Specified event association analysis and prediction method based on graph model

Abstract

The invention relates to the technical field of computer data processing, and discloses a graph model-based specified event association analysis and prediction method, which comprises the following steps of: obtaining multi-modal event data from different message event sources in a specified time period; performing identification processing on the multi-modal event data corresponding to each message event source to obtain corresponding text type description information; obtaining a character feature vector of the text type description information based on a natural language processing technology; according to the character feature vectors, the correlation degree between the text type description information corresponding to every two message event sources is calculated; graph model visualization processing is carried out on the multi-modal event data based on the calculation result of the correlation degree, and a correlation time graph is obtained; and gathering the associated time atlas and then sending the gathered associated time atlas to a trend prediction model, and obtaining a prediction result output by the trend prediction model, thereby reducing the pressure of manually screening a large amount of information, and improving the alarm processing and judgment speed.

Description

Technical Field

[0001] This application relates to the technical field of computer data processing, and in particular to a method for analyzing and predicting specified event associations based on a graph model. Background Art

[0002] A graph model, also known as Graphic Models, is a graph composed of points and lines used to describe a system. In modeling, graph theory can be used as a tool to analyze according to the properties of the graph, providing an effective method for studying various systems, especially complex systems. Graph models can be used to describe a large number of things and the relationships between things in nature and human society, and are widely used in natural science, engineering technology, social economy, management, etc.

[0003] The correlation analysis of police situation events is a comprehensive and systematic analysis of various factors involved in police situation events by the public security organs during the law enforcement process to determine the nature, cause, involved personnel, and possible development trends of the events. Currently, in the correlation analysis of police situation events using graph models, various information related to police situation events, including involved persons, evidence, clues, motives, etc., can be collected, sorted, and analyzed to reveal the internal connections and laws between events, providing a scientific basis for case detection and maintaining public order. This kind of analysis is of great significance for case detection, maintaining public order, and ensuring social stability.

[0004] However, the current graph model requires a large amount of manual participation and requires manual sorting of the causal relationships and logic between events. However, in the face of a large amount of police situation information, such as a large number of alarm calls, clue submissions, and media public opinion information that appear in a short period of time, it is difficult to screen and distinguish them manually, which will affect the subsequent accurate establishment of the graph model and thus affect the accuracy of subsequent trend prediction and judgment. Summary of the Invention

[0005] In order to reduce the analysis cost of message events and improve the accuracy of trend judgment, this application provides a method for analyzing and predicting specified event associations based on a graph model, adopting the following technical solutions:

[0006] A method for analyzing and predicting specified event associations based on a graph model, comprising:

[0007] Obtain multi-modal event data from different message event sources within a specified time period;

[0008] Perform identification processing on the multi-modal event data corresponding to each of the message event sources to obtain corresponding text-based description information;

[0009] Obtain the text feature vector of the text-based description information based on natural language processing technology;

[0010] Calculate the correlation degree between the text-based description information corresponding to each pair of the message event sources according to the text feature vector;

[0011] Perform graph model visualization processing on the multi-modal event data based on the calculation result of the correlation degree to obtain a correlation time graph;

[0012] Aggregate the correlation time graphs and send them to a trend prediction model to obtain the prediction result output by the trend prediction model;

[0013] Among them, the correlation time graph includes entity graph elements and connection graph elements. The entity graph elements include the geographical coordinates where the event occurs, the occurrence time, type attributes, and feature content. The connection graph elements include a connection main body and a logical main body. The connection main body is used to connect the entity graph elements, and the logical main body uses an arrow to represent the time relationship before and after. The prediction result includes at least 3 stepped hazard levels.

[0014] By adopting the above technical solution, the multi-modal event data obtained from multiple message event sources can be analyzed and identified to obtain the correlation degree between the events described by each message event source. The correlation time graphs of the multi-modal event data with a correlation degree exceeding a specified threshold can be aggregated and sent to a trend prediction model, thereby reducing the pressure on manual screening of a large amount of information and improving the speed of police situation handling and judgment.

