Policy influence quantitative analysis method and device

Abstract

The invention discloses a policy influence quantitative analysis method and device. The method comprises the steps of obtaining a policy to be evaluated and a policy data set; based on the policy data set, determining references in a time dimension, a space dimension and a theme dimension; calculating association degrees of the policy to be evaluated in a time dimension, a space dimension and a theme dimension based on the reference basis; and based on the correlation degree, calculating the comprehensive influence degree of the policy to be evaluated through a fusion analysis model. The whole analysis process realizes automatic processing, and the efficiency and objectivity of policy impact assessment are improved.

Description

Technical Field

[0001] This invention relates to the field of computer data processing and policy analysis technology, and in particular to a method, apparatus, electronic device and storage medium for quantitative analysis of policy impact. Background Technology

[0002] With the modernization of the national governance system, government departments at all levels issue a vast number of policy documents, which are diverse in type and frequently updated, forming a complex and dynamic policy ecosystem. Against this backdrop, objectively and accurately assessing the influence, importance, or status of a specific policy within this ecosystem has become a core requirement in policy research, industry analysis, and corporate strategic planning.

[0003] Current methods for assessing policy impact often focus on analyzing a single dimension, such as time or space. While these one-sided analyses can reveal certain local characteristics of a policy, they lead to limited perspectives and fail to comprehensively and accurately reflect the overall impact of a policy within a complex policy environment. Summary of the Invention

[0004] To overcome the deficiencies in the prior art, the first objective of this invention is to provide a method for quantitative analysis of policy impact, the second objective of this invention is to provide a device for quantitative analysis of policy impact, the third objective of this invention is to provide an electronic device, and the fourth objective of this invention is to provide a computer-readable storage medium.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A method for quantitative analysis of policy impact, comprising:

[0007] Obtain the policies to be evaluated and the policy data set;

[0008] Based on the aforementioned policy data set, reference benchmarks are determined in the time, space, and thematic dimensions;

[0009] Based on the aforementioned benchmark, the correlation of the policy to be evaluated in terms of time, space, and theme is calculated.

[0010] Based on the aforementioned correlation, the comprehensive impact of the policy to be evaluated is calculated using a fusion analysis model.

[0011] Optionally, a reference benchmark in the time dimension may be determined, including:

[0012] The policy data in the policy dataset are sorted according to their release time to form a policy time series;

[0013] A sliding window is used to perform policy quantity statistics on the policy time series to obtain a policy quantity series;

[0014] Based on the policy quantity sequence, a peak period for policy release is determined using a peak identification algorithm, which serves as a time reference benchmark.

[0015] Optionally, a reference datum in spatial dimensions may be determined, including:

[0016] Each policy data point in the policy data set is mapped to its corresponding administrative unit in a predefined multi-level administrative hierarchy, based on the administrative division information of the issuing agency.

[0017] For each administrative unit, its weight values ​​are aggregated based on the release level of the policy data mapped to that administrative unit to calculate the coverage strength of that administrative unit, wherein the weight values ​​are determined based on the release level of the policy data;

[0018] The coverage intensity of each administrative unit is used to form an intensity set, which serves as a spatial reference benchmark.

[0019] Optionally, the weight value is further adjusted based on the release time of the policy data, and the adjustment steps include:

[0020] Determine the analysis time point to be used as the current time base;

[0021] For each piece of policy data, calculate the time difference between its release time and the analysis time point;

[0022] Based on the time difference, a time decay factor is calculated using a preset decay function;

[0023] The initial weight value determined based on the release level is multiplied by the time decay factor to obtain the final weight value after time adjustment.

[0024] Optionally, a benchmark for reference in the subject dimension may be determined, including:

[0025] The policy data set is trained using an LDA topic model to generate a policy topic list and obtain the topic probability distribution of each policy data item in the policy data set corresponding to each policy topic. The policy topic list and the topic probability distribution of all policy data together constitute a topic reference benchmark.

[0026] Optionally, the correlation of the policy to be evaluated over time is calculated, including:

[0027] From the time reference benchmark, determine the target peak time period closest to the release time of the policy to be evaluated;

[0028] Calculate the time difference between the release time of the policy to be evaluated and the target peak period;

[0029] Based on the time difference, a time similarity score is calculated by mapping using a preset Gaussian kernel function, which serves as the time correlation degree.

[0030] Optionally, the correlation of the policy to be evaluated in the spatial dimension is calculated, including:

[0031] Based on the administrative division information corresponding to the issuing agency of the policy to be evaluated, determine at least one target administrative unit that it maps to;

[0032] The coverage intensity of each target administrative unit is obtained from the spatial reference datum.

[0033] Based on the obtained coverage intensity, the spatial correlation value is calculated through weighted aggregation.

[0034] The spatial correlation value is normalized to a preset numerical range to obtain the spatial correlation degree.

[0035] Optionally, the relevance of the policy to be evaluated along the thematic dimension is calculated, including:

[0036] The policy text of the policy to be evaluated is input into the trained LDA topic model to obtain the topic probability distribution of the policy to be evaluated belonging to each policy topic in the policy topic list, which is used as the topic vector of the policy to be evaluated.

[0037] Based on the topic probability distribution of all policy data in the aforementioned topic reference benchmark, a policy set average topic vector is calculated;

[0038] The cosine similarity between the policy topic vector to be evaluated and the average topic vector of the policy set is calculated as the topic relevance.

[0039] Optionally, the step of calculating the comprehensive impact of the policy to be evaluated based on the correlation degree using a fusion analysis model includes:

[0040] Construct a policy association network, where nodes represent the policy to be evaluated and policy data in the policy data set, edges represent the association between policy data, and the weight of the edges is determined based on the semantic similarity or the strength of the reference relationship between policy data.

[0041] Based on the temporal relevance, spatial relevance, and thematic relevance of the policy to be evaluated, the prior node scores corresponding to the policy to be evaluated are calculated.

[0042] Construct a jump probability vector, wherein the probability value corresponding to the policy node to be evaluated in the jump probability vector is positively correlated with the prior score of the node, and the probability values ​​of all other nodes are equal;

[0043] The PageRank algorithm is run on the policy association network, and the jump probability vector is used as the random jump distribution of the PageRank algorithm. The PageRank value of the policy node to be evaluated after the algorithm converges is used as its comprehensive influence.

[0044] Optionally, the calculation of the node prior score corresponding to the policy to be evaluated includes:

[0045] The temporal correlation, spatial correlation, and thematic correlation are concatenated into vectors;

[0046] The concatenated vector is input into an attention network, and the attention network outputs dynamic weights corresponding to time, space and topic relevance, respectively.

