Public security-oriented situation analysis method, system and equipment and medium

Abstract

The invention provides a situation analysis method, system and device for public security and a medium, and the method comprises the steps: determining a target keyword for collecting situation data according to real-time preference, and obtaining a situation data set; constructing sentiment analysis and situation classification sub-models for the pre-trained large language model through knowledge distillation, and identifying a target event in the situation data set by adopting the sub-models; performing event cluster classification on the target event through streaming clustering to update an event cluster; and defining an early warning index to divide a target event in the event cluster, and outputting an event title and an event description of the target event through a large language model.

Description

Technical Field

[0001] The present invention relates to the technical field of public safety monitoring, and in particular to a situation analysis method, system, equipment and medium for public safety. Background Art

[0002] With the popularization of various smart terminals and the development of Internet technology, people can share current news, their own opinions and other information on various social platforms anytime and anywhere. Especially in some crowded areas such as university towns, science and technology parks, financial circles, business districts, etc., monitoring of public opinion trends on social networks is of great significance to public safety.

[0003] With the continuous development of artificial intelligence and big data technology, the existing public opinion analysis system has begun to use relevant technologies to process massive amounts of data, so as to warn and intervene in events that may cause public safety based on the results of the analysis. However, in the existing technology, public opinion analysis is usually targeted at a specific field. If the field is not limited, a large amount of manual intervention in massive data is required to solve the problem of poor generalization. In the data processing part, the existing unsupervised clustering method is less effective in processing the dynamic crawling network situation and cannot sensitively warn of events. The time clusters after clustering still use description methods such as keywords and subject analysis, which makes it difficult to make a simple and intuitive summary of the event and prevent the monitoring party from quickly understanding the content of the event. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention proposes a situation analysis method, system, device and medium for public safety.

[0005] The first aspect of the present invention discloses a situation analysis method for public safety, comprising:

[0006] Determine the target keywords for collecting situation data based on real-time preferences and obtain situation data sets;

[0007] The pre-trained large language model is used to construct sub-models for sentiment analysis and situation classification through knowledge distillation, and the sub-models are used to identify target events in the situation dataset;

[0008] The target event is classified into event clusters through streaming clustering to update the event clusters;

[0009] Define early warning indicators to divide target events within event clusters, and output the event title and event description of the target event through a large language model.

[0010] In an optional embodiment, determining the target keyword for collecting situation data according to the real-time preference includes:

[0011] According to historical situation information and prior experience, the situation options of the target area are determined, and the situation options are input into the large language model to generate situation-related words. According to the situation preference of the target area, the situation-related words are ranked and fused into topk contexts, which are spliced ​​into the situation options and then input into the large language model to generate target keywords.

[0012] In an optional embodiment, the acquiring of the situation data set includes:

[0013] Develop a multimodal data fusion strategy to fuse data of different modalities from Internet mobile phones, including text data, audio data, image data, and video data, into a unified text data representation.

[0014] In an optional embodiment, constructing a sub-model for sentiment analysis and situation classification by knowledge distillation on the pre-trained large language model includes:

[0015] The teacher models for sentiment analysis and situation classification are pre-trained by converting multimodal situation data into large-scale text data, so that the teacher model can learn the sentiment tendency and situation semantics in the context of situation data.

[0016] Through knowledge distillation, the knowledge of the teacher model is transferred to the new student model to construct sub-models for sentiment analysis and situation classification.

[0017] In an optional embodiment, the adopting of the sub-model to identify the target event in the situation data set includes:

[0018] Define situation types and sentiment classifications, determine potential warning events based on the situation types and sentiment classifications identified in the situation data, save situations that are not potential warning events to the situation library to support the identification of target events, and mark situations with potential warning events as target events for streaming clustering.

[0019] In an optional embodiment, the target event is classified into an event cluster by streaming clustering to update the event cluster, including:

[0020] Customize event cluster templates based on situation preferences and add them to the event cluster library;

[0021] Identify the event elements of the situation in the target event, and recall similar historical event clusters within a custom time range based on the situation release time for the target event that meets the event elements;

[0022] Define a similarity threshold and determine whether the similarity between the target event and the historical event cluster meets the similarity threshold. If so, merge the target event and the historical event cluster into a new historical event cluster. If not, generate a new event cluster for the target event and dynamically update the cluster center for each merged historical event cluster.

