Disaster information pushing method, device and equipment based on large language model and readable medium

Abstract

The embodiment of the invention discloses a disaster information pushing method and device based on a large language model, equipment and a medium. A specific embodiment of the method comprises the following steps: acquiring environment text information; performing feature extraction processing based on the environment text information to generate a plurality of extraction features to obtain an extraction feature group; carrying out classification processing on the extracted features; preprocessing the at least one local feature and the at least one global feature; inputting the local feature group and the global feature group into a pre-trained large language model to obtain disaster information and disaster response information corresponding to the disaster information; and based on the disaster information and the disaster response information, generating a disaster push page, and pushing the disaster push page to an associated user terminal for display. According to the embodiment, the disaster grade is accurately determined, and casualties or waste of emergency response resources are avoided.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of computer technology, and more particularly, to a method, apparatus, device, and readable medium for pushing disaster information based on a large language model. Background Art

[0002] The importance of emergency response to meteorological environment in disaster prevention and mitigation is becoming increasingly prominent. Currently, when carrying out emergency response to meteorological environment, the commonly adopted method is: classifying meteorological environment data through a trained convolutional neural network, determining the corresponding meteorological environment disaster level, and generating the corresponding emergency response.

[0003] However, when the above method is used for emergency response to meteorological environment, the following technical problems often exist:

[0004] When classifying meteorological environment data through a convolutional neural network to determine the disaster level, for structured data, unstructured data, and geographical information of meteorological environment, etc., the convolutional neural network cannot effectively fuse multi-dimensional features to determine the disaster level, resulting in inaccurate determination of the disaster level, causing casualties or waste of emergency response resources.

[0005] The above information disclosed in this background art section is only used to enhance the understanding of the background of the inventive concept, and thus, it may include information that does not form the prior art known to those of ordinary skill in the art in this country. Summary of the Invention

[0006] This section of the present disclosure is used to briefly introduce concepts that will be described in detail in the subsequent Detailed Description section. This section of the present disclosure is not intended to identify the key features or essential features of the claimed technical solution, nor is it intended to be used to limit the scope of the claimed technical solution.

[0007] Some embodiments of the present disclosure propose a method, apparatus, electronic device, and computer-readable medium for pushing disaster information based on a large language model to solve one or more of the technical problems mentioned in the above background art section.

[0008] In a first aspect, some embodiments of the present disclosure provide a method for pushing disaster information based on a large language model. The method includes: obtaining environmental text information, where the environmental text information is text information received from a target terminal through a preset interface and used to represent environmental information; performing feature extraction processing based on the environmental text information to generate a plurality of extracted features and obtain an extracted feature group; for each extracted feature in the extracted feature group, classifying the extracted feature to generate at least one local feature and at least one global feature; preprocessing the at least one local feature and the at least one global feature to generate preprocessed local features and global features, and obtaining a local feature group and a global feature group; inputting the local feature group and the global feature group into a pre-trained large language model to obtain disaster information and disaster response information corresponding to the disaster information; generating a disaster push page based on the disaster information and the disaster response information, and pushing the disaster push page to an associated user terminal for display.

[0009] In a second aspect, some embodiments of the present disclosure provide a device for pushing disaster information based on a large language model. The device includes: an obtaining unit configured to obtain environmental text information, where the environmental text information is text information received from a target terminal through a preset interface and used to represent environmental information; an execution unit configured to perform feature extraction processing based on the environmental text information to generate a plurality of extracted features and obtain an extracted feature group; a classification unit configured to classify each extracted feature in the extracted feature group to generate at least one local feature and at least one global feature; a preprocessing unit configured to preprocess the at least one local feature and the at least one global feature to generate preprocessed local features and global features, and obtain a local feature group and a global feature group; an input unit configured to input the local feature group and the global feature group into a pre-trained large language model to obtain disaster information and disaster response information corresponding to the disaster information; a generating unit configured to generate a disaster push page based on the disaster information and the disaster response information, and push the disaster push page to an associated user terminal for display.

[0010] In a third aspect, some embodiments of the present disclosure provide an electronic device, including: one or more processors; a storage device having stored thereon one or more programs, which when executed by the one or more processors, cause the one or more processors to implement the method described in any implementation manner of the first aspect above.

[0011] Fourthly, some embodiments of the present disclosure provide a computer-readable medium, on which a computer program is stored, where when the program is executed by a processor, the method described in any implementation manner of the first aspect above is implemented.