[0015] Optionally, the process of calculating the correlation degree between the text-based description information corresponding to each pair of the message event sources according to the text feature vector includes:

[0016] For text-based description information A and text-based description information B, after splitting according to a preset attribute category, calculate the correlation degree for the same-attribute information;

[0017]

[0018] Among them, Z(A,B) represents the correlation degree between the text-based description information A and the text-based description information B, ZD(A,B) represents the correlation degree between the corresponding geographical coordinates, ZT(A,B) represents the correlation degree between the corresponding occurrence times, n is the total number of the same-attribute information, Zi(A,B) is the correlation degree between the i-th item of the same-attribute information, and λi is the weighting coefficient of the correlation degree between the i-th item of the same-attribute information.

[0019] Optionally, the method for obtaining the correlation degree ZD(A,B) between the corresponding geographical coordinates includes:

[0020]

[0021] Among them, |TA - TB | Indicates the time interval between the text - type description information A and the text - type description information B corresponding to the occurrence time. D ( TA - TB ) represents the maximum speculative distance interval associated with the time interval, and |DA - DB| represents the actual displacement distance between the geographical coordinates.

[0022] Optionally, the method for obtaining the correlation degree ZT(A, B) between the occurrence times includes:

[0023]

[0024] Among them, | TAC - TBC | Indicates the difference in the acquisition times of the multi - modal event data corresponding to the text - type description information A and the text - type description information B, and ΔT is the specified time period.

[0025] Optionally, if the corresponding content of the same - attribute information is text information, the method for calculating the correlation degree of the corresponding same - attribute information includes:

[0026]

[0027] Among them, when i = J, ZJ represents the correlation degree between (x AJ , x BJ ) in the J - th group of the same - attribute information of the text - type description information A and the text - type description information B, and cosθ(x AJ , x BJ ) represents the cosine value between the corresponding text vectors of the same - attribute information in the corresponding group. is the Euclidean distance.

[0028] Optionally, if the corresponding content of the same - attribute information is digital data information, the method for calculating the correlation degree of the corresponding same - attribute information includes:

[0029]

[0030] Among them, when i = K, ZK represents the correlation degree between (x AK , x BK ) in the K - th group of the same - attribute information of the text - type description information A and the text - type description information B.

[0031] Optionally, the process of performing graph - model visualization processing on the multi - modal event data based on the calculation result of the correlation degree to obtain the correlation time spectrum includes:

[0032] For the establishment of an entity graph element:

[0033] Select a primitive shape and represent the type attribute according to the primitive shape;

[0034] Convert the geographical coordinates, the occurrence time, and the feature content into primitive units that represent meanings through pixel information according to specified conversion rules.

[0035] Optionally, for one of the associated time graphs, at least two pieces of the multimodal event data with an association degree greater than a specified threshold are included, and different pieces of the multimodal event data are represented by different color themes.

[0036] In summary, the present application includes at least one of the following beneficial technical effects:

[0037] By analyzing and identifying the multimodal event data obtained from multiple message event sources, the present invention obtains the association degree between the events described by each message event source, and can send the association time graphs of the multimodal event data with an association degree exceeding the specified threshold to a trend prediction model after aggregation, thereby reducing the pressure of manual screening of a large amount of information and improving the speed of police situation handling and judgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a schematic diagram of a specified event association analysis and prediction method in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] The following details the embodiments of the present application, and the examples of the embodiments are shown in the drawings.