[0023] In an optional embodiment, the step of defining the early warning indicator to divide the target event in the event cluster and outputting the event title and event description of the target event through the large language model includes:

[0024] Early warning indicators are defined based on the proportion of negative emotional situations and the time density of situation releases, and prevention and control measures and intervention measures are set for target events designated as early warning events. In addition, event descriptions and event titles are output for target events designated as early warning events based on the emotional tendencies and situation semantics identified by the sub-model and the event cluster library.

[0025] A second aspect of the present invention discloses a situation analysis system for public safety, the system comprising:

[0026] A data acquisition module, used to determine the target keywords for collecting situation data according to real-time preferences and obtain situation data sets;

[0027] The data analysis module is used to construct sub-models for sentiment analysis and situation classification through knowledge distillation of the pre-trained large language model, and use the sub-models to identify target events in the situation dataset;

[0028] The streaming clustering module is used to classify target events into event clusters through streaming clustering to update event clusters;

[0029] The event warning module is used to define warning indicators to divide the target events within the event cluster, and output the event title and event description of the target event through a large language model.

[0030] In an optional embodiment, the data acquisition module includes:

[0031] The keyword module is used to determine the situation options of the target area based on historical situation information and prior experience, input the situation options into the large language model to generate situation-related words, rank and fuse the situation-related words according to the situation preferences of the target area into topk contexts, splice them into the situation options, and then input them into the large language model to generate target keywords.

[0032] In an optional embodiment, the data acquisition module includes:

[0033] The multimodal data fusion module is used to formulate a multimodal data fusion strategy. It fuses the multimodal data of Internet mobile phones, including text data, audio data, image data and video data, into a unified text data representation.

[0034] In an optional embodiment, the data analysis module includes:

[0035] The knowledge transfer module is used to pre-train the teacher models for sentiment analysis and situation classification respectively by converting the multimodal situation data into large-scale text data, so that the teacher model can learn the sentiment tendency and situation semantics in the context of the situation data; and transfer the knowledge of the teacher model to the new student model through knowledge distillation to construct the sub-models of sentiment analysis and situation classification.

[0036] In an optional embodiment, the data analysis module includes:

[0037] The sentiment analysis and situation classification module is used to define situation types and sentiment classifications, determine potential warning events based on the situation types and sentiment classifications identified in the situation data, save the situations of non-potential warning events to the situation library to support the identification of target events, and mark the situations with potential warning events as target events for streaming clustering.

[0038] In an optional embodiment, the streaming clustering module includes:

[0039] The streaming clustering submodule is used to add custom event cluster templates to the event cluster library based on situation preferences; identify the event elements of the situation in the target event, and recall similar historical event clusters within the custom time range for the target event that meets the event elements based on the situation release time; define a similarity threshold, and determine whether the similarity between the target event and the historical event cluster meets the similarity threshold based on the similarity threshold. If so, merge the target event and the historical event cluster into a new historical event cluster. If not, generate a new event cluster for the target event, and dynamically update the cluster center for each merged historical event cluster.

[0040] In an optional embodiment, the event warning module includes:

[0041] The event warning submodule is used to define warning indicators based on the proportion of negative emotional situations and the time density of situation releases, set prevention and control measures and intervention measures for target events designated as warning events, and output event descriptions and event titles for target events designated as warning events based on the emotional tendencies and situation semantics identified by the submodel combined with the event cluster library.

[0042] The third aspect of the present invention discloses a situation analysis device for public safety, comprising:

[0043] at least one processor, and

[0044] a memory communicatively connected to the at least one processor; wherein,

[0045] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the public safety-oriented situation analysis method as described in any one of the first aspects of the present invention.

[0046] The fourth aspect of the present invention discloses a computer-readable storage medium, which stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the public safety-oriented situation analysis method as described in any one of the items disclosed in the first aspect of the present invention.

[0047] Compared with the prior art, the present invention has the following advantages:

[0048] The present invention proposes a situation analysis method, system, device and medium for public safety, the method comprising: determining the target keywords for collecting situation data according to real-time preferences, obtaining a situation data set; constructing a sub-model for sentiment analysis and situation classification through knowledge distillation for a pre-trained large language model, and using the sub-model to identify target events in the situation data set; classifying the target events into event clusters through streaming clustering to update the event clusters; defining warning indicators to divide the target events in the event clusters, and outputting the event titles and event descriptions of the target events through the large language model. The pre-trained knowledge of the large-scale large prediction model is used to direct the label data, greatly reducing the amount of manual labeling, and then the model is distilled to a small-scale model to improve the generalization of the model in non-specific fields, and the clustering results of the feedback situation are more accurately fed back using streaming clustering, making the event warning more sensitive, and after finding the center of the event cluster through vector distance, the large model summarizes the description of the most representative event according to the event cluster, so that the content of the event cluster can be intuitively understood, so that the monitoring party can quickly respond to the public safety events in the situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0050] Figure 1 A flow chart of a situation analysis method for public safety according to the present invention;

[0051] Figure 2 It is a schematic diagram of the situation analysis method for public safety of the present invention.

