Big data driven rural government-enterprise-citizen collaborative government office system and method

Abstract

The application discloses a big data driven rural government-enterprise-citizen collaborative government office system and method, and relates to the technical field of collaborative office. Related data about a to-be-processed problem uploaded by a villager terminal is acquired, and the related data about the to-be-processed problem is subjected to content recognition to calculate a directly locatable value of the to-be-processed problem. Whether the to-be-processed problem needs to be processed by a related department node is judged according to the directly locatable value and a preset threshold value. In this way, whether the problem uploaded by the villager has a clear object direction and processing path and whether the problem needs to be transferred by the related department node can be judged, so that some problems with clear descriptions can be directly matched with enterprises by the system and work orders can be initiated without the need of being audited by the related department node, response delay and resource waste are reduced, the processing delay of problems that can originally skip a link is reduced, and unnecessary load of the system is reduced.

Description

Technical Field

[0001] This invention relates to the field of collaborative office technology, specifically to a big data-driven rural government-enterprise-citizen collaborative government office system and method. Background Technology

[0002] With the advancement of the rural revitalization strategy, the rural governance system is gradually evolving from traditional manpower-driven approaches to digitalization and intelligentization. To break down information barriers between relevant departments, enterprises, and villagers, and to improve problem response speed and collaborative efficiency, a "Rural Government-Enterprise-Citizen Collaborative Government Office System" based on big data has emerged. This system integrates villager reporting, relevant department handling, and enterprise execution, achieving a closed-loop system for multi-party collaborative rural affairs processing. In practical application, villagers upload problems they encounter (such as infrastructure damage, environmental sanitation issues, etc.) via mobile devices. The system then transfers the problem to relevant department nodes, which identify and classify the problem, and then dispatch it to enterprises or third-party service providers with processing capabilities based on the judgment results, completing the closed-loop problem handling process.

[0003] However, this mechanism currently suffers from a significant problem: regardless of whether the issues uploaded by villagers have a clear target and processing path, all issues must be processed and judged by relevant department nodes. While this "unified pre-judgment by relevant departments" approach has a certain degree of standardization, it also introduces significant efficiency bottlenecks. This is especially true when dealing with standardized issues that are clearly described, located, and target-specific. Issues that could be directly matched with a company and have a work order initiated by the system without departmental review are instead included in redundant processes, leading to response delays, resource waste, increased processing time for issues that could have been skipped, and unnecessary system load. Summary of the Invention

[0004] The purpose of this invention is to solve the problems mentioned in the background art, which lead to response delays and resource waste, increase the delay in problem processing that could have been skipped, and cause unnecessary system load. Therefore, this invention proposes a big data-driven rural government-enterprise-citizen collaborative government office system and method.

[0005] A first aspect of this invention provides a big data-driven method for collaborative government-enterprise-citizen government office work in rural areas, the method comprising: Acquire relevant data about unresolved issues uploaded by villagers' terminals. The types of relevant data include image information, language information, video information, and text description information. Perform content recognition on the relevant data of the problem to be processed, and calculate the directly locatable value of the problem to be processed; Based on the directly locatable value and the preset threshold, determine whether the problem needs to be processed through relevant department nodes; If the problem to be handled no longer needs to be processed through relevant department nodes, the target enterprise is selected to handle the problem based on the problem handling records of each enterprise node. After the enterprise node completes the problem handling, it uploads the results to the platform, and the platform archives the results and provides feedback to the villager's terminal, thus realizing a closed loop in the problem handling process.

[0006] Optionally, the steps for calculating the directly locatable value of the problem to be processed are as follows: Content recognition is performed on the relevant data of the problem to be processed, and the multimodal content consistency value and the problem-oriented clarity value are calculated. The multimodal content consistency value and the problem-oriented clarity value are added together to obtain the directly locatable value of the problem to be processed.

[0007] Optionally, the steps for calculating the multimodal content consistency value are as follows: The image information, text description information, speech transcription information and video information of the problem to be processed are semantically parsed respectively. According to the arrangement order of semantic content in a unified semantic space, each modality is transformed into a sequence of semantic points ordered according to semantic relevance. Each semantic point is used to represent the semantic position of the corresponding modality in the unified semantic space, thereby forming image semantic path, text semantic path, speech semantic path and video semantic path respectively. Using the image semantic path as a reference path, the semantic nodes with the same sequence number as the image semantic path in the text semantic path, speech semantic path and video semantic path are compared, and the distance value between each non-image modality semantic node and the corresponding image semantic node is calculated. The text offset sequence, speech offset sequence and video offset sequence are formed according to the order of the semantic nodes. The adjacent differences of the text offset sequence, speech offset sequence, and video offset sequence are calculated to obtain the change sequence of each offset. The adjacent differences of the change sequence are calculated, and the resulting numerical sequence is used to characterize the offset curvature of the corresponding modality semantic path relative to the image semantic path. The offset curvature values ​​of each modality are accumulated to obtain the cumulative offset curvature value of the text modality, the cumulative offset curvature value of the speech modality, and the cumulative offset curvature value of the video modality. The cumulative offset curvature values ​​of text modality, voice modality, and video modality are summed to obtain the overall offset curvature value; the overall offset curvature value is divided by 3 times the preset maximum offset curvature to calculate the offset curvature normalization ratio; the final multimodal content consistency value is obtained by subtracting the calculated offset curvature normalization ratio from one.

[0008] Optionally, the steps for calculating the clarity value are as follows: Semantic encoding is performed on all semantic keywords extracted from image information, text description information, speech transcription information, and video information of the problem to be processed, and their coordinates in a unified semantic space are obtained. The geometric center position of all keywords is obtained, and the maximum distance from all keywords to the geometric center position is determined based on the geometric center. The maximum distance is used as the semantic anchor point aggregation radius. Identify location-related terms, descriptive geographic information, or location coordinates from various modal data and transform them into a set of spatial coordinate points; use the average position of all location coordinate points as the location reference center and calculate the distance from each location point to the reference center; average all distance values ​​as the geographic anchorage fluctuation range. Each semantic keyword is input into a pre-defined reverse matching model for responsible entities, and the set of responsible entities that can handle the problems represented by the keywords is obtained one by one; the number of keywords that correspond to only one responsible entity among all keywords is counted as the number of unique responsible entities, and the degree of overlap between multiple sets of responsible entities corresponding to different keywords is calculated, and the average value of all overlap degrees is used as the average cross density of the set of responsible entities; the ratio of the number of unique responsible entities to the average cross density is used as the convergence ratio of responsible entities. Multiply the semantic anchor aggregation radius by the geographical anchor fluctuation range, and use this product as the degree of non-concentration in the problem expression; then divide the degree of non-concentration by the convergence ratio of the responsible entity to obtain the ambiguity index of the problem expression; compare the ambiguity index with a constant one, and divide one by the sum of the ambiguity index and one to obtain the clarity value of the problem direction.

