Big data public opinion analysis and early warning method and system for digital rural governance

Abstract

This invention relates to the field of public opinion analysis technology, specifically to a big data public opinion analysis and early warning method and system for digital rural governance. The method includes the following steps: acquiring digital rural policy texts and rural public opinion texts; establishing a mapping relationship between policy term mapping records and weight level codes; constructing a text structure reconstruction sequence; extracting word frequency overlap difference sequences based on adjacent overlapping time windows and generating public opinion mutation time segments; and generating early warning data by combining the dispersion and occurrence range data of propagation node time records. In this invention, by associating policy clause levels with public opinion statements, the recognition of text semantic weight and preceding / following structure is enhanced. Combined with time mutation analysis and diffusion intensity determination, short-term fluctuation interference is reduced, improving the accuracy, coherence, and targeted nature of digital rural public opinion risk identification, making the output results more consistent with the actual propagation trend, change characteristics, and governance needs.

Description

Technical Field

[0001] This invention relates to the field of public opinion analysis technology, and in particular to a big data public opinion analysis and early warning method and system for digital rural governance. Background Technology

[0002] The field of public opinion analysis technology mainly involves the collection, processing, and analysis of online information dissemination content. Its core aspects include multi-source data acquisition, text content analysis, sentiment identification, dissemination path analysis, and risk assessment. It typically revolves around text, images, and related interactive information on internet platforms. By systematically processing the information release entities, dissemination time sequences, and semantic features of the content, it achieves continuous tracking and structured expression of social public opinion dynamics. This technology plays an important supporting role in public governance, social management, and information security. Especially in the context of digital rural governance, it is necessary to combine the characteristics of information dissemination in rural areas, the dispersion of data sources, and regional attributes to conduct targeted integration and analysis of public opinion information.

[0003] Among them, the traditional big data public opinion analysis and early warning method for digital rural governance refers to a type of method for processing and risk warning of online public opinion information in rural areas. Its main problem is how to identify and judge potential public opinion from data distributed in government platforms, social media and local information. The methods usually include matching and filtering text content one by one by setting a fixed keyword list, sorting and archiving the collected information according to time order, classifying and labeling the information according to word frequency statistics, counting the occurrence of specific keywords according to a pre-set threshold, triggering an early warning when the count exceeds the set threshold, and combining manual review and marking of the screening results to form a basis for public opinion judgment.

[0004] Traditional early warning methods rely on fixed word lists for filtering and threshold counting. Text-by-text matching can only extract isolated word information. Differences in the hierarchy and semantic weight of policy clauses are difficult to incorporate into the judgment process. Content from scattered sources remains in a superficial state after sorting and archiving. There is a lack of unified sorting of the relationship between sentences and the evolution of events. Time changes are triggered by frequency fluctuations alone. Short-term fluctuations and continuous escalation are easily confused. The pace of dissemination and the scope of diffusion lack coordinated assessment, resulting in delayed risk identification and insufficient stability and directionality of early warning conclusions. Summary of the Invention

[0005] To address the technical problems existing in the current technology, this invention provides a big data public opinion analysis and early warning method for digital rural governance, comprising the following steps: To achieve the above objectives, the present invention adopts the following technical solution: a big data public opinion analysis and early warning method for digital rural governance, comprising the following steps: S1: Obtain the digital rural policy text, extract the clause level number and convert it into a weight level code, obtain the rural public opinion text statement, perform word segmentation and extraction on the rural public opinion text statement, generate policy vocabulary mapping record, construct the mapping relationship between the policy vocabulary mapping record and the weight level code, and generate statement code binding data; S2: Extract multiple weight level codes from the statement code binding data, perform numerical comparison, select the weight level code corresponding to the maximum value as the main weight value, calculate the difference to generate a weight gradient distribution sequence, and combine it with the continuity of clause level numbering to construct a text structure reconstruction sequence; S3: Construct adjacent overlapping time windows according to a preset time step, extract the occurrence frequency of public opinion words from the text structure reconstruction sequence and calculate the difference, generate a word frequency overlap difference sequence, perform trend analysis using a differential autoregressive moving average model, aggregate the position of symbol change nodes, and generate public opinion change time segments. S4: Extract the propagation node time records included in the public opinion change time segment, generate the serialized time interval, calculate the offset segment ratio, compare it with the total number of network nodes, and generate event diffusion intensity classification labels. S5: Extract the event diffusion intensity classification label, map the risk level value and the occurrence range data of the public opinion change time segment, perform risk judgment, and generate digital rural public opinion early warning data.

[0006] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Obtain authorization to publish the digital rural policy text, extract the embedded clause level number in the digital rural policy text, perform character identifier decomposition operation on the clause level number, extract the structured level identifier, assign decreasing values ​​to the structured level identifier according to the preset level descending order logic, aggregate the structured level identifier and the decreasing values ​​to establish a level mapping dictionary, and generate weight level code. S102: Obtain locally stored rural public opinion text statements, perform character segmentation processing on the rural public opinion text statements to extract independent words, compare the independent words with the preset policy proper noun library to extract matching text nodes, extract the coordinate positions of the text nodes in the rural public opinion text statements, bind the coordinate positions and text nodes to form a position vector sequence, establish a vector mapping association, and generate a policy term mapping record. S103: Based on the weight level code and policy term mapping record, extract the clause instruction label carried in the policy term mapping record, compare the clause instruction label with the identifier sequence covered by the weight level code to locate the associated weight, assign the associated weight to the corresponding text node to construct a numerical scalar element, merge the numerical scalar elements of the same single statement with the statement index sequence to construct a multidimensional key-value pair, integrate all multidimensional key-value pairs to perform topological merging processing, and generate statement code binding data.

[0007] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Extract the weight level code attached inside the statement code binding data, perform traversal for multiple weight level codes, filter the node with the maximum value of the numerical dimension, extract the weight level code associated with the maximum value node independently, and generate the main weight value. S202: Based on the core weight value and the remaining weight level code, the core weight value is used as the base of the subtraction term. The remaining weight level code is subtracted one by one to obtain discrete difference components. The mapping between discrete difference components and text node spatial positions is recorded. Linear arrangement is performed on all discrete difference components according to the position sorting logic to establish a weight decay structure and generate a weight gradient distribution sequence. S203: Obtain the context sequence of rural public opinion text statements, extract the clause level numbers associated with adjacent rural public opinion text statements, calculate the level index difference for adjacent clause level numbers, compare the level index difference with the preset number association range, lock the internal position interval of the weight gradient distribution sequence that meets the association boundary, superimpose preset continuous augmented values ​​on the internal position interval, reorganize the overall vector matrix, and generate a text structure reconstruction sequence.

[0008] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Obtain the time axis data range, perform sliding segmentation on the time axis data range according to the preset time step, construct adjacent overlapping time windows with overlapping time features, call the text structure reconstruction sequence, extract the key public opinion words mapped by the text structure reconstruction sequence, count the frequency of key public opinion words in each adjacent overlapping time window, calculate the difference of adjacent occurrence frequency pair by pair, aggregate all differences, and generate a word frequency overlap difference sequence. S302: Input the word frequency overlap difference sequence into the differential autoregressive moving average model, call the differential operator of the differential autoregressive moving average model to perform a stationary transformation on the difference, extract the positive and negative sign states of the sequence after the stationary transformation, compare the positive and negative sign states of adjacent time nodes, filter the target time nodes corresponding to the positive and negative flips, extract the coordinates of the target time nodes, and generate the sign change node positions. S303: For the symbol change node position, extract the local preset merging distance parameter, perform a subtraction operation on the adjacent symbol change node positions to obtain the time span metric, compare the time span metric with the local preset merging distance parameter, perform interval closure processing on the symbol change node positions corresponding to the local preset merging distance parameter, determine the starting boundary and ending time boundary after aggregation processing, and generate the public opinion change time segment.

[0009] As a further embodiment of the present invention, the method for obtaining the merging distance parameter is as follows: obtaining the text release time series in the evolution process of historical digital rural public opinion events, extracting the time interval sample data of adjacent fermentation cycles in the text release time series, calculating the statistical distribution mean of the time interval sample data, and multiplying the distribution mean with a preset time tolerance coefficient, and using the resulting product value as the merging distance parameter.

[0010] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Extract the propagation node time records included in the time segment of the sudden change in public opinion, perform linear sorting processing on all propagation node time records according to the time sequence logic, reorganize the record sequence, perform timestamp subtraction operation on adjacent propagation node time records, extract the subtraction difference value, aggregate all subtraction difference values ​​according to the sorting correspondence, form a discrete set, and generate a serialized time interval. S402: Call the local preset Gini coefficient solver, substitute the serialized time interval into the preset Gini coefficient solver to calculate the cumulative distribution probability data, solve the time distribution imbalance index parameter based on the cumulative distribution probability data, map the time distribution imbalance index parameter to the standard numerical space range to construct a single-dimensional scalar, and generate the time offset dispersion value. S403: For the time offset dispersion value, obtain the preset overall average interval parameter and the total number of local network nodes, calculate the absolute deviation value between the time offset dispersion value and the preset overall average interval parameter, extract the division quotient value between the absolute deviation value and the preset overall average interval parameter, construct the offset segment ratio, compare the offset segment ratio with the total number of local network nodes, determine the diffusion boundary range, match the local classification mapping dictionary to determine the character identifier item, and generate the event diffusion intensity classification label.

