A social group industry data monitoring method, system, medium and product

Abstract

This application provides a method, system, medium, and product for monitoring data on social organizations, relating to the field of network monitoring technology. The method includes: first, constructing a standard knowledge graph based on a compliant social organization database, using social organization nodes as the core to associate multi-dimensional attribute nodes, forming an authoritative reference benchmark; then, extracting target text segments containing the name of the target social organization from online speech texts, identifying accompanying entities co-occurring with the name; next, filtering suspected target social organization nodes through literal similarity matching, and extracting their corresponding attribute nodes; then, calculating the structural matching score between the accompanying entity and the attribute nodes; if the score is below a threshold, it is determined to be a counterfeit entity and data isolation is performed; if the score meets the threshold, the text and node are associated and bound, and included in the monitoring statistics of the corresponding entity. This method accurately separates genuine and fake social organizations while ensuring the absolute purity and high fidelity of the monitoring statistics of real social organization online speech.

Description

Technical Field

[0001] This application relates to the field of network monitoring technology, and in particular to a method, system, medium and product for monitoring data of social organizations and industries. Background Technology

[0002] Social organizations are non-profit social organizations voluntarily formed by citizens to realize the common will of their members and carry out activities in accordance with their bylaws. As an important component of contemporary social governance and political life, their standardized operation is of great significance to safeguarding the legitimate rights and interests of the public. In recent years, the internet has become the core channel for the public to obtain information about social organizations, express related opinions, and provide feedback. Monitoring internet speech by social organizations is not only an important tool for industry regulators to grasp the operational dynamics of social organizations and regulate the industry's development order, but also a key link in maintaining the credibility of legitimate social organizations and preventing the spread of false and illegal information that could lead to social speech risks. Efficient and accurate speech monitoring is of irreplaceable importance in maintaining a healthy ecosystem for social organizations and implementing standardized regulatory requirements.

[0003] Existing technologies typically employ named entity recognition and string similarity matching techniques based on Natural Language Processing (NLP). Specifically, word segmentation and part-of-speech tagging algorithms are first used to extract entity words resembling suspected organization names from the captured unstructured online text. Then, a string edit distance algorithm or a word vector model is introduced to calculate the text-level similarity between the extracted name strings and standard names from a database of officially provided legitimate social organizations. When the calculated similarity score exceeds a preset matching threshold, the monitoring system logically binds the online statement text to the specific legitimate social organization and includes it in the entity's monitoring archive and data quantification reports for aggregation and output.

[0004] However, internet data is not only rife with non-standard abbreviations (such as abbreviating "China XX Protection Association" to "Zhong XX Association"), but also harbors a large number of unregistered counterfeit organizations (such as "China XX Protection Federation"). These counterfeiters deliberately construct highly similar naming features to conduct illegitimate business. Existing technologies rely solely on one-dimensional text character similarity or contextual word vectors for entity calculation, making them susceptible to being deceived by such deliberately constructed morphological disguises. This makes it difficult to effectively distinguish these homogeneous entities with different names, leading to node alignment errors in the knowledge graph. These alignment errors can trigger a chain reaction in the network topology. That is, when an abnormal evaluation record and violation event edge of a counterfeit organization are incorrectly mapped and bound to the main node of a legitimate organization, it will pollute all downstream relation edges and attribute calculations of that node, causing a systematic bias in the final evaluation and potentially leading to an ineffective waste of resources. Summary of the Invention

[0005] This application provides a method, system, medium, and product for monitoring social organization industry data, which is used to eliminate interference from counterfeit social organization-related statements in order to improve the accuracy of monitoring data on social organization online statements.

[0006] Firstly, this application provides a method for monitoring industry data of social organizations, applied to a monitoring system. The method includes: constructing a standard knowledge graph based on an existing database of compliant social organizations, the standard knowledge graph including social organization nodes and attribute nodes associated with each social organization node; extracting target text segments containing the target name of a target social organization from online speech texts, and identifying accompanying entities co-occurring with the target name from the target text segments, the accompanying entities including at least one of a receiving account entity, an event venue, and associated individuals, the target name including both a full name and an abbreviation; and, based on the target name, in... The standard knowledge graph identifies target social group nodes with a literal similarity greater than a preset similarity threshold; it extracts the target attribute nodes corresponding to these target social group nodes; it calculates the structural matching score between the associated entity and the target attribute node; if the structural matching score is less than a preset matching score threshold, it determines that the target text segment points to a counterfeit community, and the target text segment is data isolated from the target social group node; if the structural matching score is greater than or equal to the preset matching score threshold, it associates and binds the target text segment with the target social group node, and includes the target text segment in the monitoring statistics of the target social group node.

[0007] By adopting the above technical solutions, it is evident that counterfeit communities with identical or highly similar names often exist in the online environment, making misjudgment highly likely when relying solely on name matching. This solution introduces a deep comparison between associated entities and standard attribute nodes in the knowledge graph. While counterfeit organizations can forge names, it is difficult for them to simultaneously possess genuine official accounts and associated core figures. This multi-dimensional cross-validation mechanism mandates that the text conforms to the real verification data in both structure and fact. Therefore, this solution fundamentally avoids the misidentification problem caused by simple "literal matching," accurately separating genuine and counterfeit communities. It effectively intercepts the infiltration of counterfeit community data while ensuring the absolute purity and high fidelity of the monitoring and statistical data of genuine communities.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, the attribute nodes in the standard knowledge graph carry time validity identifiers, which record the effective time and expiration time of the attribute node. The step of extracting the target attribute node corresponding to the target social group node further includes: extracting time information from the target text segment, which includes an explicitly marked timestamp or an estimated time determined by the publication time of the online speech text; filtering out valid attribute nodes whose time validity identifiers cover the time information from all attribute nodes associated with the target social group node; and using the valid attribute node as the target attribute node.