[0009] Optionally, the steps for determining whether a problem needs further processing through relevant department nodes based on directly locatable values ​​and preset thresholds are as follows: The platform compares the directly locatable value of the problem to be processed with a preset threshold. If the directly locatable value is not less than the preset threshold, it is determined that the problem to be processed does not need to go through the relevant department nodes for processing. The platform analyzes the problem type of the problem to be processed and selects the target enterprise node that best matches the problem based on the processing records of each enterprise node in the historical problem processing process, the processing success rate, and the service area range. The problem to be processed is then directly dispatched to the target enterprise node in the form of a work order for processing. When the directly locatable value is less than the preset threshold, the system will transfer the problem to the relevant department node for manual or semi-automatic identification, judgment and redistribution. The node will determine the final target enterprise node based on the problem attributes and complete the dispatch operation.

[0010] A second aspect of this invention provides a big data-driven rural government-enterprise-citizen collaborative government office system, the system comprising: Problem Data Module: Acquires relevant data about unresolved issues uploaded by villagers' terminals. The types of relevant data include image information, language information, video information, and text description information. Direct location module: performs content recognition on the relevant data of the problem to be processed and calculates the direct location value of the problem to be processed; Judgment module: Determines whether the problem to be processed needs to be further processed through relevant department nodes based on the directly locatable value and the preset threshold; Collaborative office module: If the problem to be handled no longer needs to be processed through relevant department nodes, the target enterprise can be selected to handle the problem based on the problem handling records of each enterprise node. After the enterprise node completes the problem handling, it will upload the handling result to the platform, and the platform will complete the result archiving and feedback to the villager terminal, thus realizing a closed loop in the problem handling process.

[0011] Optionally, the direct positioning includes: The direct location value module performs content recognition on the relevant data of the problem to be processed, calculates the multimodal content consistency value and the problem-oriented clarity value, and adds the multimodal content consistency value and the problem-oriented clarity value to obtain the direct location value of the problem to be processed.

[0012] Optionally, the direct positioning capability further includes: Semantic Path Module: Semantic parsing is performed on the image information, text description information, speech transcription information and video information of the problem to be processed. According to the arrangement order of semantic content in a unified semantic space, each modality is converted into a sequence of semantic points ordered according to semantic relevance. Each semantic point is used to represent the semantic position of the corresponding modality in the unified semantic space, thereby forming image semantic path, text semantic path, speech semantic path and video semantic path respectively. Offset module: Using the image semantic path as the reference path, it compares the semantic nodes with the same sequence number in the text semantic path, speech semantic path and video semantic path as the image semantic path, calculates the distance value between each non-image modality semantic node and the corresponding image semantic node, and forms the text offset sequence, speech offset sequence and video offset sequence according to the order of the semantic nodes. Offset curvature module: Calculates the adjacent differences for the text offset sequence, speech offset sequence, and video offset sequence to obtain the change sequence of each offset; calculates the adjacent differences for the change sequence, and the resulting numerical sequence is used to characterize the offset curvature of the corresponding modality semantic path relative to the image semantic path; and accumulates the offset curvature values ​​for each modality to obtain the cumulative offset curvature value for the text modality, the cumulative offset curvature value for the speech modality, and the cumulative offset curvature value for the video modality. Multimodal content consistency value module: sums the cumulative offset curvature values ​​of text modality, voice modality and video modality to obtain the overall offset curvature value; divides the overall offset curvature value by 3 times the preset maximum offset curvature to calculate the offset curvature normalization ratio; and subtracts the calculated offset curvature normalization ratio from one to obtain the final multimodal content consistency value.

[0013] Optionally, the direct positioning capability further includes: Aggregation radius module: Semantically encode all semantic keywords extracted from image information, text description information, speech transcription information and video information of the problem to be processed, and obtain their coordinates in a unified semantic space; calculate the geometric center position of all keyword coordinates, and determine the maximum distance from all keywords to the geometric center position based on the geometric center, and use the maximum distance as the semantic anchor point aggregation radius; The geographic anchoring fluctuation amplitude module identifies location-related terms, descriptive geographic information, or location coordinates from various modal data and converts them into a set of spatial coordinate points; it uses the average position of all location coordinate points as the location reference center and calculates the distance from each location point to the reference center; it then averages all distance values ​​to obtain the geographic anchoring fluctuation amplitude. The responsibility subject convergence ratio module: Input each semantic keyword into the preset responsibility subject back matching model, and obtain the set of responsibility subjects that can handle the problem represented by the keyword one by one; count the number of keywords that correspond to only a single responsibility subject among all keywords as the number of unique responsibility subjects, and calculate the degree of overlap between multiple responsibility subject sets corresponding to different keywords, and take the average of all overlap degrees as the average cross density of the responsibility subject sets; use the ratio of the number of unique responsibility subjects to the average cross density as the responsibility subject convergence ratio; The Problem Clarity Module: Multiply the semantic anchor aggregation radius by the geographical anchor fluctuation range, and use this product as the degree of non-concentration in the problem expression; then divide the degree of non-concentration by the convergence ratio of the responsible entity to obtain the ambiguity index of the problem expression; compare the ambiguity index with a constant 1, and divide 1 by the sum of the ambiguity index and 1 to obtain the problem clarity value.

[0014] Optionally, the determination module includes: The first office module compares the directly locatable value of the problem to be processed with a preset threshold. If the directly locatable value is not less than the preset threshold, it is determined that the problem to be processed does not need to go through the relevant department nodes for further processing. The platform analyzes the problem type of the problem to be processed and selects the target enterprise node that best matches the problem based on the processing records of each enterprise node in the historical problem processing process, the processing success rate, and the service area range. The problem to be processed is then directly dispatched to the target enterprise node in the form of a work order for processing. The second office module: When the directly locatable value is less than the preset threshold, the system will transfer the problem to the relevant department node for manual or semi-automatic identification, judgment and redistribution. The node will determine the final target enterprise node based on the problem attributes and complete the dispatch operation.