[0012] The above various embodiments of the present disclosure have the following beneficial effects: Through the disaster information push method based on the large language model in some embodiments of the present disclosure, the disaster level is accurately determined, avoiding casualties or waste of emergency response resources. Specifically, the reasons for inaccurate determination of the disaster level and resulting in casualties or waste of emergency response resources are as follows: When classifying meteorological environment data through a convolutional neural network to determine the disaster level, for structured data, unstructured data, and geographical information of the meteorological environment, etc., the convolutional neural network cannot effectively fuse multi-dimensional features to determine the disaster level, resulting in inaccurate determination of the disaster level and causing casualties or waste of emergency response resources. Based on this, in some embodiments of the disaster information push method based on the large language model of the present disclosure, first, environmental text information is obtained. Thus, meteorological text data can be obtained from the meteorological monitoring center. Secondly, based on the above environmental text information, feature extraction processing is performed to generate multiple extracted features, obtaining an extracted feature group. Thus, data features in the text data can be extracted. Then, for each extracted feature in the above extracted feature group, the above extracted features are classified to generate at least one local feature and at least one global feature. Thus, global features and local features can be distinguished. Then, the above at least one local feature and the above at least one global feature are preprocessed to generate preprocessed local features and global features, obtaining a local feature group and a global feature group. Thus, the local features and global features can be preprocessed. After that, the above local feature group and the above global feature group are input into a pre-trained large language model to obtain disaster information and disaster response information corresponding to the above disaster information. Thus, through the large language model's fusion analysis of multi-dimensional data such as structured data, unstructured data, and geographical information of the object environment, the disaster level corresponding to the meteorological environment can be accurately determined, thereby avoiding casualties or waste of emergency response resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In combination with the accompanying drawings and with reference to the following specific embodiments, the above and other features, advantages, and aspects of the various embodiments of the present disclosure will become more obvious. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic and the elements and elements are not necessarily drawn to scale.

[0014] Figure 1 is a flowchart of some embodiments of the model parameter tuning method based on the large language model according to the present disclosure;

[0015] Figure 2 is a schematic structural diagram of some embodiments of a model parameter tuning device based on a large language model according to the present disclosure;

[0016] Figure 3 is a schematic structural diagram of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed implementation manners

[0017] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although some embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. On the contrary, these embodiments are provided to more thoroughly and completely understand the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not used to limit the protection scope of the present disclosure.

[0018] In addition, it should be noted that for the sake of convenience of description, only parts related to the relevant invention are shown in the drawings. Without conflict, the embodiments in the present disclosure and the features in the embodiments can be combined with each other.

[0019] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order of functions executed by these devices, modules or units or their interdependent relationships.

[0020] It should be noted that the modifications of "one" and "multiple" mentioned in the present disclosure are illustrative rather than restrictive. Those skilled in the art should understand that unless otherwise clearly specified in the context, it should be understood as "one or more".

[0021] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only for illustrative purposes and are not used to limit the scope of these messages or information.

[0022] The present disclosure will be described in detail below with reference to the drawings and in combination with the embodiments.

[0023] Figure 1 shows a process 100 of some embodiments of a disaster information push method based on a large language model according to the present disclosure. The disaster information push method based on a large language model includes the following steps:

[0024] Step 101, obtain environmental text information.

[0025] In some embodiments, an execution entity (such as a server) of a disaster information push method based on a large language model may obtain environmental text information. The environmental text information is text information for characterizing environmental information received from a target terminal through a preset interface. The preset interface may be a preset interface for obtaining environmental text information from the target terminal. The target terminal is a meteorological terminal of a meteorological bureau.

[0026] Step 102: Based on the environmental text information, perform feature extraction processing to generate multiple extracted features and obtain an extracted feature group.

[0027] In some embodiments, the execution entity may perform feature extraction processing based on the environmental text information to generate multiple extracted features and obtain an extracted feature group. The feature extraction processing may be to extract each feature in the environmental text information through an attention mechanism.

[0028] In the process of adopting technical solutions to solve the technical problems mentioned in the background art, the following technical problems often accompany: Some meteorological monitoring devices are deployed in areas with poor communication conditions such as mountainous areas, resulting in a relatively high delay in meteorological data transmission, being unable to determine the disaster level in a timely manner, causing casualties or waste of emergency response resources. In response to the above technical problems, the conventional solution is generally to enhance and improve the data transmission devices corresponding to the meteorological monitoring devices in the mountains, or erect more signal towers to reduce the data transmission delay. However, the above conventional solutions still have the following problems: The cost of data transmission is relatively high by enhancing and improving the data transmission devices or erecting more signal towers. Considering the problems of the above conventional solutions, in the face of the above technical problems and combining the existing technical status, the following solution can be decided.

[0029] Optionally, after step 102, the following steps may further be included:

[0030] First step: Determine the network status corresponding to the environmental text information.

[0031] In some embodiments, the execution entity may determine the network status corresponding to the environmental text information. Here, the network status of the meteorological monitoring device corresponding to the environmental text information may be determined as the network status corresponding to the environmental text information.

[0032] Second step: In response to the network status satisfying a first preset condition, load a preset feature configuration information set.

[0033] In some embodiments, the above-mentioned execution entity may load a preset feature configuration information set in response to the above-mentioned network status satisfying a first preset condition. Among them, the preset feature configuration information in the terminal preset feature configuration information set may be the weight value preset for each extracted feature. The above-mentioned first preset condition may be a network interruption state.

[0034] Thirdly, based on the above-mentioned preset feature configuration information set and the above-mentioned extracted feature set, generate a meteorological emergency evaluation corresponding to the above-mentioned extracted feature set.

