Scientific culture dissemination feasibility assessment system based on regionalization analysis

Abstract

The application discloses a scientific and cultural dissemination feasibility evaluation system based on regionalization analysis, and comprises the following steps: acquiring initial regionalization data in a specific region; constructing a four-dimensional feature label including geography, time, cultural characteristics and scientific correlation degree; performing multi-dimensional feature extraction on the initial regionalization data based on the four-dimensional feature label; performing sentiment analysis on the feature regionalization data based on a RoBERTa fine-tuning model; generating a regional language feature map in combination with a dialect sensitive word library; establishing a regionalization hybrid evaluation model based on GWR geographic weighted regression, XGBoost and GNN graph neural network; inputting deep regionalization data into the regionalization hybrid evaluation model to generate a dissemination potential heat map and a conflict early warning index. The cultural dissemination possibility of regional characteristics such as geographical environment, cultural context and time law is fully considered, and the feasibility and accuracy of cultural dissemination are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, in particular to a scientific culture dissemination feasibility evaluation system based on regional analysis. BACKGROUND

[0002] Scientific knowledge and scientific methods, as an important bridge connecting scientific research innovation and social cognition, their dissemination efficiency directly affects the public scientific literacy and the social transformation of scientific and technological achievements. However, traditional scientific dissemination strategies rely on one-way content output and general evaluation indicators, lacking deep analysis of regional heterogeneity, resulting in unbalanced allocation of dissemination resources, insufficient cultural adaptation and other problems. SUMMARY

[0003] The purpose of the present application is to solve the above problems, and a scientific culture dissemination feasibility evaluation system based on regional analysis is designed.

[0004] To achieve the above purpose, the technical scheme of the present application is that, further, in the scientific culture dissemination feasibility evaluation system based on regional analysis, the scientific culture dissemination feasibility evaluation system comprises the following modules:

[0005] A regional data acquisition module is used to acquire scientific research databases, social media, open data and dialect data in a specific region, obtain regional data, perform data preprocessing on the regional data, and obtain initial regional data;

[0006] A multi-dimensional feature extraction module is used to construct four-dimensional feature labels including geography, time, cultural characteristics and scientific correlation degree, perform multi-dimensional feature extraction on the initial regional data based on the four-dimensional feature labels, and obtain feature regional data;

[0007] A language deep analysis module is used to perform sentiment analysis on the feature regional data based on a RoBERTa fine-tuning model, generate a regional language feature map in combination with a dialect sensitive word library, and obtain deep regional data;

[0008] An evaluation model establishment module is used to establish a regional hybrid evaluation model based on GWR geographic weighted regression, XGBoost and GNN graph neural network, introduce a spatiotemporal Transformer model to process sequence data with strong spatiotemporal dependence, and obtain a target regional hybrid evaluation model;

[0009] A feasibility evaluation module is used to input the deep regional data into the target regional hybrid evaluation model, and generate a dissemination potential heat map and a conflict early warning index.

[0010] Further, in the scientific culture dissemination feasibility evaluation system based on regional analysis, the regional data acquisition module comprises the following sub-modules:

[0011] The scientific research data acquisition submodule is used to acquire publicly available experimental data from databases and scientific research institutions to obtain regional scientific research subset data.

[0012] The media data acquisition submodule is used to collect text, image, and video content data from social media platforms within the region to obtain social media subset data.

[0013] The open data acquisition submodule is used to connect to the public data platform to acquire data on population distribution, education level and cultural facility layout in the region, and obtain open subset data.

[0014] The dialect data acquisition submodule is used to acquire dialect speech data and language data from dialect databases in the region to obtain dialect subset data.

[0015] The data integration submodule is used to integrate the regional scientific research subset data, social media subset data, open subset data, and dialect subset data to obtain regionalized data.

[0016] Furthermore, in the aforementioned feasibility assessment system for the dissemination of science and culture based on regional analysis, the regional data acquisition module further includes the following sub-modules:

[0017] The cleaning and deduplication submodule is used to acquire regionalized data, remove duplicate and invalid data from the regionalized data, and filter noise content unrelated to the dissemination of science and culture in the regionalized data through an NLP natural language model to obtain cleaned regionalized data.

[0018] The standardization processing submodule is used to unify the time field in the cleaning regional data to the ISO standard, convert the geographic coordinates to the WGS84 coordinate system, and perform vectorization encoding on the unstructured data in the cleaning regional data to obtain standard regional data.

[0019] The missing value filling submodule is used to fill in the missing fields in the standard regionalized data by interpolating neighboring regional data and using the ARIMA time series prediction model to obtain the initial regionalized data.

[0020] Furthermore, in the aforementioned feasibility assessment system for the dissemination of science and culture based on regionalization analysis, the multi-dimensional feature extraction module includes the following sub-modules:

[0021] The spatial feature extraction submodule is used to construct four-dimensional feature labels including geographical, temporal, cultural features and scientific relevance. Based on the GIS geographic information system, the initial regionalized data is divided into grids, and the statistical values ​​of geographical attributes in each grid are calculated to generate a spatial feature matrix.