[0015] The beneficial effects of this invention are: This invention proposes a big data-driven rural government-enterprise-citizen collaborative government office system and method. It acquires relevant data about pending issues uploaded by villagers' terminals, performs content recognition on this data, and calculates the direct location value of the pending issues. Based on the direct location value and a preset threshold, it determines whether the pending issues need further processing through relevant department nodes. If the pending issues do not require further processing through relevant department nodes, the system selects a target enterprise to process the pending issues based on the issue processing records of each enterprise node. After the enterprise node completes the issue processing, it uploads the results to the platform, which then archives the results and provides feedback to the villagers' terminals, thus achieving a closed-loop issue processing process. This allows the system to analyze whether the issues uploaded by villagers have a clear target and processing path, and whether they need to be processed through relevant department nodes. For clearly described issues, the system can directly match enterprises and initiate work orders without requiring review by relevant department nodes, reducing response delays and resource waste, minimizing processing time that could otherwise be skipped, and reducing unnecessary system load. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a big data-driven collaborative government-enterprise-citizen office method for rural government affairs provided in an embodiment of the present invention; Figure 2 This is a framework diagram of a big data-driven rural government-enterprise-citizen collaborative government office system provided in an embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention in order to achieve the intended purpose, the following detailed description is provided in conjunction with the accompanying drawings and preferred embodiments, based on the specific implementation methods, structures, features and effects of the present invention.

[0018] The embodiments of the invention provide a big data-driven method for collaborative government-enterprise-citizen government affairs in rural areas. See also... Figure 1 , Figure 1 A flowchart illustrating a big data-driven collaborative government-enterprise-citizen approach to government affairs in rural areas, provided in an embodiment of the present invention. The method includes the following steps: S1: Obtain relevant data about the problem to be processed uploaded by the villager's terminal. The types of relevant data include at least image information, language information, video information, and text description information. S2: Perform content recognition on the relevant data of the problem to be processed, and calculate the directly locatable value of the problem to be processed; S3: Determine whether the problem to be processed needs to be processed through relevant department nodes based on the directly locatable value and the preset threshold; S4: If the problem to be handled no longer needs to be processed through relevant department nodes, select the target enterprise to handle the problem based on the problem handling records of each enterprise node. After the enterprise node completes the problem handling, it uploads the handling result to the platform, and the platform completes the result archiving and feedback to the villager terminal, thus realizing the closed loop of the problem handling process.

[0019] Based on the big data-driven rural government-enterprise-citizen collaborative government office method provided by the embodiments of the present invention, the above-mentioned method can analyze whether the problems uploaded by villagers have a clear target and processing path, and whether they need to be transferred and judged by relevant department nodes. This allows the system to directly match enterprises and initiate work orders for some clearly described problems without the need for review by relevant department nodes, thereby reducing response delays and resource waste, reducing the delay in problem processing that could have been skipped, and reducing unnecessary system load.

[0020] In one embodiment, S1: Obtain relevant data about the problem to be processed uploaded by the villager's terminal, and the types of the relevant data include at least image information, text description information, and location information; It's important to note that the system first acquires raw data about the problems actively uploaded by villagers from their terminals. This data encompasses multiple modalities, including image information, audio information, video information, text descriptions, and location information, comprehensively reflecting the actual situation of the problems encountered by villagers. Image information typically comes from on-site photos, visually demonstrating the problem itself (e.g., damaged road surface, leaking water pipes, broken wires). Audio information includes voice descriptions input by villagers, which the system converts into structured text using speech recognition technology, suitable for elderly users or those who have difficulty typing. Video information provides dynamic scene awareness, such as videos of flashing streetlights at night taken by villagers, helping to identify detailed features in complex scenes. Text descriptions are text directly entered by villagers, providing semantic clues such as the location, time, and changes of the problem. Location information is usually obtained automatically from GPS coordinates via the mobile phone's positioning system, or manually entered by the villager as a spatial anchor point for problem location. This multimodal data can be uploaded as a single modality or a combination of modalities during collection. The system will uniformly perform data fusion processing to recreate the realistic context of the problem scene as accurately as possible. For example, a villager notices that the streetlights at the village entrance are off. They upload an image of the streetlight taken in the dark to their terminal, along with a voice description: "It goes out here every night at 7 pm." The system automatically records the precise location coordinates using GPS. In this case, the problem data includes four modalities: image, voice, text, and location. This helps the system accurately determine the problem's attributes and possible solutions, laying the information foundation for calculating the "directly locatable value." In one embodiment, S2: Perform content recognition on the relevant data of the problem to be processed, and calculate the directly locatable value of the problem to be processed; In one implementation, the steps for calculating the directly locatable value of the problem to be processed are as follows: Content recognition is performed on the relevant data of the problem to be processed, and the multimodal content consistency value and the problem-oriented clarity value are calculated. The multimodal content consistency value and the problem-oriented clarity value are added together to obtain the directly locatable value of the problem to be processed.

[0021] In one implementation, the steps for calculating the multimodal content consistency value are as follows: The image information, text description information, speech transcription information and video information of the problem to be processed are semantically parsed respectively. According to the arrangement order of semantic content in a unified semantic space, each modality is transformed into a sequence of semantic points ordered according to semantic relevance. Each semantic point is used to represent the semantic position of the corresponding modality in the unified semantic space, thereby forming image semantic path, text semantic path, speech semantic path and video semantic path respectively. Using the image semantic path as a reference path, the semantic nodes with the same sequence number as the image semantic path in the text semantic path, speech semantic path and video semantic path are compared, and the distance value between each non-image modality semantic node and the corresponding image semantic node is calculated. The text offset sequence, speech offset sequence and video offset sequence are formed according to the order of the semantic nodes. The adjacent differences of the text offset sequence, speech offset sequence, and video offset sequence are calculated to obtain the change sequence of each offset. The adjacent differences of the change sequence are calculated to characterize the degree of change of the direction of the offset with the sequence number. The obtained numerical sequence is used to characterize the offset curvature of the corresponding modality semantic path relative to the image semantic path. The offset curvature values ​​of each modality are accumulated to obtain the cumulative offset curvature value of the text modality, the cumulative offset curvature value of the speech modality, and the cumulative offset curvature value of the video modality.