[0011] As a further aspect of the present invention, the process of determining character identifiers by matching the local classification mapping dictionary specifically involves: obtaining a preset active node connectivity coefficient; multiplying the preset active node connectivity coefficient with the total number of local network nodes to extract the baseline value of the basic node; setting the preset active node connectivity coefficient based on the median of the ratio distribution of the number of edges to the number of vertices in the historical network topology graph; performing numerical mapping processing on the baseline value of the basic node using a logarithmic operator to obtain a smooth comparison benchmark; extracting the offset segment ratio; calculating the difference ratio between the offset segment ratio and the smooth comparison benchmark; dividing the diffusion boundary range according to the polarity and absolute value of the difference ratio value; calling the local classification mapping dictionary, which records the binding structure between multiple numerical boundary sub-intervals and predefined character vectors; comparing the coverage relationship between the diffusion boundary range and multiple numerical boundary sub-intervals; locking the intersecting numerical boundary sub-intervals; extracting the bound predefined character vectors as character identifiers; and concatenating the character identifiers to construct an event diffusion intensity classification label.

[0012] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Retrieve the local preset risk level mapping table, compare the event diffusion intensity classification label with the built-in associated label item of the preset risk level mapping table, extract the hierarchical numerical scalar of the matching built-in associated label item, set the hierarchical numerical scalar as the quantitative hazard assessment parameter, establish mapping conversion logic, and generate risk classification value. S502: Analyze the spatial location metadata carried by the time segment of the sudden change in public opinion, extract the network propagation platform node identifier and the physical geographic area coordinates included in the spatial location metadata, perform merging and splicing processing on the network propagation platform node identifier and the physical geographic area coordinates, construct a boundary polygon structure, establish a two-dimensional spatial expansion boundary, and generate occurrence range data. S503: Calculate the product of the risk level value and the area covered by the occurrence range data to obtain the comprehensive threat level, extract the preset alarm trigger threshold, compare the comprehensive threat level with the preset alarm trigger threshold, and when it is determined that the comprehensive threat level exceeds the preset alarm trigger threshold, retrieve the emergency linkage instruction, combine the emergency linkage instruction and the occurrence range data, construct a formatted data packet, and generate digital rural public opinion early warning data.

[0013] As a further aspect of the present invention, the process of setting the alarm trigger threshold specifically involves: acquiring locally stored historical records of similar events; calculating the arithmetic mean of the threat levels included in the historical records of similar events to generate a historical average threat level; extracting the base of the actual total number of public opinion posts and the base of the normalized total number of posts within a predetermined historical time period; calculating the division ratio of the actual total number of public opinion posts to the base of the normalized total number of posts to generate a sensitive emotion fluctuation coefficient; multiplying the historical average threat level and the sensitive emotion fluctuation coefficient; and setting the numerical result obtained from the multiplication operation as the alarm trigger threshold.

[0014] A big data-driven public opinion analysis and early warning system for digital rural governance, the system comprising: The policy semantic association module obtains the digital rural policy text, extracts the clause level number and converts it into a weight level code, obtains rural public opinion text statements, performs word segmentation and extraction on the rural public opinion text statements, generates policy vocabulary mapping records, constructs the mapping relationship between policy vocabulary mapping records and weight level codes, and generates statement code binding data. The hierarchical weighting module extracts multiple weight level codes from the statement code binding data, performs numerical comparison, selects the weight level code corresponding to the maximum value as the main weight value, calculates the difference to generate a weight gradient distribution sequence, and combines the continuity of clause hierarchical numbering to construct a text structure reconstruction sequence. The temporal fluctuation identification module constructs adjacent overlapping time windows according to a preset time step, extracts the occurrence frequency of public opinion words from the reconstructed text structure sequence and calculates the difference, generates a word frequency overlap difference sequence, performs trend analysis using a differential autoregressive moving average model, aggregates the position of symbol change nodes, and generates public opinion change time segments. The propagation trend judgment module extracts the propagation node time records included in the public opinion change time segment, generates a serialized time interval, calculates the offset segment ratio, compares it with the total number of network nodes, and generates an event diffusion intensity classification label. The hierarchical linkage early warning module extracts the event spread intensity classification labels, maps the risk classification values ​​and the occurrence range data of public opinion change time segments, performs risk assessment, and generates digital rural public opinion early warning data.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by converting policy text clause numbers into weighted level codes and establishing a mapping relationship with the word segmentation results of public opinion statements, text judgment shifts from isolated keyword comparison to hierarchical identification under the constraints of policy context. The structure of the preceding and following statements is reconstructed using the main weight and gradient difference, supplementing the correlation clues between scattered expressions. Furthermore, word frequency overlap difference and symbol change nodes are extracted within adjacent overlapping time windows. Combined with the dispersion of propagation time intervals and the scope of occurrence, joint judgment is carried out, enabling the identification of abnormal fluctuations, the differentiation of diffusion strength, and the risk output to form a continuous link. The early warning results are more in line with the real propagation situation and governance needs in the digital village scenario. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0019] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0020] Please see Figure 1 This invention provides a big data-driven public opinion analysis and early warning method for digital rural governance, comprising the following steps: S1: Obtain the digital rural policy text, extract the clause level number of the digital rural policy text, convert the clause level number into a weight level code, obtain the rural public opinion text statement, perform word segmentation and extraction on the rural public opinion text statement to generate policy vocabulary mapping record, construct the mapping relationship between the policy vocabulary mapping record and the weight level code, and generate statement code binding data; S2: Extract multiple weight level codes from the statement code binding data and perform numerical comparison. Select the weight level code corresponding to the maximum value as the main weight value. Calculate the difference between the remaining weight level codes and the main weight value to generate a weight gradient distribution sequence. Extract the clause level numbers corresponding to adjacent rural public opinion text statements and perform continuous comparison. Apply continuous augmentation values ​​to the weight gradient distribution sequence that meets the preset number range conditions to construct a text structure reconstruction sequence. S3: Construct adjacent overlapping time windows according to a preset time step, extract the occurrence count of public opinion words in each adjacent overlapping time window from the text structure reconstruction sequence, calculate the difference of occurrence counts for each pair to generate a word frequency overlap difference sequence, use a differential autoregressive moving average model to perform trend analysis on the word frequency overlap difference sequence to extract the symbol change node position, perform aggregation processing on multiple symbol change node positions that meet the preset merging distance parameter, and generate public opinion change time segment; S4: Extract the propagation node time records contained in the time segment of public opinion change, calculate the interval between adjacent propagation node time records in chronological order to generate a serialized time interval, use the Gini coefficient function to perform dispersion calculation on the serialized time interval to generate a time offset dispersion value, calculate the offset difference between the time offset dispersion value and the overall average interval to generate the offset segment ratio, compare the offset segment ratio with the total number of network nodes to generate event diffusion intensity classification labels; S5: Extract the risk classification value mapped by the event diffusion intensity classification label, extract the occurrence range data of the public opinion change time segment, combine the risk classification value and occurrence range data to perform risk judgment calculation, and generate digital rural public opinion early warning data; The statement encoding binding data includes clause semantic feature graphs, node weight distribution matrices, and policy association index lists. The text structure reconstruction sequence includes hierarchical weight gradient vectors, statement topology structure trees, and syntactic component dependency networks. The public opinion mutation time segments include outbreak start timestamps, high-frequency oscillation duration cycles, and recession critical time points. The event diffusion intensity classification labels include centralized outbreak markers, sporadic propagation markers, and multi-source concurrent evolution markers. The digital rural public opinion early warning data includes emergency response level instructions, risk impact heat maps, and control resource scheduling databases.

