An international related information dynamic monitoring and early warning method and system based on artificial intelligence

Abstract

The application discloses an international related information dynamic monitoring and early warning method and system based on artificial intelligence. After preprocessing of multi-source international related information data, the method completes exclusive sentiment polarity quantization and event extraction of a target entity, constructs an international relationship knowledge graph, realizes incremental update of relationship strength, extracts multi-dimensional related information heat characteristics by fusing propagation behavior data, constructs a situation characteristic vector, calculates a risk score through a time sequence modeling network, and realizes graded early warning of related information in combination with situation mutation detection. The application solves the problems of low monitoring efficiency, poor multi-language adaptability, accumulated analysis error and insufficient early warning precision of traditional international related information monitoring, and can be widely applied to scenes such as international related information research and judgment and international relationship research.

Description

Technical Field

[0001] This invention relates to the field of information monitoring technology, specifically to a method and system for dynamic monitoring and early warning of international relevant information based on artificial intelligence. Background Technology

[0002] Driven by the continuous advancement of globalization and the rapid development of new media technologies, the dissemination ecosystem of international information has undergone fundamental changes. Currently, international information is characterized by its massive volume, multilingual nature, and real-time nature. Traditional information processing methods that rely on manual collection and organization can no longer meet the needs of practical applications, and their limitations are becoming increasingly apparent.

[0003] First, the problem of information overload is particularly prominent. Globally, millions of pieces of information, including international news and social media posts, are generated daily. Manually processing this massive amount of information is not only extremely inefficient but also suffers from significant delays. Often, by the time manual analysis is completed, the relevant international events have already undergone new changes, rendering the analysis results worthless.

[0004] Secondly, language barriers are difficult to overcome. International information involves multiple languages ​​such as English, Russian, Arabic, French, and Spanish, and traditional manual processing methods cannot achieve comprehensive coverage of multilingual information. Even with a professional multilingual research team, the labor costs would be extremely high, and it would be difficult to guarantee the timeliness and comprehensiveness of information processing.

[0005] Secondly, the timeliness is severely lacking. Manually collecting, analyzing, and writing reports on relevant information takes a considerable amount of time. From information acquisition to the formation of a complete analysis report, it usually takes several days. For sudden international information-related events, this approach often cannot respond in a timely manner, leading to missed opportunities for optimal assessment.

[0006] Furthermore, manual analysis is highly subjective. Different analysts, due to differences in knowledge background, political stance, and personal practical experience, may reach drastically different conclusions regarding the same international information-related event. This subjective judgment lacks a unified standardized evaluation system, making it difficult to form replicable and verifiable standardized research results, thus affecting the objectivity and accuracy of relevant information analysis.

[0007] At the same time, the correlation analysis of related information is quite difficult. International relations are intricate and complex, and a single international related information event often involves multiple countries, multiple issues, and multiple levels. It is difficult to uncover the deep-seated connections behind the event by relying solely on manual methods.

[0008] In summary, existing information processing methods and systems have many limitations and cannot meet the needs of international political research for efficient, comprehensive, accurate, and timely processing of relevant information. Therefore, developing an intelligent information monitoring system specifically designed for the needs of international political research has become an urgent technical problem to be solved. Summary of the Invention

[0009] To address these issues, this invention proposes a method and system for dynamic monitoring and early warning of internationally relevant information based on artificial intelligence.

[0010] According to one aspect of the present invention, a method for dynamic monitoring and early warning of international relevant information based on artificial intelligence is proposed, comprising the following steps: S1, acquire multi-source international related information data and preprocess it to obtain standardized multi-source related information text; S2, perform entity recognition on the standardized multi-source related information text, generate a target entity representation vector based on the entity recognition result, and assign attention weights to the semantic representation of each word in the standardized multi-source related information text through a target-guided attention network, using the target entity representation vector as a condition, and extract contextual information related to the sentiment of the target entity to calculate the corresponding sentiment polarity score. S3, extract events from the standardized multi-source related information text to obtain corresponding structured events and event feature vectors, obtain the target entity and the relationship edge between the target entity corresponding to each international related information event based on the structured events, construct the corresponding international relations knowledge graph, calculate the relationship strength of the relationship edge, and update the relationship strength of the relationship edge through incremental update to obtain the corresponding relationship strength increment; S4, obtain the dissemination behavior data corresponding to the multi-source international related information data, and obtain the corresponding multi-dimensional related information heat feature vector based on the dissemination behavior data; S5. Construct a situation feature vector based on the sentiment polarity score, the event feature vector, the relationship strength increment, and the multidimensional related information heat feature vector, and calculate the change in the situation feature vector. Input the situation feature vector into a time-series modeling network to obtain a hidden state vector. Input the hidden state vector into a pre-trained risk scoring network to obtain a risk score value. Based on the change in the situation feature vector and the risk score value, perform graded early warning of the related information situation.