[0022] The cumulative offset curvature values ​​of text, speech, and video modalities are summed to obtain the overall offset curvature value. This overall offset curvature value is then divided by three times the preset maximum offset curvature to calculate the offset curvature normalization ratio. Finally, the multimodal content consistency value is obtained by subtracting this normalization ratio from one. It should be noted that in the calculation of the multimodal content consistency value mentioned above, all the calculation data involved comes from the original multimodal problem information uploaded by villagers' terminals. This data is obtained through standardized processing using the system's preset semantic parsing model, modality mapping rules, and spatial vector encoding method, ensuring that all modal data can be uniformly mapped to the same semantic space. Specifically, the image information required in the first step is obtained by villagers taking and uploading pictures of the problem at the scene. The system uses image object detection models (such as YOLO, Mask R-CNN, etc.) to identify key entities appearing in the pictures and converts the recognition results into several image semantic label nodes. The text description information is generated by villagers manually inputting or transcribing speech. The system uses text keyword extraction and semantic word segmentation algorithms (such as TF-IDF, BERT embedding) to extract keywords representing the problem and sorts them to form a text semantic path. The speech-transcribed information is converted into structured text by a speech recognition module (such as the ASR model), and semantic nodes are extracted using the same method as for the text information. The video information comes from video files uploaded by villagers. The system extracts keyframes from the video and applies the same recognition algorithm as for the images to extract semantic target sequences. All the above modalities map semantic nodes to a unified high-dimensional semantic space through a unified semantic encoder, forming image semantic paths, text semantic paths, speech semantic paths, and video semantic paths. The semantic offset calculation data needed in the second step, i.e., the spatial distance between each non-image modal semantic node and the image semantic node with the same index, is obtained by calculating the Euclidean distance between their coordinate points in the unified semantic space. The input data is the coordinate information of each semantic node in space, directly obtained from the semantic encoding in the previous step. For example, the first node of the image modality is "street lamp," with coordinates (0.2, 0.7, 0.1) in the semantic space, and the first node of the text modality is "the lamp is broken," with coordinates (0.21, 0.72, 0.09). The distance between them is the Euclidean distance, representing their semantic offset degree. The data used to calculate the change sequence in the third step is the offset sequence obtained in the previous step. The offset change sequence is obtained by subtracting its adjacent elements, and then the offset curvature change value is obtained by subtracting the adjacent elements in the offset change sequence. All data are numerical differences derived from the obtained offset sequence. In the fourth step, the curvature change values ​​of each modality are summed to form a cumulative value, which is a summation of the obtained data and does not involve external input. The "preset maximum offset curvature" value used in the fifth step is a constant set by system engineering experience, for example, a value of 10. It is a built-in standard parameter of the system and does not depend on external data. All the original data required for the calculation of the final multimodal content consistency value comes from the active upload of the villagers. It is uniformly converted into a structured semantic path by steps such as image recognition, speech recognition, text parsing and video processing. The whole process does not depend on uncontrollable external data and has complete feasibility.For example, a villager uploads a picture of a "broken manhole cover," a video of someone saying "The cover here is broken, it's dangerous at night," and a short text "The manhole cover at the village entrance is broken." The system extracts semantic tags such as "manhole cover," "broken," "night," and "village entrance." After the above steps, a consistency value is finally formed. If the result is 0.88, it means that the information consistency is very high, and it can directly enter the enterprise's order dispatch process.

[0023] It's important to note that the multimodal content consistency value is a quantitative indicator used to measure whether the various information modalities (including images, text, audio, and video) uploaded by villagers regarding the issue to be addressed maintain a high degree of semantic consistency when expressing the same problem. A higher value indicates smaller differences in the expression goals, descriptive content, semantic theme, and positioning direction among the different modalities; in other words, the multimodal information tends to point to the same clear core problem. The core value of this value lies in reflecting the system's difficulty in understanding the problem information: when the semantic content conveyed by all modalities tends to be consistent, unambiguous, and free from mutual interference, the system can accurately identify the problem object, type, and attribution without complex logical reasoning or manual judgment. This allows for direct matching of the work order to the target enterprise node responsible for that type of problem, bypassing the relevant department's review or attribution judgment process, saving time and reducing system load. Conversely, when there are conflicts or deviations in the modal expressions (for example, an image shows a street lamp, but the text describes it as a drainage problem, and the voice mentions "slippery at night"), the system will face high offset and high uncertainty during semantic aggregation. This makes it impossible to determine which category the problem belongs to or to accurately match the processing subject. In this case, relevant department nodes need to intervene to identify and judge to avoid misassignment or omission, thereby ensuring processing accuracy. For example, if a villager uploads a photo of a fallen street lamp, with the text description "the street lamp was blown down by the wind last night," the voice transcription being "at the village entrance, it was dark all night," and the video clearly showing the broken lamp post, then the modal information revolves entirely around the problem of "the street lamp at the village entrance fallen," without ambiguity or information conflict. The system can quickly identify it as a "lighting facility malfunction," automatically associate it with "public lighting maintenance" enterprise resources, and initiate a work order without entering the relevant department's judgment process, achieving rapid closure. Because the higher the consistency of multimodal information, the shorter the reasoning chain required for system understanding, the more certain the semantic merging, and the clearer the problem attribution, the more feasible it is to directly assign problems with higher multimodal content consistency values ​​to enterprises. This makes it more suitable to skip relevant departmental nodes, thereby saving a lot of manual judgment and order assignment time and achieving efficient collaboration between government, enterprises, and the public.

[0024] The core advantage of calculating multimodal content consistency values ​​using the above method lies in its approach. Unlike conventional methods that simply compare semantic labels for similarity or vector similarity, it dynamically captures the entire process of how each modality gradually approaches or deviates from a reference modality (such as an image) within the semantic space by constructing continuous semantic offset trajectories between multimodal semantic paths. This allows for a more accurate and comprehensive measurement of the consistency of content expressed across different modalities. This method not only considers the static similarity of "every point" between modalities but also focuses on the "direction and fluctuation of overall semantic change." It reflects the consistency trend and the severity of offset in the modal expression process through the second-order difference of adjacent offset change sequences, resulting in higher discrimination accuracy and stronger anti-interference capabilities. When the multimodal content consistency value is high, it indicates that the information from different modalities has reached a high degree of consensus on expressing the same issue. For example, an image shows a broken manhole cover, the text description is "The manhole cover at the village entrance is broken," the voice says "The cover is cracked and it's dangerous," and the video clearly records the details of the crack. Although these modalities come from different input methods, their semantic trajectories almost overlap or show a stable convergence in a unified space. The system can accurately understand the core of the problem and its responsibility without additional reasoning, and can directly assign tasks to the corresponding enterprises without confirmation from relevant departments. In contrast, conventional methods that rely solely on keyword overlap or similarity calculations often struggle to distinguish between semantically consistent scenarios with different expression styles, or are easily misled by local interference information, leading to system misjudgments or uncertainties. This necessitates manual intervention from relevant departments, prolonging the processing flow. Therefore, the above calculation method, by constructing cross-modal semantic paths, calculating the curvature of offset trajectories, and normalizing to reflect the consistency level, characterizes the degree of expression consistency from the perspective of data structure changes. It is a more comprehensive, dynamic, and reliable multimodal consistency calculation mechanism that can accurately delegate authority to enterprises when information is clearly expressed, improving response speed and resource scheduling efficiency.

[0025] In one implementation, the problem refers to the calculation steps for the clarity value as follows: All semantic keywords extracted from the image information, text description information, speech transcription information, and video information of the problem to be processed are semantically encoded to obtain their coordinates in a unified semantic space. The geometric center position of all keywords is obtained, and the maximum distance from all keywords to the geometric center position is determined based on the geometric center. The maximum distance is used as the semantic anchor point aggregation radius to characterize whether the semantics of all modal expressions are concentrated on the same problem focus.