[0052] Figure 3 It is a schematic diagram of the situation analysis system for public safety of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Based on the embodiments provided in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0054] Obviously, the drawings described below are only some examples or embodiments of the present application. For ordinary technicians in this field, the present application can also be applied to other similar scenarios based on these drawings without creative work. In addition, it can also be understood that although the efforts made in this development process may be complicated and lengthy, for ordinary technicians in this field related to the content disclosed in this application, some changes in design, manufacturing or production based on the technical content disclosed in this application are just conventional technical means, and should not be understood as insufficient content disclosed in this application.

[0055] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0056] Unless otherwise defined, the technical terms or scientific terms involved in this application should be understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the" and the like involved in this application do not indicate a quantitative limitation, and may represent the singular or plural. The terms "include", "comprise", "have" and any of their variations involved in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there may be three relationships, for example, "A and / or B" can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0057] Example 1

[0058] See also Figure 1-Figure 2 The embodiment of the present invention discloses a situation analysis method for public safety, including:

[0059] 101. Determine the target keywords for collecting situation data according to the real-time preference and obtain the situation data set;

[0060] It should be noted that usually more attention should be paid to public safety issues in places where people gather densely, such as university towns, science and technology parks, financial circles, business districts and other crowded areas. By collecting content on social networks, such as social media, news websites, forums, BBS, some social apps, etc., we can extract the public's emotional tendencies and reflect the overall picture of public opinion in the region as much as possible.

[0061] In an optional embodiment, determining the target keyword for collecting situation data according to the real-time preference includes:

[0062] According to historical situation information and prior experience, the situation options of the target area are determined, and the situation options are input into the large language model to generate situation-related words. According to the situation preference of the target area, the situation-related words are ranked and fused into topk contexts, which are spliced ​​into the situation options and then input into the large language model to generate target keywords.

[0063] Specifically, taking the university town as an example, the collection area is mainly composed of students. The scope of the collected situation data can be narrowed according to the students' preference for using social networks. For example, current college students are keen on expressing their personal emotions on social apps such as Weibo and Xiaohongshu, and some public opinion trends displayed by other netizens participating in discussions, as well as screening keywords through prior experience. For example, the theft of non-motor vehicles and takeout incidents often occur in university towns, so the target keywords can be locked. Furthermore, in order to avoid the problem of poor generalization of the obtained data, the basic situation of the target area is understood through the large language model, and then more optional situation keywords are tried to be generated. In order to avoid the large language model from paying too much attention to a certain situation keyword, the input is updated twice through ranking fusion, and finally keywords that are easy to discover potential situation dangers are generated as much as possible to obtain situation data. This process can purify the collected data and improve the collection efficiency, and reduce the interference of white noise data.

[0064] In an optional embodiment, the acquiring of the situation data set includes:

[0065] Develop a multimodal data fusion strategy to fuse data of different modalities from Internet mobile phones, including text data, audio data, image data, and video data, into a unified text data representation.

[0066] Specifically, in order to avoid missing some situation data and thus failing to detect potential warning events, in addition to formulating keywords, it is also necessary to consider the type of situation data to be obtained, such as the audio data in the comment area of ​​a social APP, the existence of invisible dangers in the captions of images, or the appearance of negative or even anti-social emotions in the released videos. In order to facilitate the subsequent model to process the situation data, the multimodal data is fused and unified into text data. For example, a speech-to-text tool can be used for audio data, while image data and video data can be converted into text data through feature extraction. Since the target keywords are determined before data collection, the fused text data does not need to be manually labeled, reducing resource consumption.

[0067] 102. Use knowledge distillation to construct sub-models for sentiment analysis and situation classification for the pre-trained large language model, and use the sub-models to identify target events in the situation dataset;

[0068] In an optional embodiment, constructing a sub-model for sentiment analysis and situation classification by knowledge distillation on the pre-trained large language model includes:

[0069] The teacher models for sentiment analysis and situation classification are pre-trained by converting multimodal situation data into large-scale text data, so that the teacher model can learn the sentiment tendency and situation semantics in the context of situation data.