[0047] Based on the dynamic weights, the temporal correlation, spatial correlation, and thematic correlation are weighted and summed, and the result is used as the prior score of the node.

[0048] Secondly, a policy impact quantification analysis device includes:

[0049] The data acquisition module is used to acquire the policies to be evaluated and the policy data set;

[0050] The benchmark determination module is used to determine reference benchmarks in the time dimension, spatial dimension, and thematic dimension based on the policy data set.

[0051] The correlation calculation module is used to calculate the correlation of the policy to be evaluated in terms of time, space and theme dimensions based on the reference benchmark.

[0052] The fusion analysis module is used to calculate the comprehensive impact of the policy to be evaluated based on the correlation degree through a fusion analysis model.

[0053] Firstly, a method for quantitative analysis of policy impact includes:

[0054] Thirdly, an electronic device includes:

[0055] One or more processors;

[0056] Memory;

[0057] One or more computer programs;

[0058] The one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and the one or more computer programs are configured to perform the above-described policy impact quantification analysis method.

[0059] Fourthly, a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the aforementioned method for quantitative analysis of the impact of policies.

[0060] Due to the application of the above technical solution, the present invention has the following advantages compared with the prior art:

[0061] 1. This application overcomes the limitations of existing technologies that only analyze from a single dimension by constructing a multi-dimensional analysis environment that includes time, space and thematic dimensions. It can comprehensively and systematically reflect the statistical characteristics and distribution patterns of policy data sets from multiple perspectives, providing richer reference benchmarks for policy impact assessment.

[0062] 2. By calculating the correlation between the policy to be evaluated and the benchmark policy data set in terms of time, space and subject, the policy to be evaluated is placed in the macro policy environment for comparative analysis. This makes the evaluation results not only reflect the policy's own attributes, but also reflect its relative position and correlation characteristics in the overall policy ecosystem.

[0063] 3. By employing a fusion analysis model, temporal, spatial, and thematic relevance are organically integrated. A policy correlation network is constructed, and prior node scores are calculated based on these correlations to construct a jump probability vector. The PageRank algorithm is then used to calculate the overall impact. This model effectively integrates multi-dimensional features, improving the overall rationality and accuracy of impact calculation.

[0064] 4. By using an attention network, the weights of temporal relevance, spatial relevance, and thematic relevance in the fusion process are dynamically calculated, enabling the importance of different dimensional features to be adaptively adjusted according to the specific policy context, further improving the accuracy and scenario adaptability of comprehensive impact measurement.

[0065] 5. The entire analysis process is automated, from policy data acquisition, determination of multi-dimensional benchmarks, correlation calculation to fusion analysis, reducing the need for manual intervention, improving the efficiency and objectivity of quantitative analysis of policy impact, and providing a feasible technical solution for large-scale policy analysis.

[0066] To make the above and other objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0067] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0068] Figure 1 This is a schematic diagram of the policy impact quantification analysis method in an embodiment of the present invention. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] Example: See Figure 1 As shown, a method for quantitative analysis of policy impact includes:

[0071] S1. Obtain the policy to be evaluated and the policy data set.

[0072] It should be noted that the policy data set refers to a structured dataset comprising multiple policy data items. The size of the policy data set can be determined based on the actual application scenario, and its purpose is to provide a representative policy environment background for subsequent analysis. Each policy data item in the policy data set is obtained from the public information platforms of government departments at all levels through a data collection system, and each policy data item includes at least the policy text, release date, issuing agency, and associated administrative division information.

[0073] To illustrate more clearly, the policy data may include, but is not limited to, the following types of policy documents: national-level administrative regulations, departmental rules (e.g., the "XXXX Development Plan"), opinions (e.g., the "XXXX Guiding Opinions"), and notices (e.g., the "XXXX Notice") issued by the State Council or its ministries; provincial-level local regulations, plans, and implementation schemes (e.g., the "XXXX Implementation Scheme") issued by provincial-level people's governments or their constituent departments; and municipal-level normative documents, working rules, and action plans (e.g., the "XXXX Several Policies") issued by municipal-level people's governments. These policy documents, after being acquired through a data collection system, are deconstructed and populated into structured data templates to form the policy data available for analysis.

[0074] The policy text refers to the complete content of the policy, including its title, body, appendices, and other components; the release date refers to the date the policy was officially released; the issuing agency refers to the full, standardized name of the government department that formulated and issued the policy; and the associated administrative division information refers to the standard administrative division code corresponding to the geographical scope to which the policy applies.

[0075] In one optional implementation, due to the diversity of data sources, the initial data directly obtained through the data acquisition system is often inconsistent in format and standardization. Therefore, it is necessary to construct the policy data set, which mainly includes three steps: data acquisition, data cleaning, and data standardization. The data acquisition step automatically retrieves policy documents and related metadata from government portals at all levels using a configured web crawler system. The data cleaning step processes the collected policy documents and related metadata, including deduplication, format validation, and handling of missing values. The data standardization step converts the policy release date to the ISO8601 standard format (e.g., "YYYY-MM-DD"), maps the issuing agency name to the standardized agency name in the GB / T2260 administrative division code system, and converts the policy text to UTF-8 encoded plain text format. After these preprocessing steps, a policy data set with a unified structure and controllable quality is finally formed, providing standardized basic data support for subsequent determination of multi-dimensional reference benchmarks.

[0076] Similarly, the policy to be evaluated, as the target policy for analysis and evaluation, also includes complete information with the same structure as the policy data in the policy dataset, namely, the policy text, issuance date, issuing agency, and associated administrative division information. In practice, the policy to be evaluated can be selected from a single policy data item in the policy dataset or can be imported separately from an external source. When the policy to be evaluated is selected from the policy dataset, it will participate in the calculation as both the analysis object and a component of the policy dataset in the subsequent analysis process. When the policy to be evaluated is imported separately from an external source, it is necessary to ensure that its data format is consistent with the policy data in the cleaned and standardized policy dataset for unified processing.

[0077] S2. Based on the policy data set, determine the reference benchmarks in the time dimension, spatial dimension and thematic dimension.

[0078] Specifically, the reference benchmarks include a time reference benchmark in the time dimension, a spatial reference benchmark in the spatial dimension, and a thematic reference benchmark in the thematic dimension, all determined based on the policy data set. These three benchmarks together constitute the multi-dimensional analysis environment.

[0079] Specifically, the process of determining the reference benchmark includes analysis in the following three dimensions:

[0080] S21. Determine the reference benchmark in the time dimension.