[0021] Please see Figure 2 The specific steps of S1 are as follows: S101: Obtain authorization to publish the digital rural policy text, extract the embedded clause level number in the digital rural policy text, perform character identifier decomposition operation on the clause level number, extract the structured level identifier, assign decreasing values ​​to the structured level identifier according to the preset level descending order logic, aggregate the structured level identifier and the decreasing values ​​to establish a level mapping dictionary, and generate weight level code. The process of assigning decreasing values ​​to structured hierarchical identifiers based on a preset descending hierarchical logic is as follows: First, the clause hierarchical numbers are segmented according to a preset character delimiter to extract hierarchical sequences with depth indices. Second, a depth truncation threshold is obtained, set based on the probability parameters of nested hierarchies in historical digital rural policy texts. Third, the depth indices in the hierarchical sequences with depth indices are compared with the depth truncation threshold; segments exceeding the depth truncation threshold are removed, and segments not exceeding the threshold are retained as structured hierarchical identifiers. Fourth, initial weights and hierarchical decay coefficients are called; the initial weights are set based on the administrative level of the policy issuing entity, and the hierarchical decay coefficients are set based on the binding force ratio of adjacent hierarchical clauses. Fifth, a numerical decay operation is performed by combining the initial weights, the depth indices of the structured hierarchical identifiers, and the hierarchical decay coefficients; the result of the numerical decay operation is then assigned as a decreasing value to the corresponding structured hierarchical identifier. The process involves calling the data interface of the government data open platform to obtain formatted policy documents. A unified character encoding standard is applied to the documents for parsing, removing full-width and half-width spaces, and extracting the embedded clause hierarchy numbers within the digital rural policy text. Preset character separators are obtained, with commas and periods set as the delimiters. String splitting is performed on the clause hierarchy numbers to extract independent hierarchy numbers and a hierarchy sequence with a deep index. By accessing the local historical digital rural policy text database, past document format parameters are obtained, and the ratio of the total frequency of each nested level to the total number of clauses is calculated and used as a probability parameter. When obtaining the probability distribution data of nested levels in historical texts, a big data analysis mechanism is introduced. Through in-depth mining of massive amounts of government documents, universally applicable nested hierarchical patterns are accurately extracted. The probability distribution data of nested levels in historical texts is compared with a preset truncation confidence lower limit. The deepest level position exceeding the truncation confidence lower limit is extracted and set as the depth truncation threshold. By comparing the depth index and depth truncation threshold in the hierarchical sequence with depth index, segments exceeding the depth truncation threshold are removed, while segments within the threshold are retained and converted into structured hierarchical identifiers. The issuing agency registration table is retrieved to find the administrative level of the current policy issuing entity, and initial weights are set based on a mapping relationship of 10 for national level, 8 for provincial level, and 6 for municipal level. The proportion of binding judgments supported by adjacent hierarchical clauses in historical judicial practice is retrieved, and the statistical median of this proportion is set as the hierarchical decay coefficient. A numerical decay operation is performed by combining the initial weights, the depth index of the structured hierarchical identifiers, and the hierarchical decay coefficient; the result is the decreasing value. The decreasing value is assigned to the corresponding structured hierarchical identifier. The structured hierarchical identifiers and their corresponding decreasing values ​​are aggregated, and the data is stored using a hash mapping structure. A hierarchical mapping dictionary is established, and the key-value pairs in the dictionary are concatenated into hexadecimal strings to generate weight level codes. The issuing entity is set to a provincial-level institution, with an initial weight of 8, a median binding force ratio of 0.8, a hierarchy decay coefficient of 0.8, and a depth index of the current structured hierarchy identifier of 2. Numerical decay calculation is then performed, the process of which is as follows: Decreasing value = Initial weight * Level decay coefficient * Depth index; Calculation results: Decreasing value = 8 * 0.8 * 2 = 12.8; The value 12.8 is assigned as a decreasing value to the structured hierarchical identifier of the second level. The advantage of this operation is that, through the joint product of the hierarchical decay coefficient and the depth index, the degree of effectiveness decay during the policy implementation process is accurately quantified.

[0022] S102: Obtain locally stored rural public opinion text statements, perform character segmentation processing on the rural public opinion text statements to extract independent words, compare the independent words with the pre-set policy proper noun library to extract matching text nodes, extract the coordinate positions of the text nodes in the rural public opinion text statements, bind the coordinate positions and text nodes to form a position vector sequence, establish vector mapping association, and generate policy term mapping records. The system retrieves rural public opinion log files from the local storage server, extracts rural public opinion text statements, and, facing massive and complex rural public opinion log files, leverages a big data analysis framework to achieve throughput and accurate feature mapping for unstructured text. It calls a word segmentation operator, performing character segmentation on the rural public opinion text statements based on the maximum matching word segmentation mechanism, filtering out semantically meaningless function words and conjunctions according to a pre-set stop word dictionary, and extracting independent word units. It then retrieves a local pre-set policy proper noun database, performing a character-by-character comparison between the independent word units and entries in the database. When the character similarity reaches a perfect match, the matching text node is extracted. A position calculation pointer is initiated, setting an initial index value of 0 for the starting character of the rural public opinion text statement, and recording the starting and ending position indices of matching text nodes character by character, extracting the coordinate positions of the text nodes within the rural public opinion text statement. The starting and ending position indices are integrated into a two-dimensional array, and the two-dimensional array is merged with the character content of the matching text nodes to form a position vector sequence. A two-dimensional association table structure is allocated in memory. The identifier keys of the position vector sequences are paired with the term classification codes in a pre-set policy terminology database to establish a vector mapping association. The paired records are written to a read-only storage block to generate policy terminology mapping records. The starting position index of a specific term in a rural public opinion text is set to 15, the ending position index to 22, and the total character length of the text node is 8. The calculation process for verifying the validity of the coordinate positions is as follows: Valid position identifier = End position index - Start position index + 1; Calculation results: Valid location identifier = 22 - 15 + 1 = 8; By comparing the valid position identifier with the total character length of the text node, a match confirms that the coordinate extraction is correct. The advantage of this operation logic is that by calculating the absolute difference between the start and end positions in conjunction with the total character length verification, it eliminates positional offset errors caused by the mixing of full-width and half-width characters during the segmentation process.

[0023] Table 1. Policy Vocabulary Mapping Space Parameters

[0024] As shown in Table 1, the absolute position range and length attribute of the text nodes corresponding to different identifier keys in the original sentence were extracted and recorded.

[0025] S103: Based on the weight level coding and policy term mapping record, extract the clause instruction label carried in the policy term mapping record, compare the clause instruction label with the identifier sequence covered by the weight level coding to locate the associated weight, assign the associated weight to the corresponding text node to construct numerical scalar elements, merge the numerical scalar elements of the same single statement with the statement index sequence to construct multidimensional key-value pairs, integrate all multidimensional key-value pairs to perform topological merging processing, and generate statement code binding data; The process involves retrieving the weight level codes and policy terminology mapping records from the bus, parsing the underlying tag frames of the policy terminology mapping records, and extracting the clause instruction tags carried by the policy terminology mapping records. For the identifier sequence covered by the weight level codes, a bit-by-bit comparison operation is performed. When the string content of the clause instruction tag matches the string content of an item in the identifier sequence covered by the weight level codes, the associated weight is located. The numerical type of the associated weight is converted to floating-point, and the floating-point associated weight is assigned to the attribute field of the corresponding text node, constructing a numerical scalar element containing positional and weight information. Line breaks and periods are retrieved from rural public opinion text statements. Statement boundaries are defined according to period markers, and incrementally increasing statement sequence numbers are assigned as statement index sequences. Numerical scalar elements with the same statement index sequence are collected, and numerical scalar elements belonging to the same single statement are merged with the statement index sequence, stored in a key-value pair structure, constructing a multi-dimensional key-value pair. All multidimensional key-value pairs are imported into the graph database engine. Using the statement index sequence as the central node and numerical scalar elements as edge nodes, a topology merging process is performed. This process fully utilizes the graph inference capabilities of big data analysis, ensuring effective dimensionality reduction and feature preservation of complex semantic networks. Duplicate edge nodes are removed, and the association weights of edge nodes with the same name are accumulated. The topology graph data is serialized into a JSON stream, generating statement-encoded binding data. Assuming a single statement contains three numerical scalar elements, with the first element having an association weight of 12.8, the second scalar element having an association weight of 6.4, and the third scalar element having an association weight of 3.2, the calculation process for merging the association weights of elements belonging to the same single statement is as follows: The total weight of a single sentence = the weight of the first association + the weight of the second association + the weight of the third association; Calculation results: The total weight of a single sentence = 12.8 + 6.4 + 3.2 = 22.4; 22.4 is used as the comprehensive feature after the topological merging of the index sequence of this statement. The advantage of this operation logic is that by accumulating the association weights of all scalar elements within a single sentence, the density of the entire statement at the policy response level can be intuitively quantified.