[0040] In the description of this specification, the description referring to the terms "certain embodiments", "one embodiment", "some embodiments", "schematic embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic descriptions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0041] The embodiments of the present application disclose a specified event association analysis and prediction method based on a graph model. Refer to Figure 1 , including:

[0042] Obtain multimodal event data from different message event sources within a specified time period;

[0043] Perform identification processing on the multimodal event data corresponding to each of the message event sources to obtain corresponding text-based description information;

[0044] Obtain the text feature vector of the text description information based on natural language processing technology;

[0045] Calculate the correlation degree between each pair of the message event sources corresponding to the text description information according to the text feature vector;

[0046] Perform graph model visualization processing on the multi-modal event data based on the calculation result of the correlation degree to obtain a correlation time spectrum;

[0047] Send the set of the correlation time spectra into a trend prediction model to obtain the prediction result output by the trend prediction model;

[0048] Wherein, the correlation time spectrum includes entity graph elements and connection graph elements. The entity graph elements include the geographical coordinates, occurrence time, type attributes and characteristic content of the event occurrence. The connection graph elements include a connection main body and a logical main body. The connection main body is used to connect the entity graph elements, and the logical main body uses an arrow to represent the front-back time relationship; the prediction result includes at least 3 stepped hazard levels.

[0049] In the present embodiment of the present invention, the message event source can be an independent individual such as an alarm caller, a self-media, a clue provider, etc. Correspondingly, the multi-modal time data can be an alarm call, an alarm text message, a screenshot of the alarm time, a short video, and a public article, etc. In this embodiment, an existing multi-modal large model can be used to analyze and identify the multi-modal event data obtained from multiple message event sources to obtain the meaning description of each multi-modal event data, and at least summarize the geographical coordinates, occurrence time, type attributes and characteristic content of the event occurrence. Then, the correlation degree between each pair of multi-modal event data can be measured, and all multi-modal event data with a correlation degree exceeding the specified threshold can be extracted. For each part of the multi-modal event data, a correlation time spectrum is established, and after the establishment of the correlation time spectrum is completed and set, finally, it is sent to the trend prediction model to obtain primary, intermediate, advanced and special-level risk prediction results, thereby reducing the pressure of manual screening of a large amount of information and improving the speed of police situation handling and judgment.

[0050] The multi-modal large model in this embodiment can be designed by using a vision-language model (VLM), or can also use MiniGPT-4, which can analyze the content in the image and is helpful for identifying the picture evidence sent by the alarm caller to improve information screening. For data such as alarm calls, alarm text messages, screenshots of alarm time, short videos, and public articles, the meaning summary description of the multi-modal event data can be completed through speech recognition and text recognition (frames can be used for text recognition in videos).

[0051] In this embodiment, the process of visually processing the multi-modal event data by using the calculation result based on the correlation degree and establishing a correlation time graph for the multi-modal event data includes:

[0052] Suggest at least two entity graphic elements, and then connect the entity graphic elements through a connection subject.

[0053] Establishment of an entity graphic element:

[0054] The graphic element shape can be selected, and the type attribute is represented according to the graphic element shape; if the type attribute is reported information, the graphic element shape can be represented by a triangle, and if the type attribute is alarm information, the graphic element shape can be represented by a circle;

[0055] Convert the geographical coordinates, the occurrence time, and the feature content into graphic element units representing meanings through pixel information according to the specified conversion rules.

[0056] In this embodiment, charts can be established in advance for the meanings of Arabic numerals, text nouns, mathematical units, etc.; for example, "1" can correspond to a pixel arrangement image of "white white white white white white white black", and "mall" can correspond to a pixel arrangement image of "white white black white white black white black". The above pixel arrangement images can all be collectively referred to as graphic element units.

[0057] In addition, when aggregating two correlation time graphs of two multi-modal event data with a correlation degree greater than a specified threshold, different multi-modal event data can be represented by different color themes, which is convenient for the training and recognition of the trend prediction model; when clarifying the correlation content, the entity graphic elements in the two correlation time graphs can be logically connected manually by humans.

[0058] In this embodiment, the trend prediction model can adopt Convolutional Neural Network (CNN), which is a deep learning architecture and is particularly suitable for processing image data. Its core idea is to extract local features in the image through convolutional operations, and these features are crucial for tasks such as image classification, object detection, and image segmentation.