[0070] Through knowledge distillation, the knowledge of the teacher model is transferred to the new student model to construct sub-models for sentiment analysis and situation classification.

[0071] Specifically, traditional methods of sentiment analysis and situation classification are mainly based on statistical and machine learning methods. These methods often rely on feature engineering and manually extracted features when dealing with sentiment analysis and situation classification tasks. However, these methods may encounter challenges in dealing with complex language phenomena. In contrast, large language models have shown significant advantages in the fields of sentiment analysis and situation classification. Large language models can learn weak emotional tendencies in contexts through pre-training on large-scale text data. The situation analysis system for public security and public safety uses a large amount of situation data to distill knowledge into a small-sized large language model, while ensuring the model effect and improving processing efficiency. The large language model extracts rich situation semantic information to increase the accuracy of situation judgment. Like sentiment analysis, the situation category analysis process also distills knowledge into a small-sized large language model, while enhancing the situation category judgment effect and processing speed.

[0072] In an optional embodiment, the adopting of the sub-model to identify the target event in the situation data set includes:

[0073] Define situation types and sentiment classifications, determine potential warning events based on the situation types and sentiment classifications identified in the situation data, save situations that are not potential warning events to the situation library to support the identification of target events, and mark situations with potential warning events as target events for streaming clustering.

[0074] Specifically, still taking the situation analysis of the university town area as an example, the situation type can be defined as including but not limited to mental health, campus safety, public safety, natural disasters and social hotspots, and the sentiment classification can be defined as including but not limited to positive emotions, negative emotions, positive emotions, negative emotions and neutral emotions. During identification, the sub-model identifies potential threats in the situation data through the learned sentiment tendencies and situation semantics. For example, if there are negative emotions and negative emotions in the campus safety situation and the proportion is high, the campus safety situation can be marked as a target event. Since the situation is temporal and spatial, that is, the fermentation of the event is not completed at a certain moment, nor is it fermented at a certain location, therefore, the traditional clustering method cannot meet the needs of such scenarios and needs to be processed through streaming clustering.

[0075] 103. The target event is classified into event clusters through streaming clustering to update the event cluster;

[0076] In an optional embodiment, the target event is classified into an event cluster by streaming clustering to update the event cluster, including:

[0077] Customize event cluster templates based on situation preferences and add them to the event cluster library;

[0078] Identify the event elements of the situation in the target event, and recall similar historical event clusters within a custom time range based on the situation release time for the target event that meets the event elements;

[0079] Define a similarity threshold and determine whether the similarity between the target event and the historical event cluster meets the similarity threshold. If so, merge the target event and the historical event cluster into a new historical event cluster. If not, generate a new event cluster for the target event and dynamically update the cluster center for each merged historical event cluster.

[0080] Specifically, when a situation enters the event aggregation module, it needs to go through the following steps: determine whether the situation has event elements; if the situation meets the event elements, recall the most similar historical event cluster within the custom time range based on the release time of the situation; if the similarity between the recalled most similar historical event cluster and the current situation is less than the custom threshold (the larger the value, the more similar), a new event cluster will be generated; otherwise, if the similarity between the recalled most similar historical event cluster and the current situation is greater than the custom threshold, the current situation and the historical event cluster will be merged into one cluster; dynamically update the cluster center of each cluster for the next round of recalling historical clusters; the situation entering the event aggregation step will repeat the above steps until it is assigned to a cluster. For the dynamic clustering of the continuously flowing situations, the situation clustering results can be monitored more accurately, thereby making event warnings more sensitive.

[0081] Furthermore, custom event clusters allow users to merge situations they are concerned about into an event cluster, which can be quickly responded to and classified into the event cluster of concern during streaming clustering, which will be conducive to the prevention and control of events other than automatically generated events.

[0082] 104. Define warning indicators to divide target events within event clusters, and output event titles and event descriptions of target events through a large language model.

[0083] In an optional embodiment, the step of defining the early warning indicator to divide the target event in the event cluster and outputting the event title and event description of the target event through the large language model includes:

[0084] Early warning indicators are defined based on the proportion of negative emotional situations and the time density of situation releases, and prevention and control measures and intervention measures are set for target events designated as early warning events. In addition, event descriptions and event titles are output for target events designated as early warning events based on the emotional tendencies and situation semantics identified by the sub-model and the event cluster library.