[0011] According to one aspect of the present invention, an international information dynamic monitoring and early warning system based on artificial intelligence is proposed, comprising the following modules according to any one of the first aspects: The relevant information data processing module is configured to acquire multi-source international relevant information data and preprocess it to obtain standardized multi-source relevant information text. The sentiment polarity assessment module is configured to perform entity recognition on the standardized multi-source related information text, generate a target entity representation vector based on the entity recognition result, and assign attention weights to the semantic representation of each word in the standardized multi-source related information text through a target-guided attention network, using the target entity representation vector as a condition, and extract contextual information related to the sentiment of the target entity to calculate the corresponding sentiment polarity score. The relation strength calculation module is configured to extract events from the standardized multi-source related information text to obtain corresponding structured events and event feature vectors, obtain the target entity and the relation edge between the target entity corresponding to each international related information event based on the structured events, construct the corresponding international relation knowledge graph, calculate the relation strength of the relation edge, and update the relation strength of the relation edge through incremental update to obtain the corresponding relation strength increment. The relevant information heat assessment module is configured to acquire the dissemination behavior data corresponding to the multi-source international relevant information data, and to acquire the corresponding multi-dimensional relevant information heat feature vector based on the dissemination behavior data. The relevant information situation assessment module is configured to construct a situation feature vector based on the sentiment polarity score, the event feature vector, the relationship strength increment, and the multidimensional relevant information heat feature vector, and calculate the change in the situation feature vector. The situation feature vector is then input into a time-series modeling network to obtain a hidden state vector. The hidden state vector is then input into a pre-trained risk scoring network to obtain a risk score value. Based on the change in the situation feature vector and the risk score value, the relevant information situation is classified and warned.

[0012] According to one aspect of the present invention, a computer program product is provided having a computer program stored thereon, which, when executed by a processor, performs the method as described in the first aspect.

[0013] The advantages of this invention are: By using a sentiment analysis mechanism based on target entity perception, we can achieve precise sentiment quantification for specific international actors, thus avoiding the problem of insufficient accuracy in general sentiment analysis. Establish an event-driven incremental update mechanism for international relations knowledge graphs to dynamically depict the evolution of relationships between entities and provide structured support for related information analysis; It provides a hierarchical early warning mechanism that integrates multi-dimensional information heat features and time series modeling, and combined with situational change detection, it greatly improves the accuracy and timeliness of early warning; The fully intelligent processing significantly reduces labor costs and provides standardized technical support for the analysis of relevant international information. Attached Figure Description

[0014] The accompanying drawings are included to provide a further understanding of the embodiments and are incorporated in and constitute a part of this specification. The drawings illustrate embodiments and, together with the description, serve to explain the principles of the invention. Other embodiments and many anticipated advantages of the embodiments will be readily recognized as they become better understood through reference to the following detailed description. Elements in the drawings are not necessarily to scale. The same reference numerals refer to corresponding similar parts.

[0015] Figure 1 A flowchart illustrating a method for dynamic monitoring and early warning of international relevant information based on artificial intelligence according to the present invention is shown. Figure 2 A schematic diagram of the structure of an international information dynamic monitoring and early warning system based on artificial intelligence according to the present invention is shown; Figure 3 A schematic diagram of a computer system architecture suitable for implementing the embodiments of this application is shown. Detailed Implementation

[0016] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0017] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0018] Figure 1 This paper presents a method for dynamic monitoring and early warning of international relevant information based on artificial intelligence, including the following steps: S1, acquire multi-source international related information data and preprocess it to obtain standardized multi-source related information text; S2, perform entity recognition on the standardized multi-source related information text, generate a target entity representation vector based on the entity recognition result, and assign attention weights to the semantic representation of each word in the standardized multi-source related information text through a target-guided attention network, using the target entity representation vector as a condition, and extract contextual information related to the sentiment of the target entity to calculate the corresponding sentiment polarity score. S3, extract events from the standardized multi-source related information text to obtain corresponding structured events and event feature vectors, obtain the target entity and the relationship edge between the target entity corresponding to each international related information event based on the structured events, construct the corresponding international relations knowledge graph, calculate the relationship strength of the relationship edge, and update the relationship strength of the relationship edge through incremental update to obtain the corresponding relationship strength increment; S4, obtain the dissemination behavior data corresponding to the multi-source international related information data, and obtain the corresponding multi-dimensional related information heat feature vector based on the dissemination behavior data; S5. Construct a situation feature vector based on the sentiment polarity score, the event feature vector, the relationship strength increment, and the multidimensional related information heat feature vector, and calculate the change in the situation feature vector. Input the situation feature vector into a time-series modeling network to obtain a hidden state vector. Input the hidden state vector into a pre-trained risk scoring network to obtain a risk score value. Based on the change in the situation feature vector and the risk score value, perform graded early warning of the related information situation.

[0019] In a specific implementation, the collection of multi-source international information data covers four core compliant data sources: international mainstream news media platforms with broad global influence, global public social networking platforms, official release channels of governments and international organizations, and publicly released research reports and analytical documents from authoritative international think tanks and professional research institutions. Through a pre-defined collection strategy, the above data sources are monitored in real time and periodically targeted for capture, achieving comprehensive coverage of international information across all channels.