[0035] In some embodiments, the above-mentioned execution entity may generate a meteorological emergency evaluation corresponding to the above-mentioned extracted feature set based on the above-mentioned preset feature configuration information set and the above-mentioned extracted feature set. In practice, the product of each extracted feature in the above-mentioned extracted feature set and the corresponding preset feature configuration information may be determined as a sub-meteorological emergency evaluation, and the sum of the normalized sub-meteorological emergency evaluations may be determined as the meteorological emergency evaluation.

[0036] Fourthly, based on the above-mentioned meteorological emergency evaluation, generate a meteorological emergency warning, and push the above-mentioned meteorological emergency warning to the associated user terminals.

[0037] In some embodiments, the above-mentioned execution entity may generate a meteorological emergency warning based on the above-mentioned meteorological emergency evaluation, and push the above-mentioned meteorological emergency warning to the associated user terminals.

[0038] Fifthly, based on the historical query information set, select each extracted feature that meets the preset query condition from the above-mentioned extracted feature set, and mark each selected extracted feature to generate marked extracted features.

[0039] In some embodiments, the above-mentioned execution entity may select each extracted feature that meets the preset query condition from the above-mentioned extracted feature set based on the historical query information set, and mark each selected extracted feature to generate marked extracted features. Among them, the above-mentioned preset query condition may be that the number of historical query information corresponding to the extracted feature is greater than or equal to the preset query number.

[0040] Sixthly, in response to detecting that the network status satisfies a second preset condition, determine the network transmission speed corresponding to the above-mentioned network status.

[0041] In some embodiments, the above-mentioned execution entity may determine the network transmission speed corresponding to the above-mentioned network status in response to detecting that the network status satisfies a second preset condition. Among them, the above-mentioned second preset condition may be a network connection state.

[0042] Step 7: Based on the above network transmission speed, select a preset compression algorithm corresponding to the above network transmission speed from a preset compression algorithm library, and perform compression processing on the feature extraction of each tag and the feature extraction set obtained by removing the feature extraction of each tag.

[0043] In some embodiments, the above execution entity may select a preset compression algorithm corresponding to the above network transmission speed from a preset compression algorithm library, and perform compression processing on the feature extraction of each tag and the feature extraction set obtained by removing the feature extraction of each tag. Here, the above preset compression algorithm library may be an algorithm database preset for storing compression algorithms.

[0044] Step 8: Transmit the compressed tag extraction features to the target server according to a preset transmission protocol.

[0045] In some embodiments, the above execution entity may transmit the compressed tag extraction features to the target server according to a preset transmission protocol. Among them, the above preset transmission protocol may be an incremental update protocol.

[0046] The above first to eighth steps, as an inventive point of the embodiments of the present disclosure, in combination with the optional steps below, solve the technical problem: "Some meteorological monitoring devices are deployed in areas such as mountainous areas with poor communication conditions, resulting in high delays in meteorological data transmission, making it impossible to determine the disaster level in a timely manner, causing casualties or waste of emergency response resources." The reasons for causing casualties or waste of emergency response resources are as follows: Some meteorological monitoring devices are deployed in areas such as mountainous areas with poor communication conditions, resulting in high delays in meteorological data transmission, making it impossible to determine the disaster level in a timely manner, causing casualties or waste of emergency response resources. If the above factors are solved, the situation of casualties or waste of emergency response resources can be avoided. To achieve this effect, the present disclosure first determines the network status corresponding to the above environmental text information. Thus, it can be determined whether the current network can perform real-time data transmission. Second, in response to the above network status satisfying the first preset condition, load the preset feature configuration information set; based on the above preset feature configuration information set and the above extraction feature set, generate a meteorological emergency evaluation corresponding to the above extraction feature set. Third, based on the above meteorological emergency evaluation, generate a meteorological emergency warning and push the above meteorological emergency warning to the associated user terminal. Thus, based on edge computing, a temporary emergency warning can be generated and the emergency warning can be pushed. Fourth, based on the historical query information set, select each extraction feature that meets the preset query condition from the above extraction feature set, and mark each selected extraction feature to generate a marked extraction feature. Thus, the extraction features with a higher query frequency can be selected. Fifth, in response to detecting that the network status satisfies the second preset condition, determine the network transmission speed corresponding to the above network status; based on the above network transmission speed, select the preset compression algorithm corresponding to the above network transmission speed from the preset compression algorithm library, and perform compression processing on each marked extraction feature and the extraction feature set after removing each marked extraction feature; transmit the compressed marked extraction features to the target server according to the preset transmission protocol. Thus, different compression algorithms can be selected according to different network transmission speeds, so as to more quickly transmit the extraction features to the server for more accurate calculation. In combination with the optional steps below, based on the above disaster response information, control the associated emergency response device to execute the corresponding disaster emergency response task. Thus, the negative impact caused by network delay can be reduced, and the situation of casualties or waste of emergency response resources can be avoided

[0047] Step 103: For each extraction feature in the extraction feature group, classify the extraction feature to generate at least one local feature and at least one global feature.

[0048] In some embodiments, the above execution subject may classify each extraction feature in the above extraction feature group to generate at least one local feature and at least one global feature.

[0049] Step 104: Preprocess at least one local feature and at least one global feature to generate preprocessed local features and global features, obtaining a local feature group and a global feature group.

