A potential risk hidden danger identification method, device and equipment based on multi-dimensional information and event causal analysis, medium and product

Abstract

The application discloses a potential risk hidden danger identification method and device based on multi-dimensional information and event causal analysis, equipment, medium and product, relates to the field of artificial intelligence. The method comprises the following steps: obtaining multi-modal data from a subscribed public opinion monitoring object and performing feature extraction, and extracting multi-modal information features; performing data structural analysis based on the multi-modal information features, and extracting key attribute structured data of each event; performing clustering analysis on all events, and dividing all events into a series of large event sets; performing information deduplication and integration analysis on each small event in the large event set, obtaining key attribute structured data of each large event set, and performing causal chain analysis on the occurrence order and causal relationship of the series of large event sets, obtaining an event causal chain, and performing potential risk hidden danger identification of the event based on the event causal chain, which can improve the comprehensiveness and timeliness of public opinion monitoring and the accuracy of public opinion analysis.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of artificial intelligence, in particular to a potential risk hidden danger identification method and device based on multi-dimensional information and event causal analysis, equipment, medium and product. BACKGROUND

[0002] Network public opinion refers to the network public opinion of different views on social events on the network, which is a form of social public opinion. In recent years, the influence of network public opinion on the order of life and social stability is increasing day by day, and some major network public opinion events have made people begin to realize the great role of the network in social supervision. At the same time, with the development of the big data era, the demand for public opinion monitoring is getting higher and higher, and the monitoring of network public opinion state becomes particularly important.

[0003] The potential risk hidden danger identification of public opinion monitoring refers to integrating internet information collection technology and information intelligent processing technology, automatically grabbing, automatically classifying and clustering, theme detection, and special topic focusing on massive internet information, realizing the information needs of user network public opinion monitoring and hot event tracking, forming analysis results such as briefing, report and chart, and providing analysis basis for identifying potential risk hidden dangers of network public opinion events and making correct public opinion guidance.

[0004] At present, the potential risk hidden danger identification method of public opinion monitoring is mostly to screen public opinion by public opinion analysts, and then track the public opinion, so as to roughly judge the influence of the event on the public opinion state. However, this method has the problems of insufficient comprehensive monitoring data source, inability to realize real-time analysis and monitoring, and inability to intuitively understand the public opinion state and risk level of network public opinion, thereby leading to the inability to timely and accurately obtain effective information to assist business decision-making. SUMMARY

[0005] The purpose of the present application is to provide a potential risk hidden danger identification method, device, equipment, medium and product based on multi-dimensional information and event causal analysis, so as to improve the comprehensiveness and timeliness of public opinion monitoring and the accuracy of public opinion analysis.

[0006] To achieve the above purpose, the present application provides the following solutions.

[0007] In a first aspect, the present application provides a potential risk hidden danger identification method based on multi-dimensional information and event causal analysis, comprising:

[0008] Obtaining multi-modal data from a subscribed public opinion monitoring object, including audio data, video data and text data;

[0009] Performing feature extraction on the multi-modal data to extract multi-modal information features including tone features, face features and text information features;

[0010] Based on the multi-modal information feature, data structured analysis is performed to extract key attribute structured data of each event, wherein the key attribute structured data of each event includes time, place, involved person, event specific information and risk index;

[0011] Based on the key attribute structured data of each event, clustering analysis is performed on all events to divide all events into a series of large event sets;

[0012] Information deduplication and integration analysis are performed on each small event in the large event set to obtain key attribute structured data of each large event set;

[0013] Based on the key attribute structured data of each large event set, the occurrence order and causal relationship of the series of large event sets are analyzed to obtain an event causal chain of public opinion monitoring;

[0014] Based on the event causal chain of public opinion monitoring, potential risk hazards of the event are identified.

[0015] Optionally, the feature extraction on the multi-modal data extracts multi-modal information features including timbre features, face features and text information features, specifically including:

[0016] The timbre features of the audio data are extracted using the audio processing library Librosa, and the audio data extraction is converted into text data;

[0017] The face features in the video data are identified using a deep convolutional neural network, and the recognition ability is enhanced using an ArcFace loss function;

[0018] The text information features in the text data are extracted using a large language model, including time, place, involved person and event specific information.