[0026] Identify location-related vocabulary, descriptive geographic information, or location coordinates from various modal data and transform them into a set of spatial coordinate points. Use the average position of all location coordinate points as the location reference center and calculate the distance from each location point to the reference center. Take the average of all distance values ​​as the geographic anchoring fluctuation amplitude to characterize the degree of concentration of uploaded issues in spatial representation.

[0027] Each semantic keyword is input into a pre-defined reverse matching model for responsible entities, and a set of responsible entities capable of handling the problem represented by the keyword is obtained one by one. The number of keywords that correspond to only one responsible entity among all keywords is counted as the number of unique responsible entities, and the degree of overlap between multiple sets of responsible entities corresponding to different keywords is calculated. The average value of all overlap degrees is used as the average cross density of the set of responsible entities. The ratio of the number of unique responsible entities to the average cross density is used as the convergence ratio of responsible entities, which is used to characterize whether the attribution of responsibility for the problem is clear.

[0028] Multiply the semantic anchor aggregation radius by the geographical anchor fluctuation range, and use this product as the non-concentration degree of the problem expression; then divide the non-concentration degree by the aforementioned convergence ratio of the responsible parties to obtain the ambiguity index of the problem expression; compare the ambiguity index with a constant one, and divide one by the sum of the ambiguity index and one to obtain the final problem orientation clarity value. The value of the problem orientation clarity value ranges from zero to one. The closer the value is to one, the clearer the object orientation of the problem is, and the closer it is to zero, the less clear the object orientation of the problem is.

[0029] It should be noted that in the calculation of the clarity value of the above-mentioned problem, the data required for each calculation comes directly from the multimodal problem information uploaded by villagers through the terminal, including images, text descriptions, audio and video files. The system performs structured processing on this data through the built-in semantic parsing module. First, in the calculation of the "semantic anchor point aggregation radius", the system identifies target object labels (such as streetlights, utility poles, etc.) from the image content, extracts keyword phrases (such as "streetlights are not working" and "manhole covers are broken") from the text, converts the speech into text through a speech recognition model and then extracts keywords, extracts keyframes from the video and executes object detection algorithms to obtain semantic entity labels. All keywords extracted from the modalities are mapped to a unified semantic space through a unified semantic encoder and their coordinates are obtained, which are used to calculate the geometric center and maximum radius of the semantic anchor point. For example, if a villager uploads a photo of a streetlight that has fallen over, with the caption "blown over by the wind last night", the voice says "the streetlight at the village entrance collapsed", and the video shows a construction site, semantic labels such as "streetlight", "collapsed", and "village entrance" are extracted from these modalities, encoded to form a coordinate set, and the spatial aggregation degree is calculated. Secondly, in the calculation of "geographic anchoring fluctuation range", the system extracts location words (such as "west entrance of the village" or "next to the electrical box") appearing in each modality or calls the GPS coordinate data attached when the terminal is uploaded. If the place name is not explicitly stated in the text description, the system attempts to infer the location point through buildings, directional words, etc. appearing in the text and convert them into spatial coordinates. After all coordinates are combined into a set, the average center is calculated and the average distance to the center is calculated. For example, if the text mentions "next to the convenience store", the voice says "near the village entrance", and the picture is located at GPS coordinates (X1, Y1), the system will uniformly parse this information into a coordinate set for geographic aggregation judgment. Secondly, in calculating the "convergence ratio of responsible entities," the system inputs all the aforementioned semantic keywords one by one into the built-in responsibility matching model. This model determines which enterprises or departments typically handle each keyword based on a pre-established knowledge graph or rule base. If a keyword matches only one entity, it is recorded as a unique responsibility. If the set of responsible entities corresponding to multiple keywords has a high degree of overlap, it indicates good convergence. For example, the keyword "streetlight" only matches "lighting maintenance company A," which is a unique responsibility. "Electric wire" and "overturned" match "power company" and "emergency management" respectively, and their intersection is empty, indicating ambiguous responsibility. Finally, in the comprehensive step, all intermediate values ​​are calculated by the system based on the above structured data, without manual input or reliance on uncontrollable external information, ensuring that the calculation process is stable, reliable, and repeatable. This fully reflects the comprehensive clarity of the villagers' problems in terms of "whether the semantic expression is focused," "whether the location is concentrated," and "whether the responsibility is clear."

[0030] It should be noted that the "problem clarity value" is a composite indicator used to measure whether the problems uploaded by villagers are clear and determinate in three dimensions: semantic expression, spatial location, and responsibility attribution. The higher the value, the clearer the object, location, and attribution of the problem. The reason why a larger value indicates that the corresponding enterprise can be directly found for work order assignment without going through relevant department nodes is that when a problem is highly focused at the semantic level (i.e., the semantic anchor aggregation radius is small), highly concentrated in terms of location expression (i.e., the geographical anchor fluctuation range is small), and the set of responsible parties involved is clear and has good convergence (i.e., multiple keywords point to the same or highly overlapping responsible party), then the system can achieve a closed loop of the entire process from semantic recognition and location positioning to processing path matching without relying on manual judgment, and has sufficient conditions for automatic work order assignment. For example, if a villager uploads a photo clearly showing a "broken manhole cover at the village entrance," with the caption "The manhole cover at the village entrance is broken and dangerous," and the voice message adding "Someone almost fell in last night," the system recognizes keywords such as "manhole cover," "broken," and "village entrance." These keywords have a high degree of semantic aggregation, their spatial coordinates are concentrated in the same area, and by matching keywords with a historical responsibility database, it finds that 90% of such issues are the responsibility of the "village infrastructure maintenance company." Therefore, the system can directly match the issue with the company, eliminating the need for relevant departments to first determine attribution or intervene. This significantly improves response efficiency, saves human resources for relevant departments, and shortens the processing path. Thus, the clarity of the problem's location is essentially a measure of whether the problem can be automatically identified in terms of responsibility and path determination. A higher value indicates that the problem requires less intervention and is more suitable for direct closed-loop processing.

[0031] The advantage of calculating the problem-directed value in the above way is that it comprehensively considers the concentration of semantic expression, the stability of spatial positioning, and the convergence of responsibility attribution. It adopts a progressive structured calculation logic, which can comprehensively characterize the "decidability" of the problem from three of the most critical and quantifiable dimensions. Compared with traditional methods, it is more detailed and discriminative, and does not rely on any fuzzy reasoning or human intervention, making it suitable for big data-driven automated dispatching systems. This method quantifies the focus of a problem's expression in the semantic space by constructing a semantic anchor aggregation radius, avoiding misjudgments caused by fuzzy keyword matching. It reflects the clarity of the problem's location through geographical anchor fluctuations, enabling the system to accurately pinpoint the actual occurrence point and improving spatial matching efficiency. Furthermore, it measures the clarity of the processing unit pointed to by the keyword through the responsibility subject convergence ratio, effectively eliminating system interruptions or dispatch delays caused by unclear multiple attributions. The combined computational structure is logically rigorous, and all input data can be automatically extracted directly from user-uploaded information, avoiding reliance on expert annotation or complex model training. Therefore, this method is not only suitable for intelligent decision-making systems in complex, multi-source, heterogeneous data environments but also possesses engineering feasibility and interpretability. It is an efficient and reliable way to achieve rapid problem assignment and automatic decision-making on processing paths in government-enterprise-citizen collaborative scenarios.