[0085] Specifically, in event aggregation, the situations describing the same event within the same time range are aggregated, but not all events need to be warned, so multiple indicators will be defined to warn events, such as the proportion of negative emotional situations in the event cluster, the time density of situation releases, etc. Once an event cluster is upgraded to a potential preventable event, external forces will be used for intervention, and at the same time, the event that best represents the event cluster will be found for large-scale model summary, making the clustered events more intuitive.

[0086] The present invention proposes a situation analysis method, system, device and medium for public safety, the method comprising: determining the target keywords for collecting situation data according to real-time preferences, obtaining a situation data set; constructing a sub-model for sentiment analysis and situation classification through knowledge distillation for a pre-trained large language model, and using the sub-model to identify target events in the situation data set; classifying the target events into event clusters through streaming clustering to update the event clusters; defining warning indicators to divide the target events in the event clusters, and outputting the event titles and event descriptions of the target events through the large language model. The pre-trained knowledge of the large-scale large prediction model is used to direct the label data, greatly reducing the amount of manual labeling, and then the model is distilled to a small-scale model to improve the generalization of the model in non-specific fields, and the clustering results of the feedback situation are more accurately fed back using streaming clustering, making the event warning more sensitive, and after finding the center of the event cluster through vector distance, the large model summarizes the description of the most representative event according to the event cluster, so that the content of the event cluster can be intuitively understood, so that the monitoring party can quickly respond to the public safety events in the situation.

[0087] like Figure 3 As shown, the second aspect of the present invention discloses a situation analysis system for public safety, the system comprising:

[0088] A data acquisition module, used to determine the target keywords for collecting situation data according to real-time preferences and obtain situation data sets;

[0089] The data analysis module is used to construct sub-models for sentiment analysis and situation classification through knowledge distillation of the pre-trained large language model, and use the sub-models to identify target events in the situation dataset;

[0090] The streaming clustering module is used to classify target events into event clusters through streaming clustering to update event clusters;

[0091] The event warning module is used to define warning indicators to divide the target events within the event cluster, and output the event title and event description of the target event through a large language model.

[0092] In an optional embodiment, the data acquisition module includes:

[0093] The keyword module is used to determine the situation options of the target area based on historical situation information and prior experience, input the situation options into the large language model to generate situation-related words, rank and fuse the situation-related words according to the situation preferences of the target area into topk contexts, splice them into the situation options, and then input them into the large language model to generate target keywords.

[0094] In an optional embodiment, the data acquisition module includes:

[0095] The multimodal data fusion module is used to formulate a multimodal data fusion strategy. It fuses the multimodal data of Internet mobile phones, including text data, audio data, image data and video data, into a unified text data representation.

[0096] In an optional embodiment, the data analysis module includes:

[0097] The knowledge transfer module is used to pre-train the teacher models for sentiment analysis and situation classification respectively by converting the multimodal situation data into large-scale text data, so that the teacher model can learn the sentiment tendency and situation semantics in the context of the situation data; and transfer the knowledge of the teacher model to the new student model through knowledge distillation to construct the sub-models of sentiment analysis and situation classification.

[0098] In an optional embodiment, the data analysis module includes:

[0099] The sentiment analysis and situation classification module is used to define situation types and sentiment classifications, determine potential warning events based on the situation types and sentiment classifications identified in the situation data, save the situations of non-potential warning events to the situation library to support the identification of target events, and mark the situations with potential warning events as target events for streaming clustering.

[0100] In an optional embodiment, the streaming clustering module includes:

[0101] The streaming clustering submodule is used to add custom event cluster templates to the event cluster library based on situation preferences; identify the event elements of the situation in the target event, and recall similar historical event clusters within the custom time range for the target event that meets the event elements based on the situation release time; define a similarity threshold, and determine whether the similarity between the target event and the historical event cluster meets the similarity threshold based on the similarity threshold. If so, merge the target event and the historical event cluster into a new historical event cluster. If not, generate a new event cluster for the target event, and dynamically update the cluster center for each merged historical event cluster.

[0102] In an optional embodiment, the event warning module includes:

[0103] The event warning submodule is used to define warning indicators based on the proportion of negative emotional situations and the time density of situation releases, set prevention and control measures and intervention measures for target events designated as warning events, and output event descriptions and event titles for target events designated as warning events based on the emotional tendencies and situation semantics identified by the submodel combined with the event cluster library.

[0104] The third aspect of the present invention discloses a situation analysis device for public safety, comprising:

[0105] at least one processor, and

[0106] a memory communicatively connected to the at least one processor; wherein,

[0107] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the public safety-oriented situation analysis method as described in any one of the first aspects of the present invention.