[0026] Please see Figure 3 The specific steps of S2 are as follows: S201: Extract the weight level codes attached to the data bound to the statement code, perform traversal for multiple weight level codes, filter the node with the maximum value of the numerical dimension, extract the weight level codes associated with the maximum value node independently, and generate the main weight values. The system reads statement-encoded data files from the local cache, parses the JSON stream structure, removes non-numeric identifier characters, and extracts the weight level codes embedded within the statement-encoded data. For the massive amounts of statement-encoded data files, big data analytics is used to accelerate the filtering and location of maximum value nodes, improving data processing throughput in high-concurrency scenarios. A floating-point comparison pointer is started, initialized to a maximum value of 0, and a linear loop traversal operation is performed on multiple weight level codes. During the traversal, the numerical part of the currently read weight level code is subtracted from the maximum value recorded by the comparison pointer. When the subtraction result is greater than 0, the comparison pointer is updated to the current value until the traversal ends, filtering out the maximum value node in the numerical dimension. Based on the memory physical address of the maximum value node, the original data structure is located, and the weight level codes associated with the maximum value node are extracted independently from the original sequence. The independently extracted weight level codes are assigned to specific high-priority variable registers, and precision alignment and zero-padding are performed to generate the core weight values. The current maximum value of the comparison pointer during the traversal is set to 22.4, and the newly read weight level code value is set to 35.6. The comparison process is transformed into a difference calculation: Difference state = New read weight - Current maximum value of comparison pointer; Calculation results: Difference state = 35.6 - 22.4 = 13.2; If the difference state is greater than 0, then 35.6 is set as the new backbone weight value. The advantage of this operation logic is that, through continuous difference state determination, it ensures that no absolute peak nodes in the full data are missed.

[0027] S202: Based on the core weight values ​​and the remaining weight level codes, the core weight values ​​are used as the base of the minuend, and the remaining weight level codes are subtracted one by one to obtain discrete difference components. The mapping between discrete difference components and text node spatial positions is recorded. According to the position sorting logic, all discrete difference components are linearly arranged to establish a weight decay structure and generate a weight gradient distribution sequence. Extract the fixed backbone weight values ​​from the register and the remaining weight level codes temporarily stored in the queue. Start the batch processing arithmetic logic unit, using the backbone weight values ​​as the base for the subtraction, and pop the remaining weight level codes one by one from the queue as subtraction items. Subtract the remaining weight level codes from the base of each subtraction item. Save the absolute value obtained from each subtraction to obtain the discrete difference components. Retrieve the starting position index of the corresponding text node in the rural public opinion text statement, establish a corresponding concatenated string between the discrete difference components and this starting position index, and record the spatial position mapping between the discrete difference components and the text node. Read the starting position indices of all records, and perform a linear arrangement operation on all discrete difference components according to the position sorting logic, following an ascending order rule. The arranged discrete difference components are sequentially filled into a one-dimensional floating-point array. One-way pointer references are established between adjacent elements, and a weight decay structure is created. This weight decay structure, combined with a dispersion statistical algorithm from big data analysis, makes the spatial location mapping of each text node more statistically significant. The entire memory block of this one-dimensional floating-point array is read and converted into a fixed-length floating-point sequence stream to generate a weight gradient distribution sequence. The main weight is set to 35.6, and the first residual weight level is encoded as 12.8, corresponding to a text node starting index of 15. The calculation process for obtaining the discrete difference components is as follows: Discrete difference component = Main weight value - Residual weight level code; Calculation results: Discrete difference component = 35.6 - 12.8 = 22.8; 22.8 is bound to the starting position index 15 and placed into the corresponding sequence slot of the weight decay structure. The advantage of this operation logic is that by subtracting the root of the main body from the remaining codes one by one, the degree of deviation between each part of the statement and the core focus weight is clearly mapped.

[0028] S203: Obtain the context sequence of rural public opinion text statements, extract the clause level numbers associated with adjacent rural public opinion text statements, calculate the level index difference for adjacent clause level numbers, compare the level index difference with the preset number association range, lock the internal position interval of the weight gradient distribution sequence that meets the association boundary, superimpose the preset continuous augmented value on the internal position interval, reorganize the overall vector matrix, and generate the text structure reconstruction sequence. The system retrieves the timestamps of the reading order of rural public opinion texts stored locally. It then sequentially captures text statements before and after the original statements, obtaining a context sequence of rural public opinion text statements. The metadata in the context sequence is parsed to extract the clause hierarchy numbers associated with adjacent rural public opinion text statements. The clause hierarchy number of the preceding statement is read as the minuend, and the clause hierarchy number of the following statement is read as the subtrahend. Subtraction is performed on adjacent clause hierarchy numbers to obtain their absolute values, calculating the hierarchy index difference. The configuration parameter table is read to obtain the upper and lower limits of the preset number association range, setting the upper limit to 3 and the lower limit to 0. The hierarchy index difference is compared with the preset number association range. When the hierarchy index difference is between the upper and lower limits, the position coordinates of the current adjacent statement are recorded, locking the internal position interval of the weight gradient distribution sequence that conforms to the association boundary. A preset continuous augmentation value is retrieved and set to 5.0. Addition is performed on all elements within the internal position interval, superimposing the preset continuous augmentation value. The superimposed local sequence is placed back into the original one-dimensional floating-point array, reconstructing the overall vector matrix. The recombined overall vector matrix undergoes normalization division. This recombining and normalization process deeply aligns with the standardization requirements of multidimensional matrices in big data analysis, laying a solid data foundation for subsequent mining of temporal patterns and generating a text structure reconstruction sequence. The clause level number of the preceding statement is set to 5, the clause level number of the following statement to 2, the upper limit of the preset number association range is 3, the preset continuous augmentation value is 5.0, and the first element value of the internal position interval is 22.8. The calculation process for the hierarchical index difference and superposition augmentation operation is as follows: Hierarchical index difference = Previous hierarchical number - Next hierarchical number; Enhanced element value = original element value + preset continuous augmentation value; Calculation results: Hierarchical index difference = 5 - 2 = 3; The difference in the hierarchical index equals the upper limit of 3, which meets the boundary requirements.

[0029] The enhanced element value = 22.8 + 5.0 = 27.8; Update 27.8 to the overall vector matrix. The advantage of this operation is that it numerically enhances the semantically coherent context by comparing hierarchical index differences and applying quantitative augmentation values.

[0030] Please see Figure 4 The specific steps of S3 are as follows: S301: Obtain the time axis data range, perform sliding segmentation on the time axis data range according to the preset time step, construct adjacent overlapping time windows with overlapping time features, call the text structure reconstruction sequence, extract the key public opinion words mapped by the text structure reconstruction sequence, count the frequency of key public opinion words in each adjacent overlapping time window, calculate the difference of adjacent occurrence frequency pair by pair, aggregate all differences, and generate a word frequency overlap difference sequence. The system connects to the server's clock log to extract the times of the first and last public opinion record, thus obtaining the timeline data range. It reads the preset time step parameter, setting it to 24 hours and the window length to 48 hours. A sliding segmentation operation is performed across the entire timeline data range using a 24-hour step, constructing data blocks that are consecutive with a 24-hour overlap in between, creating adjacent overlapping time windows. This process of constructing adjacent overlapping time windows is a typical method for processing continuous time series data in big data analysis, effectively capturing the micro-dynamic characteristics of public opinion evolution. The system calls the generated text structure to reconstruct the sequence, using a character pattern matching operator to extract key public opinion terms mapped from the reconstructed sequence. For each text record within a data block, the total occurrence count of key public opinion terms is accumulated, and the frequency of key public opinion terms in each adjacent overlapping time window is calculated. The algorithm reads the frequency of occurrence in overlapping time windows with earlier timestamps, and the frequency of occurrence in adjacent overlapping time windows with later timestamps. It then performs the operation of subtracting the earlier frequency from the later frequency for each pair, calculating the frequency difference between adjacent occurrences. All generated differences are stored in a one-dimensional difference array in chronological order. All differences are then aggregated to generate a word frequency overlap difference sequence. For example, a key public opinion term is assumed to appear 156 times in the preceding overlapping time window, and 234 times in the following adjacent overlapping time window. The calculation process for the frequency difference between adjacent occurrences is as follows: The frequency difference between adjacent occurrences = the frequency of the later occurrence - the frequency of the earlier occurrence; Calculation results: The frequency difference between adjacent occurrences = 234 - 156 = 78; The difference of 78 is appended as a discrete data point to the end of the one-dimensional difference array. The advantage of this operation is that by directly subtracting the preceding frequency from the subsequent frequency, the background noise of the popularity is filtered out, highlighting the absolute increment of public opinion evolution.