[0009] By adopting the above technical solution, the attribute status of real social organizations evolves dynamically over time, such as office address relocation, legal representative change, or renewal of payment account upon expiration. If only static attribute node comparison is used, it is easy to misjudge historically compliant text that has undergone genuine changes as counterfeit text. This solution precisely aligns the extracted text time information with the time validity identifier of the attribute nodes, forcing the system to verify only within slices that overlap on the timeline. This spatiotemporal verification mechanism, which introduces a time dimension, enables the system to dynamically adapt to the historical changes of community information, completely eliminating structural matching failures caused by information expiration or update misalignment, and significantly improving the robustness and rigor of authenticity determination results across complex time spans.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, the step of extracting a target text segment containing the target name corresponding to a target social group from online speech text specifically includes: scanning and locating the first and last character positions matching the target name in a continuous character sequence of the online speech text; performing a character-by-character traversal scan from the first character position towards the beginning of the online speech text, and stopping the scan when a preset event trigger word is matched or the scan span from the first character position reaches a first preset maximum character span, and determining the corresponding stopping position as the left boundary position, wherein the preset event trigger word is associated with the associated entity; performing a character-by-character traversal scan from the last character position towards the end of the online speech text, and stopping the scan when a preset event trigger word is matched or the scan span from the last character position reaches a second preset maximum character span, and determining the corresponding stopping position as the right boundary position; extracting all characters located between the left boundary position and the right boundary position to generate the target text segment as an analysis window.

[0011] By adopting the above technical solution, long online texts often contain multiple mixed themes. If a fixed length or direct extraction of the entire text is used for analysis, a large number of interfering entities related to other subjects will be introduced. This solution cleverly uses the beginning and end positions of the target name as anchor points, combined with "preset event trigger words" that are strongly related to the accompanying entities, to perform bidirectional dynamic detection. This allows the extracted analysis window to adaptively fit the actual business context of the community. At the same time, the introduction of a maximum character span limit can effectively prevent out-of-bounds scanning caused by long and complex sentences. This mechanism accurately removes redundant context-irrelevant noise, making the extracted accompanying entities have extremely high subject dependence, directly ensuring the signal-to-noise ratio and accuracy of subsequent structural matching calculations.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, the step of calculating the structural matching score between the accompanying entity and the target attribute node specifically includes: classifying the accompanying entity into the corresponding business feature dimension according to a preset static label mapping table; extracting the standard feature set from the target attribute node that is consistent with each of the business feature dimensions; calculating the data overlap between the character sequence of the accompanying entity and the character sequence in the standard feature set under each individual business feature dimension; counting the number of character intervals between each accompanying entity and the target name in the target text segment; assigning a corresponding positional distance weight to the accompanying entity according to a preset distance decay rule; and using the positional distance weight, performing a weighted summation of the data overlap calculated under each of the business feature dimensions to obtain the structural matching score.

[0013] By adopting the above technical solution, some entities appearing in the text may only be mentioned occasionally and not be the true attributes of the target community. This solution not only divides entities into specific business dimensions for fine-grained data overlap calculation, but more importantly, it introduces a "distance decay rule" based on the number of character intervals. In natural language expression logic, the closer an entity is to the target name, the higher its semantic probability of modifying the target. By using positional distance weights to weight the overlap of each dimension, marginal entities that deviate from the topic are de-weighted. This mathematical modeling method, which transforms spatial physical distance into semantic association strength, perfectly quantifies the subordinate relationship between entities and names, making the final structure matching score highly consistent with the real context and greatly reducing the misjudgment rate caused by coincidental co-occurrence.

[0014] In combination with some embodiments of the first aspect, in some embodiments, the step of matching a target social group node in the standard knowledge graph with a literal similarity greater than a preset similarity threshold according to the target name specifically includes: extracting key semantic components in the target name, querying a preset confusion character mapping library for the key semantic components to obtain replacement characters with similar visual forms or the same phonetic codes; replacing the original characters in the target name with the replacement characters to generate a first type of variant name; detecting and deleting non-semantic characters in the target name to generate a normalized name as a second type of variant name; summarizing the target name, the first type of variant name, and the second type of variant name to form the candidate name set; using each name in the candidate name set to perform matching in the standard knowledge graph respectively.

[0015] By adopting the above technical solution, in order to avoid conventional literal monitoring, counterfeit societies often use similar-looking characters (such as "协" and "办"), homophonic and异形 characters, or insert interfering punctuation to confuse the public. This solution takes the initiative to act, generates the first type of deformation at the visual and phonetic levels by combining the confusion character mapping library for key semantic components, and cleans non-semantic characters to generate the second type of variant. This is essentially an anti-deception mechanism of "attacking instead of defending". The system anticipates in advance the bypass strategies of malicious entities and spelling mistakes in network texts. Using a multi-dimensional candidate name set to perform parallel matching of the standard knowledge graph completely breaks the rigid limitation of single-string matching, greatly broadens the recall rate for hidden variant names, and realizes comprehensive and non-blind-spot monitoring of various network variant disguises.

[0016] In combination with some embodiments of the first aspect, in some embodiments, the method further includes: assigning a semantic importance level to each original character in the key semantic components; only performing replacement operations on the original characters with a semantic importance level lower than a preset importance threshold; counting the number of character differences between the generated first type of variant name and the target name; if the number of character differences exceeds a preset difference upper limit, discarding the first type of variant name; assigning a variant confidence level to the target social group node matched by the variant name, and the variant confidence level has a negative correlation with the number of character differences; when calculating the structure matching score, using the variant confidence level as a correction factor to participate in the calculation.