[0019] Optionally, the data structured analysis based on the multi-modal information features extracts key attribute structured data of each event, specifically including:

[0020] Based on the text information features, the time, place and event specific information corresponding to each event are extracted;

[0021] The involved person of each event is extracted by integrating the timbre features, face features and text information features;

[0022] Based on the time, place, event specific information and involved person corresponding to each event, the risk index is calculated.

[0023] Optionally, the calculation of the risk index based on the time, place, event specific information and involved person corresponding to each event specifically includes:

[0024] Calculate the time distance of each event from the current time t, and according to the formula Calculate the time risk index R_t;

[0025] Calculate the linear distance s of each event point from the public opinion monitoring center of the region, and according to the formula Calculate the spatial risk index R_s;

[0026] Based on the event specific information and the involved person of each event, according to the formula Calculate the public opinion risk index R_p; Where p is the total number of events involved, including reading volume, comment number, forwarding number and like number; When the involved person includes public figures or the event specific information involves social hotspots, flag=true, otherwise flag=false;

[0027] According to the formula R=εt×R_t+εs×R_s+εp×R_p, the risk index R is calculated; Where εt, εs and εp are the weights of the time risk index, the spatial risk index and the public opinion risk index respectively.

[0028] Optionally, the clustering analysis of all events based on the key attribute structured data of each event divides all events into a series of large event sets, specifically including:

[0029] Based on the key attribute structured data of each event, a feature vector corresponding to each event is constructed;

[0030] According to the feature vector corresponding to each event, the Euclidean distance between each pair of events is calculated;

[0031] According to the Euclidean distance between each pair of events, the similarity between each pair of events is calculated;

[0032] Based on the similarity between each pair of events, a similarity matrix is constructed;

[0033] Based on the similarity matrix, an adjacency matrix of a graph is constructed, and spectral clustering algorithm is used for clustering, so as to divide all events into a series of large event sets; Each large event set includes multiple small events.

[0034] Optionally, the information deduplication and integration analysis of each small event in the large event set is carried out to obtain the key attribute structured data of each large event set, specifically including:

[0035] The time of each big event set is the earliest time of all small events in the big event set; the place of each big event set includes the places of all small events in the big event set; the involved person of each big event set includes the involved persons of all small events in the big event set; the event specific information of each big event set is the integration of the event specific information of all small events in the big event set; and the risk index of each big event set is the maximum value of the total risk coefficient of all small events in the big event set.

[0036] In a second aspect, the present application provides a potential risk hidden danger identification device based on multi-dimensional information and event causal combing, comprising:

[0037] A multi-modal data acquisition module is configured to acquire multi-modal data, including audio data, video data and text data, from a subscribed public opinion monitoring object;

[0038] A data feature extraction module is configured to extract multi-modal information features including tone features, face features and text information features from the multi-modal data;

[0039] A data structured analysis module is configured to perform data structured analysis based on the multi-modal information features to extract key attribute structured data of each event; the key attribute structured data of each event includes time, place, involved person, event specific information and risk index;

[0040] An event clustering analysis module is configured to perform clustering analysis on all events based on the key attribute structured data of each event, and divide all events into a series of big event sets;

[0041] An information deduplication and integration module is configured to perform information deduplication and integration analysis on each small event in the big event set to obtain key attribute structured data of each big event set;

[0042] An event causal chain combing module is configured to perform causal chain combing on the occurrence order and causal relationship of the series of big event sets based on the key attribute structured data of each big event set to obtain a public opinion monitoring event causal chain;

[0043] A potential risk hidden danger identification module is configured to identify potential risk hidden dangers of events based on the public opinion monitoring event causal chain.

[0044] In a third aspect, the present application provides a computer device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the potential risk hidden danger identification method based on multi-dimensional information and event causal combing.

[0045] In a fourth aspect, the present application provides a computer readable storage medium, having stored thereon a computer program, which, when executed by a processor, implements the method for identifying potential risk hidden dangers based on multi-dimensional information and event causal analysis.

[0046] In a fifth aspect, the present application provides a computer program product comprising a computer program which, when executed by a processor, implements the method for identifying potential risk hidden dangers based on multi-dimensional information and event causal analysis.