[0031] S302: Input the word frequency overlap difference sequence into the differential autoregressive moving average model, call the differential operator of the differential autoregressive moving average model to perform a stationary transformation on the difference, extract the positive and negative sign states of the sequence after the stationary transformation, compare the positive and negative sign states of adjacent time nodes, filter the target time nodes corresponding to the positive and negative flips, extract the coordinates of the target time nodes, and generate the sign change node positions. The overlapping term frequency difference sequence data stream is retrieved and input into the data receiver of the differential autoregressive moving average model. The internal structure of this model consists of three feedforward operation layers. The input layer is the difference operation layer, which sets a delayed node to perform a staggered subtraction operation on the sequence elements. The intermediate layer is the autoregressive calculation layer, which deploys five neurons with linear activation functions to receive the difference data from the past five time steps. The historical data is multiplied by pre-trained coefficients and summed using a fully connected structure. The output layer is the moving average layer, which deploys three neurons to receive the output residuals from the autoregressive calculation layer. The residual terms are processed using a mean squared error penalty function and summed to obtain the final fitted value. The difference operator in the input layer is called to perform the subtraction of the previous term from the next term on the difference sequence, completing the stationary transformation operation. The sign state of the sequence elements after the stationary transformation is extracted; values ​​greater than 0 are marked as positive, and values ​​less than 0 are marked as negative. By comparing the sign states of adjacent time points, when a positive sign followed by a negative sign, or vice versa, is detected, the target time point corresponding to the sign reversal is selected. The absolute offset scale of this target time point on the entire time axis is extracted, i.e., the target time point coordinates are extracted. For the extraction and reconstruction of the time point offset scale, a time series abrupt change detection engine from big data analysis is fully utilized, significantly enhancing the ability to capture sign reversals in non-stationary time series. The extracted absolute offset scales are summarized and arranged to generate the sign change node positions. The data obtained from the autoregressive calculation layer one step before the current time is set to 78, with a pre-training coefficient of 0.6; the data two steps before the current time is set to 45, with a pre-training coefficient of 0.4. The calculation process of the autoregressive fitted value is as follows: Autoregressive fit value = previous step data * previous step pre-training coefficient + previous two steps data * previous two steps pre-training coefficient; Calculation results: The autoregressive fit value = 78 * 0.6 + 45 * 0.4 = 46.8 + 18.0 = 64.8; The 64.8 is passed to the moving average layer for residual correction. The advantage of this operation is that by using a weighted summation of multi-step historical data and pre-trained coefficients, it effectively smooths out the high-frequency noise caused by short-term fluctuations.

[0032] Table 2 Model Structure Parameter Configuration Table

[0033] As shown in Table 2, the node layout, function selection, and coefficient constraint boundaries of each level of the model are clearly defined.

[0034] S303: For the symbol change node position, extract the local preset merging distance parameter, perform a subtraction operation on the adjacent symbol change node positions to obtain the time span metric, compare the time span metric with the local preset merging distance parameter, perform interval closure processing on the symbol change node positions corresponding to the local preset merging distance parameter, determine the starting boundary and ending time boundary after aggregation processing, and generate the public opinion change time segment. The method for obtaining the merging distance parameter is as follows: obtain the text release time series in the evolution process of historical digital rural public opinion events, extract the time interval sample data of adjacent fermentation cycles in the text release time series, calculate the statistical distribution mean of the time interval sample data, and multiply the distribution mean with the preset time tolerance coefficient. The resulting product value is used as the merging distance parameter.

[0035] The system retrieves the set of symbol change node positions from memory and connects to the local storage unit to obtain the preset merging distance parameter. This preset merging distance parameter is extracted; its data type is an integer variable. For the sorted symbol change node positions, following the index order, the absolute offset scale of the subsequent symbol change node position is subtracted from the absolute offset scale of the previous symbol change node position, and the subtraction operation is performed to obtain the time span measure. The time span measure is compared with the local preset merging distance parameter. When the time span measure is smaller than the local preset merging distance parameter, the corresponding two adjacent symbol change node positions are grouped into the same cluster, and interval closure processing is performed. Within the cluster, the smallest absolute offset scale is determined as the starting boundary after aggregation, and the largest absolute offset scale is determined as the ending time boundary. These two boundary values ​​are merged to generate the public opinion change time segment. The merge distance parameter is obtained by reading the historical digital rural public opinion event database, extracting the text release time series during the event evolution process. When processing the historical digital rural public opinion event database, a big data analysis platform was used to perform a full feature backtracking on the text release time series accumulated over many years, ensuring the objectivity of the sample data distribution mean. Time interval sample data between adjacent fermentation cycles were extracted from the text release time series. All time interval sample data were summed and divided by the total number of samples to calculate the statistical distribution mean. A preset time tolerance coefficient was read, and the distribution mean was multiplied by the preset time tolerance coefficient. The resulting product was assigned to a variable as the merge distance parameter. The total time interval sample data of historical adjacent fermentation cycles was set to 480 hours, the total number of samples was 12, and the preset time tolerance coefficient was 1.5. The calculation process of the merge distance parameter is as follows: Distribution mean = sum of sample data / total number of samples; Merge distance parameter = distribution mean * time tolerance coefficient; Calculation results: Distribution mean = 480 / 12 = 40; Merging distance parameter = 40 * 1.5 = 60; The merging distance parameter is set to 60 hours. The advantage of this calculation logic is that it dynamically adapts to changes in the time scale caused by differences in the rhythm of different historical events by calculating the product of the distribution mean and the time tolerance coefficient.

[0036] Please see Figure 5 The specific steps of S4 are as follows: S401: Extract the propagation node time records included in the time segment of public opinion change, perform linear sorting on all propagation node time records according to the time sequence logic, reorganize the record sequence, perform timestamp subtraction operation on adjacent propagation node time records, extract the subtraction difference value, aggregate all subtraction difference values ​​according to the sorting correspondence, form a discrete set, and generate a serialized time interval. The underlying code of the obtained public opinion change time period is analyzed, and the timestamp information of network forwarding and commenting operations is extracted. The time records of propagation nodes within the public opinion change time period are also extracted. Faced with massive network forwarding and commenting operation logs, relying on the distributed computing advantages of big data analysis components, extremely fast linear sorting of propagation node time records is achieved. Using a fast sorting algorithm engine, based on timestamp values ​​as the comparison benchmark, all propagation node time records are sorted in ascending linear order according to chronological logic. Contiguous memory address space is reallocated, and the record sequence is reassembled. In the reassembled record sequence, a bidirectional sliding window is set. For two adjacent propagation node time records, the timestamp of the later record is subtracted from the timestamp of the earlier record to perform a timestamp subtraction operation. The non-negative integer value generated after the operation is read, and the difference value is extracted. A dynamically expanding one-dimensional array structure is established. According to the index correspondence in the original sorting process, all the difference values ​​are sequentially pushed into this one-dimensional array for aggregation, forming a discrete set. The data is encapsulated and output as a data stream in a fixed format, generating a serialized time interval. The time record of the previous propagation node is set to 16200 seconds, and the time record of the next adjacent propagation node is set to 16540 seconds. The calculation process for the difference is as follows: The difference between the two propagation nodes is equal to the time record of the next propagation node minus the time record of the previous propagation node. Calculation results: The difference between the subtraction and subtraction values ​​= 16540 - 16200 = 340; The 340-second difference is pressed into the corresponding sequence position of the discrete set. The advantage of this operation is that by obtaining the difference value through direct subtraction of timestamps, the interference of absolute time coordinates on propagation frequency analysis is completely eliminated.

[0037] S402: Call the local preset Gini coefficient solver, substitute the serialized time interval into the preset Gini coefficient solver to calculate the cumulative distribution probability data, solve the time distribution imbalance index parameter based on the cumulative distribution probability data, map the time distribution imbalance index parameter to the standard numerical space range to construct a single-dimensional scalar, and generate the time offset dispersion value. The process retrieves a pre-built Gini coefficient calculation operator from the local server. It passes the pointers to the serialized time intervals generated in memory to the input of the Gini coefficient calculation operator. Internally, the serialized time intervals are sorted from smallest to largest, the ratio of each interval to the total number of intervals is calculated, and these ratios are accumulated to calculate the cumulative probability distribution data. This step fully integrates the Gini coefficient calculation model from big data analysis, accurately analyzing the time distribution imbalance index parameters. Based on the cumulative probability distribution data, a Lorenz curve is plotted, and the area between the diagonal line of absolute equality and the Lorenz curve is calculated. This area is multiplied by 2 to solve for the time distribution imbalance index parameters. The minimum value of the standard numerical space range is set to 0, and the maximum value to 100. The obtained time distribution imbalance index parameters are multiplied by 100, a linear amplification operation is performed, and the data is mapped to the standard numerical space range to construct a single-dimensional scalar. This single-dimensional scalar is converted into a character data structure, a fixed classification prefix field is added, and a time offset dispersion value is generated. The area between the absolute equal diagonal line and the Lorenz curve is set to 0.32. The calculation process for the time distribution imbalance index parameters and the single-dimensional scalar is as follows: The time distribution imbalance index parameter = area between curves * 2; Single-dimensional scalar = Time distribution imbalance index parameter * 100; Calculation results: The time distribution imbalance index parameter = 0.32 * 2 = 0.64; One-dimensional scalar = 0.64 * 100 = 64; The value 64 is assigned as a single-dimensional scalar and converted into an offset dispersion value. The advantage of this operation logic is that by doubling the area value and multiplying it by one hundred, a small probability distribution deviation is transformed into a significant and intuitive scalar with a scale of one hundred times.