[0032] In one embodiment, S3: The step of determining whether the problem to be processed needs to be further processed through relevant department nodes based on the directly locatable value and the preset threshold is as follows: The platform compares the directly locatable value of the problem to be processed with a preset threshold. If the directly locatable value is not less than the preset threshold, it is determined that the problem does not need to be processed by relevant department nodes. The platform analyzes the problem type and selects the target enterprise node that best matches the problem based on the types, success rates, and service areas of each enterprise node in the historical problem processing process. The problem is then directly dispatched to the target enterprise node in the form of a work order for processing. After the enterprise node completes the actual processing of the problem, it uploads the problem processing results and related feedback information to the platform through the system. The platform archives the problem processing data and feeds back the archived processing results to the corresponding villager terminal, realizing an automatic closed loop in the problem processing process. When the directly locatable value is less than the preset threshold, the system will transfer the problem to the relevant department node for manual or semi-automatic identification, judgment and redistribution. The node will determine the final target enterprise node based on the problem attributes and complete the dispatch operation.

[0033] It should be noted that the directly locatable value corresponding to the problem to be processed is compared with a threshold preset by the platform. This threshold can be set based on the system's historical experience or strategy, for example, 0.75, to distinguish whether the problem meets the conditions for direct assignment. When the directly locatable value is greater than or equal to this threshold, it means that the problem uploaded by the villager is clear and specific enough in terms of semantic expression, spatial location, and responsibility attribution. The system determines that it does not need further manual judgment or classification by relevant departments. At this time, the platform will automatically identify the problem type, including keyword matching, semantic vector analysis, and image recognition result classification, such as identifying standard problem types such as "damaged lighting facilities" and "sewage pipe blockage". Subsequently, the system matches the historical problem handling records of each enterprise node according to the problem type, including but not limited to whether the enterprise has handled similar problems, whether the success rate is high, and whether it covers the geographical area of ​​the current village. For example, if an enterprise has successfully handled "damaged street lights" problems many times in the past and covers the area where the village is located, it will be selected as the target enterprise node first. After successful matching, The system automatically generates work orders and pushes information such as the problem content, location, and estimated processing time to the target enterprise node. The enterprise completes the problem handling according to the standard processing procedure. After the handling is completed, the enterprise needs to upload a processing report, before and after photos, or other feedback information through the platform. The platform archives the information and simultaneously sends the processing conclusion and pictures back to the original villager's terminal, indicating that the problem has been resolved, forming a closed-loop processing process. When the directly locatable value is lower than the threshold, the system considers the problem content to be unclear or the responsibility to be unclear, such as vague descriptions, unclear images, semantic conflicts, or large geographical deviations. At this time, the system automatically submits the problem to the relevant department node. The staff of that node or the auxiliary AI system manually or semi-automatically judges and classifies the problem, identifies the specific type of problem and which type of enterprise should be responsible, and finally the relevant department node determines the target enterprise node and dispatches a work order, ensuring that even with ambiguous information, the problem can be effectively identified and handled. For example, if a problem is described as "something is broken at the village entrance" and the image is blurry and the object cannot be identified, the platform will transfer it to the relevant department node for handling.

[0034] Based on the same inventive concept, embodiments of the present invention also provide a big data-driven rural government-enterprise-citizen collaborative government office system. See also Figure 2 , Figure 2 A framework diagram of a big data-driven rural government-enterprise-citizen collaborative government office system provided in this embodiment of the invention includes: Problem Data Module: Acquires relevant data about unresolved issues uploaded by villagers' terminals. The types of relevant data include image information, language information, video information, and text description information. Direct location module: performs content recognition on the relevant data of the problem to be processed and calculates the direct location value of the problem to be processed; Judgment module: Determines whether the problem to be processed needs to be further processed through relevant department nodes based on the directly locatable value and the preset threshold; Collaborative office module: If the problem to be handled no longer needs to be processed through relevant department nodes, the target enterprise can be selected to handle the problem based on the problem handling records of each enterprise node. After the enterprise node completes the problem handling, it will upload the handling result to the platform, and the platform will complete the result archiving and feedback to the villager terminal, thus realizing a closed loop in the problem handling process.

[0035] Based on the big data-driven rural government-enterprise-citizen collaborative government office system provided by the embodiments of the present invention, the above-mentioned method can analyze whether the problems uploaded by villagers have a clear target and processing path, and whether they need to be transferred and judged by relevant department nodes. This allows the system to directly match enterprises and initiate work orders for some clearly described problems without going through the review of relevant department nodes, reducing response delays and resource waste, reducing the delay in problem processing that could have been skipped, and reducing unnecessary system load.

[0036] In one embodiment, the direct positioning includes: The direct location value module performs content recognition on the relevant data of the problem to be processed, calculates the multimodal content consistency value and the problem-oriented clarity value, and adds the multimodal content consistency value and the problem-oriented clarity value to obtain the direct location value of the problem to be processed.

[0037] In one embodiment, the direct positioning capability further includes: Semantic Path Module: Semantic parsing is performed on the image information, text description information, speech transcription information and video information of the problem to be processed. According to the arrangement order of semantic content in a unified semantic space, each modality is converted into a sequence of semantic points ordered according to semantic relevance. Each semantic point is used to represent the semantic position of the corresponding modality in the unified semantic space, thereby forming image semantic path, text semantic path, speech semantic path and video semantic path respectively. Offset module: Using the image semantic path as the reference path, it compares the semantic nodes with the same sequence number in the text semantic path, speech semantic path and video semantic path as the image semantic path, calculates the distance value between each non-image modality semantic node and the corresponding image semantic node, and forms the text offset sequence, speech offset sequence and video offset sequence according to the order of the semantic nodes. Offset curvature module: Calculates the adjacent differences for the text offset sequence, speech offset sequence, and video offset sequence to obtain the change sequence of each offset; calculates the adjacent differences for the change sequence, and the resulting numerical sequence is used to characterize the offset curvature of the corresponding modality semantic path relative to the image semantic path; and accumulates the offset curvature values ​​for each modality to obtain the cumulative offset curvature value for the text modality, the cumulative offset curvature value for the speech modality, and the cumulative offset curvature value for the video modality. Multimodal content consistency value module: sums the cumulative offset curvature values ​​of text modality, voice modality and video modality to obtain the overall offset curvature value; divides the overall offset curvature value by 3 times the preset maximum offset curvature to calculate the offset curvature normalization ratio; and subtracts the calculated offset curvature normalization ratio from one to obtain the final multimodal content consistency value.