[0050] In some embodiments, the above-mentioned execution entity may preprocess the above-mentioned at least one local feature and the above-mentioned at least one global feature to generate preprocessed local features and global features, obtaining a local feature group and a global feature group.

[0051] In practice, the preprocessing can be performed through the following steps:

[0052] First step: Perform synonym replacement on the minority classes among the above-mentioned at least one local feature and the above-mentioned at least one global feature.

[0053] Second step: Add noise to each local feature and each global feature to generate local features and global features after adding noise.

[0054] Step 105: Input the local feature group and the global feature group into a pre-trained large language model to obtain disaster information and disaster response information corresponding to the disaster information.

[0055] In some embodiments, the above-mentioned execution entity may input the above-mentioned local feature group and the above-mentioned global feature group into a pre-trained large language model to obtain disaster information and disaster response information corresponding to the disaster information. Among them, the above-mentioned large language model includes: an input processing layer, an encoder layer, a decoder layer, and an output layer.

[0056] In practice, the above-mentioned output layer can be trained through the following steps:

[0057] First step: Receive at least one output vector sent by the decoder included in the large language model. Among them, the above-mentioned decoder may be composed of 28 Transformer blocks, and each Transformer block includes a self-attention mechanism and a feed-forward neural network. The output vectors among the above-mentioned at least one output vector correspond to dimension information. The above-mentioned dimension information may characterize the feature dimension.

[0058] Second step: Based on the above-mentioned feature fusion module, perform enhancement processing on the above-mentioned at least one output vector to generate an enhanced output vector set. Among them, the above-mentioned enhancement processing may be feature fusion processing. The above-mentioned feature fusion module includes a local feature unit and a global feature unit.

[0059] In practice, the above-mentioned at least one output vector can be enhanced through the following steps:

[0060] In the first step, input at least one of the above output vectors into the local feature unit included in the above feature fusion module to obtain the local feature set corresponding to the above output vector. In practice, the local feature can be obtained through the following formula:

[0061] a i = softmax(W q ·h i ).

[0062] where a i represents the i-th local feature. W q represents a learnable query matrix of 4096×4096 dimensions. h i represents the i-th output vector.

[0063] In the second step, based on the above local feature set and the above global feature unit, generate a global feature set. Here, the global feature can be obtained by weighting. In practice, the global feature can be generated through the following formula:

[0064] g = ∑a i ·h i .

[0065] where g represents the global feature.

[0066] In the third step, for each local feature in the above local feature set, select the global feature corresponding to the above local feature from the above global feature set, and perform feature fusion processing on the above local feature and the above global feature to generate a fused feature as the enhanced output vector. Here, the following formula can be used to perform feature fusion processing on the global feature and the local feature:

[0067] f = LayerNorm(W f ·[h; g] + b f ).

[0068] where W f represents a fusion matrix of 4096×8192 dimensions. b f represents a bias vector of 4096 dimensions. f represents the enhanced output vector. [h; g] can be a concatenated feature obtained by concatenating the local feature and the global feature.

[0069] In the process of adopting the technical solution to solve the above technical problem, the following technical problem often accompanies: When fusing the local feature with the flattened or pooled global feature in the channel dimension, redundancy may be introduced, resulting in the fused feature not matching the actual required feature, and further causing the determined disaster level to be inaccurate, resulting in casualties or waste of emergency response resources.

[0070] Optionally, the above-mentioned local features and the above-mentioned global features can be subjected to feature fusion processing through the following sub-steps:

[0071] The first sub-step is to perform upsampling on the above-mentioned global features to generate upsampled global features. Here, bilinear interpolation or transposed convolution can be used for upsampling.

[0072] The second sub-step is to determine the number of feature channels of the above-mentioned global features and the above-mentioned local features.

[0073] The third sub-step is to, in response to determining that the number of feature channels of the above-mentioned upsampled global features and the above-mentioned local features is inconsistent, adjust the number of channels corresponding to the above-mentioned upsampled global features so that the number of channels of the above-mentioned upsampled global features and the above-mentioned local features is the same. Here, the number of channels of the upsampled global features can be adjusted through 1×1 convolution.

[0074] The fourth sub-step is to perform a first convolution on the above-mentioned local features to generate first-convolved local features. In practice, 1×1 convolution can be performed on the above-mentioned local features.

[0075] The fifth sub-step is to perform element-wise addition on the above-mentioned first-convolved local features and the above-mentioned upsampled global features to generate added features.

[0076] The sixth sub-step is to perform a second convolution on the above-mentioned added features to generate convolved fusion features, obtaining convolved features. In practice, the second convolution can be performed through 3×3 convolution to obtain convolved features.

[0077] The seventh sub-step is to perform batch normalization on the above-mentioned convolved features to generate batch-normalized convolved features as the enhanced output vector.