[0047] According to the specific embodiments provided by the present application, the present application discloses the following technical effects:

[0048] The present application provides a method, device, equipment, medium and product for identifying potential risk hidden dangers based on multi-dimensional information and event causal analysis, which can comprehensively integrate multi-modal data such as text, audio and video related to public opinion events, significantly improve the accuracy and robustness of public opinion information processing; through the structural analysis of multi-modal information features such as tone features, face features and text information features, the key attribute structured data of each event (including time, place, involved person, event specific information and risk index) is extracted, the clustering analysis method can be dynamically adjusted to achieve more accurate clustering effect, thereby improving the accuracy of event analysis, which is helpful for quickly identifying and responding to the key nodes of public opinion events; through the analysis and understanding of the causal development context of events, the event development trend and potential risk hidden dangers are more effectively analyzed, the depth and breadth of complex event analysis are improved, which can better predict and handle complex public opinion events, and provide a strong basis for quickly responding to various emergencies and social events. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of these drawings.

[0050] Figure 1 A flowchart of a method for identifying potential risk hidden dangers based on multi-dimensional information and event causal analysis according to the present application. DETAILED DESCRIPTION

[0051] The technical solutions in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0052] In view of the defects in the prior art, the present application proposes a potential risk hidden danger identification method and device based on multi-dimensional information and event causal analysis, equipment, medium and product, so as to improve the comprehensiveness and timeliness of public opinion monitoring and the accuracy of public opinion analysis.

[0053] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below in combination with the drawings and specific embodiments.

[0054] In one exemplary embodiment, as shown in Figure 1 A potential risk hidden danger identification method based on multi-dimensional information and event causal analysis is provided, including the following steps 1 to step 7.

[0055] Step 1: Obtain multi-modal data from subscribed public opinion monitoring objects, including audio data, video data and text data.

[0056] The subscribed public opinion monitoring object can be a specific enterprise, group, individual, etc. It should be noted that the acquisition of multi-modal data in the present application is carried out in accordance with the corresponding data protection regulations and policies of the place of residence, and with the authorization of the corresponding device owner.

[0057] By regularly crawling multi-modal data such as audio, video and text from different channels / platforms, the latest multi-dimensional information is obtained to ensure the comprehensiveness and timeliness of the data. Further, the format of the crawled data (multi-dimensional information) is identified, and the data type is identified by the file extension of the crawled data, for example, MP4, AVI, MOV type files are classified into video data, MP3, WAV type files are classified into audio data, and HTML, TXT, PDF type files or web text are classified into text data.

[0058] Step 2: Feature extraction is performed on the multi-modal data to extract multi-modal information features including tone features, face features and text information features.

[0059] The features of audio, video, text, etc. data are extracted and the feature information is integrated. For example, for audio data, audio processing library Librosa is used to extract timbre features such as MFCCs, and then voiceprint comparison and voiceprint recognition can be performed. For video data, deep convolutional neural networks (DCNNs) are used for deep feature learning of face recognition, face features are extracted, and ArcFace loss function is used to enhance the recognition ability. For text data, large language models (LLM) are used to analyze important text information features such as time, place, involved characters, event specific information, etc. in detail and classify and organize. Then the extracted above feature information is integrated into local files (such as csv files, etc.).

[0060] Step 3: Based on multi-modal information features, data structure analysis is performed to extract key attribute structured data of each event; the key attribute structured data of each event includes time, place, involved characters, event specific information and risk index.

[0061] In the context of social governance and risk management, the analysis and processing of each event need to be based on structured data to ensure the accuracy and operability of information. Therefore, event data needs to include the following five key attributes: time, place, involved characters, event specific information and risk index, and arrange each event according to the risk index to make a table for further analysis.

[0062] The definition and use of key attribute structured data of each event are as follows.(a) Time: recorded in standard time format (such as ISO 8601ss), used to assess time risk and analyze the timing relationship of related events.(b) Place: the geographical location of the event, used for spatial risk assessment to measure the geographical impact range of the event and the proximity to the target area of public opinion monitoring.(c) Involved characters: key characters or entities involved in the event, used to determine the public opinion influence and social attention of the event.(d) Event specific information: detailed description of the event, as the basis for qualitative analysis, to help identify the nature and potential consequences of the event.(e) Risk index: the comprehensive risk score obtained after quantitative analysis of specific information, to help quickly identify and handle high-risk events.