[0081] In an optional implementation, the process includes:

[0082] S211. Sort the policy data in the policy data set according to their release time to form a policy time series.

[0083] The policy time series is a sequence of policy data arranged chronologically, with each policy data point containing a standard-formatted release date. It should be noted that the release date is a date string that has been uniformly converted to the ISO 8601 standard format during the aforementioned data standardization steps, and its format can be "YYYY-MM-DD", for example, 2023-05-10. Using a unified time format ensures the accuracy and reliability of time comparison and sorting operations.

[0084] In practice, the sorting operation can be performed in chronological order. In one optional implementation, the policy data are arranged in ascending order from earliest to latest according to their release date; in another optional implementation, the policy data are arranged in descending order from latest to earliest according to their release date. Regardless of the chronological order used, the goal is to organize the discrete policy data into a sequence with a clear order in the time dimension, i.e., the policy time series. The policy time series reflects the chronological relationship of policy releases, laying the foundation for subsequent time-series-based analysis.

[0085] In an optional implementation, the sorting operation can be implemented programmatically. For example, in a Python environment, the release times of all policy data can be converted into comparable date-time objects, and then a built-in sorting function can be called to sort the list containing these policy data. After sorting, the index position of the policy data in the list corresponds to its temporal position in the policy time series. The resulting policy time series, as a basic data structure, will be directly used for subsequent sliding window statistics and peak period identification analysis.

[0086] It should be further explained that each element in the policy time series not only includes the release time information of the policy data, but also maintains the association with the original policy data, so that complete information of the policy data can be obtained at any time in subsequent analysis, thereby ensuring the traceability between the time series analysis results and the specific policy data.

[0087] S212. A sliding window is used to perform policy quantity statistics on the policy time series to obtain the policy quantity series.

[0088] The sliding window is a fixed-length time interval that slides across the policy time series. The window size and sliding step size of the sliding window can be configured according to the analysis needs. For example, when analyzing the pattern of policy releases within a year, the window size can be set to one month and the sliding step size can be set to half a month; when analyzing long-term trends, the window size can be set to one quarter and the sliding step size can be set to one month.

[0089] In practice, starting from the beginning of the policy time series, the sliding window is moved gradually along the time axis with a set sliding step size. For each time interval covered by the sliding window, the number of policy data items falling within that time interval is counted, and this number is used as the policy quantity value corresponding to the center or starting position of that window. Arranging the policy quantity values ​​calculated from all windows in chronological order constitutes the policy quantity sequence. The policy quantity sequence is a time series data, reflecting the fluctuations in the number of policy releases over time.

[0090] S213. Based on the policy quantity sequence, determine the peak time period for policy release using a peak identification algorithm, and use it as a time reference benchmark.

[0091] The peak identification algorithm is a calculation method for identifying significantly high-value points in sequence data. In this application, the peak identification algorithm refers to a calculation method configured to identify time windows in the policy quantity sequence where the number of policy releases is significantly higher than the usual level. The peak policy release period refers to a specific time interval identified by the peak identification algorithm that represents an abnormally active policy release activity.

[0092] In an optional implementation, the peak identification algorithm can employ a statistical distribution-based method. First, the mean and standard deviation of the policy quantity sequence are calculated. The mean reflects the average level of policy releases, while the standard deviation reflects the dispersion of policy releases. Then, a preset statistical threshold is set, which can be the mean plus a certain multiple of the standard deviation, for example, the mean plus 1.5 times the standard deviation. It should be noted that the preset statistical threshold is a configurable parameter, and its value can be adjusted according to the strictness of the peak definition. Next, the policy quantity sequence is traversed, and windows corresponding to policy quantity values ​​exceeding the preset statistical threshold are marked as candidate peak windows. Finally, considering that a policy release peak may last for a period of time, adjacent candidate peak windows can be merged. For example, if two candidate peak windows are consecutive or overlap in time, they are merged into a larger peak window, and the time interval of the merged window is ultimately identified as a policy release peak time period.

[0093] In another alternative implementation, the peak identification algorithm can also employ a peak detection-based method. This method identifies peak policy release periods by finding local maxima in the policy quantity sequence—that is, windows whose policy quantity values ​​are simultaneously greater than the policy quantity values ​​of the preceding and following windows—and combining this with an absolute quantity threshold or a relative percentage threshold.

[0094] It should be noted that, regardless of the specific algorithm used, the peak policy release periods all identify specific periods when policy release activities are exceptionally active in the time dimension. These periods serve as a reference benchmark in the time dimension, providing a reference for subsequent calculations of the time-dimensional relevance of the policies to be evaluated.

[0095] S22. Determine the reference datum in the spatial dimension.

[0096] The spatial dimension analysis is used to quantify the distribution density and impact of policy data in geospatial space. In a specific embodiment of this application, the geographic framework upon which the spatial dimension analysis relies is constructed based on the following core concepts:

[0097] Administrative level: refers to the hierarchical category of administrative divisions, such as national level, provincial level, municipal level, etc., which represents the relative position of administrative units in the administrative management system.

[0098] A predefined multi-level administrative hierarchy system refers to a complete structural framework consisting of multiple administrative levels arranged according to their hierarchical relationships, such as the system adopted in this application that includes national, provincial, and municipal levels.

[0099] Administrative unit: refers to a specific entity within the predefined multi-level administrative hierarchy, which is an administrative division with clearly defined geographical boundaries and administrative functions. Examples include national-level administrative units, provincial-level administrative units, and municipal-level administrative units.

[0100] In subsequent processing, each policy data item in the policy data set will be mapped to the corresponding administrative unit in the multi-level administrative hierarchy based on the administrative division information of its issuing agency. Policy coverage intensity is calculated and aggregated using administrative units as the basic unit, thereby forming a spatial reference benchmark.

[0101] Specifically, the analysis process includes the following steps:

[0102] S221. Map each policy data in the policy data set to the corresponding administrative unit in the predefined multi-level administrative hierarchy system, based on the administrative division information of the issuing agency.

[0103] A single policy data point may be associated with a single administrative unit or multiple administrative units. In the case of association with multiple administrative units, the policy data will be included in the statistics of all associated administrative units.

[0104] In practice, the mapping process is based on the standardized administrative division coding system established during the data standardization step. The associated administrative division information for each policy data item is uniformly processed into a standard administrative division code, which is consistent with the coding in the GB / T2260 standard. For example, the associated administrative division information for a policy data item may be standardized into standard codes such as "110000" (Beijing) and "440100" (Guangzhou).