[0078] The above first sub-step to seventh sub-step are an inventive point of the embodiment of the present disclosure. Combined with the optional steps below, it solves the technical problem: "When splicing and fusing local features with flattened or pooled global features in the channel dimension, redundancy may be introduced, resulting in the fused features not matching the actual required features, thereby causing the determined disaster level to be inaccurate, resulting in casualties or waste of emergency response resources." The reasons for casualties or waste of emergency response resources are as follows: When splicing and fusing local features with flattened or pooled global features in the channel dimension, redundancy may be introduced, resulting in the fused features not matching the actual required features, thereby causing the determined disaster level to be inaccurate, resulting in casualties or waste of emergency response resources. If the above factors are solved, the situation of casualties or waste of emergency response resources can be avoided. To achieve this effect, the present disclosure first performs upsampling on the above global features to generate upsampled global features. Thus, the resolution of the global features can be enlarged. Second, determine the number of feature channels of the above global features and the above local features; in response to determining that the number of feature channels of the above upsampled global features and the above local features are inconsistent, adjust the number of channels corresponding to the above upsampled global features so that the number of channels of the above upsampled global features and the above local features are the same. Thus, the number of channels of the global features and the local features can be adjusted to be the same. Third, perform first convolution processing on the above local features to generate first-convolved local features; perform element-wise addition processing on the above first-convolved local features and the above upsampled global features to generate added features. Thus, the global features and the local features can be fused by element-wise addition, thereby enhancing the response of the key area. Fourth, perform second convolution processing on the above added features to generate convolved fused features, obtaining convolved features. Thus, the fused features can be further convolved to integrate local details and global semantics to generate the final feature map. Fifth, perform batch normalization processing on the above convolved features to generate batch-normalized convolved features as the enhanced output vector. Thus, the fusion of local features and global features is completed, and the robustness of the model is improved through batch normalization processing. Combined with the step "step 106" below, based on the disaster information and disaster response information, generate a disaster push page, and push the disaster push page to the associated user terminal for display. Thus, the disaster level can be accurately determined, thereby avoiding casualties or waste of emergency response resources.

[0079] Fourth, determine the generated enhanced output vectors as the enhanced output vector set.

[0080] In the third step, through the above-mentioned feature fusion module, noise suppression processing is performed on each enhanced output vector in the above-mentioned enhanced output vector set to generate a suppressed enhanced output vector as the suppressed output feature, and a suppressed output feature set is obtained.

[0081] In some optional implementation manners of some embodiments, the above-mentioned execution subject may perform noise suppression processing through the following steps:

[0082] In the first step, for each enhanced output vector in the above-mentioned enhanced output vector set, the following suppression steps are performed:

[0083] In the first suppression step, according to the gating unit included in the above-mentioned feature fusion module, the feature importance value corresponding to the above-mentioned enhanced output vector is determined. Here, the feature importance value corresponding to the above-mentioned enhanced output vector can be determined through the following formula:

[0084] n = σ(W n ·f + b n )。

[0085] Where, W n represents a 4096×4096 dimensional transformation matrix. b n represents the bias term. f represents the enhanced output vector.

[0086] In the second suppression step, based on the above-mentioned feature importance, feature screening processing is performed on the above-mentioned enhanced output vector to generate a screened output vector. Among them, the above-mentioned feature screening processing can be expressed as f′ = g⊙f. d

[0087] In the third suppression step, the above-mentioned enhanced output vector and the above-mentioned screened output vector are combined to generate a combined vector as the suppressed output feature. Among them, the above-mentioned combination processing can be expressed as o = f + f′.

[0088] Step 104, based on the hierarchical classification module, classify each suppressed output feature in the suppressed output feature set to generate a classified output feature set.

[0089] In some embodiments, the above-mentioned execution subject may classify each suppressed output feature in the above-mentioned suppressed output feature set based on the above-mentioned hierarchical classification module to generate a classified output feature set. Among them, the classified output features in the above-mentioned classified output feature set correspond to classification information. The above-mentioned hierarchical classification module includes a main classifier and a fine-grained classifier.

[0090] In practice, the following steps can be used to classify each suppressed output feature in the above-mentioned suppressed output feature set:

[0091] First step, for each of the suppressed output features in the above-mentioned suppressed output feature set, perform the following processing steps:

[0092] First processing step, perform dimensionality reduction on the above-mentioned suppressed output feature to generate a dimensionality-reduced feature. In practice, the dimensionality reduction of the above-mentioned suppressed output feature can be performed through the linear layer included in the main classifier of the above-mentioned hierarchical classification module:

[0093] FC1(x) = ReLU(W1·x + b1).

[0094] Among them, W1 represents a 2048×4096-dimensional matrix. x represents the suppressed output feature. b1 represents the bias term in the FC1 function.

[0095] Second processing step, map the above-mentioned dimensionality-reduced feature into a preset category space to generate a mapped feature. Here, the above-mentioned dimensionality-reduced feature can be mapped into the preset category space through the following formula:

[0096] FC2(FC1(x)) = W2·FC1(x) + b2.

[0097] Among them, W2 represents a custom-dimensional matrix, and the dimension of the custom-dimensional matrix is set according to the number of main categories. b2 represents the bias term in the FC2 function.

[0098] Third processing step, based on the fine-grained classifier included in the above-mentioned hierarchical classification module, perform fine-grained classification processing on the above-mentioned mapped feature to generate a fine-grained classified mapped feature as the classified output feature.

