.

5. The method for sending early warning text messages according to claim 4, characterized in that, After obtaining the binary classification of whether or not a warning SMS was pushed, as output by the warning SMS classification model, the process includes: If it is determined that a warning SMS message has been sent, check the cache to see if the warning SMS message has been sent. If the warning SMS has been sent, then query the cache to see if the warning SMS has been sent to every subscriber in the overlapping area. If there are target subscribers who have not received the warning SMS, then send the warning SMS to those target subscribers. If the warning SMS is not sent, then the warning SMS will be sent to every subscriber in the overlapping area.

6. The method for sending early warning text messages according to claim 4, characterized in that, After obtaining the binary classification of whether or not a warning SMS was pushed, as output by the warning SMS classification model, the process includes: Periodically record and query the cache to see if the meteorological disaster early warning information exists; If it does not exist, the warning SMS will be pushed to every subscriber in the overlapping area, the meteorological disaster warning information will be recorded in the cache, and the validity period of the meteorological disaster warning information will be set to a preset duration. If they exist, then query whether each subscriber in the overlapping area has received the warning SMS message; If there are target subscribers who have not received the alert, the alert SMS will be pushed to the target subscribers, and the identifier of the target subscribers will be recorded in the cache.

7. A warning SMS push device, characterized in that, include: The detailed text extraction module is used to extract detailed text from meteorological disaster early warning information. The warning SMS binary classification module is used to: input the detailed text and the first meteorological disaster characteristics into the warning SMS classification model, and obtain the binary classification of whether a warning SMS has been pushed out by the warning SMS classification model; The first meteorological disaster feature is extracted manually from the meteorological disaster early warning information; The warning SMS classification model is trained based on any classification model, using historical detailed text, historical first meteorological disaster characteristics and their corresponding historical binary classification labels; Each of the classification models includes a deep learning network and a feature fusion module. The deep learning network is used to extract a second meteorological disaster feature from the meteorological disaster early warning information, and the feature fusion module is used to fuse the first meteorological disaster feature and the second meteorological disaster feature.

8. An electronic device comprising a processor and a memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the early warning SMS push method according to any one of claims 1 to 6.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the early warning SMS push method according to any one of claims 1 to 6.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the early warning SMS push method according to any one of claims 1 to 6.

Images