[0022] The time feature extraction submodule is used to perform sliding window analysis on the time series data in the initial regionalized data to extract trend items, periodic items and outliers, and obtain a time feature matrix.

[0023] The cultural feature extraction submodule is used to extract cultural keywords and scientific keywords from the initial regionalized data based on the LDA topic model, calculate the word frequency co-occurrence frequency and the region specificity index, and obtain the scientific and cultural feature matrix.

[0024] The feature integration submodule is used to integrate the spatial feature matrix, temporal feature matrix, and scientific and cultural feature matrix to obtain feature regionalization data.

[0025] Furthermore, in the aforementioned feasibility assessment system for the dissemination of science and culture based on regionalization analysis, the language depth analysis module includes the following sub-modules:

[0026] The emotion recognition submodule is used to score the positive and negative emotions of social media texts using a pre-trained RoBERTa fine-tuning model, identify the public's emotional attitude towards scientific topics, calculate the emotion intensity value, distinguish between mild concern and strong emotional expression, and obtain emotion data.

[0027] The language recognition submodule is used to build regional dialect speech recognition models. It uses an LSTM deep neural network to convert dialect audio into text, thus obtaining deep regionalized data.

[0028] Furthermore, in the aforementioned feasibility assessment system for the dissemination of science and culture based on regional analysis, the assessment model building module includes the following sub-modules:

[0029] The weight determination submodule is used to construct a local regression model using geographic coordinates as weights and generate a spatial weight matrix.

[0030] The data input submodule is used to input preprocessed multidimensional features, uses XGBoost's tree model to handle high-order interactions between features, and outputs a ranking of feature importance.

[0031] The model building submodule treats each geographic unit as a graph node. Node attributes include spatiotemporal and cultural features, while edge attributes are defined as the degree of correlation between regions. The propagation dependencies between regions are learned through GCN graph convolution operations to obtain a regionalized hybrid evaluation model.

[0032] Furthermore, in the aforementioned feasibility assessment system for the dissemination of science and culture based on regional analysis, the assessment model building module also includes the following sub-modules:

[0033] The data partitioning submodule is used for stratified sampling based on geographical location, dividing the data into training, validation, and test sets.

[0034] The model optimization submodule is used to adjust the bandwidth parameter of WR, the learning rate and tree depth of XGBoost, and the number of layers and embedding dimension of GNN using Bayesian optimization to obtain a regionalized hybrid evaluation model.

[0035] Furthermore, in the method for implementing a feasibility assessment system for the dissemination of science and culture based on regional analysis, the method includes the following steps:

[0036] Acquire scientific research databases, social media, open data, and dialect data from a specific region to obtain regionalized data; perform data preprocessing on the regionalized data to obtain initial regionalized data.

[0037] A four-dimensional feature label including geographical, temporal, cultural features and scientific relevance is constructed. Based on the four-dimensional feature label, multi-dimensional feature extraction is performed on the initial regionalized data to obtain feature regionalized data.

[0038] Sentiment analysis is performed on the feature regionalized data based on the RoBERTa fine-tuning model, and regional language feature maps are generated by combining the dialect sensitive word library to obtain deep regionalized data.

[0039] A regionalized hybrid evaluation model is established based on GWR geographic weighted regression, XGBoost and GNN graph neural network. A spatiotemporal Transformer model is introduced to process sequence data with strong spatiotemporal dependence, and the target regionalized hybrid evaluation model is obtained.

[0040] The deep regionalized data is input into the target regionalized hybrid assessment model to generate a heat map of propagation potential and a conflict early warning index.

[0041] Furthermore, in the method for implementing a feasibility assessment system for the dissemination of science and culture based on regional analysis, the method includes the following steps:

[0042] Obtain publicly available experimental data from databases and research institutions to obtain regional research subset data;

[0043] Collect text, image, and video content data from social media platforms within the region to obtain social media subset data;

[0044] By connecting to publicly available data platforms, data on population distribution, education level, and cultural facility layout in the region can be obtained, resulting in open subset data.

[0045] Obtain dialect speech data and language data from dialect databases in the region to obtain dialect subset data;

[0046] The regionalized data is obtained by integrating the regional scientific research subset data, social media subset data, open subset data, and dialect subset data.

[0047] Furthermore, in the method for implementing a feasibility assessment system for the dissemination of science and culture based on regional analysis, the method includes the following steps:

[0048] Acquire regionalized data, remove duplicate and invalid data from the regionalized data, and filter out noise content unrelated to the dissemination of science and culture from the regionalized data using an NLP natural language model to obtain cleaned regionalized data;

[0049] The time field in the cleaned regional data is standardized to ISO standard, the geographic coordinates are converted to WGS84 coordinate system, and the unstructured data in the cleaned regional data is vectorized to obtain standard regional data.

[0050] The missing fields in the standard regionalized data are filled by interpolating neighboring regional data and using the ARIMA time series prediction model to obtain the initial regionalized data.