[0099] Fourth step, input the above-mentioned classified output feature set into the above-mentioned loss function module to obtain a set of feature loss values. Among them, the above-mentioned loss function module includes: a main classification loss function, a fine-grained classification loss function, and a feature consistency loss function.

[0100] In some optional implementation manners of some embodiments, the above-mentioned main classification loss function, fine-grained classification loss function, and feature consistency loss function can be expressed by the following formula:

[0101] Main classification loss function:.

[0102]

[0103] Among them, α can be set to 0.25. γ can be initially set to 2. p represents the probability value predicted by the model. Here, the class weight α uses the reciprocal of the class frequency as the initial weight and is smoothed by the exponential function: Normalization ensures that the sum of the weights is 1. γ increases with the number of training rounds for adaptively adjusting the weights of hard samples.

[0104] Fine-grained classification loss function:

[0105]

[0106] Among them, ∈ can be set to 0.1. y true represents the true label. N represents the total number of categories.

[0107] Feature consistency loss function:

[0108]

[0109] Among them, F local and F global respectively represent the probability distributions of local features and global features.

[0110] The total loss is obtained through weighted combination:

[0111]

[0112] Among them, each weight coefficient in the above total loss is determined by grid search optimization.

[0113] Step 5: Based on the above feature loss value set, perform optimization and parameter adjustment on the above large language model.

[0114] Optionally, the above output layer can be trained through the following strategy: First, set the output layer batch size to 32, use the AdamW optimizer, set the initial learning rate to 2×10 -5 , adopt a linear learning rate decay strategy with warmup, set the weight decay coefficient to 0.01 for regularization, set the dropout rate to 0.1, and perform early stopping through the performance of the validation set during training, with patience set to 3 epochs. There is no restriction on the selection of key parameters, and they can be parameters obtained through experiments.

[0115] Step 106: Based on the disaster information and disaster response information, generate a disaster push page and push the disaster push page to the associated user terminals for display.

[0116] In some embodiments, the above execution entity can generate a disaster push page based on the above disaster information and the above disaster response information, and push the above disaster push page to the associated user terminals for display. Among them, the above associated user terminals can be users within a preset area range who are connected to the above execution entity through wired or wireless connections. The above preset area range can be a certain administrative division set in advance.

[0117] Optionally, after step 106, based on the above-mentioned disaster response information, control the associated emergency response equipment to perform the corresponding disaster emergency response tasks.

[0118] In some embodiments, the above-mentioned execution entity may control the associated emergency response equipment to perform the corresponding disaster emergency response tasks based on the above-mentioned disaster response information. Among them, the above-mentioned emergency response equipment may be equipment for coping with a certain disaster emergency response task. For example, if the above-mentioned disaster information is that a wildfire occurs due to high temperature weather, and the above-mentioned disaster emergency response task is to extinguish the fire in a region, then the above-mentioned emergency response equipment may be a sprinkler drone.

[0119] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: Through the disaster information push method based on a large language model in some embodiments of the present disclosure, the disaster level is accurately determined, avoiding casualties or waste of emergency response resources. Specifically, the reasons for inaccurate determination of the disaster level, resulting in casualties or waste of emergency response resources are as follows: When classifying meteorological environment data through a convolutional neural network to determine the disaster level, for structured data, unstructured data, and geographical information data of the meteorological environment, the convolutional neural network cannot effectively fuse multi-dimensional features to determine the disaster level, resulting in inaccurate determination of the disaster level and causing casualties or waste of emergency response resources. Based on this, in some embodiments of the present disclosure, the disaster information push method based on a large language model first obtains environmental text information. Thus, meteorological text data can be obtained from a meteorological monitoring center. Secondly, based on the above-mentioned environmental text information, perform feature extraction processing to generate multiple extracted features, obtaining an extracted feature group. Thus, data features in the text data can be extracted. After that, for each extracted feature in the above-mentioned extracted feature group, classify the above-mentioned extracted feature to generate at least one local feature and at least one global feature. Thus, global features and local features can be distinguished. Then, preprocess the above-mentioned at least one local feature and the above-mentioned at least one global feature to generate preprocessed local features and global features, obtaining a local feature group and a global feature group. Thus, the local features and global features can be preprocessed. Then, input the above-mentioned local feature group and the above-mentioned global feature group into a pre-trained large language model to obtain disaster information and disaster response information corresponding to the above-mentioned disaster information. Thus, through the large language model, multi-dimensional data such as structured data, unstructured data, and geographical information of the object environment are fused and analyzed, so that the disaster level corresponding to the meteorological environment can be accurately determined, thereby avoiding casualties or waste of emergency response resources.

[0120] Further reference Figure 2, as an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a disaster information push device based on a large language model. These device embodiments correspond to Figure 1 the method embodiments shown. The disaster information push device based on the large language model can be specifically applied to various electronic devices.