[0020] In terms of language coverage, the system is pre-set to cover six major international languages: Chinese, English, Russian, Arabic, French, and Spanish. These languages ​​fully cover the core information dissemination links of major geopolitical regions around the world, enabling comprehensive collection of mainstream international information flows.

[0021] At the level of data collection triggering rules, the system has constructed a keyword database dedicated to the field of international information. The keyword database adopts a hierarchical architecture: the basic layer includes basic and general terms such as the names of sovereign states, international organizations, and core geopolitical regions; the professional layer includes geopolitical terms, diplomatic normative expressions, and professional terms corresponding to specific international issues, which can achieve precise and targeted collection of international information.

[0022] The pre-processed multilingual text is input to a pre-trained multilingual language model for unified encoding. The models used include pre-trained multilingual models such as mBERT and XLM-R, which, trained on large-scale multilingual corpora, can map text from different languages ​​to a unified semantic vector space. The text is encoded in the model as a set of high-dimensional context vectors, which simultaneously contain word semantics, contextual relationships, and cross-lingual alignment information.

[0023] Based on cross-language unified semantic representation, the system takes the shared encoding results output by the shared multilingual encoding network as input and executes three types of information extraction tasks in parallel: named entity recognition, relation extraction, and event extraction. This avoids the problem of error accumulation caused by the propagation of errors from previous tasks in the traditional serial processing flow.

[0024] Named entity recognition is used to identify various entities related to the field of international information from multilingual related information texts, including at least sovereign states, geopolitical regions, international organizations, government agencies, public figures, and key geographical areas, and to match each identified entity with a corresponding standardized entity type label.

[0025] The system incorporates a target entity perception subnetwork and a target-guided attention subnetwork, which communicate sequentially to collaboratively extract and quantify the sentiment features specific to the target entity. The target entity perception subnetwork, based on the standardized entity recognition results output from the preceding named entity recognition stage, performs feature aggregation processing on the contextual semantic representations corresponding to the target entity in the text, generating a target entity representation vector that fully characterizes the semantic features of the target entity. The target-guided attention subnetwork, using the target entity representation vector output by the target entity perception subnetwork as a constraint, assigns differentiated attention weights to the semantic representations corresponding to each word in the sentence. Based on the weighted aggregation of attention weights, it extracts contextual semantic information highly relevant to the sentiment expression of the target entity, ultimately outputting the sentiment polarity classification result and sentiment intensity quantification result for that target entity.

[0026] In one embodiment, the emotional polarity score in S2 is defined as , This indicates the probability that the standardized multi-source related information text presents a positive sentiment towards the target entity. This indicates the probability that the standardized multi-source related information text presents a negative sentiment towards the target entity. This represents the emotional intensity coefficient for the target entity, where, when When, it indicates that the emotional tendency towards the target entity is positive. When, it indicates a negative emotional tendency towards the target entity. When this occurs, it indicates that the emotional tendency toward the target entity is close to neutral or that positive and negative emotions cancel each other out.

[0027] For an event e containing multiple relevant texts within the same time window, the sentiment polarity scores corresponding to each text can be weighted and aggregated to obtain an event-level sentiment polarity score: ; in: M represents the number of texts involving the target entity within the current event or time window; This represents the sentiment polarity score corresponding to the m-th text; The weight of the m-th text can be determined based on at least one of the following: the credibility of the text source, its dissemination influence, and the recentity of its publication time.

[0028] Relation extraction is used to identify semantic relationships between different entities in a text, including at least job affiliation, organizational affiliation, cooperative interaction, and adversarial relationships. The output is a standardized combination of entity pairs and corresponding relationship types.

[0029] Event extraction is used to identify internationally relevant events described in multilingual related information texts and extract the core key elements corresponding to the event, including at least the event type, event trigger words, event participants, time of occurrence, location of occurrence, and event development outcome, and output standardized structured event information.

[0030] For example, for text containing information related to diplomatic meetings, the system can accurately identify the corresponding individuals, institutions, and regions within the text, extract the corresponding job titles and national affiliations between entities, and further identify the type of diplomatic interaction event corresponding to the text, extracting core elements such as the participants, time, and location of the event. The standardized and structured output results obtained through the above information extraction can provide unified and standardized data input for the subsequent construction and in-depth analysis of international relations knowledge graphs.

[0031] In one embodiment, the event feature vector in S3 specifically includes: the feature vector of the type of internationally related information event and the number of internationally related information events within the same time window.

[0032] The system uses standardized entities obtained through information extraction as nodes and semantic relationships between entities as edges to construct an international relations knowledge graph for the field of international information. When the system identifies a new entity not yet included in the graph, it creates a standardized node corresponding to that entity in the knowledge graph and simultaneously configures the entity's basic attribute information. When the system identifies an existing entity already included in the graph, it updates the attribute information of the node corresponding to that entity in real time. The updated content includes at least the entity's latest appearance time, related international information event records, and changes in related entities.