[0105] In one optional implementation, a mapping table from administrative division codes to administrative levels can be established. The mapping table defines the national, provincial, and municipal administrative levels to which each administrative division code belongs. During mapping, the system reads the administrative division code of each policy data item and, by querying the mapping table, determines the administrative unit at each level to which the policy data item belongs. For example, for policy data coded "440100", querying the mapping table determines that it belongs to the national administrative unit "China", the provincial administrative unit "Guangdong Province", and the municipal administrative unit "Guangzhou City".

[0106] For policy data that relates to multiple administrative units, the system will repeatedly count it in each related administrative unit. For example, a national policy may relate to all provincial-level administrative units, in which case the policy data will be repeatedly counted in all 31 provincial-level administrative units (excluding Hong Kong, Macao, and Taiwan). This processing method ensures the integrity of the policy coverage intensity calculation and can truly reflect the actual spatial scope of the policy's impact.

[0107] After the mapping is completed, each administrative unit is associated with a subset of policy data, which contains all policy data mapped to that administrative unit. These policy data subsets provide the data foundation for subsequent calculations of the policy coverage intensity of each administrative unit.

[0108] S222. For each administrative unit, based on the release level of the policy data mapped to that administrative unit, aggregate its weight values ​​to calculate the coverage strength of that administrative unit, wherein the weight values ​​are determined based on the release level of the policy data.

[0109] The weight value is a quantitative measure of the relative importance of a single piece of policy data. The weight value is determined based on the policy data's publication level; the basic principle is that policy data published at a higher level receives a greater weight value.

[0110] In practice, different base weight values ​​can be preset for policy data at different release levels. For example, the base weight value for national-level policy data can be set to 1.0, for provincial-level policy data to 0.7, and for municipal-level policy data to 0.5. In actual applications, the base weight values ​​can be adjusted according to actual analysis needs. It should be noted that the specific values ​​mentioned above are only examples and can be configured according to different analysis scenarios in actual applications.

[0111] For an administrative unit, the weight values ​​of all its associated policy data are summed, and the sum is the coverage strength of that administrative unit.

[0112] In an optional implementation, the weight value can also be adjusted based on the release time of the policy data to reflect the timeliness of the policy. Specifically, the steps for adjusting the weight value according to the release time include:

[0113] S2221. Determine the analysis time point as the current time reference.

[0114] The analysis time point is a reference time point used to calculate time differences. It is typically set to the system's current date at the time of the analysis, or a specific historical date set according to the analysis requirements. For example, if the analysis objective is to assess the policy coverage strength up to the end of 2023, the analysis time point can be set to "2023-12-31". The purpose of this analysis time point is to serve as a unified benchmark for calculating how far away the policy data release time is from the present.

[0115] S2222. For each piece of policy data, calculate the time difference between its release time and the analysis time point.

[0116] The time difference refers to the length of time between the release time of the policy data and the analysis time point.

[0117] In practice, the time difference can be calculated in days. For example, if the release date of a policy data item is "2023-06-01" and the analysis date is "2023-12-31", then the time difference is 213 days. It should be noted that since the release dates of all policy data have been uniformly converted to the ISO 8601 standard format during the data standardization process, accurate time difference calculations can be performed directly.

[0118] S2223. Based on the time difference, a time decay factor is calculated using a preset decay function.

[0119] The decay function is a mathematical function that maps time differences to values ​​in the range [0,1], and its function is to quantify the influence of time factors on policy weights. The time decay factor is a value between 0 and 1, representing the degree of weight decay caused by time factors.

[0120] In an optional implementation, the decay function can be in the form of an exponential decay function, the mathematical expression of which can be expressed as: ,in It is a decay coefficient greater than 0, used to control the decay rate. For example, when λ is 0.001, it means that every 1000 days, the weight of the policy data will decay to about 37% of its original value. The decay coefficient λ can be adjusted according to actual analysis needs to reflect different timeliness requirements.

[0121] S2224. Multiply the initial weight value determined based on the release level with the time decay factor to obtain the final weight value after time adjustment.

[0122] The initial weight values ​​are determined solely based on the release level of the policy data. For example, the basic weight value for national-level policy data is 1.0, the basic weight value for provincial-level policy data is 0.7, and the basic weight value for municipal-level policy data is 0.5.

[0123] The final weight value is a comprehensive weight value that takes into account both the publication level and time factors. For example, a national policy data published on "2022-01-01" has an initial weight value of 1.0. Assuming that the time decay factor calculated at the analysis time point "2023-12-31" is 0.8, then its final weight value is 1.0 × 0.8 = 0.8.

[0124] By adopting the above approach, both the release level of policy data and the timeliness of policy data are considered, so that recently released high-level policy data have a higher weight value, thereby more accurately reflecting their actual influence in the spatial dimension.

[0125] S223. The coverage intensity of each administrative unit is used to form an intensity set, which serves as a spatial reference benchmark.

[0126] The intensity set is a data structure containing all administrative units and their corresponding coverage intensity values, such as a dictionary or database table, where the key is the administrative unit identifier (e.g., administrative division code) and the value is the coverage intensity of that administrative unit. This intensity set serves as a spatial reference benchmark, providing a reference for subsequent calculations of the spatial relevance of the policies to be evaluated.

[0127] S23. Determine the reference benchmark in the thematic dimension.

[0128] The topic dimension analysis described above is used to quantify the areas of interest and their distribution in a policy data set from a textual semantic perspective. Its core lies in constructing a semantic space, transforming textual content into a computable vector representation. This semantic space consists of several latent topics automatically identified through a Latent Dirichlet Allocation (LDA) topic model. Each topic is represented as a set of keywords that frequently co-occur in the policy text, i.e., a policy topic. The policy topic list is the set of all these policy topics, defining the dimension of the semantic space. The topic probability distribution of each policy data item in the policy data set corresponding to each policy topic describes its specific location within this semantic space.

[0129] In practice, an LDA topic model is used to train all policy texts in the policy dataset to generate a list of policy topics, and the topic probability distribution of each policy data item in the policy dataset corresponding to each policy topic is obtained. The policy topic list and the topic probability distributions of all policy data together constitute a topic reference benchmark.

[0130] Latent Dirichlet Allocation (LDA) is a commonly used topic model that automatically discovers potential topic structures from a document collection. In its implementation, the first step is to preprocess all policy texts, including word segmentation and stop word removal, and then construct a bag-of-words model based on this. The bag-of-words model is a mathematical model that represents text as the frequency of word occurrences. It does not consider the order of words in the document but effectively captures the statistical distribution characteristics of vocabulary, making it suitable for LDA topic modeling. Then, the processed policy text is trained using the LDA topic model.