[0017] By adopting the above technical solution, low-weight character replacement operations are restricted based on semantic importance levels to maintain the core semantic integrity of the name. Simultaneously, the validity of variants is controlled by the number of character differences, eliminating excessively altered variant data with no reference value. Furthermore, the confidence level of differentiated variants is bound and incorporated as a correction factor into the structural matching score calculation. The greater the character difference, the lower the confidence level, which inversely constrains the overall scoring results, offsetting the inflated matching bias caused by malformed variant names. This achieves a two-way balance between broad variant recall and accurate score correction, taking into account both comprehensive retrieval and rigorous scoring, further reducing the technical error space for misjudgments and omissions.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, the method further includes: counting the number of target text segments identified as counterfeit communities for the same target social group node within a preset time period; when the number exceeds a preset threshold, generating risk warning information matching the target social group node; and pushing the risk warning information to the regulatory terminal.

[0019] By adopting the above technical solution, the total amount of text flagged for counterfeiting by a single compliant community is aggregated at fixed time periods. Risk triggering standards are defined based on quantified quantity thresholds, forming a routine data review and statistical mechanism. When the number of counterfeit statements exceeds a critical value, a targeted risk warning is automatically triggered, directly pushing control information to the regulatory terminal, thus establishing a linkage between algorithmic judgment and manual supervision. This eliminates the need for manual review and screening of massive amounts of monitoring data, shortening the detection and response cycle for high-risk counterfeit statements, enabling rapid detection of batch-based counterfeit and abnormal behavior, and improving the real-time performance and control efficiency of batch prevention and control of abnormal statements from social groups.

[0020] In a second aspect, this application provides a monitoring system comprising: one or more processors and a memory; the memory being coupled to the one or more processors, the memory being used to store computer program code including computer instructions, the one or more processors invoking the computer instructions to cause the monitoring system to perform the methods described in the first aspect and any possible implementation thereof.

[0021] Thirdly, this application provides a computer-readable storage medium including instructions that, when executed on a monitoring system, cause the monitoring system to perform the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, this application provides a computer program product, including a computer program that, when run on a monitoring system, causes the monitoring system to perform the method described in the first aspect and any possible implementation thereof.

[0023] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0024] 1. By employing techniques that accurately extract co-occurring entities from the extracted text segments and perform deep structural matching calculations between these co-occurring entities and the corresponding official attribute nodes of the community in the standard knowledge graph, and finally strictly enforce data isolation or association binding based on preset matching thresholds, this method effectively solves the technical problem of existing technologies that rely solely on the literal similarity of names for coarse matching, making it difficult to identify counterfeit communities with identical or highly similar names, and easily leading to irrelevant or false data being mixed into the real community statistics database. This achieves multi-dimensional cross-validation and precise separation of true and false data. This feature mandates that online speech strictly align with official verification data in its core structure and facts, overcoming the misidentification defects caused by the "homophone" phenomenon, blocking interference from counterfeit data at the source of information, and resulting in statistical data included in the monitoring system exhibiting extremely high purity and logical rigor.

[0025] 2. By employing a technique that maps associated entities to specific business feature dimensions using static labels to calculate their data overlap, and meticulously counts the number of character intervals between associated entities and target names, and assigns positional distance weights based on this, and finally uses these distance weights to perform a weighted summation of the overlap of each dimension to obtain the structural matching score, this technique effectively solves the technical problem in existing technologies that, when processing complex long texts, typically extract entity features indiscriminately for comparison, ignoring the differences in semantic subordinate relationships caused by the physical spatial span of entities, and easily leading to distorted evaluation scores due to coincidental co-occurrence of unrelated edge entities. This technique thus achieves a refined measurement model that transforms textual spatial distance into semantic association strength.

[0026] 3. By employing a technique that extracts key semantic components of the target name and combines them with an obfuscated character mapping library to generate first-type variant names that are visually or phonetically consistent, while actively removing interfering non-semantic characters to generate second-type variant names, and then summarizing and constructing a multi-dimensional candidate name set for concurrent graph matching, this approach effectively solves the technical problem of existing technologies that tend to rely on fixed one-dimensional string matching. This makes it difficult to effectively handle the concealed and disguised forms created in the network environment through methods such as replacing similar-looking characters, homophone obfuscation, or adding meaningless punctuation, leading to a large number of missed detections and losses of deep target data. Therefore, it achieves a high-order anti-obfuscation mechanism with strong robustness that surpasses simple literal comparison. This design accurately predicts conventional strategies for unpacking and concealment, actively broadening the retrieval dimensions while maintaining the core semantic boundaries. Attached Figure Description

[0027] Figure 1This is a schematic diagram of a scenario for the social organization industry data monitoring method in the embodiments of this application;

[0028] Figure 2 This is a flowchart illustrating a method for monitoring industry data of social organizations in this application.

[0029] Figure 3 This is a schematic diagram of the physical device structure of a monitoring system in an embodiment of this application. Detailed Implementation

[0030] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. The singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0031] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0032] To facilitate understanding, the method provided in this implementation is described in a scenario below. Please refer to [link / reference]. Figure 1 This is a schematic diagram of a scenario for the social organization industry data monitoring method in this application embodiment.

[0033] exist Figure 1 As can be seen from this, the monitoring system in this application embodiment starts full-process monitoring based on two types of core data sources: one is the existing database of compliant social organizations, and the other is the online speech texts that are publicly disseminated in the network environment.