[0033] To characterize the dynamic evolution of relationships between entities, each relationship edge is assigned a relationship strength weight, which reflects the frequency, sentiment, and temporal impact of interactions between entities. For newly extracted entity relationships, the system does not directly include them in the knowledge graph. Instead, it uses a pre-defined multi-source cross-validation mechanism to screen the relationship for validity. When the corresponding relationship of the same entity pair appears multiple times in different independent text sources, or is cross-validated by multiple independent information sources within the same pre-defined statistical time window, the relationship strength is updated accordingly. Only when the relationship strength exceeds a pre-defined threshold is the relationship formally included as a valid edge in the international relations knowledge graph.

[0034] In one embodiment, calculating the relationship strength of the relation edge in S3, and updating the relationship strength of the relation edge using an incremental update method to obtain the corresponding relationship strength increment specifically includes: The relationship strength of the relationship edge is calculated based on the event frequency factor, the sentiment polarity factor, and the time decay factor, and the calculation formula is as follows: , in, Represents the target entity With the target entity The strength of the relationship between them This indicates the target entity involved within the statistical time window. With the target entity The number of events, Indicates the first Event weighting factors for internationally relevant information events. Indicates the first The emotional polarity factor corresponding to each internationally relevant information event. Indicates the first The timing of international related information events Indicates the current time. Indicates the time decay coefficient; Among them, the emotional polarity factor The value is determined based on the sentiment analysis results between target entities in standardized multi-source relevant text, and the rules for its selection are as follows: The preset emotion adjustment coefficient: , The relation strength of the relation edges is updated incrementally, and the update formula is as follows: , in, Indicates the strength of the relationship at the previous point in time. This indicates the time interval between the current update time and the last update time. and These represent the event weight factor and the sentiment polarity factor corresponding to the new event, respectively.

[0035] Based on the structured information and international relations knowledge graph already obtained, the scale of international information events, the hierarchy of the dissemination links and key dissemination nodes are quantitatively evaluated, and the relevant information heat index, dissemination breadth index, dissemination depth index and key KOL identification results are output.

[0036] In one embodiment, the dissemination behavior data in S4 specifically includes the publishing relationships, forwarding relationships, citation relationships, commenting relationships, dissemination timestamps, publishing account attributes, and source media types of the multi-source international related information data. A dissemination behavior graph is constructed based on this data, with all accounts participating in the dissemination of the international related information event as nodes, where the root node is the initial account. Forwarding, citation, and commenting behaviors between accounts are used as connecting edges to obtain corresponding multi-dimensional related information popularity feature vectors. These multi-dimensional related information popularity feature vectors specifically include: dissemination breadth. Dissemination depth indicators , growth rate of dissemination Event popularity index Indicators of change in behavioral activity and key KOL engagement metrics .

[0037] In one embodiment, the breadth of propagation is defined as ,in, Indicates international related information events The number of unique accounts participating in the dissemination (number of accounts posting / forwarding / commenting / quoting). Indicates international related information events The number of duplicates across platforms (deduplication counts for news sites, social media platforms, etc.). Indicates international related information events The number of languages ​​involved in deduplication. Indicates international related information events The number of regions involved (statistics based on the account's location / the country of media source). This indicates the preset weighting coefficients; The propagation depth index is defined as ,in, , For international related information events The propagation behavior diagram, The propagation behavior diagram is the first one. The number of propagation nodes in the layer; The propagation growth rate term is defined as , For time slices The cumulative number of transmissions; The event popularity index is defined as follows: , This indicates the preset weighting coefficients; The behavioral activity change index Defined as ,in This represents the activity level of target entity i within the time window t. This indicates the activity level of target entity i within the previous time window. It is a smoothing constant. , This represents the number of news texts mentioning target entity i within a time window t. This represents the number of social media posts related to target entity i within a time window t. This represents the number of forwards related to target entity i within the time window t. This represents the number of comments related to target entity i within a time window t. This represents the number of structured events that target entity i participates in within the time window t. This indicates the preset weighting coefficients; The key KOL engagement metric is used to characterize account nodes that play a crucial role in driving the spread of the message in the propagation graph, and is defined as follows: ,in, Indicates international related information events Key KOL collection, Indicates participation in international related information events The complete set of all account nodes used for propagation. Indicates account node For international related information events within the current time window The intensity of the transmission behavior, Indicates account node For international related information events within the current time window The intensity of the propagation behavior is obtained by weighted summation of the account node's posting, forwarding, commenting, and quoting behaviors within the current time window.