[0131] It should be further noted that the number of topics in the LDA topic model is a hyperparameter that needs to be preset. In an optional implementation, the optimal number of topics can be determined by optimizing the perplexity index and the topic consistency index. The perplexity index measures the model's predictive ability on unknown data; a lower value generally indicates better generalization ability. The topic consistency index evaluates the semantic relevance of keywords within a topic; a higher value indicates stronger interpretability. Specifically, multiple LDA topic models can be trained within a preset range of candidate topics (e.g., from 10 to 50). Then, the perplexity index and topic consistency index of each model are calculated. Finally, the number of topics that perform well in both perplexity and topic consistency indices is selected as the final number of topics by combining these two indices (e.g., selecting values ​​near the inflection point of the perplexity curve and with high topic consistency scores). After training, the LDA topic model outputs the keyword distribution for each policy topic, i.e., the probability of occurrence of each keyword under each policy topic. These keyword distributions constitute the specific content of the policy topic list.

[0132] For each policy data item in the policy dataset, its policy text is input into a pre-trained LDA topic model. The LDA topic model outputs the probability distribution of the policy text across various policy topics, i.e., the topic probability distribution. The topic probability distribution is a numerical vector, where each dimension of the vector corresponds to a policy topic, and its value represents the probability that the policy text belongs to the corresponding policy topic. The sum of the probability values ​​of all dimensions is 1.

[0133] The policy topic list defines the dimensions of the policy semantic space, while the topic probability distribution of all policy data describes the position of each policy within that semantic space. Together, they constitute the benchmark for the topic dimension, providing a semantic reference for subsequent calculations of the relevance of the policies to be evaluated along the topic dimension.

[0134] S3. Based on the reference benchmark, calculate the correlation of the policy to be evaluated in the time dimension, spatial dimension and thematic dimension.

[0135] The correlation degree is a numerical indicator used to quantify the degree of correlation between the policy to be evaluated and the policy data set across various dimensions. Specifically, this correlation degree is not calculated directly with all the data in the policy data set, but rather based on reference benchmarks extracted from the policy data set (i.e., time reference benchmark, spatial reference benchmark, and thematic reference benchmark). This step specifically includes calculations across three dimensions:

[0136] S31. Calculate the correlation of the policy to be evaluated over time.

[0137] In one optional implementation, the specific steps include:

[0138] S311. By calculating the time difference between the release time of the policy to be evaluated and each peak time period in the time reference benchmark, determine the nearest target peak time period and the corresponding time difference.

[0139] In practice, all peak policy release periods identified in step S213 are iterated through, and a representative time point is determined for each peak period to quantify the closest proximity. In a preferred embodiment, the representative time point is the center time point of the peak period.

[0140] It should be noted that using the central time point as the representative time point is a preferred calculation method, not the only limitation. In other alternative implementations, the start or end time point of the peak period can also be used as the representative time point.

[0141] By calculating the absolute time interval (usually in days) between the release time of the policy to be evaluated and the time point represented by each peak period, and comparing all the calculation results, the peak period with the smallest corresponding time difference is determined as the target peak period, and the smallest time difference is recorded.

[0142] S312. Based on the time difference, a time similarity score is calculated by mapping using a preset Gaussian kernel function, which serves as the time correlation degree.

[0143] This step is based on the minimum time difference determined in step S311. The mathematical expression for the Gaussian kernel function can be expressed as:

[0144] in It is the bandwidth parameter of the Gaussian kernel function, a preset constant greater than zero, used to control the rate at which similarity decays over time. The larger the value, the slower the similarity decays over time; The smaller the value, the faster the decay. Using the Gaussian kernel function, the smaller the time difference, the closer the calculated time similarity score is to 1; the larger the time difference, the closer the score is to 0, thus obtaining a quantization result that changes continuously in the interval [0,1]. This quantization result is used as the time correlation degree.

[0145] S32. Calculate the correlation of the policy to be evaluated in the spatial dimension.

[0146] In one optional implementation, the specific steps include:

[0147] S321. Based on the administrative division information corresponding to the issuing agency of the policy to be evaluated, determine at least one target administrative unit that it maps to.

[0148] Specifically, the policy to be evaluated, as the object of analysis, has its issuing agency's corresponding administrative division information processed into a standardized administrative division code during the data standardization process, for example, a coding format consistent with the GB / T2260 standard. This administrative division code is used to map the policy to be evaluated to specific administrative units within a predefined multi-level administrative hierarchy. This multi-level administrative hierarchy includes, for example, national, provincial, and municipal administrative units, each with clearly defined geographical boundaries and administrative functions. The mapping process is achieved by querying a pre-established mapping table between administrative division codes and administrative levels. For example, if the issuing agency of the policy to be evaluated corresponds to code "110000," then the mapping table can determine that it is mapped to the national administrative unit "China," the provincial administrative unit "Beijing," and the municipal administrative unit "Beijing." It should be noted that a policy may be associated with multiple administrative units; for example, a national policy may be mapped to all provincial administrative units. In this case, the policy to be evaluated will be mapped to multiple target administrative units.

[0149] S322. Obtain the coverage intensity of each target administrative unit from the spatial reference datum.

[0150] In this step, for each target administrative unit determined in step S321, a query is performed from the spatial reference benchmark based on its administrative unit identifier to obtain the corresponding coverage strength value. Specifically, the spatial reference benchmark uses the administrative unit identifier as the search key, and the administrative unit identifier adopts a standardized administrative division code; by matching the administrative division code of the target administrative unit with the key in the spatial reference benchmark, the coverage strength value mapped to that administrative unit can be obtained.

[0151] For example, if the policy to be evaluated is mapped to the provincial administrative unit "Guangdong Province" and the municipal administrative unit "Guangzhou City", and their corresponding administrative division codes are "440000" and "440100" respectively, then the system obtains the coverage intensity values ​​associated with the codes "440000" and "440100" respectively by querying the spatial reference benchmark, and uses them as the coverage intensity of the target administrative unit.

[0152] S323. Based on the obtained coverage intensity, the spatial correlation value is obtained through weighted aggregation calculation.

[0153] This step is used to aggregate the coverage intensity of multiple target administrative units obtained in step S322 into a comprehensive evaluation value, namely the spatial correlation value. The weighted aggregation is used to assign a weight coefficient to the coverage intensity of each target administrative unit, and the weight coefficient is determined based on the relative importance of the administrative unit in the predefined multi-level administrative hierarchy.