[0034] For the database of compliant social organizations, the system first performs a standard knowledge graph construction operation, transforming the scattered registration information into a structured graph structure system: generating independent social organization nodes with each compliant social organization as the core, and associating multi-dimensional attribute nodes with each node to form an authoritative reference benchmark covering the core characteristics of the subject, providing a unified standard for subsequent subject matching and feature verification.

[0035] For online speech texts, the system first performs target text segment extraction and associated entity recognition operations: accurately locate the relevant statements of social groups from the lengthy online text, extract the target text segment with the target name (including full name and abbreviation) as the core, and at the same time identify the associated entities (such as account holders, activity locations, related personnel, etc.) that co-occur with the target name in the text, thus completing the extraction and structured transformation of the core information of online speech.

[0036] Subsequently, the system initiates the subject matching process: on the one hand, based on the extracted target name, it performs literal similarity matching with social group nodes in the standard knowledge graph, and filters out target social group nodes whose similarity exceeds a preset threshold; on the other hand, it extracts the corresponding target attribute nodes from the matched target social group nodes to obtain authoritative reference data for feature verification.

[0037] After completing the initial screening of the main body and the extraction of attributes, the system performs structural matching score calculation, compares the accompanying entities in the network text with the target attribute nodes in multiple dimensions, and integrates the content overlap and positional relevance to generate a quantitative structural matching score, realizing in-depth verification from the literal name to the associated features.

[0038] Finally, the system makes a final judgment based on the structure matching score: if the score does not reach the preset threshold, it is determined that the text segment points to a counterfeit entity, and data isolation operation is performed to prevent abnormal information from polluting the monitoring data of compliant entities; if the score meets the threshold requirements, the text segment is associated and bound with the target social group node, and included in the monitoring statistics of the corresponding entity to complete the accurate collection of compliant speech and form a complete monitoring closed loop.

[0039] The following describes the process of the method provided in this implementation, using the above scenario as an example. Please refer to [link / reference]. Figure 2 This is a flowchart illustrating a method for monitoring industry data of social organizations in this application.

[0040] S101. Construct a standard knowledge graph based on the existing database of compliant social groups. The standard knowledge graph includes social group nodes and attribute nodes associated with each social group node.

[0041] Among them, the compliant social organization database refers to a structured data set collected and organized by the industry's competent authority to represent the registration information of social organization entities. It refers to a standardized data source containing social organization names, entity identifiers, operation-related information, business scope, funding sources, etc. The standard knowledge graph is used to represent a structured knowledge system organized in the form of nodes and relationships, which carries social organization entity information and its ancillary characteristic information. Social organization nodes refer to the basic entity units in the knowledge graph used to independently represent a social organization entity, corresponding to a single social organization entity in a real-world scenario. Attribute nodes are used to represent ancillary nodes that exist on top of social organization nodes and are used to represent the relevant characteristic information of the social organization entity, including at least the entity's responsible person node, business scope node, and funding source channel node.

[0042] This step is triggered when the monitoring system completes initial loading, receives an update command for the social organization database, or enters a preset knowledge graph construction cycle. The monitoring system first calls the stored compliant social organization database, parsing and structurally splitting each piece of social organization information. Each set of independent entity information is then used to generate an independent social organization node, ensuring that each node has a unique identifier to avoid confusion or duplication of entity nodes. After creating the social organization nodes, the monitoring system matches and creates corresponding subordinate attribute nodes for each node according to preset node association rules. It maps the responsible person information recorded in the database to the main responsible person node, business-related content to the business scope node, and funding-related information to the funding source node. Finally, it establishes subordinate connections between social organization nodes and various attribute nodes through the internal relationships of the knowledge graph, forming a hierarchical and fully interconnected standard knowledge graph.

[0043] During the construction process, the monitoring system will remove duplicate information and redundant fields to ensure the simplicity and effectiveness of the data in the knowledge graph. At the same time, it will mark and store the relationships between nodes to facilitate quick access and retrieval matching in the future.

[0044] The standard knowledge graph constructed through this step can transform scattered social group registration information into a structured and interconnected knowledge system, providing a unified and complete reference benchmark for subject matching and feature comparison in subsequent online texts, and solving the matching deviation problem caused by data dispersion and inconsistent reference standards in traditional technologies.

[0045] S102. Extract the target text segment containing the target name corresponding to the target social group from the online speech text, and identify the co-occurring entity with the target name from the target text segment;

[0046] Among them, online speech text refers to unstructured text information consisting of continuous characters that is disseminated through internet channels, including text content published on platforms such as social media, news pages, and interactive communities.

[0047] The monitoring system first converts the input online speech text into an indexable continuous character sequence, assigning a unique position index value to each character. Then, it performs a full-text search within this character sequence, using a combination of string matching, fuzzy matching, and semantic matching to locate the region where the target name appears. It precisely marks the index corresponding to the first character of the target name as the first character position and the index corresponding to the last character as the last character position, thus accurately anchoring the target name. After location positioning, the system scans character by character from the first character position along the text header. During the scan, it performs two real-time checks: first, it checks whether the currently scanned character matches a preset event trigger word. Preset event trigger words are indicative words related to associated entities such as account information, venue information, and personnel information; if a match is found, scanning to the left immediately stops. Second, it calculates the scanning span from the first character position to the current scanning position in real time. If this span reaches a pre-configured first preset maximum character span, scanning to the left is forcibly terminated to prevent unlimited expansion and excessively long text fragments.