[0108] The computer device may be a terminal, and the computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a situation analysis method for public safety is implemented. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0109] The fourth aspect of the present invention discloses a computer-readable storage medium, which stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the public safety-oriented situation analysis method as described in any one of the items disclosed in the first aspect of the present invention.

[0110] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing related hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments of the situation analysis method for public safety. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0111] Alternatively, if the above modules of the present invention are implemented in the form of software function modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the relevant technology. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, terminal, or network device, etc.) to execute all or part of the methods of each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, RAM, ROM, magnetic disks or optical disks.

[0112] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A situation analysis method for public safety, characterized in that: The method comprises: Determine the target keywords for collecting situation data based on real-time preferences and obtain situation data sets; The pre-trained large language model is used to construct sub-models for sentiment analysis and situation classification through knowledge distillation, and the sub-models are used to identify target events in the situation dataset; The target event is classified into event clusters through streaming clustering to update the event clusters; Define early warning indicators to divide the target events within the event cluster, and output the event title and event description of the target event through a large language model.

2. The public safety-oriented situation analysis method according to claim 1, characterized in that: Determining target keywords for collecting situation data according to real-time preferences includes: According to historical situation information and prior experience, the situation options of the target area are determined, and the situation options are input into the large language model to generate situation-related words. According to the situation preference of the target area, the situation-related words are ranked and fused into topk contexts, which are spliced ​​into the situation options and then input into the large language model to generate target keywords.

3. The public safety-oriented situation analysis method according to claim 1, characterized in that: The step of acquiring a situation data set includes: Develop a multimodal data fusion strategy to fuse data of different modalities from Internet mobile phones, including text data, audio data, image data, and video data, into a unified text data representation.

4. The public safety-oriented situation analysis method according to claim 1, characterized in that: The pre-trained large language model is used to construct a sub-model for sentiment analysis and situation classification through knowledge distillation, including: The teacher models for sentiment analysis and situation classification are pre-trained by converting multimodal situation data into large-scale text data, so that the teacher model can learn the sentiment tendency and situation semantics in the context of situation data. Through knowledge distillation, the knowledge of the teacher model is transferred to the new student model to construct sub-models for sentiment analysis and situation classification.

5. The public safety-oriented situation analysis method according to claim 1, characterized in that: The method of using a sub-model to identify a target event in a situation data set includes: Define situation types and sentiment classifications, determine potential warning events based on the situation types and sentiment classifications identified in the situation data, save situations that are not potential warning events to the situation library to support the identification of target events, and mark situations with potential warning events as target events for streaming clustering.

6. The public safety-oriented situation analysis method according to claim 1, characterized in that: The target event is classified into event clusters through streaming clustering to update the event cluster, including: Customize event cluster templates based on situation preferences and add them to the event cluster library; Identify the event elements of the situation in the target event, and recall similar historical event clusters within a custom time range based on the situation release time for the target event that meets the event elements; Define a similarity threshold and determine whether the similarity between the target event and the historical event cluster meets the similarity threshold. If so, merge the target event and the historical event cluster into a new historical event cluster. If not, generate a new event cluster for the target event and dynamically update the cluster center for each merged historical event cluster.

7. The public safety-oriented situation analysis method according to claim 1, characterized in that: The definition of early warning indicators divides the target events in the event cluster, and outputs the event title and event description of the target event through the large language model, including: Early warning indicators are defined based on the proportion of negative emotional situations and the time density of situation releases, and prevention and control measures and intervention measures are set for target events designated as early warning events. In addition, event descriptions and event titles are output for target events designated as early warning events based on the emotional tendencies and situation semantics identified by the sub-model and the event cluster library.

8. A situation analysis system for public safety, characterized in that: The system comprises: A data acquisition module, used to determine the target keywords for collecting situation data according to real-time preferences and obtain situation data sets; The data analysis module is used to construct sub-models for sentiment analysis and situation classification through knowledge distillation of the pre-trained large language model, and use the sub-models to identify target events in the situation dataset; The streaming clustering module is used to classify target events into event clusters through streaming clustering to update event clusters; The event warning module is used to define warning indicators to divide the target events within the event cluster, and output the event title and event description of the target event through a large language model.

9. A situation analysis device for public safety, characterized in that: include: at least one processor, and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the public safety-oriented situation analysis method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the public safety-oriented situation analysis method as described in any one of claims 1 to 7.

Images