[0121] As Figure 2 shown, the disaster information push device 200 based on the large language model in some embodiments includes: a receiving unit 201, an enhancement unit 202, a noise suppression unit 203, a classification unit 204, an input unit 205, and an optimization parameter adjustment unit 206. Among them, the receiving unit 201 is configured to receive at least one output vector sent by a decoder included in the large language model, where the output vectors in the at least one output vector correspond to dimension information; the enhancement unit 202 is configured to perform enhancement processing on the at least one output vector based on the above-mentioned feature fusion module to generate an enhanced output vector set; the noise suppression unit 203 is configured to perform noise suppression processing on each enhanced output vector in the enhanced output vector set through the above-mentioned feature fusion module to generate a suppressed enhanced output vector as a suppressed output feature, and obtain a suppressed output feature set; the classification unit 204 is configured to perform classification processing on each suppressed output feature in the suppressed output feature set based on the above-mentioned hierarchical classification module to generate a classified output feature set, where the classified output features in the classified output feature set correspond to classification information; the input unit 205 is configured to input the classified output feature set into the above-mentioned loss function module to obtain a set of feature loss values; the optimization parameter adjustment unit 206 is configured to perform optimization parameter adjustment processing on the large language model based on the set of feature loss values.

[0122] It can be understood that the units described in the disaster information push device 200 based on the large language model correspond to Figure 1 the respective steps in the method described with reference to. Thus, the operations, features, and beneficial effects described above for the method also apply to the disaster information push device 200 based on the large language model and the units included therein, and will not be repeated here.

[0123] Next, refer to Figure 3 , which shows a schematic structural diagram of an electronic device 300 suitable for implementing some embodiments of the present disclosure. The electronic devices in some embodiments of the present disclosure may include, but are not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Tablet Computers), PMPs (Portable Multimedia Players), in-vehicle terminals (such as in-vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 3The illustrated electronic device is merely an example and should not impose any limitations on the functions and usage scope of the embodiments of the present disclosure.

[0124] As Figure 3 shown, the electronic device 300 may include a processing device 301 (such as a central processing unit, a graphics processing unit, etc.), which may perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic device 300 are also stored. The processing device 301, the ROM 302, and the RAM 303 are connected to each other through a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0125] Generally, the following devices may be connected to the I / O interface 305: an input device 306 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 308 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 309. The communication device 309 may allow the electronic device 300 to communicate with other devices wirelessly or wiredly to exchange data. Although Figure 3 the electronic device 300 with various devices is shown, it should be understood that it is not required to implement or include all the shown devices. Instead, more or fewer devices may be implemented or included. Figure 3 Each block shown in

[0126] particular, according to some embodiments of the present disclosure, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, some embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes program codes for performing the methods shown in the flowcharts. In such some embodiments, the computer program may be downloaded and installed from a network through the communication device 309, or installed from the storage device 308, or installed from the ROM 302. When the computer program is executed by the processing device 301, the above functions defined in the methods of some embodiments of the present disclosure are performed.

[0127] It should be noted that the computer-readable medium described in some embodiments of the present disclosure may be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0128] In some embodiments, the client and the server may communicate using any currently known or future-developed network protocol such as HTTP (HyperText Transfer Protocol), and may be interconnected with digital data communication in any form or medium (e.g., a communication network). Examples of communication networks include local area networks ("LANs"), wide area networks ("WANs"), the Internet (e.g., the Internet), and end-to-end networks (e.g., ad hoc end-to-end networks), as well as any currently known or future-developed networks.

[0129] The above computer-readable medium may be included in the above electronic device; or it may exist separately without being assembled into the electronic device. The above computer-readable medium carries one or more programs. When the above one or more programs are executed by the electronic device, the electronic device is caused to: obtain environmental text information, where the above environmental text information is text information for characterizing environmental information received from a target terminal through a preset interface. Based on the above environmental text information, perform feature extraction processing to generate a plurality of extracted features, obtaining an extracted feature group. For each extracted feature in the above extracted feature group, classify the above extracted feature to generate at least one local feature and at least one global feature. Preprocess the above at least one local feature and the above at least one global feature to generate preprocessed local features and global features, obtaining a local feature group and a global feature group. Input the above local feature group and the above global feature group into a pre-trained large language model to obtain disaster information and disaster response information corresponding to the above disaster information. Based on the above disaster information and the above disaster response information, generate a disaster push page, and push the above disaster push page to an associated user terminal for display.

[0130] Computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages or combinations thereof. The above programming languages include object-oriented programming languages - such as Java, Smalltalk, C++; and also include conventional procedural programming languages - such as the "C" language or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network - including a local area network (LAN) or a wide area network (WAN) - or may be connected to an external computer (e.g., by using an Internet service provider to connect through the Internet).

[0131] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a part of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in an order different from that marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, as well as combinations of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0132] The units described in some embodiments of the present disclosure can be implemented in software or in hardware. The described units can also be provided in a processor. For example, it can be described as: a processor includes an acquisition unit, an execution unit, a classification unit, a preprocessing unit, an input unit, and a generation unit. Among them, the names of these units do not constitute a limitation to the unit itself in some cases. For example, the acquisition unit can also be described as "the unit for acquiring environmental text information".

[0133] The functions described above can be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that can be used include: Field Programmable Gate Array (FPGA), Application Specific Integrated Circuit (ASIC), Application Specific Standard Product (ASSP), System on Chip (SOC), Complex Programmable Logic Device (CPLD), and so on.