[0038] S403: For the time offset dispersion value, obtain the preset overall average interval parameter and the total number of local network nodes, calculate the absolute deviation value between the time offset dispersion value and the preset overall average interval parameter, extract the division quotient value between the absolute deviation value and the preset overall average interval parameter, construct the offset segment ratio, compare the offset segment ratio with the total number of local network nodes, determine the diffusion boundary range, match the local classification mapping dictionary to determine the character identifier item, and generate the event diffusion intensity classification label; The process of determining character identifiers by matching the local classification mapping dictionary is as follows: First, obtain the connectivity coefficient of pre-set active nodes. Then, multiply the pre-set active node connectivity coefficient by the total number of local network nodes to extract the baseline value of the basic nodes. The pre-set active node connectivity coefficient is set based on the median of the ratio of edges to vertices in the historical network topology graph. Next, perform numerical mapping processing on the baseline value of the basic nodes using a logarithmic operator to obtain a smooth comparison baseline. Extract the offset segment ratio and calculate the difference ratio between the offset segment ratio and the smooth comparison baseline. Then, divide the diffusion boundary range based on the polarity and absolute value of the difference ratio value. Call the local classification mapping dictionary, which records the binding structure between multiple numerical boundary sub-intervals and predefined character vectors. Compare the coverage relationship between the diffusion boundary range and multiple numerical boundary sub-intervals. Lock the intersecting numerical boundary sub-intervals and extract the bound predefined character vectors as character identifiers. Concatenate the character identifiers to construct the event diffusion intensity classification label.

[0039] Retrieve the temporarily stored time offset dispersion value. Read the basic configuration file to obtain the preset overall average interval parameter. Connect to the network topology management database to count the total number of currently active node entries and obtain the total number of local network nodes. When counting the total number of currently active node entries, real-time snapshots of the entire network dynamic topology are captured using big data analysis technology to ensure the absolute accuracy of the underlying benchmark for calculating the offset segment ratio. Subtract the preset overall average interval parameter from the time offset dispersion value, and then take the absolute value of the result to calculate the absolute deviation value. Use the absolute deviation value as the dividend and the preset overall average interval parameter as the divisor to perform a division operation and extract the division quotient. Keep four decimal places to construct the offset segment ratio. Compare the offset segment ratio with the threshold conditions converted from the total number of local network nodes to determine the diffusion boundary range. Start the key-value pair query service to match the local classification mapping dictionary. Extract the character records bound to the diffusion boundary range from the dictionary, determine the character identifier items, and concatenate them into the event archive metadata to generate event diffusion intensity classification tags. The system obtains the preset active node connectivity coefficient, multiplies it by the total number of local network nodes, and extracts the baseline value for the basic nodes. This coefficient is set based on the median of the ratio distribution of the number of edges to the number of vertices in the historical network topology graph. A logarithmic operator with a base of the natural constant is used to operate on the baseline value of the basic nodes to obtain a smoothing comparison benchmark. The offset segment ratio is extracted, and the smoothing comparison benchmark is subtracted from the offset segment ratio to calculate the difference ratio value. The diffusion boundary range is defined based on the polarity (positive or negative) and magnitude of the difference ratio value. The local classification mapping dictionary is called, which contains multiple intervals of lower and upper limits and mappings of predefined characters. The intersection of the diffusion boundary range with these sub-intervals is compared, the intersecting sub-intervals are identified, the corresponding character identifiers are extracted, and concatenation is used to construct the event diffusion intensity classification label. The time offset dispersion value is set to 64, the preset overall average interval parameter is 40, the total number of local network nodes is 5000, and the preset active node connectivity coefficient is 0.05. The process of calculating the absolute deviation value, the offset segment ratio, and the base node reference value is as follows: Subtraction of absolute deviation value = absolute value calculation (time offset dispersion value - preset overall average interval parameter); Offset segment ratio = Subtraction of absolute deviation value / Preset overall average interval parameter; Baseline node value = Total number of local network nodes * Preset active node connectivity coefficient; Calculation results: Subtraction of absolute deviation value = absolute value calculation (64-40) = 24; Offset segment ratio = 24 / 40 = 0.60; Base node baseline value = 5000 * 0.05 = 250; The baseline value of 250 for the basic node is logarithmically smoothed before being used for subsequent polarity determination. The advantage of this calculation logic is that by calculating the product of node connectivity using the ratio of absolute deviation to average interval, the absolute influence of network size on the offset metric is eliminated.

[0040] Please see Figure 6 The specific steps of S5 are as follows: S501: Retrieve the local preset risk level mapping table, compare the event diffusion intensity classification label with the built-in associated label item of the preset risk level mapping table, extract the hierarchical numerical scalar of the matching built-in associated label item, set the hierarchical numerical scalar as the quantitative hazard assessment parameter, establish the mapping conversion logic, and generate the risk classification value. The system calls the local database management query interface to retrieve the pre-set risk level mapping table stored in the disk drive. The event diffusion intensity classification label is passed to the retrieval parameter column. The system scans the character information in the primary key column of the mapping table, comparing the string sequence of the event diffusion intensity classification label with the built-in associated label item in the pre-set risk level mapping table. This scanning and comparison of the character information in the primary key column of the mapping table, thanks to the large-scale distributed key-value matching technology under the big data analysis system, effectively shortens the mapping conversion cycle of risk level values. When the comparison result is true, the system reads the record and extracts the hierarchical value scalar attached to the row containing the matching built-in associated label item. A floating-point conversion variable is declared in the memory runtime environment, and the hierarchical value scalar is assigned to this conversion variable to establish the mapping conversion logic. A forced formatting process of multiplying the conversion variable by 1.0 is performed to generate the risk level value, which is then written to the global status register. The hierarchical value scalar attached to the matching built-in associated label item matched by the event diffusion intensity classification label is set to 4, and the forced formatting constant is set to 1.0. The calculation process for generating the risk level value is as follows: Risk rating value = hierarchical numerical scalar * forced formatting processing constant; Calculation results: Risk rating value = 4 * 1.0 = 4.0; 4.0 is used as the base value for the quantified hazard level. The advantage of this operational logic is that it directly converts discrete hierarchical scalars into a recognizable standard floating-point number format through multiplicative formatting.

[0041] S502: Analyze the spatial location metadata carried by the time segment of sudden public opinion changes, extract the network dissemination platform node identifier and physical geographic area coordinates included in the spatial location metadata, perform merging and splicing processing on the network dissemination platform node identifier and physical geographic area coordinates, construct a boundary polygon structure, establish a two-dimensional spatial expansion boundary, and generate occurrence range data; The system captures additional metadata blocks from memory representing time periods of sudden public opinion events. Following a JSON path depth-first traversal mechanism, it parses nested data containing spatial locations. Platform attribute fields are extracted from the nested data to obtain the network propagation platform node identifier. A geographic information interface is invoked to perform inverse latitude and longitude parsing of address text, extracting entity geographic region coordinates. In the inverse latitude and longitude parsing stage, a spatial big data analysis engine is deeply applied to ensure the efficiency and accuracy of entity geographic region coordinate extraction. A string builder is declared, using the network propagation platform node identifier as a prefix, connected by underscores followed by the entity geographic region coordinates, and then merged and concatenated. In the geographic information cache, based on a vertex sequence composed of multiple entity geographic region coordinates, lines are connected clockwise to construct a closed shape, creating a boundary polygon structure. A fixed outer buffer distance of 5 kilometers is set, and equidistant parallel boundary lines are extended outwards from the boundary polygon structure to establish a two-dimensional spatial extension boundary. The polygon coordinate set, including the extension boundary, is encapsulated and output to generate the occurrence range data. The network propagation platform node identifier is set to the numeric code 1024, the geographical coordinates of the entity are 116.40 (longitude) and 39.90 (latitude). The string operation process of merging and concatenating is transformed into logical combination: The merged and spliced ​​result = node identifier connection character plus longitude character plus latitude character; Results Display: The merged and spliced ​​result is 1024_116.40_39.90.

[0042] This combined string is used as a globally unique identifier for the geographic region. The advantage of this logic is that by directly concatenating the node identifier with latitude and longitude, it ensures the unique indexing characteristic of multi-dimensional spatial entities in a flattened database.