[0059] Optionally, the process of calculating the correlation degree between the text-type description information corresponding to the two message event sources according to the text feature vectors includes:

[0060] For text-type description information A and text-type description information B, after splitting according to the preset attribute categories, calculate the correlation degree for the information with the same attribute;

[0061]

[0062] Among them, Z(A,B) represents the correlation degree between the text description information A and the text description information B, ZD(A,B) represents the correlation degree between the corresponding geographical coordinates, ZT(A,B) represents the correlation degree between the corresponding occurrence times, n is the total number of same-attribute information, Zi(A,B) is the correlation degree between the i-th item of same-attribute information, and λi is the weighting coefficient of the correlation degree between the i-th item of same-attribute information.

[0063] Optionally, the method for obtaining the correlation degree ZD(A,B) between the corresponding geographical coordinates includes:

[0064]

[0065] Among them, | TA - TB | represents the time interval between the occurrence times corresponding to the text description information A and the text description information B, D ( TA - TB ) represents the maximum speculative distance interval associated with the time interval, | DA - DB represents the actual displacement distance between the geographical coordinates. Designed in this way, it can try to overcome the exclusion of situations where the time is similar but the locations are far apart; for example, caller A and caller B call within 5 minutes. Caller A said that a suspicious person with a red backpack walking was just found in XX Square, and caller B said that a suspicious person with a red backpack walking was just found in YY Shopping Mall, and XX Square and YY Shopping Mall are 30 kilometers apart. At this time D ( TA - TB ) can be set to 3 times the normal walking distance of a normal person within 5 minutes. In this way, the correlation degree ZD(A,B) between the geographical coordinates obtained by calculation will be significantly reduced, thereby improving the accuracy of the correlation degree calculation.

[0066] Optionally, the method for obtaining the correlation degree ZT(A,B) between the occurrence times includes:

[0067]

[0068] Among them, | TAC - TBC | represents the difference in the acquisition times of the multi-modal event data corresponding to the text description information A and the text description information B, and ΔT is the specified time period.

[0069] Similarly, for the calculation of the correlation degree with geographical coordinates, the smaller the time difference described by two alarm reporters when reporting the incident, or the smaller the time difference when the multi-modal event data is made public, the greater the correlation degree can be considered; in addition, a specified time period can be designated, such as 1 hour. If the specified time period is re-designated as a different value, all correlation degrees need to be recalculated.

[0070] Optionally, if the corresponding content of the same-attribute information is text information, the correlation degree calculation method for the corresponding same-attribute information includes:

[0071]

[0072] where when i = J, ZJ represents the correlation degree between the Jth group of same-attribute information of the text description information A and the text description information B for (x AJ , x BJ ), and cosθ(x AJ , x BJ ) represents the cosine value between the corresponding text vectors of the same-attribute information in the corresponding group, is the Euclidean distance.

[0073] Specifically, if the cosine values of two text vectors are closer to 1, it means that their directions in the vector space are closer, that is, the text contents they represent are more similar. On the contrary, if the cosine value is closer to -1, it means that their directions in the vector space are more opposite, and the differences in the text contents they represent are greater. The Euclidean distance is also known as the Euclidean metric, which refers to the actual distance between two points in an m-dimensional space, or the natural length of a vector. In a two-dimensional space, the Euclidean distance between two points is the actual distance between the two points on the plane; in a multi-dimensional space, this concept can be extended to multiple dimensions.

[0074] Optionally, if the corresponding content of the same-attribute information is digital data information, the correlation degree calculation method for the corresponding same-attribute information includes:

[0075]

[0076] where when i = K, ZK represents the correlation degree between the Kth group of same-attribute information of the text description information A and the text description information B for (x AK , x BK ), and |Kax - Kin| represents the difference between the maximum value and the minimum value of the corresponding attribute information. For example, if the alarm reporter A said that the male suspect's estimated weight was 80 kg and the alarm reporter B said that the male suspect's estimated weight was 60 kg, then | Kax - Kin |The corresponding male value is 30 kg; if the suspect is female, since the visually estimated weight value of females is more accurate and has less fluctuation, thus Kax-Kin | The corresponding female value is 10 kg.