[0048] The system determines the left boundary position as the character position at which the scanning finally stops. Similarly, starting from the last character position, the system performs a character-by-character traversal scan along the end of the text, stopping at either a preset event trigger word or the second preset maximum character span, and determines the right boundary position at the stopping point. After determining the left and right boundary positions, the system extracts all characters within that interval and combines them to form continuous text, which is the target text segment used for subsequent analysis.

[0049] After acquiring the target text segment, the monitoring system invokes its built-in entity recognition model to parse the text content field by field, identifying and extracting associated entities such as account holder information, event venue information, and related personnel information that are contextually related to the target name. Simultaneously, the system records the location and representation of each associated entity within the target text segment. During the recognition process, the monitoring system standardizes and converts ambiguous and abbreviated expressions to improve the completeness and accuracy of associated entity extraction.

[0050] This step employs a dual-boundary dynamic constraint mechanism, which not only fully preserves the target name and its associated contextual entities, but also effectively eliminates irrelevant paragraphs, redundant modifiers, and interfering statements, significantly improving the accuracy of subsequent entity recognition and reducing unnecessary computational overhead. At the same time, it solves the problems of missing key information or excessive noise caused by traditional fixed-length truncation, making the analysis window more closely match the actual semantic distribution and providing a stable and reliable text foundation for subsequent structure matching and subject identification.

[0051] S103. Based on the target name, match the target social group node in the standard knowledge graph with a literal similarity greater than the preset similarity threshold.

[0052] This step is executed when the monitoring system completes the extraction of the target name and enters the knowledge graph matching and retrieval stage. Its purpose is to fully cover various name variations in online text, such as similar characters, similar pronunciations, and non-standard formats, to improve the retrieval recall rate and avoid the problem that legitimate entities cannot be matched due to differences in name expression.

[0053] The monitoring system first performs semantic decomposition on the target name. Based on preset keyword weighting rules, it filters and extracts key semantic components that carry the main identifier function, excluding auxiliary characters with no actual distinguishing meaning, ensuring that subsequent variant generation revolves around the core recognition part. Then, the system searches a preset obfuscation character mapping library for the extracted key semantic components. According to glyph similarity rules and speech encoding rules, it matches the corresponding replacement characters and replaces the original key characters in the target name one by one with the replacement characters, forming several first-class variant names with glyph or speech similarity features, covering common glyph and speech similarity rewriting expressions in the network. Simultaneously, the system performs non-semantic character detection on the original target name, identifying and removing spaces, special symbols, redundant conjunctions, and other characters that do not affect the main identifier, obtaining a standardized name with a regular format and concise expression, which is used as the second-class variant name. After generating both types of variant names, the system summarizes and integrates the original target name, all first-class variant names, and second-class variant names, deduplicating them to form a complete candidate name set, avoiding resource waste caused by repeated searches.

[0054] Subsequently, the system iterates through each name data in the candidate name set, compares it with the names of each social group node in the standard knowledge graph at the character level, calculates the literal similarity score between each pair, and compares each score with a preset similarity threshold. All social group nodes with a literal similarity greater than the threshold are selected as target social group nodes for subsequent verification.

[0055] By using parallel retrieval of multiple name variants, this step broadens the matching coverage and can effectively address the diverse and non-standardized name expressions in the network environment. It reduces the problem of missed matches caused by name rewriting, abbreviation, and misspelling. At the same time, it provides sufficient and reasonable candidate subjects for subsequent fine-grained verification based on associated entities and attribute nodes, taking into account both the comprehensiveness of retrieval and the accuracy of judgment, thereby improving the reliability and stability of the entire monitoring process from the front end.

[0056] In some embodiments, during the process of generating the first type of variant name based on key semantic components and performing knowledge graph matching, the indiscriminate replacement of similar-looking or similar-sounding characters in the name may result in some variant names being too different from the original target name, thus matching social group nodes with weak associations. This can lead to abnormally high or low subsequent structural matching scores, affecting the accuracy of subject identification. In this case, the following steps can be performed:

[0057] Before performing character replacement on key semantic components, the monitoring system first assigns a corresponding semantic importance level to each original character based on its contribution to the uniqueness of the subject name. Characters categorized in the core area are given higher levels, while auxiliary characters are given lower levels. Subsequently, the system compares the semantic importance level of each original character with a preset importance threshold. Only non-core characters with a level below the threshold are replaced with similar-looking or similar-sounding characters, while core key characters with a level above the threshold are retained. This prevents the variant name from completely deviating from its original referent due to the modification of core characters.

[0058] After generating the first variant name, the system compares the variant name character by character with the target name, counts the number of character differences between them, and compares this number with a preset difference limit. If the number of differences exceeds the limit, it means that the variant name has deviated excessively from the original expression and has no reasonable reference value. The system then discards the variant name and no longer includes it in the candidate name set. If the number of differences is within the allowable range, the variant name is retained to continue participating in subsequent matching. When a target social group node is matched through a variant name, the system assigns a variant confidence score to the matching result based on the number of character differences corresponding to the variant name. The smaller the number of differences, the higher the variant confidence score; the larger the number of differences, the lower the variant confidence score, so that the confidence score can truly reflect the reliability of the matching source.

[0059] When calculating the structural matching score between the associated entity and the target attribute node in the subsequent execution, the system introduces the confidence of this variant as a correction factor into the calculation process to perform weighted correction on the basic structural matching score, so that the final score can take into account both the degree of feature matching and the confidence of name matching.

[0060] Through the above steps, the system can perform fine-grained control over the generation and application of variant names, avoiding the distortion of matching objects caused by unlimited replacement, and achieving reasonable calibration of matching results through confidence correction. This effectively balances the comprehensiveness of name retrieval with the accuracy of the judgment results, and improves the stability and reliability of the overall monitoring process.