[0043] S503: Calculate the product of the risk level value and the area covered by the occurrence range data to obtain the comprehensive threat level, extract the preset alarm trigger threshold, compare the comprehensive threat level with the preset alarm trigger threshold, and when it is determined that the comprehensive threat level exceeds the preset alarm trigger threshold, retrieve the emergency linkage instruction, combine the emergency linkage instruction and the occurrence range data, construct a formatted data package, and generate digital rural public opinion early warning data. The process of retrieving emergency response instructions when the overall threat level exceeds the preset alarm trigger threshold is as follows: Extract the latitude and longitude coordinates of each entity's geographic area included in the occurrence range data; substitute these coordinates into the spherical distance calculation logic to obtain the boundary span distance between adjacent latitude and longitude coordinates; calculate the geometric space mapping area based on the closed geometric figure formed by the latitude and longitude coordinates and the boundary span distance, and determine the geometric space mapping area as the area covered by the occurrence range; multiply the area covered by the occurrence range by the risk classification value, and extract the product as the overall threat level; retrieve locally stored historical records of similar events, calculate the arithmetic mean of the threat levels included in the historical records of similar events, and generate the historical average threat level. The system extracts the actual total number of public opinion posts and the baseline of the normalized total number of posts within a predetermined historical time period, calculates the division ratio of the actual total number of public opinion posts to the baseline of the normalized total number of posts, and generates a sensitive emotion fluctuation coefficient. It then multiplies the historical average threat level with the sensitive emotion fluctuation coefficient, setting the result of the multiplication operation as a preset alarm trigger threshold. Finally, it compares the overall threat level with the preset alarm trigger threshold; when the overall threat level exceeds the preset alarm trigger threshold, it calculates the excess value of the overall threat level minus the preset alarm trigger threshold; it compares the excess value with the preset difference interval boundary in the local hierarchical response strategy library; and extracts the processing plan text associated with the difference interval boundary into which the excess value falls, setting the processing plan text as an emergency response instruction. The system retrieves the risk classification value from the bus and the area covered by the occurrence range data output by the polygon coordinate calculation operator. Using the risk classification value as the multiplier and the area covered as the multiplicand, it performs a multiplication calculation to obtain the product value and thus the overall threat level. It queries the early warning configuration table to extract the preset alarm trigger threshold. A logical judgment function compares the overall threat level with the preset alarm trigger threshold. When the overall threat level exceeds the preset alarm trigger threshold, the trigger condition is deemed met, and a request code is sent to the scheduling module to retrieve the emergency response command. The text content of the emergency response command and the coordinate set of the occurrence range data are packaged into the same XML document format for combination, constructing a formatted data packet. This data packet is then sent to the terminal to generate digital rural public opinion early warning data. During the judgment process, the latitude and longitude coordinates of each entity's geographic area included in the occurrence range data are extracted. The differences in adjacent longitude and latitude values ​​are substituted into the semi-sine spherical distance calculation logic to obtain the boundary span distance between adjacent latitude and longitude points. Based on a closed geometric figure formed by latitude and longitude points and boundary span distances, the geometric spatial mapping area is calculated, and this area is determined as the area covered by the occurrence range. The area is multiplied by the risk value to obtain the comprehensive threat level. Historical records of similar events stored locally are retrieved. When retrieving these records and calculating the arithmetic mean of threat levels, a multi-dimensional big data analysis model is fully activated, enabling intelligent comparison between the normalized posting base and actual public opinion fluctuations. Multiple threat levels in the records are summed and divided by the total number of records to calculate the arithmetic mean of threat levels, generating the historical average threat level. The total amount of actual public opinion posts crawled within a predetermined time period is extracted and compared with the locally configured normalized posting base. The actual total is used as the dividend, and the normalized base is used as the divisor to calculate the division ratio, generating a sensitive sentiment fluctuation coefficient. The historical average threat level and the sensitive sentiment fluctuation coefficient are multiplied, and the result is set as a preset alarm trigger threshold. When the comprehensive threat level exceeds the threshold, the excess value is calculated by subtracting the preset alarm trigger threshold from the comprehensive threat level. Compare the difference value with the difference range in the strategy library, and extract the contingency plan text as the instruction. The scope of the incident is set to cover an area of ​​150 square kilometers, with a risk rating of 4.0. The total historical threat level is 1200, the total number of records is 100, the total number of actual public opinion posts is 8000, and the base number of normalized posts is 5000. The calculation process for the comprehensive threat level, historical average threat level, sensitive emotion fluctuation coefficient, and preset alarm trigger threshold is as follows: Overall threat level = Area covered by the incident * Risk level score; Historical average threat level = Total historical threat level / Total number of records; Sensitive Emotion Fluctuation Coefficient = Total Actual Number of Public Opinion Posts / Base Number of Regular Posts; Preset alarm trigger threshold = historical average threat level * sensitive emotion fluctuation coefficient; Exceeding the difference value = Overall threat level - Preset alarm trigger threshold; Calculation results: Overall threat level = 150 * 4.0 = 600; Historical average threat level = 1200 / 100 = 12; Sensitive emotion fluctuation coefficient = 8000 / 5000 = 1.6; The preset alarm trigger threshold is 12 * 1.6 = 19.2. The difference exceeds the limit by 600 - 19.2 = 580.8; Because 600 is greater than 19.2, an alarm is triggered, and a higher-level emergency command is matched based on the difference value of 580.8. The advantage of this calculation logic is that by adjusting the ratio of the product of area and risk value to the total number of posts, a two-dimensional dynamic quantification of spatial breadth and public opinion heat is achieved.

[0044] Table 3 Risk Threat Assessment Data Table

[0045] As shown in Table 3, the basic input quantities of each key indicator and their computational role configuration in the final judgment are listed in detail.

[0046] A big data-driven public opinion analysis and early warning system for digital rural governance includes: The policy semantic association module obtains the digital rural policy text, extracts the clause level number and converts it into a weight level code, obtains rural public opinion text statements, performs word segmentation and extraction on the rural public opinion text statements, generates policy vocabulary mapping records, constructs the mapping relationship between policy vocabulary mapping records and weight level codes, and generates statement code binding data. The hierarchical weighting module extracts multiple weight level codes from the statement code binding data, performs numerical comparisons, selects the weight level code corresponding to the maximum value as the main weight value, calculates the difference to generate a weight gradient distribution sequence, and combines the continuity of clause hierarchical numbering to construct a text structure reconstruction sequence. The temporal fluctuation identification module constructs adjacent overlapping time windows according to a preset time step, extracts the occurrence frequency of public opinion words from the text structure reconstruction sequence and calculates the difference, generates a word frequency overlap difference sequence, performs trend analysis using a differential autoregressive moving average model, aggregates the position of symbol change nodes, and generates public opinion change time segments. The propagation trend judgment module extracts the propagation node time records included in the time segment of public opinion change, generates the serialized time interval, calculates the proportion of the offset segment, compares it with the total number of network nodes, and generates event diffusion intensity classification labels. The tiered linkage early warning module extracts event spread intensity classification labels, maps risk classification values ​​and the occurrence range data of public opinion change time periods, performs risk assessment, and generates digital rural public opinion early warning data.

[0047] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the described technical solutions.

Claims

1. A big data-driven public opinion analysis and early warning method for digital rural governance, characterized in that, Includes the following steps: S1: Obtain the digital rural policy text, extract the clause level number and convert it into a weight level code, obtain the rural public opinion text statement, perform word segmentation and extraction on the rural public opinion text statement, generate policy vocabulary mapping record, construct the mapping relationship between the policy vocabulary mapping record and the weight level code, and generate statement code binding data; S2: Extract multiple weight level codes from the statement code binding data, perform numerical comparison, select the weight level code corresponding to the maximum value as the main weight value, calculate the difference to generate a weight gradient distribution sequence, and combine it with the continuity of clause level numbering to construct a text structure reconstruction sequence; S3: Construct adjacent overlapping time windows according to a preset time step, extract the occurrence frequency of public opinion words from the text structure reconstruction sequence and calculate the difference, generate a word frequency overlap difference sequence, perform trend analysis using a differential autoregressive moving average model, aggregate the position of symbol change nodes, and generate public opinion change time segments. S4: Extract the propagation node time records included in the public opinion change time segment, generate the serialized time interval, calculate the offset segment ratio, compare it with the total number of network nodes, and generate event diffusion intensity classification labels. S5: Extract the event diffusion intensity classification label, map the risk level value and the occurrence range data of the public opinion change time segment, perform risk judgment, and generate digital rural public opinion early warning data.

2. The big data public opinion analysis and early warning method for digital rural governance according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Obtain authorization to publish the digital rural policy text, extract the embedded clause level number in the digital rural policy text, perform character identifier decomposition operation on the clause level number, extract the structured level identifier, assign decreasing values ​​to the structured level identifier according to the preset level descending order logic, aggregate the structured level identifier and the decreasing values ​​to establish a level mapping dictionary, and generate weight level code. S102: Obtain locally stored rural public opinion text statements, perform character segmentation processing on the rural public opinion text statements to extract independent words, compare the independent words with the preset policy proper noun library to extract matching text nodes, extract the coordinate positions of the text nodes in the rural public opinion text statements, bind the coordinate positions and text nodes to form a position vector sequence, establish a vector mapping association, and generate a policy term mapping record. S103: Based on the weight level code and policy term mapping record, extract the clause instruction label carried in the policy term mapping record, compare the clause instruction label with the identifier sequence covered by the weight level code to locate the associated weight, assign the associated weight to the corresponding text node to construct a numerical scalar element, merge the numerical scalar elements of the same single statement with the statement index sequence to construct a multidimensional key-value pair, integrate all multidimensional key-value pairs to perform topological merging processing, and generate statement code binding data.