[0063] Specifically, the publishing time of the text can be directly obtained from the text information feature as the event time, and the sentence describing the geographical position in the text can be found as the event location. The event participants appearing in the event can be comprehensively judged according to the face features extracted from the video data, the tone features in the audio data, and the names appearing in the text data as the involved characters. The audio data is also integrated into the text data, combined with the directly obtained text data, and the general meaning of the extracted text is summarized as the event specific information through a large language model. Further, based on the key attribute information of the integrated event, such as time, location, character, and event specific information, the risk index of the event can be calculated, and the steps of calculating the risk index are as follows.

[0064] (3.1) Calculate the time risk index R_t

[0065] Let the number of hours from the time corresponding to each event (also referred to as event time) to the current time be t hours, and the time risk index of the event be R_t. The time development of public opinion is divided into incubation period, outbreak period, sustained period, and decline period, and the corresponding values of R_t in each stage are given:

[0066] Incubation period (0≤t<10), the event just occurred, at this time the risk index is low, but there is a risk of rapid spread. Let the time risk index in this stage be R_t=0.5.

[0067] Outbreak period (10≤t<60), the public opinion event has fermented for a period of time, and the trend is not yet clear. At this time, the risk index is very high, which may trigger negative public opinion and chain reactions. Let the time risk index in this stage be R_t=1.

[0068] Sustained period (60≤t<120), the public opinion event is at risk, and the risk growth trend slows down or tends to be stable. At this time, the risk index is relatively high but stable. Let the time risk index in this stage be R_t=0.75.

[0069] Decline period (t≥120), the public opinion event is starting to decline. At this time, the risk index gradually decreases, but subsequent effects still need to be concerned. Let the time risk index in this stage be R_t=0.25.

[0070] In summary, the calculation formula of the time risk index R_t can be expressed as follows:

[0071]

[0072] (3.2) Calculate the space risk index R_s

[0073] Here, take the local public opinion response processing as an example, record the straight-line distance (s kilometers) between the location of the corresponding point of each event (referred to as the event location) and the regional public opinion monitoring center. The occurrence range of social events is divided into high-risk areas and low-risk areas, and the corresponding numerical value of R_s in each area is given. If the exact occurrence of an event is not considered, the spatial risk index of such event is not considered, and the weight of the spatial risk index of such event is εs=0.

[0074] where, high-risk area (0≤s<5), the event location is close to the regional public opinion monitoring center, the risk is larger and the change with distance is smaller, and a quadratic function about s=0 is used to describe the attenuation of the spatial risk index in this area, that is

[0075] Low-risk area (s≥5), the event location is far away from the regional public opinion monitoring center, the risk is smaller and the initial decay is faster with the growth of distance, and the decay is slower in the later period. A power function is used to describe the decay of the spatial risk index in this area, that is

[0076] In summary, the calculation formula of the spatial risk index R_s can be expressed as:

[0077]

[0078] (3.3) Calculate the public opinion risk index R_p

[0079] Let the sum of the number of events involved (including reading volume, comment number, forwarding number, like number, etc.) be p. Based on the face features and event specific information in the multi-modal information features, compared with the face database, if a public figure is identified or the event specific information involves a social hot spot by using LLM, flag=true (true), otherwise flag=false (false). flag is a label given for convenience. The exponential function is used to describe the numerical relationship between the event involved person and the public opinion risk index, that is:

[0080]

[0081] (3.4) Calculate the total risk index

[0082] The total risk index of an event is obtained by multiplying R_t, R_s, and R_p by the corresponding weight and then adding them up, which is denoted as the risk index R of the event:

[0083] R=εt×R_t+εs×R_s+εp×R_p (4)

[0084] where εt, εs, and εp are the weights of the time risk index, the spatial risk index, and the public opinion risk index, respectively.

[0085] After the collection of structured data of key attributes of each event (referred to as event data), the event data is sorted and stored in a structured manner for subsequent analysis. Among them, (a) data formatting: all event data is entered into a structured database according to the defined format, and a relational database (such as MySQL) or a NoSQL database (such as MongoDB) is used for storage. (b) Data verification and cleaning: through the combination of automatic tools such as OpenRefine and Weka and manual inspection, the accuracy and consistency of the data are ensured.