[0154] In an optional implementation, the process of obtaining spatial correlation values ​​through weighted aggregation includes:

[0155] A weight coefficient is determined for each target administrative unit. This weight coefficient is preset based on the administrative level to which the target administrative unit belongs. It should be noted that this weight coefficient is a secondary weighting based on the administrative unit level, used to coordinate the relative contribution of the coverage intensity of administrative units at different levels to the final spatial correlation value. This complements the initial weighting based on the policy release level when calculating coverage intensity, together constructing a complete spatial importance assessment system. In an optional implementation, following the principle of determining weights based on the release level in step S222, basic weight coefficients can be preset for administrative units at different levels. For example, the basic weight coefficient for national-level administrative units can be set to 1.0; the basic weight coefficient for provincial-level administrative units can be set to 0.7; and the basic weight coefficient for municipal-level administrative units can be set to 0.5. It should be noted that the above specific values ​​are only examples and can be configured according to different analysis scenarios in actual applications.

[0156] The calculation is performed using determined weighting coefficients. It is assumed that the policy to be evaluated is mapped to... The target administrative unit, for the first The coverage intensity of each target administrative unit is denoted as... The corresponding weighting coefficient is denoted as Then the spatial correlation value It is calculated using the following formula:

[0157] ;

[0158] That is, the spatial correlation value is the sum of the product of the coverage intensity of each target administrative unit and its corresponding weight coefficient.

[0159] Through the above weighted aggregation method, the coverage intensity of high-level and high-weight administrative units contributes more to the final spatial correlation value, thereby enabling the calculation results to more accurately characterize the degree of correlation between the policy to be evaluated and the overall coverage intensity of the policy data set in the spatial dimension.

[0160] S324. Normalize the spatial correlation value to a preset numerical range to obtain the spatial correlation degree.

[0161] The purpose of normalization is to scale the spatial correlation value to a range comparable to the temporal and thematic correlation values ​​calculated in other steps, such as the [0,1] interval. The normalization method can be the min-max normalization method. Specifically, The minimum and maximum coverage intensities are taken from the entire intensity set in the spatial reference benchmark, i.e., the minimum and maximum coverage intensities of all administrative units. Through normalization, the spatial correlation becomes a standardized value; the closer the value is to 1, the more correlated the policy to be evaluated is with the coverage intensity distribution of the policy data set in the spatial dimension; the closer it is to 0, the lower the correlation.

[0162] S33. Calculate the relevance of the policy to be evaluated in terms of thematic dimensions.

[0163] In one optional implementation, the specific steps include:

[0164] S331. Input the policy text of the policy to be evaluated into the trained LDA topic model to obtain the topic probability distribution of the policy to be evaluated belonging to each policy topic in the policy topic list, and use this topic probability distribution as the policy topic vector to be evaluated.

[0165] Using the LDA topic model trained in step S231, the policy text of the policy to be evaluated is inferred to obtain its topic probability distribution vector, which is the topic vector of the policy to be evaluated.

[0166] S332. Based on the topic probability distribution of all policy data in the topic reference benchmark, calculate an average topic vector of the policy set.

[0167] The arithmetic mean of the thematic probability distribution vectors of all policy data in the policy dataset is calculated to obtain the average thematic vector of the policy dataset. This average thematic vector represents the central tendency of the entire policy dataset in the thematic dimension.

[0168] S333. Calculate the cosine similarity between the policy topic vector to be evaluated and the average topic vector of the policy set, and use it as the topic correlation degree.

[0169] Specifically, let the vector of the policy themes to be evaluated be... in The total number of topics in the policy topic list, vector components This indicates that the policy to be evaluated belongs to the first category. The probability of each policy topic. Let the average topic vector of the policy set be... , where vector components This represents the probability distribution of the themes of all policy data in the policy dataset. The arithmetic mean of the probability values ​​corresponding to each policy theme.

[0170] Then, the topic relevance Calculated using the following cosine similarity formula:

[0171] ;

[0172] Due to the policy theme vector to be evaluated Average topic vector of policy set All components are non-negative, and the topic relevance is... The value range is [0,1]. The closer the value is to 1, the more similar the distribution of the policy to be evaluated in the thematic dimension is to the overall thematic distribution of the policy set; the closer the value is to 0, the less similar it is.

[0173] S4. Based on the correlation, calculate the comprehensive impact of the policy to be evaluated using a fusion analysis model.

[0174] The fusion analysis model is a mathematical model that integrates the correlations across three dimensions—time, space, and theme—and calculates influence based on a graph ranking algorithm. This model constructs a policy correlation network, transforming the quantifiable information of the multi-dimensional correlations of the policy to be evaluated into prior scores for its corresponding nodes within the network. An improved PageRank algorithm is then used to simulate the propagation of influence within the network, ultimately converging to a stable comprehensive influence score. The process specifically includes the following steps:

[0175] S41. Construct a policy-related network.

[0176] The policy correlation network is used to represent the interrelationships between policies. Its construction process includes the following steps:

[0177] S411. Determine the set of nodes.

[0178] The node set includes the node corresponding to the policy to be evaluated and the node corresponding to each piece of policy data in the policy data set. Each node represents a specific policy, and the node identifier uses a unique identifier for the policy, such as a policy document number or a unique ID assigned internally by the system.

[0179] S412. Determine the set of edges.

[0180] Based on the set of nodes, an edge set is established to represent the relationships between policies. For any two policy nodes, if there is semantic similarity between them, or if there is a clear relationship such as referencing, updating, or implementation extracted by the policy relationship identification module, then a directed edge is established between them. The direction of the edge is determined by the semantics of the relationship; for example, in a referencing relationship, the edge points from the referencing policy node to the referenced policy node.

[0181] S413, Calculate the weight of the edge.

[0182] Each edge in the edge set is assigned a weight value to quantify the strength of the relationship between policies. The weight values ​​are calculated based on the semantic similarity or the strength of the reference relationship between policy data.

[0183] In an optional implementation, the semantic similarity-based weight calculation is achieved by calculating the cosine similarity between the policy text embedding vectors of two policy nodes. The policy text embedding vector is a distributed numerical representation of the policy text. Specifically, a pre-trained model such as BGE-m3 can be used and fine-tuned using policy domain data to generate vectors with greater domain discriminative power. The cosine similarity value between two policy vectors, with a range of [0,1], serves as the weight value for the corresponding edge. The closer this value is to 1, the more semantically similar the policy texts are, and the stronger the association.