[0038] In one embodiment, for each account node in the propagation behavior graph, the structural centrality component, interaction contribution component, and cross-domain diffusion component corresponding to each account node are calculated, and the structural centrality component is defined as follows: Interactive contribution weight Transdomain diffusion component Each account node has a pre-defined weighting coefficient. The three components of each account node are multiplied by their respective weighting coefficients, and then summed to obtain the overall influence score for each account node. , The weighting coefficient is used. All account nodes within the propagation behavior graph are sorted from highest to lowest based on their comprehensive influence scores. A predetermined number of account nodes at the top of the sorting are selected as the key KOL set for the international related information event. The structural centrality components are calculated using the weighted PageRank algorithm or by calculating the in-degree centrality index. Taking weighted PageRank as an example: ; in For edge weights (such as forwarding strength, reference strength). It is the damping factor; This represents all nodes pointing to the account in the event propagation subgraph. The set of neighboring nodes, i.e., those related to the account node. There is a set of account nodes with incoming edges; Indicates account node Pointing to account node The edge weights are used to characterize the edge weights of nodes. To the node The propagation correlation strength; the edge weight can be determined based on at least one of the following: number of forwards, number of citations, and number of comment responses; Indicates account node The weighted PageRank value in the event propagation subgraph is used to characterize the nodes. The structural importance in the propagation network and the magnitude of its influence that can be transmitted to neighboring nodes.

[0039] The interactive contribution component is obtained by summing the total number of reposts, comments, and citations triggered by the international information events posted by the corresponding account node, and then performing a logarithmic operation on the result after adding 1 to the sum. ; The cross-domain diffusion component is obtained by separately counting the number of cross-platform diffusions and cross-language diffusions caused by the content published by the corresponding account node, adding 1 to each of the two quantities, performing a logarithmic operation on the sum, and then summing the results. ; in The number of cross-platform disseminations triggered by this account, The number of cross-language diffusions it has caused.

[0040] Based on the output of relevant information heat analysis, sentiment assessment, and event extraction, risk assessment and trend prediction of international relevant information situation are carried out. It outputs relevant information risk level and early warning results by integrating and modeling multi-dimensional information such as relevant information heat, sentiment tendency, key events and actor dynamics.

[0041] To characterize the temporal evolution of relevant information situation, the situation feature vector within a continuous time window is... Input the temporal modeling network in chronological order.

[0042] The temporal modeling network adopts a recurrent neural network structure, and its update process is as follows: , in The hidden state vector is the hidden state from the previous time step. The risk score is obtained by inputting the pre-trained risk scoring network. , It is the Sigmoid activation function. , These are the trainable parameters for the risk scoring network.

[0043] The following formula is used to classify and issue early warnings regarding the status quo of relevant information: , in, This is the threshold for risk classification.

[0044] Meanwhile, to enhance sensitivity to sudden international information events, an anomaly detection mechanism is introduced based on risk prediction.

[0045] Calculate the change in the situation feature vector: ; The following conditions are met when a signal is identified as a change in relevant information: ; in This is a preset abnormal threshold.

[0046] When a mutation signal occurs, the corresponding risk score is given. It is either upgraded or directly triggered to determine an extremely high risk warning.

[0047] In one embodiment, the method proposed in this invention captures a news text from a mainstream international media outlet concerning diplomacy. The text describes a formal meeting between high-level diplomatic officials from two target countries to discuss bilateral relations and regional security issues. During the information processing phase, named entity recognition is first performed to accurately identify entities such as the diplomatic officials' positions, corresponding individuals, and the location of the meeting, and then assigns standardized type tags to each entity. Subsequently, event type identification is performed, determining that the text corresponds to a diplomatic meeting. Based on the formal consultation statements and overall narrative tone, the system uses sentiment analysis to determine that the text's sentiment is neutral to slightly positive. Simultaneously, the system translates the foreign language text into a preset unified analysis language, ensuring standardized processing of multilingual information.

[0048] During the knowledge graph update process, the system performs the following operations: 1. Add this diplomatic meeting as a new event node to the main chain of diplomatic relations between the target two countries; 2. Simultaneously record the event's core attributes, such as the time of occurrence, location, and participating entities; 3. If the text mentions specific issues of bilateral concern, then establish a semantic association between the event node and the corresponding issue node; 4. Update the relationship strength corresponding to the bilateral diplomatic relations between the target countries in real time.

[0049] In the relevant information analysis phase, the system generates a standardized relevant information analysis report based on all collected data within a preset time window. The report content includes at least: Related information dissemination statistics: The total amount of related information reports concerning the target bilateral relationship within a preset time window, and the change in the amount of reports within this statistical period compared to the previous period; Sentiment Distribution: Statistically analyze the sentiment distribution of all relevant information texts, clarify the proportion of positive, neutral, and negative texts, and mark the core focus areas corresponding to each type of sentiment text; Hot topic sub-topic analysis: Deconstruct the core hot topics of current relevant information, count the proportion of relevant information for each sub-topic, and clarify the core focus of each sub-topic, including bilateral science and technology interaction, bilateral economic and trade exchanges, cross-sector pragmatic cooperation, regional security issues, etc. Key Opinion Leader Viewpoint Analysis: Identify key opinion leaders who have a core influence in the dissemination of relevant information, and analyze and present their core viewpoints on the target bilateral relationship and the data on their dissemination influence.

[0050] In the early warning assessment stage, the system outputs prediction results based on the Long Short-Term Memory Network (LSTM) time series prediction model, outputs sentiment trend prediction results, and finally outputs early warning signals of corresponding levels based on risk scores, while simultaneously outputting corresponding response suggestions, including: closely monitoring the dynamics of official policy releases of the target country, developing corresponding relevant information response plans for potential policy changes, and strengthening the tracking and analysis of the positions of relevant third-party entities.