[0061] S104. Extract the target attribute nodes corresponding to the target social group node;

[0062] Among them, target attribute nodes refer to the set of attribute nodes that belong to the target social group nodes and are used for feature comparison with the associated entities, including the main person in charge node, the scope of business operations node, and the funding source channel node.

[0063] This step is executed during the monitoring system's process of identifying target social group nodes and preparing to extract corresponding attribute nodes for structural matching. It addresses the issue of attribute information updating over time and the inability of static attributes to adapt to comparisons of content from different periods, ensuring that the attribute nodes involved in the matching are consistent with the time period of the online discourse, thus improving the authenticity and accuracy of feature comparison. When constructing the standard knowledge graph, the monitoring system assigns a time validity identifier to each attribute node, recording its effective and expiration times to identify the applicable time range of the attribute information. This allows for the differentiated storage of characteristic information of social groups at different stages.

[0064] During attribute node extraction, the monitoring system first performs time-dimensional analysis on the target text segment to identify whether a specific time is directly stated within the text. If so, it is used as a clearly marked timestamp. If the text does not directly contain time information, the system retrieves the publication time corresponding to the online statement text and uses that publication time as an inferred time to determine the time information corresponding to the text content. After determining the time information, the system iterates through all attribute nodes associated with the target social group node, reading the effective and expiration times carried by each attribute node one by one, and determining whether the time period limited by the time validity identifier can cover the previously extracted time information. For attribute nodes whose time range covers the text's time information, the system marks them as valid attribute nodes; for attribute nodes that exceed the time range or do not complete coverage, they are excluded. Finally, the system uses the selected valid attribute nodes as target attribute nodes for subsequent comparison calculations with associated entities.

[0065] By introducing time validity identification and time-series filtering mechanisms, this step can dynamically adapt to the updates and changes in social group attribute information, avoid using expired or ineffective attribute data to participate in feature matching, and enable the structure matching score to reflect the real correlation within the corresponding time period of the text, effectively improving the rationality and reliability of monitoring and judgment, and providing a more accurate reference basis for subsequent identification of counterfeit entities and determination of data ownership.

[0066] S105. Calculate the structural matching score between the associated entity and the target attribute node;

[0067] This step is executed after the monitoring system completes the identification of accompanying entities and the screening of target attribute nodes. It is used to achieve multi-dimensional, weighted, and accurate feature matching, avoid the judgment bias caused by single character comparison, and enable the structure matching score to truly reflect the degree of correlation between the content mentioned in the network text and the main features of the knowledge graph.

[0068] The monitoring system first calls a pre-defined static label mapping table to classify accompanying entities into corresponding business feature dimensions based on their type and meaning. This ensures precise alignment between different types of entities and their corresponding attribute dimensions, guaranteeing that subsequent comparisons are conducted within the same dimensional framework. Subsequently, the system extracts a set of standard features from the target attribute nodes that corresponds one-to-one with each business feature dimension, providing standardized reference data for direct comparison for each type of accompanying entity. Under each independent business feature dimension, the system compares the character sequence of the accompanying entity segment by segment with the character sequences in the standard feature set, calculating the proportion of identical characters or words and determining the data overlap for that dimension. A higher overlap indicates a closer similarity in the features of that dimension.

[0069] Meanwhile, the system counts the number of character intervals between each accompanying entity and the target name within the target text segment; fewer intervals indicate a stronger semantic connection. Following a preset distance decay rule, the system assigns positional distance weights to accompanying entities with different interval numbers: closer to the target name, higher weight; farther away, lower weight. After calculating and assigning weights for each dimension of overlap, the system uses the positional distance weights as weighting coefficients to sum the data overlap across each business feature dimension, merging content similarity and positional relevance into a unified quantitative indicator, ultimately yielding a structural matching score.

[0070] By employing a multi-level calculation method involving dimensional alignment, dimensional comparison, and positional weighting, this step effectively improves the precision and rationality of feature matching, weakens the interference of weakly related entities at long distances, and enables the structural matching score to more accurately reflect the true degree of fit between the text content and the main features, providing a reliable scoring basis for subsequent identification of counterfeit entities and determination of data ownership.

[0071] S106. If the structure matching score is less than the preset matching score threshold, it is determined that the target text segment points to the fake community, and the target text segment is isolated from the target social group node.

[0072] Among them, the preset matching threshold is a critical score set in advance by the monitoring system to distinguish the degree of matching of related information; the counterfeit subject is a subject whose name is similar to that of an existing social group, but whose related characteristic information does not match the filing characteristic information; data isolation refers to the operation of delinking the target text segment from the target social group node, not including the text segment in the statistical system of the social group node, and storing it separately in an independent data space.

[0073] The monitoring system first compares the calculated structure matching score with a preset matching score threshold. When the structure matching score is lower than the preset threshold, it indicates that the accompanying entity information in the target text segment has a low degree of fit with the target attribute node information corresponding to the target social group node. There is a significant difference between the two in the core association features. Based on this, the monitoring system determines that the subject pointed to by the target text segment is not the subject corresponding to the target social group node, but a counterfeit subject with a similar name.

[0074] After the determination is completed, the monitoring system performs data isolation. First, it removes the temporary association mapping between the target text segment and the target social group node, prohibiting the text segment from participating in any data statistics and summary calculations of the target social group node. Then, it adds relevant tags of the impersonating entity to the target text segment and stores it in a separate dedicated data storage space, achieving both physical and logical isolation from the monitoring data of legitimate social groups. During the data isolation process, the monitoring system simultaneously records the source information, target name information, matching deviation characteristics, etc. of the text segment, facilitating subsequent data traceability and feature analysis.