3. The big data public opinion analysis and early warning method for digital rural governance according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Extract the weight level code attached inside the statement code binding data, perform traversal for multiple weight level codes, filter the node with the maximum value of the numerical dimension, extract the weight level code associated with the maximum value node independently, and generate the main weight value. S202: Based on the core weight value and the remaining weight level code, the core weight value is used as the base of the subtraction term. The remaining weight level code is subtracted one by one to obtain discrete difference components. The mapping between discrete difference components and text node spatial positions is recorded. Linear arrangement is performed on all discrete difference components according to the position sorting logic to establish a weight decay structure and generate a weight gradient distribution sequence. S203: Obtain the context sequence of rural public opinion text statements, extract the clause level numbers associated with adjacent rural public opinion text statements, calculate the level index difference for adjacent clause level numbers, compare the level index difference with the preset number association range, lock the internal position interval of the weight gradient distribution sequence that meets the association boundary, superimpose preset continuous augmented values ​​on the internal position interval, reorganize the overall vector matrix, and generate a text structure reconstruction sequence.

4. The big data public opinion analysis and early warning method for digital rural governance according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Obtain the time axis data range, perform sliding segmentation on the time axis data range according to the preset time step, construct adjacent overlapping time windows with overlapping time features, call the text structure reconstruction sequence, extract the key public opinion words mapped by the text structure reconstruction sequence, count the frequency of key public opinion words in each adjacent overlapping time window, calculate the difference of adjacent occurrence frequency pair by pair, aggregate all differences, and generate a word frequency overlap difference sequence. S302: Input the word frequency overlap difference sequence into the differential autoregressive moving average model, call the differential operator of the differential autoregressive moving average model to perform a stationary transformation on the difference, extract the positive and negative sign states of the sequence after the stationary transformation, compare the positive and negative sign states of adjacent time nodes, filter the target time nodes corresponding to the positive and negative flips, extract the coordinates of the target time nodes, and generate the sign change node positions. S303: For the symbol change node position, extract the local preset merging distance parameter, perform a subtraction operation on the adjacent symbol change node positions to obtain the time span metric, compare the time span metric with the local preset merging distance parameter, perform interval closure processing on the symbol change node positions corresponding to the local preset merging distance parameter, determine the starting boundary and ending time boundary after aggregation processing, and generate the public opinion change time segment.

5. The big data public opinion analysis and early warning method for digital rural governance according to claim 4, characterized in that, The method for obtaining the merging distance parameter is as follows: obtain the text release time series in the evolution process of historical digital rural public opinion events, extract the time interval sample data of adjacent fermentation cycles in the text release time series, calculate the statistical distribution mean of the time interval sample data, and multiply the distribution mean with the preset time tolerance coefficient, and use the resulting product value as the merging distance parameter.

6. The big data public opinion analysis and early warning method for digital rural governance according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Extract the propagation node time records included in the time segment of the sudden change in public opinion, perform linear sorting processing on all propagation node time records according to the time sequence logic, reorganize the record sequence, perform timestamp subtraction operation on adjacent propagation node time records, extract the subtraction difference value, aggregate all subtraction difference values ​​according to the sorting correspondence, form a discrete set, and generate a serialized time interval. S402: Call the local preset Gini coefficient solver, substitute the serialized time interval into the preset Gini coefficient solver to calculate the cumulative distribution probability data, solve the time distribution imbalance index parameter based on the cumulative distribution probability data, map the time distribution imbalance index parameter to the standard numerical space range to construct a single-dimensional scalar, and generate the time offset dispersion value. S403: For the time offset dispersion value, obtain the preset overall average interval parameter and the total number of local network nodes, calculate the absolute deviation value between the time offset dispersion value and the preset overall average interval parameter, extract the division quotient value between the absolute deviation value and the preset overall average interval parameter, construct the offset segment ratio, compare the offset segment ratio with the total number of local network nodes, determine the diffusion boundary range, match the local classification mapping dictionary to determine the character identifier item, and generate the event diffusion intensity classification label.

7. The big data public opinion analysis and early warning method for digital rural governance according to claim 6, characterized in that, The process of determining character identifiers by matching the local classification mapping dictionary specifically involves: obtaining a preset active node connectivity coefficient; multiplying the preset active node connectivity coefficient by the total number of local network nodes to extract the baseline value of the basic node; setting the preset active node connectivity coefficient based on the median of the ratio distribution of the number of edges to the number of vertices in the historical network topology graph; performing numerical mapping processing on the baseline value of the basic node using a logarithmic operator to obtain a smooth comparison benchmark; extracting the offset segment ratio; calculating the difference ratio between the offset segment ratio and the smooth comparison benchmark; dividing the diffusion boundary range based on the polarity and absolute value of the difference ratio; calling the local classification mapping dictionary, which records the binding structure between multiple numerical boundary sub-intervals and predefined character vectors; comparing the coverage relationship between the diffusion boundary range and multiple numerical boundary sub-intervals; locking the intersecting numerical boundary sub-intervals; extracting the bound predefined character vectors as character identifiers; and concatenating the character identifiers to construct an event diffusion intensity classification label.

8. The big data public opinion analysis and early warning method for digital rural governance according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Retrieve the local preset risk level mapping table, compare the event diffusion intensity classification label with the built-in associated label item of the preset risk level mapping table, extract the hierarchical numerical scalar of the matching built-in associated label item, set the hierarchical numerical scalar as the quantitative hazard assessment parameter, establish mapping conversion logic, and generate risk classification value. S502: Analyze the spatial location metadata carried by the time segment of the sudden change in public opinion, extract the network propagation platform node identifier and the physical geographic area coordinates included in the spatial location metadata, perform merging and splicing processing on the network propagation platform node identifier and the physical geographic area coordinates, construct a boundary polygon structure, establish a two-dimensional spatial expansion boundary, and generate occurrence range data. S503: Calculate the product of the risk level value and the area covered by the occurrence range data to obtain the comprehensive threat level, extract the preset alarm trigger threshold, compare the comprehensive threat level with the preset alarm trigger threshold, and when it is determined that the comprehensive threat level exceeds the preset alarm trigger threshold, retrieve the emergency linkage instruction, combine the emergency linkage instruction and the occurrence range data, construct a formatted data packet, and generate digital rural public opinion early warning data.

9. The big data public opinion analysis and early warning method for digital rural governance according to claim 8, characterized in that, The specific process for setting the alarm trigger threshold is as follows: obtain historical records of similar events stored locally; calculate the arithmetic mean of the threat levels included in the historical records of similar events to generate a historical average threat level; extract the base of the actual total number of public opinion posts and the base of the normalized total number of posts within a predetermined historical time period; calculate the division ratio of the actual total number of public opinion posts and the base of the normalized total number of posts to generate a sensitive emotion fluctuation coefficient; multiply the historical average threat level and the sensitive emotion fluctuation coefficient; and set the numerical result obtained from the multiplication operation as the alarm trigger threshold.

10. A big data public opinion analysis and early warning system for digital rural governance, characterized in that: The system is used to implement the big data public opinion analysis and early warning method for digital rural governance as described in any one of claims 1-9, and the system includes: The policy semantic association module obtains the digital rural policy text, extracts the clause level number and converts it into a weight level code, obtains rural public opinion text statements, performs word segmentation and extraction on the rural public opinion text statements, generates policy vocabulary mapping records, constructs the mapping relationship between policy vocabulary mapping records and weight level codes, and generates statement code binding data. The hierarchical weighting module extracts multiple weight level codes from the statement code binding data, performs numerical comparison, selects the weight level code corresponding to the maximum value as the main weight value, calculates the difference to generate a weight gradient distribution sequence, and combines the continuity of clause hierarchical numbering to construct a text structure reconstruction sequence. The temporal fluctuation identification module constructs adjacent overlapping time windows according to a preset time step, extracts the occurrence frequency of public opinion words from the reconstructed text structure sequence and calculates the difference, generates a word frequency overlap difference sequence, performs trend analysis using a differential autoregressive moving average model, aggregates the position of symbol change nodes, and generates public opinion change time segments. The propagation trend judgment module extracts the propagation node time records included in the public opinion change time segment, generates a serialized time interval, calculates the offset segment ratio, compares it with the total number of network nodes, and generates an event diffusion intensity classification label. The hierarchical linkage early warning module extracts the event spread intensity classification labels, maps the risk classification values ​​and the occurrence range data of public opinion change time segments, performs risk assessment, and generates digital rural public opinion early warning data.

Images