[0051] According to one aspect of the present invention, an artificial intelligence-based dynamic monitoring and early warning system for international related information is proposed, wherein the method described in any one of the first aspects, such as... Figure 2 As shown, it includes the following modules: The relevant information data processing module 201 is configured to acquire multi-source international relevant information data and preprocess it to obtain standardized multi-source relevant information text; The sentiment polarity assessment module 202 is configured to perform entity recognition on the standardized multi-source related information text, generate a target entity representation vector based on the entity recognition result, and assign attention weights to the semantic representation of each word in the standardized multi-source related information text through a target-guided attention network, using the target entity representation vector as a condition, and extract contextual information related to the sentiment of the target entity to calculate the corresponding sentiment polarity score. The relation strength calculation module 203 is configured to extract events from the standardized multi-source related information text to obtain corresponding structured events and event feature vectors, obtain the target entity and the relation edge between the target entity corresponding to each international related information event based on the structured events, construct the corresponding international relation knowledge graph, calculate the relation strength of the relation edge, and update the relation strength of the relation edge through incremental update to obtain the corresponding relation strength increment. The relevant information heat assessment module 204 is configured to acquire the dissemination behavior data corresponding to the multi-source international relevant information data, and acquire the corresponding multi-dimensional relevant information heat feature vector based on the dissemination behavior data. The relevant information situation assessment module 205 is configured to construct a situation feature vector based on the sentiment polarity score, the event feature vector, the relationship strength increment, and the multidimensional relevant information heat feature vector, and calculate the change in the situation feature vector. The situation feature vector is then input into a time-series modeling network to obtain a hidden state vector. The hidden state vector is then input into a pre-trained risk scoring network to obtain a risk score value. Based on the change in the situation feature vector and the risk score value, the relevant information situation is classified and warned.

[0052] The following is for reference. Figure 3 It shows a schematic diagram of the structure of a computer system 300 suitable for implementing electronic devices according to embodiments of the present application. Figure 3The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0053] like Figure 3 As shown, the computer system 300 includes a central processing unit (CPU) 301, which performs various appropriate actions and processes based on programs stored in read-only memory (ROM) 302 or programs loaded from storage section 309 into random access memory (RAM) 304. RAM 304 also stores various programs and data required for the operation of system 300. CPU 301, ROM 302, ROM 303, and RAM 304 are interconnected via bus 305. Input / output (I / O) interface 306 is also connected to bus 305.

[0054] The following components are connected to I / O interface 306: an input section 307 including a keyboard, mouse, etc.; an output section 308 including a liquid crystal display (LCD) and speakers, etc.; a storage section 309 including a hard disk, etc.; and a communication section 310 including a network interface card such as a LAN card and a modem, etc. The communication section 310 performs communication processing via a network such as the Internet. A drive 311 is also connected to I / O interface 306 as needed. A removable medium 312, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 311 as needed so that computer programs read from it can be installed into storage section 309 as needed.

[0055] Specifically, according to embodiments of this disclosure, the processes described above with reference to the flowcharts are implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program is downloaded and installed from a network via communication section 310, and / or installed from removable medium 312. When the computer program is executed by central processing unit (CPU) 301, it performs the functions defined in the methods of this application.

[0056] It should be noted that the computer-readable storage medium of this application is a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium is, for example—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium is any tangible medium that contains or stores a program used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium includes a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals take various forms, including, but not limited to, electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium or any computer-readable storage medium other than a computer-readable storage medium may transmit, propagate, or transfer a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wireless, wireline, optical fiber, RF, etc., or any suitable combination thereof.

[0057] Computer program code for performing the operations of this application is written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code executes entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer is connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or connected to an external computer (e.g., via the Internet using an Internet service provider).

[0058] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram represents a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually execute substantially in parallel, and they may sometimes execute in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, is implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0059] The modules described in the embodiments of this application are implemented in software or hardware.

[0060] On the other hand, this application also provides a computer-readable storage medium included in the electronic device described in the above embodiments; it also exists independently and is not assembled into the electronic device. The computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to: S1, acquire multi-source international related information data and preprocess it to obtain standardized multi-source related information text; S2, perform entity recognition on the standardized multi-source related information text, generate a target entity representation vector based on the entity recognition result, and, using the target entity representation vector as a condition, assign attention weights to the semantic representations of each word in the standardized multi-source related information text through a target-guided attention network, extract contextual information related to the target entity's sentiment, and calculate the corresponding sentiment polarity score; S3, extract events from the standardized multi-source related information text to obtain corresponding structured events and event feature vectors, and obtain the target related event for each international related information event based on the structured events. S4. Construct a corresponding international relations knowledge graph by identifying the relationship edges between the target entity and the target entity, calculate the relationship strength of the relationship edges, and update the relationship strength of the relationship edges through incremental updates to obtain the corresponding relationship strength increment; S5. Obtain the dissemination behavior data corresponding to the multi-source international related information data, and obtain the corresponding multi-dimensional related information heat feature vector based on the dissemination behavior data; S6. Construct a situation feature vector based on the sentiment polarity score, the event feature vector, the relationship strength increment, and the multi-dimensional related information heat feature vector, and calculate the change in the situation feature vector. Input the situation feature vector into a temporal modeling network to obtain a hidden state vector, input the hidden state vector into a pre-trained risk scoring network to obtain a risk score value, and perform graded early warning of the related information situation based on the change in the situation feature vector and the risk score value.