[0184] In another optional implementation, the weight calculation based on the strength of the reference relationship is achieved by using the relationship confidence score output by the policy association identification module as the weight value. This confidence score is the probability value of the classification model's determination of a specific relationship type (such as a reference), with a value range of [0,1]. A higher confidence score indicates a higher degree of credibility in the existence of the reference relationship, and the corresponding edge weight value is also higher.

[0185] In practical implementation, priority rules can be set. For example, when there is a high-confidence reference relationship between two policy nodes, the weight calculated based on the strength of the reference relationship should be used first; otherwise, the weight calculated based on semantic similarity should be used.

[0186] S42. Based on the temporal correlation, spatial correlation, and thematic correlation of the policy to be evaluated, calculate the prior score of the node corresponding to the policy to be evaluated.

[0187] The node prior score is a scalar value used in subsequent graph ranking algorithms to characterize the inherent importance of the policy node being evaluated based on its multidimensional association characteristics. The calculation process specifically includes:

[0188] S421. Standardize the temporal correlation, spatial correlation and thematic correlation before merging.

[0189] The purpose of this step is to provide a consistent numerical benchmark for subsequent multi-dimensional correlation calculations. Although the theoretical ranges for temporal, spatial, and thematic correlation are all closed intervals from zero to one, their values ​​may exhibit different distribution characteristics in actual analysis due to differences in calculation methods and data sources. For example, temporal correlation values ​​may be concentrated in the higher range of 0.7-1.0, while thematic correlation values ​​may be concentrated in the lower range of 0-0.3. Without standardization, dimensions with generally higher values ​​will receive disproportionate influence in subsequent fusion calculations, introducing evaluation bias.

[0190] To achieve standardized processing, the specific implementation method is as follows:

[0191] The overall sample is the correlation coefficients calculated across all dimensions for all policy data to be compared (including all policies in the policy dataset and the policies to be evaluated). The following operations are performed separately for the time, spatial, and thematic dimensions:

[0192] Let the original set of correlation values ​​for a certain dimension (taking time as an example) be . ,in This represents the total sample size. Calculate the minimum value of all correlation scores for this dimension. With the maximum value For any correlation value under this dimension (Including the relevance of the policies to be evaluated), its standardized value Calculated using the following minimum-maximum normalization formula: .

[0193] This calculation is performed independently for the three dimensions of time, space, and theme. After this standardization process, for any given dimension, the correlation values ​​of all policy samples (including the policy to be evaluated) in that dimension will be linearly transformed to the range of zero to one, and the numerical distribution will cover the entire range of zero to one. This process eliminates the scale differences caused by the different original numerical distribution ranges between different dimensions, ensuring that the correlation values ​​of each dimension have a consistent numerical benchmark in the fusion calculation.

[0194] S422, Vector concatenation and attention weight allocation.

[0195] The standardized temporal correlation, spatial correlation, and thematic correlation values ​​are treated as three independent components and concatenated in a fixed order of time, space, and theme (i.e., vector concatenation) to form a three-dimensional feature vector. This three-dimensional feature vector is... ,in , , These represent the standardized temporal relevance, spatial relevance, and thematic relevance, respectively.

[0196] Subsequently, the three-dimensional feature vector The input is fed into an attention network. This attention network is a simple feedforward neural network that learns a mapping function through training, enabling it to adaptively evaluate the relative importance of the three dimensions based on the specific numerical patterns of the input feature vector. The network's terminals employ a Softmax activation function, ensuring that its output consists of three dynamic weights greater than 0 that sum to 1, corresponding to the time, space, and topic dimensions, respectively.

[0197] These dynamic weights can dynamically reflect the relative contribution of the three-dimensional correlation to the prior scores of the computation nodes in the context of the current specific policy to be evaluated.

[0198] S423, Weighted summation and fractional normalization.

[0199] Based on the dynamic weights output by the attention network, the standardized temporal correlation, spatial correlation, and topic correlation are weighted and summed to obtain an initial prior score.

[0200] To make this score applicable to the construction of subsequent probability distributions, the initial prior score is normalized again using the Softmax function, transforming it into a probability value between 0 and 1, which is then used as the final node prior score, denoted as . This step ensures that the prior scores of the policy to be evaluated have clear probabilistic significance.

[0201] S43. Construct the jump probability vector.

[0202] Construct a length of The vector representing the total number of nodes in the policy-related network (i.e., the total number of nodes in the network) serves as the random jump probability distribution in the subsequent PageRank algorithm, and is called the jump probability vector, denoted as . The jump probability vector The construction process is as follows:

[0203] Initialize the jump probability vector , all of them The value of the cell element is set to 0. Get the index of the policy node to be evaluated in the node list. And the node prior scores calculated via step S423 are Prior scores of nodes. Assign to the jump probability vector The index is The element that executes The remaining probability values Distribute equally to the rest Each node. Specifically, for the jump probability vector Each index (in ), set its value to That is, execution .

[0204] The above steps ensure the jump probability vector is accurate. The sum of all elements is 1. Meanwhile, the prior scores of the policy nodes to be evaluated... The higher the value, the greater the probability that it will be accessed during random redirects. The larger it is.

[0205] S44. Run the improved PageRank algorithm on the policy association network and use the PageRank value of the policy node to be evaluated after convergence as its comprehensive influence.

[0206] The policy association network constructed in step S41 and the jump probability vector constructed in step S43 are combined. and preset damping factor As input, the PageRank algorithm is run. The set of edge weights in the policy association network is... The damping factor Used to control the probability of random jumps, it is usually set to 0.85.

[0207] The standard PageRank algorithm iterative formula is as follows:

[0208]

[0209] in:

[0210] It is a node PageRank value.

[0211] It points to The set of all nodes.

[0212] It is a node The sum of the weights of all outgoing edges.

[0213] This represents the probability that, during a random jump, the user will uniformly jump to any node in the network.

[0214] In this application, the standard PageRank algorithm is improved by uniformly jumping to different terms. Replace with jump probability vector Components of the corresponding node The improved algorithm iteration formula is as follows:

[0215] ;

[0216] in, Represents the jump probability vector The Middle Each component corresponds to a node. .

[0217] According to the improved iterative formula, the PageRank value of all nodes is calculated iteratively until the sum of the absolute values ​​of the changes in the PageRank values ​​of all nodes in two consecutive iterations is less than a preset convergence threshold, at which point the algorithm converges. After convergence, the final PageRank value corresponding to the node of the policy to be evaluated is output as the comprehensive impact of the policy to be evaluated.