[0075] This step uses structural matching and threshold determination to accurately filter text related to counterfeit entities, blocking the transmission of erroneous associations caused by similar names, preventing counterfeit entity information from polluting the monitoring data of legitimate social organizations, preventing chain deviations in subsequent data statistics caused by a single node association error, and ensuring the purity and reliability of monitoring data.

[0076] S107. If the structure matching score is greater than or equal to the preset matching score threshold, then the target text segment is associated with the target social group node and the target text segment is included in the monitoring statistics of the target social group node.

[0077] Among them, association binding refers to the operation by which the monitoring system establishes a fixed attribution mapping relationship between the target text segment and the target social group node, marking that the text segment belongs to the corresponding social group entity.

[0078] This step, executed concurrently with S106, is the core component for collecting and storing compliant associated texts. When the monitoring system determines that the structural matching score is greater than or equal to a preset matching score threshold, it indicates that the accompanying entity information in the target text segment highly matches the attribute node information of the target social group node, and the subject pointed to by the text content is consistent with the subject corresponding to the target social group node. At this point, the monitoring system performs an association binding operation, establishing a unique attribution mapping relationship between the target text segment and the target social group node within the system. The system records the identification information of this mapping relationship and stores it in the association relationship database to ensure that the subject attribution corresponding to the text segment can be quickly queried subsequently.

[0079] After the association and binding are completed, the monitoring system formally incorporates the target text segment into the monitoring and statistical data set corresponding to the target social group node. It updates and statistically analyzes information such as the number of comments, association characteristics, and text sources within the dataset, while simultaneously standardizing and archiving the text segment content. During the incorporation process, the monitoring system removes duplicate text to avoid data distortion caused by repeated statistics, and also tags the feature information of the text segment to enrich the dimensions of the monitoring and statistical data.

[0080] This step achieves efficient collection of formal speech data through precise association and binding, ensuring the integrity and accuracy of relevant monitoring data of social groups. It provides reliable data support for subsequent understanding of the dynamics of online speech of social groups and analysis of speech distribution characteristics, while improving the data closed loop of the monitoring system and enhancing the systematicness and practicality of the overall monitoring work.

[0081] In this embodiment, by using a standard knowledge graph containing social group nodes and multi-dimensional attribute nodes as an authoritative reference, and combining multi-morphological variant retrieval of target names, temporal structure matching of accompanying entities and attribute nodes, and position-weighted score calculation, it is possible to further perform multi-level verification through association features, time validity, and semantic distance on the basis of name literal similarity. This enables accurate identification and reliable binding of the subject in the online text, effectively solving the problems of traditional methods that rely solely on character similarity and are easily misled by similar names, entity alignment errors causing monitoring data pollution, and key information loss or noise interference leading to distorted judgment. Thus, it achieves high-precision monitoring of online speech related to social groups, effective differentiation of impersonating entities, and pure and reliable monitoring data, providing stable and accurate data support for speech analysis and dynamic control.

[0082] In some embodiments, during the process of identifying the impersonating entity and isolating data in the target text segment, it may be difficult to pay attention to single, scattered impersonation-related texts in a timely manner, leading to the continuous spread of abnormal texts with similar names, making it impossible to form a timely perception and response to high-frequency impersonation mentions. In this case, the following steps can be performed:

[0083] The monitoring system uses preset time periods as statistical units to categorize and analyze all target text segments that have been identified within the system. It then filters out text data that is determined to point to a counterfeit entity and initially matches the same target social group node. The system then accumulates the number of such texts in real time to generate an abnormal text count for the corresponding target social group node. During the statistical process, the system deduplicates duplicate texts and data from duplicate sources, retaining only the number of valid texts that are independently published and disseminated, ensuring that the statistical results accurately reflect the scale of abnormal mentions.

[0084] After completing the statistics, the system compares the real-time count of abnormal texts with a preset threshold. When the count is greater than or equal to the threshold, it determines that the abnormal mentions of similar names targeting that social group have reached a peak, and automatically generates a risk warning message matching the target social group node. The warning message includes at least the identification information of the corresponding social group, the statistical count of abnormal texts within a preset time period, the source distribution of the abnormal texts, and a summary of relevant content for easy viewing and location later. Subsequently, the monitoring system proactively pushes the risk warning message to the corresponding regulatory terminal through an internal data transmission channel, enabling relevant parties to obtain information on the concentrated abnormal mentions as soon as possible.

[0085] Through a closed-loop mechanism of timed statistics, threshold triggering, and automatic push notifications, this step can transform scattered, single-item judgment results into perceptible, large-scale risk alerts, achieving an upgrade from single-point identification to overall risk monitoring. This significantly improves the detection efficiency of high-frequency similar name behaviors without increasing the pressure of manual verification, providing timely and reliable evidence for subsequent handling, and further enhancing the practicality and early warning capabilities of the entire monitoring system.

[0086] The monitoring system in this application embodiment is described below from a hardware processing perspective. Please refer to [link / reference]. Figure 3 This is a schematic diagram of the physical device structure of the monitoring system in this application embodiment.

[0087] It should be noted that, Figure 3 The structure of the monitoring system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0088] like Figure 3 As shown, the monitoring system includes a Central Processing Unit (CPU) 301, which can perform various appropriate actions and processes based on programs stored in Read-Only Memory (ROM) 302 or programs loaded from storage section 308 into Random Access Memory (RAM) 303, such as performing the methods described in the above embodiments. The RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An Input / Output (I / O) interface 305 is also connected to the bus 304.