[0134] The above description is only some preferred embodiments of the present disclosure and an explanation of the technical principles applied. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with the (but not limited to) technical features having similar functions disclosed in the embodiments of the present disclosure.

Claims

1. A disaster information push method based on a large language model, comprising: Obtaining environmental text information, where the environmental text information is text information for characterizing environmental information received from a target terminal through a preset interface; Performing feature extraction processing based on the environmental text information to generate a plurality of extracted features and obtaining an extracted feature set; For each extracted feature in the extracted feature set, classifying the extracted feature to generate at least one local feature and at least one global feature; Preprocessing the at least one local feature and the at least one global feature to generate preprocessed local features and global features, obtaining a local feature set and a global feature set; Inputting the local feature set and the global feature set into a pre-trained large language model to obtain disaster information and disaster response information corresponding to the disaster information; Generating a disaster push page based on the disaster information and the disaster response information, and pushing the disaster push page to an associated user terminal for display.

2. The method according to claim 1, wherein: The large language model includes: an input processing layer, an encoder layer, a decoder layer, and an output layer; and The method further includes: Controlling an associated emergency response device to execute a corresponding disaster emergency response task based on the disaster response information.

3. The method according to claim 2, wherein: The output layer includes: a feature fusion module, a hierarchical classification module, and a loss function module; And the output layer is obtained through the following steps: Receiving at least one output vector sent by a decoder included in the large language model, where the output vector in the at least one output vector corresponds to dimension information; Performing enhancement processing on the at least one output vector based on the feature fusion module to generate an enhanced output vector set; Performing noise suppression processing on each enhanced output vector in the enhanced output vector set through the feature fusion module to generate a suppressed enhanced output vector as a suppressed output feature, obtaining a suppressed output feature set; Performing classification processing on each suppressed output feature in the suppressed output feature set based on the hierarchical classification module to generate a classified output feature set, where the classified output feature in the classified output feature set corresponds to classification information; Inputting the classified output feature set into the loss function module to obtain a feature loss value set; Performing optimization parameter adjustment processing on the large language model based on the feature loss value set.

4. The method according to claim 3, wherein, The feature fusion module includes a local feature unit and a global feature unit; And The performing enhancement processing on the at least one output vector based on the feature fusion module to generate an enhanced output vector set includes: Inputting the at least one output vector into the local feature unit included in the feature fusion module to obtain a local feature set corresponding to the output vector; Generating a global feature set based on the local feature set and the global feature unit; For each local feature in the local feature set, select the corresponding global feature from the global feature set, and perform feature fusion processing on the local feature and the global feature to generate a fused feature as the enhanced output vector. Determine the generated enhanced output vectors as the enhanced output vector set.

5. The method according to claim 3, wherein The loss function module includes: a main classification loss function, a fine-grained classification loss function, and a feature consistency loss function.

6. The method according to claim 3, wherein: Through the feature fusion module, perform noise suppression processing on each enhanced output vector in the enhanced output vector set to generate a suppressed enhanced output vector as the suppressed output feature, and obtain the suppressed output feature set, including: For each enhanced output vector in the enhanced output vector set, perform the following suppression steps: Determine the feature importance value corresponding to the enhanced output vector according to the gating unit included in the feature fusion module. Based on the feature importance, perform feature screening processing on the enhanced output vector to generate a screened output vector. Combine the enhanced output vector and the screened output vector to generate a combined vector as the suppressed output feature.

7. The method according to claim 3, wherein The hierarchical classification module includes a main classifier and a fine-grained classifier; and Based on the hierarchical classification module, perform classification processing on each suppressed output feature in the suppressed output feature set to generate a classified output feature set, including: For each suppressed output feature in the suppressed output feature set, perform the following processing steps: Perform dimensionality reduction processing on the suppressed output feature to generate a dimensionality-reduced feature. Map the dimensionality-reduced feature to a preset category space to generate a mapped feature. Based on the fine-grained classifier included in the hierarchical classification module, perform fine-grained classification processing on the mapped feature to generate a fine-grained classified mapped feature as the classified output feature.

8. A disaster information push device based on a large language model, including: An acquisition unit configured to acquire environmental text information, where the environmental text information is text information for characterizing environmental information received from a target terminal through a preset interface. An execution unit configured to perform feature extraction processing based on the environmental text information to generate multiple extracted features and obtain an extracted feature group. A classification unit configured to, for each extracted feature in the extracted feature group, perform classification processing on the extracted feature to generate at least one local feature and at least one global feature. A preprocessing unit configured to preprocess the at least one local feature and the at least one global feature to generate preprocessed local features and global features and obtain a local feature group and a global feature group. An input unit configured to input the local feature group and the global feature group into a pre-trained large language model to obtain disaster information and disaster response information corresponding to the disaster information. A generation unit configured to generate a disaster push page based on the disaster information and the disaster response information, and push the disaster push page to an associated user terminal for display.

9. An electronic device, comprising: One or more processors; A storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable medium having a computer program stored thereon, wherein, When the program is executed by the processor, the method according to any one of claims 1 to 7 is implemented.

Images