[0086] In order to further optimize resource allocation and emergency response strategies, events need to be sorted according to their risk indexes, as follows: (a) Algorithm sorting: all events are sorted according to risk indexes from high to low. Efficient sorting algorithms such as quicksort are used to ensure fast processing on large data sets. (b) Table generation: the sorted event data is presented in table form, containing five fields of time, location, involved person, event specific information and risk index. The table can be exported as a commonly used analysis format (CSV file) for further analysis and modeling.

[0087] Step 4: Cluster analysis of all events based on structured data of key attributes of each event, dividing all events into a series of large event sets.

[0088] The cluster analysis of all event data includes the following steps:

[0089] (4.1) Based on the structured data of key attributes of each event, a feature vector corresponding to each event is constructed.

[0090] A feature vector xi = [xi1, xi2, xi3, xi4, xi5] corresponding to each event data is constructed. The five elements contained in the feature vector are derived from the key attribute structured data in step 3, in the order of time, location, involved person, event specific information and risk index. The obtained feature vector is used to calculate the Euclidean distance and similarity between events, and is the basis for subsequent construction of similarity matrix and event clustering.

[0091] (4.2) Calculate the Euclidean distance between each pair of events according to the feature vector corresponding to each event.

[0092] The Euclidean distance is used to compare each pair of events, and the similarity between each pair of events i and j can be calculated, and a similarity matrix is established according to the similarity. The formula for calculating the Euclidean distance between each pair of events is as follows:

[0093]

[0094] where xi k and xj k(k = 1, 2, 3, 4, 5) represent the k-th element of the feature vector xi = [xi1, xi2, xi3, xi4, xi5] and xj = [xj1, xj1, xj3, xj4, xj5] of events i and j respectively.d ij represents the Euclidean distance between events i and j.

[0095] (4.3) Calculate the similarity between each pair of events according to the Euclidean distance between them, as follows:

[0096]

[0097] where S ij represents the similarity between events i and j.

[0098] (4.4) Construct a similarity matrix based on the similarity between each pair of events.

[0099] Construct an N x N similarity matrix, where N is the number of events, and each element S ij in the matrix represents the similarity between events i and j. The similarity calculated according to (4.3) fills the matrix, and the similarity matrix is a symmetric matrix, i.e. ij = S ji , so only half of the matrix (e.g. upper triangular or lower triangular) needs to be calculated. The final similarity matrix is generated as shown in the following matrix type:

[0100]

[0101] where S ij is the similarity between events i and j, and the elements on the diagonal are all 1, indicating that the similarity of an event with itself is 1.

[0102] (4.5) Based on the similarity matrix, construct an adjacency matrix of a graph (which can be an undirected graph, a directed graph, or a weighted graph, etc.), and use the spectral clustering algorithm for clustering, thereby dividing all events into a series of large event sets M1, M2, …, M n ; n is the number of large event sets. Each large event set includes multiple small events m i , i = 1, 2, … p, p is the number of small events included in the large event set. All events in each large event set are related to this large event and are part of this large event. Through the order and causal relationship of each small event, causal chain analysis can be performed to analyze the context and risk index of the large event.

[0103] Step 5: Perform information deduplication and integration analysis on each small event in the large event set to obtain the key attribute structured data of each large event set.

[0104] Record each large event set M j = {m1, m2,..., m p}, j = 1, 2,... n, remove duplicates and merge information for m i to obtain the integrated event node M' j . M' j is the integrated large event set, and each small event m' i of the integrated large event set M' j is a thread, and the information of all m' i together constitutes the complete M' j event. Specifically, the time of each large event set M' j is the earliest time of all small events m j in the large event set M i , that is, the minimum value of time; the location of each large event set M' j includes the location of all small events in M j ; the involved person of each large event set M' j includes the involved person of all small events m j in M i ; the event specific information of each large event set M' j is the integration of the event specific information of all small events m j in M i ; the risk index of each large event set M' j is the maximum value of the total risk coefficient of all small events m j in M i .

[0105] Step 6: Based on the key attribute structure of each large event set, the occurrence order and causal relationship of a series of large event sets are sorted by causal chain, and the public opinion monitoring event causal chain is obtained.

[0106] The event node M' j is selected and sorted according to time, and the linear event causal chain, that is, the public opinion monitoring event causal chain, is obtained. The risk index of the event causal chain is the maximum value containing the risk index of the event node, and then the causal chain information is stored in the database for further analysis.

[0107] Step 7: Identify the potential risk hidden danger of the event based on the public opinion monitoring event causal chain.