[0061] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for dynamic monitoring and early warning of internationally relevant information based on artificial intelligence, characterized in that, Includes the following steps: S1, acquire multi-source international related information data and preprocess it to obtain standardized multi-source related information text; S2, perform entity recognition on the standardized multi-source related information text, generate a target entity representation vector based on the entity recognition result, and assign attention weights to the semantic representation of each word in the standardized multi-source related information text through a target-guided attention network, using the target entity representation vector as a condition, and extract contextual information related to the sentiment of the target entity to calculate the corresponding sentiment polarity score. S3, extract events from the standardized multi-source related information text to obtain corresponding structured events and event feature vectors, obtain the target entity and the relationship edge between the target entity corresponding to each international related information event based on the structured events, construct the corresponding international relations knowledge graph, calculate the relationship strength of the relationship edge, and update the relationship strength of the relationship edge through incremental update to obtain the corresponding relationship strength increment; S4, obtain the dissemination behavior data corresponding to the multi-source international related information data, and obtain the corresponding multi-dimensional related information heat feature vector based on the dissemination behavior data; S5. Construct a situation feature vector based on the sentiment polarity score, the event feature vector, the relationship strength increment, and the multidimensional related information heat feature vector, and calculate the change in the situation feature vector. Input the situation feature vector into a time-series modeling network to obtain a hidden state vector. Input the hidden state vector into a pre-trained risk scoring network to obtain a risk score value. Based on the change in the situation feature vector and the risk score value, perform graded early warning of the related information situation.

2. The method for dynamic monitoring and early warning of international relevant information based on artificial intelligence according to claim 1, characterized in that, The dissemination behavior data mentioned in S4 specifically includes the publishing relationships, forwarding relationships, citation relationships, commenting relationships, dissemination timestamps, publishing account attributes, and source media types of the multi-source international related information data. A dissemination behavior graph is constructed based on this data. This graph uses all accounts participating in the dissemination of the international related information event as nodes, and forwarding, citation, and commenting behaviors between accounts as connecting edges. A corresponding multi-dimensional related information popularity feature vector is obtained. This multi-dimensional related information popularity feature vector specifically includes: dissemination breadth. Dissemination depth indicators , growth rate of dissemination Event popularity index Indicators of change in behavioral activity and key KOL engagement metrics .

3. The method for dynamic monitoring and early warning of international relevant information based on artificial intelligence according to claim 2, characterized in that, The breadth of dissemination is defined as ,in, Indicates international related information events The number of duplicate accounts involved in the dissemination. Indicates international related information events The number of duplicates involved in the platform deduplication. Indicates international related information events The number of languages ​​involved in deduplication. Indicates international related information events The number of regions involved This indicates the preset weighting coefficients; The propagation depth index is defined as ,in, , For international related information events The propagation behavior diagram The propagation behavior diagram is the first one. The number of propagation nodes in the layer; The propagation growth rate term is defined as , For time slices The cumulative number of transmissions; The event popularity index is defined as follows: , This indicates the preset weighting coefficients; The behavioral activity change index Defined as ,in This represents the activity level of target entity i within the time window t. This indicates the activity level of target entity i within the previous time window. It is a smoothing constant. ， This represents the number of news texts mentioning target entity i within a time window t. This represents the number of social media posts related to target entity i within a time window t. This represents the number of forwards related to target entity i within the time window t. This represents the number of comments related to target entity i within a time window t. This represents the number of structured events that target entity i participates in within the time window t. This indicates the preset weighting coefficients; The key KOL engagement metric is defined as follows: ,in, Indicates international related information events Key KOL collection, Indicates participation in international related information events The complete set of all account nodes used for propagation. Indicates account node For international related information events within the current time window The intensity of the transmission behavior, Indicates account node For international related information events within the current time window The intensity of the propagation behavior is obtained by weighted summation of the account node's posting, forwarding, commenting, and quoting behaviors within the current time window.

4. The method for dynamic monitoring and early warning of international related information based on artificial intelligence according to claim 3, characterized in that, For each account node in the propagation behavior graph, the structural centrality component, interaction contribution component, and cross-domain diffusion component corresponding to each account node are calculated, and corresponding weight coefficients are preset for the structural centrality component, interaction contribution component, and cross-domain diffusion component. The three components of a single account node are multiplied by the corresponding weight coefficients and then summed to obtain the comprehensive influence score of each account node. All account nodes in the propagation behavior graph are sorted in descending order of comprehensive influence score, and a preset number of account nodes at the top of the sort are selected as the key KOL set of the international related information event. The structural centrality components are obtained by weighted PageRank algorithm or by calculating in-degree centrality index; The interactive contribution component is obtained by summing the total number of reposts, comments, and citations triggered by the international information events published by the corresponding account node, and then performing a logarithmic operation on the result after adding 1 to the sum. The cross-domain diffusion component is obtained by separately counting the number of cross-platform diffusions and cross-language diffusions caused by the content published by the corresponding account node, adding 1 to each of the two quantities, performing a logarithmic operation on the sum, and then summing them up.