[0038] In one embodiment, the direct positioning capability further includes: Aggregation radius module: Semantically encode all semantic keywords extracted from image information, text description information, speech transcription information and video information of the problem to be processed, and obtain their coordinates in a unified semantic space; calculate the geometric center position of all keyword coordinates, and determine the maximum distance from all keywords to the geometric center position based on the geometric center, and use the maximum distance as the semantic anchor point aggregation radius; The geographic anchoring fluctuation amplitude module identifies location-related terms, descriptive geographic information, or location coordinates from various modal data and converts them into a set of spatial coordinate points; it uses the average position of all location coordinate points as the location reference center and calculates the distance from each location point to the reference center; it then averages all distance values ​​to obtain the geographic anchoring fluctuation amplitude. The responsibility subject convergence ratio module: Input each semantic keyword into the preset responsibility subject back matching model, and obtain the set of responsibility subjects that can handle the problem represented by the keyword one by one; count the number of keywords that correspond to only a single responsibility subject among all keywords as the number of unique responsibility subjects, and calculate the degree of overlap between multiple responsibility subject sets corresponding to different keywords, and take the average of all overlap degrees as the average cross density of the responsibility subject sets; use the ratio of the number of unique responsibility subjects to the average cross density as the responsibility subject convergence ratio; The Problem Clarity Module: Multiply the semantic anchor aggregation radius by the geographical anchor fluctuation range, and use this product as the degree of non-concentration in the problem expression; then divide the degree of non-concentration by the convergence ratio of the responsible entity to obtain the ambiguity index of the problem expression; compare the ambiguity index with a constant 1, and divide 1 by the sum of the ambiguity index and 1 to obtain the problem clarity value.

[0039] In one embodiment, the determination module includes: The first office module compares the directly locatable value of the problem to be processed with a preset threshold. If the directly locatable value is not less than the preset threshold, it is determined that the problem to be processed does not need to go through the relevant department nodes for further processing. The platform analyzes the problem type of the problem to be processed and selects the target enterprise node that best matches the problem based on the processing records of each enterprise node in the historical problem processing process, the processing success rate, and the service area range. The problem to be processed is then directly dispatched to the target enterprise node in the form of a work order for processing. The second office module: When the directly locatable value is less than the preset threshold, the system will transfer the problem to the relevant department node for manual or semi-automatic identification, judgment and redistribution. The node will determine the final target enterprise node based on the problem attributes and complete the task dispatch operation. The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention should still fall within the scope of the claims of the present invention.

Claims

1. A big data-driven collaborative government-enterprise-citizen approach to rural government office work, characterized by: Includes the following steps: Acquire relevant data about unresolved issues uploaded by villagers' terminals. The types of relevant data include image information, language information, video information, and text description information. Perform content recognition on the relevant data of the problem to be processed, and calculate the directly locatable value of the problem to be processed; Based on the directly locatable value and the preset threshold, determine whether the problem needs to be processed through relevant department nodes; If the problem to be handled no longer needs to be processed through relevant department nodes, the target enterprise is selected to handle the problem based on the problem handling records of each enterprise node. After the enterprise node completes the problem handling, it uploads the results to the platform, and the platform archives the results and provides feedback to the villager's terminal, thus realizing a closed loop in the problem handling process.

2. The big data-driven rural government-enterprise-citizen collaborative government office method according to claim 1, characterized in that, The steps for calculating the directly locatable value of the problem to be processed are as follows: Content recognition is performed on the relevant data of the problem to be processed, and the multimodal content consistency value and the problem-oriented clarity value are calculated. The multimodal content consistency value and the problem-oriented clarity value are added together to obtain the directly locatable value of the problem to be processed.

3. The big data-driven rural government-enterprise-citizen collaborative government office method according to claim 1, characterized in that, The steps for calculating the multimodal content consistency value are as follows: The image information, text description information, speech transcription information and video information of the problem to be processed are semantically parsed separately. According to the arrangement order of semantic content in a unified semantic space, each modality is transformed into a sequence of semantic points ordered according to semantic relevance. Each semantic point is used to represent the semantic position of the corresponding modality in the unified semantic space, thereby forming image semantic path, text semantic path, speech semantic path and video semantic path respectively. Using the image semantic path as a reference path, the semantic nodes with the same sequence number as the image semantic path in the text semantic path, speech semantic path and video semantic path are compared, and the distance value between each non-image modality semantic node and the corresponding image semantic node is calculated. The text offset sequence, speech offset sequence and video offset sequence are formed according to the order of the semantic nodes. The adjacent differences of the text offset sequence, speech offset sequence and video offset sequence are calculated to obtain the change sequence of each offset; the adjacent differences of the change sequence are calculated and the resulting numerical sequence is used to characterize the offset curvature of the corresponding modal semantic path relative to the image semantic path. The offset curvature values ​​of each modality are then summed to obtain the cumulative offset curvature values ​​for the text modality, the speech modality, and the video modality. The cumulative offset curvature values ​​of the text modality, speech modality, and video modality are summed to obtain the overall offset curvature value; The offset curvature normalization ratio is calculated by dividing the total offset curvature value by 3 times the preset maximum offset curvature. The final multimodal content consistency value is obtained by subtracting the normalization ratio of the calculated offset curvature from one.

4. The big data-driven rural government-enterprise-citizen collaborative government office method according to claim 1, characterized in that, The problem refers to the following steps for calculating the sharpness value: Semantic encoding is performed on all semantic keywords extracted from image information, text description information, speech transcription information, and video information of the problem to be processed, and their coordinates in a unified semantic space are obtained. The geometric center position of all keywords is obtained, and the maximum distance from all keywords to the geometric center position is determined based on the geometric center. The maximum distance is used as the semantic anchor point aggregation radius. Identify location-related terms, descriptive geographic information, or location coordinates from various modal data and transform them into a set of spatial coordinate points; use the average position of all location coordinate points as the location reference center and calculate the distance from each location point to the reference center; average all distance values ​​as the geographic anchorage fluctuation range. Each semantic keyword is input into a pre-defined reverse matching model for responsible entities, and the set of responsible entities that can handle the problems represented by the keywords is obtained one by one; the number of keywords that correspond to only one responsible entity among all keywords is counted as the number of unique responsible entities, and the degree of overlap between multiple sets of responsible entities corresponding to different keywords is calculated, and the average value of all overlap degrees is used as the average cross density of the set of responsible entities; the ratio of the number of unique responsible entities to the average cross density is used as the convergence ratio of responsible entities. Multiply the semantic anchor aggregation radius by the geographical anchor fluctuation range, and use this product as the degree of non-concentration in the problem expression; then divide the degree of non-concentration by the convergence ratio of the responsible entity to obtain the ambiguity index of the problem expression; compare the ambiguity index with a constant one, and divide one by the sum of the ambiguity index and one to obtain the clarity value of the problem direction.