[0108] The public opinion monitoring event causal chain arranges the event nodes in chronological order, which can help analyze the cause, development context and key nodes of the public opinion event, evaluate the risk index, and has applications in event tracing, risk assessment, decision support and early warning mechanism, which can effectively identify high-risk events and predict their development direction. In dealing with complex public opinion, the causal chain provides a basis for coping strategies, by taking measures against high-risk nodes and updating feedback results in real time, the coping measures can be optimized. Cross analysis of multiple causal chains can also reveal the correlation between events and prevent similar events from repeating.

[0109] The present application can help relevant institutions to timely grasp the dynamics of public opinion by integrating and analyzing public opinion information from different social platforms, and provide strong support for decision-making. The method of the present application has important application value in government public relations, enterprise brand management, emergency response and other fields, and can significantly improve the level of social management and the ability of public security protection.

[0110] For example, in one specific embodiment, a series of discussions about "a food safety problem" are found on a social media in the city where the crawler program is used. After crawling these data, a large number of negative comments are found through analysis. These comments are associated with the original post, which describes the location and time of the food safety problem in detail. The event chain of the food safety problem is generated by using the method of the present application, and further integration and analysis reveals the potential public health risk. Related to the above-mentioned food safety problem, a large number of negative comments about the lack of government regulatory measures subsequently appear, which are also analyzed by the method of the present application and associated with relevant discussion posts, identifying the time, location and other information of these posts, and linking them into a new event chain of regulatory measures. Comprehensive analysis of the two event chains finds that the food safety problem not only poses a threat to public health, but also may have a negative impact on the trust in government. Based on this, a detailed risk assessment report is generated based on the risk index of the event causal chain, and specific suggestions such as strengthening food safety inspection, improving regulatory mechanisms, and strengthening publicity and education are put forward, and the data visualization tool is used to display them to the relevant departments, helping them to take timely action. After new posts and comments appear, the event chain will be automatically updated, and the risk assessment and strategy recommendations will be adjusted based on the latest data. For example, after a citizen's complaint is properly handled, a series of positive comments appear, and the method of the present application will incorporate this new information into the analysis, adjust the risk assessment of the event chain, and provide the latest strategy recommendations. The method of the present application ensures timely discovery and effective response to social risks through effective event chain linking and real-time data updating, maintaining social stability and public security.

[0111] In one exemplary embodiment, the application also proposes a potential risk hidden danger identification device based on multi-dimensional information and event causal combing, comprising the following functional modules. A multi-modal data acquisition module is used to acquire multi-modal data including audio data, video data and text data from subscribed public opinion monitoring objects; a data feature extraction module is used to extract multi-modal information features including tone features, face features and text information features; a data structured analysis module is used to perform data structured analysis based on multi-modal information features to extract key attribute structured data of each event; the key attribute structured data of each event includes time, place, involved person, event specific information and risk index; an event clustering analysis module is used to perform clustering analysis on all events based on the key attribute structured data of each event, and divide all events into a series of large event sets; an information deduplication and integration module is used to perform information deduplication and integration analysis on each small event in the large event set to obtain key attribute structured data of each large event set; an event causal chain combing module is used to comb the occurrence order and causal relationship of a series of large event sets based on the key attribute structured data of each large event set to obtain public opinion monitoring event causal chains; and a potential risk hidden danger identification module is used to identify potential risk hidden dangers of events based on the public opinion monitoring event causal chains.

[0112] The potential risk hidden danger identification device based on multi-dimensional information and event causal combing can crawl multi-modal data such as audio, video and text from different channels, regularly acquire the latest information, and ensure the comprehensiveness and timeliness of the data; the crawled multi-modal data is subjected to structured analysis, and different types of data are separated; the audio, video and text data are subjected to feature extraction and integration of feature information respectively; the key information such as time, place and person involved in the integrated information is extracted by a large language model, and the main context of the event involved in the information is inferred by using a deep learning algorithm, and the risk index corresponding to the event is calculated, and all events and their key attribute structured data are stored in an event set M; similarity and correlation analysis of the event set M is performed by a similarity calculation algorithm, a similarity matrix is constructed, all events in the event set M are clustered, similar events are classified into a class, and the M is divided into M1, M2……M n a series of large event sets; M1, M2……M n large event sets are subjected to multi-modal causal chain combing of the same event respectively, and then public opinion risk degree sorting is performed, corresponding event countermeasures and action plans are generated according to the risk degree order, high-risk events are processed in priority, and the processing results are evaluated and fed back, and unprocessed events are stored in a database for reference.