[0218] This embodiment also discloses a policy impact quantification analysis device, including:

[0219] The data acquisition module is used to acquire the policies to be evaluated and the policy data set;

[0220] The benchmark determination module is used to determine reference benchmarks in the time dimension, spatial dimension, and thematic dimension based on the policy data set.

[0221] The correlation calculation module is used to calculate the correlation of the policy to be evaluated in terms of time, space and theme dimensions based on the reference benchmark.

[0222] The fusion analysis module is used to calculate the comprehensive impact of the policy to be evaluated based on the correlation degree through a fusion analysis model.

[0223] This embodiment also discloses an electronic device, including:

[0224] One or more processors;

[0225] Memory;

[0226] One or more computer programs;

[0227] The one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and the one or more computer programs are configured to perform the above-described policy impact quantification analysis method.

[0228] This embodiment also discloses a computer-readable storage medium storing a computer program thereon, characterized in that the computer program, when executed by a processor, implements the above-mentioned method for quantitative analysis of the impact of policies.

[0229] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for quantitative analysis of policy impact, characterized in that, include: Obtain the policies to be evaluated and the policy data set; Based on the aforementioned policy data set, reference benchmarks are determined in the time, space, and thematic dimensions; Based on the aforementioned benchmark, the correlation of the policy to be evaluated in terms of time, space, and theme is calculated. Based on the aforementioned correlation, the comprehensive impact of the policy to be evaluated is calculated using a fusion analysis model.

2. The method for quantitative analysis of policy impact according to claim 1, characterized in that, Determine the reference point in the time dimension, including: The policy data in the policy dataset are sorted according to their release time to form a policy time series; A sliding window is used to statistically analyze the number of policies on the policy time series to obtain a policy quantity series. Based on the policy quantity sequence, a peak period for policy release is determined using a peak identification algorithm, which serves as a time reference benchmark.

3. The method for quantitative analysis of policy impact according to claim 1, characterized in that, Determining the reference datum in spatial dimensions includes: Each policy data point in the policy data set is mapped to its corresponding administrative unit in a predefined multi-level administrative hierarchy, based on the administrative division information of the issuing agency. For each administrative unit, its weight values ​​are aggregated based on the release level of the policy data mapped to that administrative unit to calculate the coverage strength of that administrative unit, wherein the weight values ​​are determined based on the release level of the policy data; The coverage intensity of each administrative unit is used to form an intensity set, which serves as a spatial reference benchmark.

4. The method for quantitative analysis of policy impact according to claim 3, characterized in that, The weight values ​​are also adjusted based on the release time of the policy data, and the adjustment steps include: Determine the analysis time point to be used as the current time base; For each piece of policy data, calculate the time difference between its release time and the analysis time point; Based on the time difference, a time decay factor is calculated using a preset decay function; The initial weight value determined based on the release level is multiplied by the time decay factor to obtain the final weight value after time adjustment.

5. The method for quantitative analysis of policy impact according to claim 1, characterized in that, Determine the benchmarks for thematic dimensions, including: The policy data set is trained using an LDA topic model to generate a policy topic list and obtain the topic probability distribution of each policy data item in the policy data set corresponding to each policy topic. The policy topic list and the topic probability distribution of all policy data together constitute a topic reference benchmark.

6. The method for quantitative analysis of policy impact according to claim 2, characterized in that, Calculating the correlation of the policy to be evaluated over time includes: By calculating the time difference between the release time of the policy to be evaluated and each peak time period in the time reference benchmark, the nearest target peak time period and the corresponding time difference are determined. Based on the time difference, a time similarity score is calculated by mapping using a preset Gaussian kernel function, which serves as the time correlation degree.

7. The method for quantitative analysis of policy impact according to claim 3, characterized in that, Calculating the spatial correlation of the policies to be evaluated includes: Based on the administrative division information corresponding to the issuing agency of the policy to be evaluated, determine at least one target administrative unit that it maps to; The coverage intensity of each target administrative unit is obtained from the spatial reference datum. Based on the obtained coverage intensity, the spatial correlation value is calculated through weighted aggregation; The spatial correlation value is normalized to a preset numerical range to obtain the spatial correlation degree.

8. The method for quantitative analysis of policy impact according to claim 5, characterized in that, Calculating the relevance of the policies to be evaluated across thematic dimensions includes: The policy text of the policy to be evaluated is input into the trained LDA topic model to obtain the topic probability distribution of the policy to be evaluated belonging to each policy topic in the policy topic list, which is used as the topic vector of the policy to be evaluated. Based on the topic probability distribution of all policy data in the aforementioned topic reference benchmark, a policy set average topic vector is calculated; The cosine similarity between the policy topic vector to be evaluated and the average topic vector of the policy set is calculated as the topic relevance.

9. The method according to claim 1, characterized in that, The calculation of the comprehensive impact of the policy to be evaluated based on the correlation degree using a fusion analysis model includes: Construct a policy association network, where nodes represent the policy to be evaluated and policy data in the policy data set, edges represent the association between policy data, and the weight of the edges is determined based on the semantic similarity or the strength of the reference relationship between policy data. Based on the temporal relevance, spatial relevance, and thematic relevance of the policy to be evaluated, the prior node scores corresponding to the policy to be evaluated are calculated. Construct a jump probability vector, wherein the probability value corresponding to the policy node to be evaluated in the jump probability vector is positively correlated with the prior score of the node, and the probability values ​​of all other nodes are equal; The PageRank algorithm is run on the policy association network, the jump probability vector is used as the random jump distribution of the PageRank algorithm, and the PageRank value of the policy node to be evaluated after the algorithm converges is used as its comprehensive influence.

10. The method according to claim 9, characterized in that, The calculation of the prior scores of the nodes corresponding to the policy to be evaluated includes: The temporal correlation, spatial correlation, and thematic correlation are concatenated into vectors; The concatenated vector is input into an attention network, and the attention network outputs dynamic weights corresponding to time, space and topic relevance, respectively. Based on the dynamic weights, the temporal correlation, spatial correlation, and thematic correlation are weighted and summed, and the result is used as the prior score of the node.

11. A device for quantitative analysis of policy impact, characterized in that, include: The data acquisition module is used to acquire the policies to be evaluated and the policy data set; The benchmark determination module is used to determine reference benchmarks in the time dimension, spatial dimension, and thematic dimension based on the policy data set. The correlation calculation module is used to calculate the correlation of the policy to be evaluated in terms of time, space and theme dimensions based on the reference benchmark. The fusion analysis module is used to calculate the comprehensive impact of the policy to be evaluated based on the correlation degree through a fusion analysis model.

Images