5. The method for dynamic monitoring and early warning of international relevant information based on artificial intelligence according to claim 1, characterized in that, The emotional polarity score described in S2 is defined as follows: , This indicates the probability that the standardized multi-source related information text presents a positive sentiment towards the target entity. This indicates the probability that the standardized multi-source related information text presents a negative sentiment towards the target entity. This represents the emotional intensity coefficient for the target entity, where, when When, it indicates that the emotional tendency towards the target entity is positive. When, it indicates a negative emotional tendency towards the target entity. When this occurs, it indicates that the emotional tendency toward the target entity is close to neutral or that positive and negative emotions cancel each other out.

6. The method for dynamic monitoring and early warning of international relevant information based on artificial intelligence according to claim 1, characterized in that, S3 calculates the relation strength of the relation edge, and updates the relation strength of the relation edge using an incremental update method to obtain the corresponding relation strength increment, specifically including: The relationship strength of the relationship edge is calculated based on the event frequency factor, the sentiment polarity factor, and the time decay factor, and the calculation formula is as follows: ， in, Represents the target entity With the target entity The strength of the relationship between them This indicates the target entity involved within the statistical time window. With the target entity The number of events, Indicates the first Event weighting factors for internationally relevant information events. Indicates the first The emotional polarity factor corresponding to each internationally relevant information event. Indicates the first The timing of international related information events Indicates the current time. Indicates the time decay coefficient; Among them, the emotional polarity factor The value is determined based on the sentiment analysis results between target entities in standardized multi-source relevant text, and the rules for its selection are as follows: The preset emotion adjustment coefficient: ， The relation strength of the relation edges is updated incrementally, and the update formula is as follows: ， in, Indicates the strength of the relationship at the previous point in time. This indicates the time interval between the current update time and the last update time. and These represent the event weight factor and the sentiment polarity factor corresponding to the new event, respectively.

7. The method for dynamic monitoring and early warning of international relevant information based on artificial intelligence according to claim 1, characterized in that, The event feature vectors described in S3 specifically include: feature vectors of internationally relevant information event types and the number of internationally relevant information events within the same time window.

8. The method for dynamic monitoring and early warning of international relevant information based on artificial intelligence according to claim 1, characterized in that, S5 specifically includes: constructing situation feature vectors. The situation feature vector Input the time series modeling network to obtain the time. Hidden state vector ,in The hidden state vector is the hidden state from the previous time step. The risk score is obtained by inputting the pre-trained risk scoring network. , It is the Sigmoid activation function. , These are the trainable parameters for the risk scoring network; The following formula is used to classify and issue early warnings regarding the status quo of relevant information: ， in, The risk classification threshold is used as the basis for calculating the change in the situation feature vector. ,when Risk score at the corresponding time It is either upgraded or directly triggered to determine an extremely high risk warning.

9. An international information dynamic monitoring and early warning system based on artificial intelligence, characterized in that, The method according to any one of claims 1-8 comprises the following modules: The relevant information data processing module is configured to acquire multi-source international relevant information data and preprocess it to obtain standardized multi-source relevant information text. The sentiment polarity assessment module is configured to perform entity recognition on the standardized multi-source related information text, generate a target entity representation vector based on the entity recognition result, and assign attention weights to the semantic representation of each word in the standardized multi-source related information text through a target-guided attention network, using the target entity representation vector as a condition, and extract contextual information related to the sentiment of the target entity to calculate the corresponding sentiment polarity score. The relation strength calculation module is configured to extract events from the standardized multi-source related information text to obtain corresponding structured events and event feature vectors, obtain the target entity and the relation edge between the target entity corresponding to each international related information event based on the structured events, construct the corresponding international relation knowledge graph, calculate the relation strength of the relation edge, and update the relation strength of the relation edge through incremental update to obtain the corresponding relation strength increment. The relevant information heat assessment module is configured to acquire the dissemination behavior data corresponding to the multi-source international relevant information data, and to acquire the corresponding multi-dimensional relevant information heat feature vector based on the dissemination behavior data. The relevant information situation assessment module is configured to construct a situation feature vector based on the sentiment polarity score, the event feature vector, the relationship strength increment, and the multidimensional relevant information heat feature vector, and calculate the change in the situation feature vector. The situation feature vector is then input into a time-series modeling network to obtain a hidden state vector. The hidden state vector is then input into a pre-trained risk scoring network to obtain a risk score value. Based on the change in the situation feature vector and the risk score value, the relevant information situation is classified and warned.

10. A computer program product, characterized in that, It stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1-8.

Images