[0113] The potential risk hidden danger identification method and device based on multi-dimensional information and event causal analysis proposed in the present application have the following characteristics: (1) the present application can efficiently fuse multi-modal data such as text, audio and video involved in social events, significantly improving the accuracy and robustness of public opinion information processing; even in the face of incomplete or noisy data, the reliability of the analysis results can be ensured, which is crucial for public opinion monitoring and risk assessment in social governance; (2) by analyzing the characteristics of multi-modal public opinion information, the calculation method of similarity can be dynamically adjusted to achieve more accurate clustering effect, thereby improving the accuracy of event analysis and helping to quickly identify and respond to key nodes of social events; (3) by analyzing and understanding the causal development context of events, the development trend of events and potential risk judgment can be more effectively analyzed, improving the depth and breadth of complex event analysis, and better predicting and handling complex public opinion events, providing strong support for social governance; (4) the technology of the present application has high portability and can quickly adapt to different application scenarios and needs, showing cross-field application potential and being able to quickly respond to various emergencies and social events; (5) the present application analyzes multi-modal data in depth, fuses tone, face and text information, constructs a comprehensive event context diagram, realizes efficient discovery of potential risks and intelligent generation of coping strategies, has wide application value, especially in government public relations, enterprise brand management and public security management, etc., which can significantly improve the level of social governance and public security protection capability.

[0114] In an exemplary embodiment, the present application also provides a computer device which can be a server or a terminal. The computer device comprises a processor, a memory, an input / output interface and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through network connection. The computer program is executed by the processor to realize the potential risk hidden danger identification method based on multi-dimensional information and event causal analysis.

[0115] In an exemplary embodiment, the present application also provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to realize the potential risk hidden danger identification method based on multi-dimensional information and event causal analysis.

[0116] In one exemplary embodiment, the present application also provides a computer program product comprising a computer program which, when executed by a processor, implements the method for identifying potential risk hidden dangers based on multi-dimensional information and event causal analysis.

[0117] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by computer program instruction related hardware. 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 embodiment methods. Any reference to memory or other medium in the embodiments provided by the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0118] It should be noted that the information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0119] The present application obtains multi-modal data such as audio, video and text in specific social hotspots from different channels by using crawler technology, and processes different types of data by using a large language model. For audio data, timbre features are extracted to identify voiceprints, and the core gist of the audio content is identified through semantic analysis. For video data, the face region in the video is identified and the corresponding face deep features are extracted, and then the characters in the video are identified. For text data, the main content is extracted and summarized. On the basis of extracting these features, the large language model is used to further identify key elements such as time, place, character and event specific information, and the deep learning algorithm is used to infer the main context of the event, and the key attribute structure of the event is stored in the database. Then all events in the database are clustered based on spectral clustering algorithm, and the evolution chain of the event is obtained by using time sequence and causal relationship, so as to realize more convenient public opinion information control and rapid public opinion development analysis, and predict and cope with potential risks of the event. The method of the present application has important role in government public relations, enterprise brand management, emergency response and other fields, and significantly improves the social management level and public safety guarantee ability.

[0120] The technical features of the above embodiments can be combined in any manner. To make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not exist contradictions, they should be considered as the scope of the present application.

[0121] The principles and implementation modes of the present application are described by using specific examples in this paper, and the above embodiment description is only used to help understand the method and core idea of the present application; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed. In conclusion, the content of the present application should not be understood as a limitation.