[0077] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present application. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application.

Claims

1. A method for analyzing and predicting association of specified events based on a graph model, characterized in that: include: Get multimodal event data from different message event sources within a specified time period; Identify and process the multimodal event data corresponding to each message event source to obtain corresponding text description information; Acquire a text feature vector of the text description information based on natural language processing technology; Calculate the correlation between the text description information corresponding to each of the message event sources according to the text feature vector; Based on the calculation result of the correlation degree, the multimodal event data is subjected to graph model visualization processing to obtain a correlation time graph; The associated time graphs are collected and sent to a trend prediction model to obtain a prediction result output by the trend prediction model; Among them, the associated time graph includes entity graph elements and connection graph elements, the entity graph elements include the geographical coordinates of the event, the time of occurrence, type attributes and characteristic content, the connection graph elements include connection subjects and logical subjects, the connection subjects are used to connect the entity graph elements, and the logical subjects use arrows to indicate the time relationship before and after; the prediction results include at least 3 stepped danger levels.

2. The method for analyzing and predicting designated event association based on a graph model according to claim 1, characterized in that: The process of calculating the correlation between the message event sources and the text description information according to the text feature vector comprises: For the text description information A and the text description information B, after splitting them according to the preset attribute categories, the correlation degree of the same attribute information is calculated; Among them, Z(A,B) represents the correlation between the text description information A and the text description information B, ZD(A,B) represents the correlation between the corresponding geographic coordinates, ZT(A,B) represents the correlation between the corresponding occurrence times, n is the total number of information with the same attribute, Zi(A,B) is the correlation between the i-th item of information with the same attribute, and λi is the weighting coefficient of the correlation between the i-th item of information with the same attribute.

3. The method for analyzing and predicting designated event associations based on a graph model according to claim 2, characterized in that: The method for obtaining the correlation degree ZD(A, B) between the corresponding geographic coordinates includes: Among them, |TA-TB| represents the time interval between the text description information A and the text description information B corresponding to the occurrence time, D(|TA-TB|) represents the maximum estimated distance interval associated with the time interval, and |DA-DB| represents the actual displacement distance between the geographic coordinates.

4. The method for analyzing and predicting designated event association based on a graph model according to claim 3 is characterized in that: The method for obtaining the correlation degree ZT(A, B) between the occurrence times includes: Among them, |TAC-TBC| represents the difference between the acquisition time of the multimodal event data corresponding to the text description information A and the text description information B, and ΔT is the specified time period.

5. The method for analyzing and predicting designated event association based on a graph model according to claim 2, characterized in that: If the corresponding content of the information with the same attribute is text information, the method for calculating the correlation degree of the information with the same attribute includes: When i=J, ZJ represents the Jth group of the same attribute information of the text description information A and the text description information B (x AJ ,x BJ ), cosθ(x AJ ,x BJ ) represents the cosine value between the corresponding text vectors of the same attribute information of the corresponding group, is the Euclidean distance.

6. The method for analyzing and predicting designated event association based on a graph model according to claim 2, characterized in that: If the corresponding content of the information with the same attribute is digital data information, the method for calculating the correlation degree of the information with the same attribute includes: When i=K, ZK represents the Kth group of the same attribute information of the text description information A and the text description information B (x AK , x BK ) between them.

7. The method for analyzing and predicting designated event association based on a graph model according to claim 1, characterized in that: The process of performing graph model visualization processing on the multimodal event data based on the calculation result of the correlation degree to obtain a correlation time graph includes: To create a solid primitive: Selecting a primitive shape, and representing the type attribute according to the primitive shape; The geographic coordinates, the occurrence time and the characteristic content are converted into a graphic primitive unit whose meaning is expressed by pixel information according to a specified conversion rule.

8. The method for analyzing and predicting designated event association based on a graph model according to claim 1, characterized in that: For one of the correlation time graphs, at least two of the multimodal event data having a correlation degree greater than a specified threshold are included, and different multimodal event data are represented by different color themes.

Images