5. The big data-driven rural government-enterprise-citizen collaborative government office method according to claim 1, characterized in that, The steps to determine whether a problem needs further processing through relevant department nodes based on directly locatable values ​​and preset thresholds are as follows: The platform compares the directly locatable value of the problem to be processed with a preset threshold. If the directly locatable value is not less than the preset threshold, it is determined that the problem to be processed does not need to go through the relevant department nodes for processing. The platform analyzes the problem type of the problem to be processed and selects the target enterprise node that best matches the problem based on the processing records of each enterprise node in the historical problem processing process, the processing success rate, and the service area range. The problem to be processed is then directly dispatched to the target enterprise node in the form of a work order for processing. When the directly locatable value is less than the preset threshold, the system will transfer the problem to the relevant department node for manual or semi-automatic identification, judgment and redistribution. The node will determine the final target enterprise node based on the problem attributes and complete the dispatch operation.

6. A big data-driven rural government-enterprise-citizen collaborative government office system, characterized in that: The system includes: Problem Data Module: Acquires relevant data about unresolved issues uploaded by villagers' terminals. The types of relevant data include image information, language information, video information, and text description information. Direct location module: performs content recognition on the relevant data of the problem to be processed and calculates the direct location value of the problem to be processed; Judgment module: Determines whether the problem to be processed needs to be further processed through relevant department nodes based on the directly locatable value and the preset threshold; Collaborative office module: If the problem to be handled no longer needs to be processed through relevant department nodes, the target enterprise can be selected to handle the problem based on the problem handling records of each enterprise node. After the enterprise node completes the problem handling, it will upload the handling result to the platform, and the platform will complete the result archiving and feedback to the villager terminal, thus realizing a closed loop in the problem handling process.

7. The big data-driven rural government-enterprise-citizen collaborative government office system according to claim 6, characterized in that, The directly locatable includes: The direct location value module performs content recognition on the relevant data of the problem to be processed, calculates the multimodal content consistency value and the problem-oriented clarity value, and adds the multimodal content consistency value and the problem-oriented clarity value to obtain the direct location value of the problem to be processed.

8. The big data-driven rural government-enterprise-citizen collaborative government office system according to claim 7, characterized in that, The ability to be directly located also includes: Semantic Path Module: Semantic parsing is performed on the image information, text description information, speech transcription information and video information of the problem to be processed. According to the arrangement order of semantic content in a unified semantic space, each modality is converted into a sequence of semantic points ordered according to semantic relevance. Each semantic point is used to represent the semantic position of the corresponding modality in the unified semantic space, thereby forming image semantic path, text semantic path, speech semantic path and video semantic path respectively. Offset module: Using the image semantic path as the reference path, it compares the semantic nodes with the same sequence number in the text semantic path, speech semantic path and video semantic path as the image semantic path, calculates the distance value between each non-image modality semantic node and the corresponding image semantic node, and forms the text offset sequence, speech offset sequence and video offset sequence according to the order of the semantic nodes. Offset curvature module: Calculates the adjacent differences for the text offset sequence, speech offset sequence, and video offset sequence to obtain the change sequence of each offset; calculates the adjacent differences for the change sequence, and the resulting numerical sequence is used to characterize the offset curvature of the corresponding modality semantic path relative to the image semantic path; and accumulates the offset curvature values ​​for each modality to obtain the cumulative offset curvature value for the text modality, the cumulative offset curvature value for the speech modality, and the cumulative offset curvature value for the video modality. Multimodal content consistency value module: sums the cumulative offset curvature values ​​of text modality, voice modality and video modality to obtain the overall offset curvature value; divides the overall offset curvature value by 3 times the preset maximum offset curvature to calculate the offset curvature normalization ratio; and subtracts the calculated offset curvature normalization ratio from one to obtain the final multimodal content consistency value.

9. The big data-driven rural government-enterprise-citizen collaborative government office system according to claim 7, characterized in that, The ability to be directly located also includes: Aggregation radius module: Semantically encode all semantic keywords extracted from image information, text description information, speech transcription information and video information of the problem to be processed, and obtain their coordinates in a unified semantic space; calculate the geometric center position of all keyword coordinates, and determine the maximum distance from all keywords to the geometric center position based on the geometric center, and use the maximum distance as the semantic anchor point aggregation radius; The geographic anchoring fluctuation amplitude module identifies location-related terms, descriptive geographic information, or location coordinates from various modal data and converts them into a set of spatial coordinate points; it uses the average position of all location coordinate points as the location reference center and calculates the distance from each location point to the reference center; it then averages all distance values ​​to obtain the geographic anchoring fluctuation amplitude. The responsibility subject convergence ratio module: Input each semantic keyword into the preset responsibility subject back matching model, and obtain the set of responsibility subjects that can handle the problem represented by the keyword one by one; count the number of keywords that correspond to only a single responsibility subject among all keywords as the number of unique responsibility subjects, and calculate the degree of overlap between multiple responsibility subject sets corresponding to different keywords, and take the average of all overlap degrees as the average cross density of the responsibility subject sets; use the ratio of the number of unique responsibility subjects to the average cross density as the responsibility subject convergence ratio; The Problem Clarity Module: Multiply the semantic anchor aggregation radius by the geographical anchor fluctuation range, and use this product as the degree of non-concentration in the problem expression; then divide the degree of non-concentration by the convergence ratio of the responsible entity to obtain the ambiguity index of the problem expression; compare the ambiguity index with a constant 1, and divide 1 by the sum of the ambiguity index and 1 to obtain the problem clarity value.

10. The big data-driven rural government-enterprise-citizen collaborative government office system according to claim 6, characterized in that, The judgment module includes: The first office module compares the directly locatable value of the problem to be processed with a preset threshold. If the directly locatable value is not less than the preset threshold, it is determined that the problem to be processed does not need to go through the relevant department nodes for further processing. The platform analyzes the problem type of the problem to be processed and selects the target enterprise node that best matches the problem based on the processing records of each enterprise node in the historical problem processing process, the processing success rate, and the service area range. The problem to be processed is then directly dispatched to the target enterprise node in the form of a work order for processing. The second office module: When the directly locatable value is less than the preset threshold, the system will transfer the problem to the relevant department node for manual or semi-automatic identification, judgment and redistribution. The node will determine the final target enterprise node based on the problem attributes and complete the dispatch operation.

Images