[0089] The following components are connected to I / O interface 305: input section 306 including audio input devices, push-button switches, etc.; output section 307 including a liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 308 including a hard disk, etc.; and communication section 309 including a network interface card such as a LAN (Local Area Network) card, modem, etc. Communication section 309 performs communication processing via a network such as the Internet. Drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

[0090] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by central processing unit (CPU) 301, it performs the various functions defined in this application.

[0091] It should be noted that specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0092] 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. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains 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 may occur in a different order than those shown in the drawings.

[0093] Specifically, the monitoring system in this embodiment includes a processor and a memory. The memory stores a computer program, and when the computer program is executed by the processor, it implements the social group industry data monitoring method provided in the above embodiment.

[0094] In another aspect, this application also provides a computer-readable storage medium, which may be included in the monitoring system described in the above embodiments; or it may exist independently and not be assembled into the monitoring system. The storage medium carries one or more computer programs, which, when executed by a processor of the monitoring system, cause the monitoring system to implement the social group industry data monitoring method provided in the above embodiments.

[0095] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0096] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0097] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A social group industry data monitoring method, characterized by, Applied to a monitoring system, the method includes: A standard knowledge graph is constructed based on an existing database of compliant social groups. The standard knowledge graph includes social group nodes and attribute nodes associated with each social group node. Extract target text segments containing the target name of the target social group from online speech texts, and identify co-occurring entities that appear with the target name from the target text segments. The co-occurring entities include at least one of the following: the account holder, the location of the event, and the associated person. The target name includes both the full name and the abbreviation. Based on the target name, match the target social group node in the standard knowledge graph with a literal similarity greater than a preset similarity threshold; Extract the target attribute nodes corresponding to the target social group nodes; Calculate the structural matching score between the associated entity and the target attribute node; If the structure matching score is less than the preset matching score threshold, it is determined that the target text segment points to the fake community, and the target text segment is isolated from the target social group node. If the structure matching score is greater than or equal to the preset matching score threshold, the target text segment is associated with the target social group node, and the target text segment is included in the monitoring statistics of the target social group node.

2. The method of claim 1, wherein, The step of extracting the target attribute node corresponding to the target social group node, wherein the attribute node in the standard knowledge graph carries a time validity identifier, the time validity identifier recording the effective time and expiration time of the attribute node, further includes: Extract time information from the target text segment, including explicitly marked timestamps or estimated times determined by the publication time of the online commentary text; Among all attribute nodes associated with the target social group node, filter out the valid attribute nodes whose time validity identifier covers the time information; The valid attribute node is used as the target attribute node.

3. The method of claim 1, wherein, The steps for extracting target text segments containing the target name of a target social group from online speech texts specifically include: In the continuous character sequence of the online speech text, scan and locate the first and last character positions that match the target name; Starting from the first character position, the text of the online speech is scanned character by character towards the beginning of the text. The scanning stops when a preset event trigger word is matched or when the scanning span from the first character position reaches the first preset maximum character span. The corresponding stopping position is determined as the left boundary position. The preset event trigger word is associated with the accompanying entity. Starting from the position of the last character, the text of the online speech is scanned character by character towards the end of the text. The scanning stops when the preset event trigger word is matched or the scanning span from the position of the last character reaches the second preset maximum character span. The corresponding stopping position is determined as the right boundary position. Extract all characters located between the left and right boundary positions to generate the target text segment, which serves as the analysis window.

4. The method of claim 1, wherein, The steps for calculating the structural matching score between the associated entity and the target attribute node specifically include: According to the preset static label mapping table, the associated entities are divided into the corresponding business feature dimensions; Extract the set of standard features from the target attribute node that are consistent with each of the business feature dimensions; For each individual business feature dimension, the data overlap between the character sequence of the associated entity and the character sequence in the standard feature set is calculated. Count the number of character intervals between each associated entity and the target name in the target text segment; According to the preset distance attenuation rule, the associated entity is assigned a corresponding positional distance weight; Using the location distance weights, the data overlap calculated under each of the business feature dimensions is weighted and summed to obtain the structure matching score.

5. The method of claim 1, wherein, The step of matching target social group nodes with a literal similarity greater than a preset similarity threshold in the standard knowledge graph based on the target name specifically includes: Extract the key semantic components from the target name, and for the key semantic components, query a preset obfuscation character mapping library to obtain replacement characters with similar visual morphology or the same speech encoding; Replace the original characters in the target name with the replacement characters to generate a first-type variant name; Detect and remove non-semantic characters from the target name, and generate a normalized name as a second type of variant name; The target name, the first type of variant name, and the second type of variant name are combined to form the candidate name set; Each name in the candidate name set is matched against the standard knowledge graph.

6. The method according to claim 5, characterized in that, The method further includes: Assign a semantic importance level to each original character in the key semantic components; Replacement is performed only on original characters whose semantic importance level is lower than a preset importance threshold; Count the number of character differences between the generated first-type variant name and the target name; If the number of character differences exceeds a preset difference limit, then the first type of variant name is discarded; A variant confidence score is assigned to the target social group node matched by the variant name, and the variant confidence score is negatively correlated with the number of character differences; When calculating the structure matching score, the variant confidence score is used as a correction factor in the calculation.

7. The method according to claim 1, characterized in that, The method further includes: Count the number of target text segments that are identified as fake social group nodes for the same target social group within a preset time period; When the number exceeds a preset threshold, a risk warning message matching the target social group node is generated; The risk warning information will be pushed to the regulatory terminal.

8. A monitoring system, characterized in that, The monitoring system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the monitoring system to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on the monitoring system, the monitoring system performs the method as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is run on the monitoring system, it causes the monitoring system to perform the method as described in any one of claims 1-7.

Images