Claims

1. A method for identifying potential risks and hazards based on multidimensional information and event causal analysis, characterized in that, include: Obtain multimodal data, including audio, video, and text data, from the public opinion monitoring targets you have subscribed to; Feature extraction is performed on multimodal data to extract multimodal information features, including timbre features, facial features, and text information features; The feature extraction process for multimodal data extracts multimodal information features including timbre features, facial features, and textual information features, specifically including: Use the audio processing library Librosa to extract the timbre features of the audio data and then convert the audio data into text data. Deep convolutional neural networks are used to identify facial features in video data, and the ArcFace loss function is used to enhance their recognition capabilities. By leveraging large language models, textual information features can be extracted from text data, including time, location, people involved, and specific event information. Data structure analysis is performed based on multimodal information features to extract key attribute structured data for each event; the key attribute structured data for each event includes time, location, people involved, specific event information, and risk index; The data structuring analysis based on multimodal information features extracts key attribute structured data for each event, specifically including: Extract the time, location, and specific information of each event based on textual information features; By combining timbre features, facial features, and textual information features, the people involved in each event are extracted; The risk index is calculated based on the time, location, specific information about the event, and the people involved in each event; The risk index is calculated based on the time, location, specific information about the event, and the individuals involved for each event, specifically including: Calculate the number of hours t between the current time and the time corresponding to each event, and then apply the formula... Calculate the time risk index R_t; Calculate the straight-line distance *s* from the location corresponding to each event point to the regional public opinion monitoring center, and then apply the formula... Calculate the spatial risk index R_s; Based on the specific information of each event and the people involved, according to the formula Calculate the public opinion risk index R_p; where p is the total number of items involved in the event, including the number of views, comments, reposts and likes; when the people involved include public figures or the specific information of the event involves social hotspots, it is marked as flag=true, otherwise it is marked as flag=false; The risk index R is calculated using the formula R = εt × R_t + εs × R_s + εp × R_p, where εt, εs, and εp are the weights of the time risk index, spatial risk index, and public opinion risk index, respectively. Cluster analysis is performed on all events based on the key attribute structured data of each event, dividing all events into a series of large event sets; The clustering analysis, based on the structured data of key attributes of each event, divides all events into a series of large event sets, specifically including: Construct a feature vector for each event based on the structured data of its key attributes; Calculate the Euclidean distance between each pair of events based on the feature vector corresponding to each event; Calculate the similarity between pairs of events based on the Euclidean distance between them. Construct a similarity matrix based on the similarity between pairs of events; Based on the similarity matrix, an adjacency matrix of the graph is constructed, and spectral clustering algorithm is used to cluster all events into a series of large event sets; each large event set includes multiple small events. The information of each sub-event in the large event set is deduplicated and integrated to obtain the key attribute structured data of each large event set; The process of deduplicating and integrating information for each sub-event in the large event set to obtain structured data of key attributes for each large event set specifically includes: The time of each major event set is the earliest occurrence time of all minor events within that major event set; the location of each major event set includes the locations of all minor events within that major event set; the people involved in each major event set include the people involved in all minor events within that major event set; the event details of each major event set are a compilation of the event details of all minor events within that major event set; the risk index of each major event set is the maximum value of the total risk coefficient of all minor events within that major event set. Based on the key attribute structured data of each large event set, the causal chain of the occurrence order and causal relationship of a series of large event sets is sorted out to obtain the causal chain of public opinion monitoring events; Identify potential risks and hazards of events based on the causal chain of public opinion monitoring.

2. A potential risk identification device based on multidimensional information and event causal analysis, used to implement the potential risk identification method based on multidimensional information and event causal analysis as described in claim 1, characterized in that, The potential risk identification device based on multidimensional information and event causality analysis includes: The multimodal data acquisition module is used to acquire multimodal data from subscribed public opinion monitoring targets, including audio data, video data, and text data; The data feature extraction module is used to extract features from multimodal data, including timbre features, facial features, and text information features. The data structuring analysis module is used to perform data structuring analysis based on multimodal information features and extract key attribute structured data for each event; the key attribute structured data for each event includes time, location, people involved, specific event information, and risk index; The event clustering analysis module is used to perform clustering analysis on all events based on the key attribute structured data of each event, dividing all events into a series of large event sets; The information deduplication and integration module is used to perform information deduplication and integration analysis on each small event in a large event set to obtain structured data of key attributes of each large event set. The event causal chain analysis module is used to analyze the causal chain of a series of large event sets based on the key attribute structured data of each large event set, and obtain the event causal chain for public opinion monitoring. The potential risk and hazard identification module is used to identify potential risks and hazard of events based on the causal chain of events monitored by public opinion.

3. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the potential risk identification method based on multidimensional information and event causal analysis as described in claim 1.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the potential risk identification method based on multidimensional information and event causal analysis as described in claim 1.

5. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the potential risk identification method based on multidimensional information and event causal analysis as described in claim 1.

Images