[0051] Its beneficial effects lie in the following: First, by acquiring scientific research databases, social media, open data, and dialect data from a specific region, regionalized data is obtained. This regionalized data undergoes data preprocessing to obtain initial regionalized data. Second, a four-dimensional feature label is constructed, encompassing geographical, temporal, cultural characteristics, and scientific relevance. Based on this four-dimensional feature label, multi-dimensional feature extraction is performed on the initial regionalized data to obtain characteristic regionalized data. Third, sentiment analysis is conducted on the characteristic regionalized data using a RoBERTa fine-tuning model, and a regional language feature map is generated by combining it with a dialect-sensitive word database, resulting in deep regionalized data. Fourth, a regionalized hybrid evaluation model is established based on GWR geographical weighted regression, XGBoost, and GNN graph neural networks. Finally, the deep regionalized data is input into the target regionalized hybrid evaluation model to generate a heatmap of dissemination potential and a conflict warning index. 1. It fully considers the cultural dissemination possibilities of regional characteristics such as geographical environment, cultural context, and temporal patterns, improving the feasibility and accuracy of cultural dissemination. 2. Combining dialect and language analysis can avoid scientific cognitive biases behind emotional metaphors, focusing on emotional fluctuation characteristics and ensuring that the dissemination content aligns with the regional cultural and psychological needs. 3. It can conduct specific scientific and cultural dissemination based on the differences in acceptance levels in regions with different economic levels, thereby improving dissemination efficiency. Attached Figure Description

[0052] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0053] Figure 1 This is a schematic diagram of the first embodiment of the feasibility assessment system for the dissemination of science and culture based on regionalization analysis in this invention.

[0054] Figure 2 This is a schematic diagram of the second embodiment of the feasibility assessment system for the dissemination of science and culture based on regionalization analysis in this invention.

[0055] Figure 3 This is a schematic diagram of the third embodiment of the feasibility assessment system for the dissemination of science and culture based on regionalization analysis in this invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0057] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0058] The present invention will now be described in detail with reference to the accompanying drawings, such as... Figure 1 As shown, the feasibility assessment system for the dissemination of science and culture based on regional analysis includes the following modules:

[0059] 101. Regionalized data acquisition module, used to acquire scientific research databases, social media, open data and dialect data in a specific region to obtain regionalized data, and to perform data preprocessing on the regionalized data to obtain initial regionalized data;

[0060] Specifically, this embodiment also includes a scientific research data acquisition submodule, which is used to acquire publicly available experimental data from databases and scientific research institutions to obtain regional scientific research subset data;

[0061] The media data acquisition submodule is used to collect text, image, and video content data from social media platforms within the region to obtain social media subset data.

[0062] The open data acquisition submodule is used to connect to the public data platform to acquire data on population distribution, education level and cultural facility layout in the region, and obtain open subset data.

[0063] The dialect data acquisition submodule is used to acquire dialect speech data and language data from dialect databases in the region to obtain dialect subset data.

[0064] The data integration submodule is used to integrate regional scientific research subset data, social media subset data, open subset data, and dialect subset data to obtain regionalized data.

[0065] The cleaning and deduplication submodule is used to acquire regionalized data, remove duplicate and invalid data from the regionalized data, and filter out noise content that is not related to the dissemination of science and culture in the regionalized data through NLP natural language model to obtain cleaned regionalized data.

[0066] The standardization processing submodule is used to unify the time field in the cleaned regionalized data to the ISO standard, convert the geographic coordinates to the WGS84 coordinate system, and perform vectorization encoding on the unstructured data in the cleaned regionalized data to obtain standard regionalized data.

[0067] The missing value imputation submodule is used to fill in missing fields in standard regionalized data by interpolating neighboring regional data and using the ARIMA time series prediction model to obtain initial regionalized data.

[0068] Constructing a three-dimensional regional characteristic profile to break down the "communication context barrier".

[0069] Multi-source data integration achieves full coverage by innovatively combining scientific research data, social media UGC, government open data, and dialect data. In particular, for dialect data, field surveys and existing databases are used to complement each other, filling the gap in regional cultural semantic analysis.

[0070] For example, when disseminating agricultural technology in the first region, dialect recognition technology is used to analyze local expressions such as "corn planting," making the popular science content more relevant to farmers' cognitive habits.

[0071] 102. Multi-dimensional feature extraction module, used to construct four-dimensional feature labels including geographical, temporal, cultural features and scientific relevance, and to extract multi-dimensional features from the initial regionalized data based on the four-dimensional feature labels to obtain feature regionalized data;

[0072] Specifically, this embodiment also includes a spatial feature extraction submodule, which is used to construct a four-dimensional feature label including geographical, temporal, cultural features and scientific relevance. Based on the GIS geographic information system, the initial regional data is divided into grids, the statistical values ​​of geographical attributes in each grid are calculated, and a spatial feature matrix is ​​generated.

[0073] The time feature extraction submodule is used to perform sliding window analysis on the time series data in the initial regionalized data to extract trend terms, periodic terms and outliers, and obtain a time feature matrix.

[0074] The cultural feature extraction submodule is used to extract cultural keywords and scientific keywords from the initial regionalized data based on the LDA topic model, calculate the word frequency co-occurrence frequency and the region specificity index, and obtain the scientific and cultural feature matrix.

[0075] The feature integration submodule is used to integrate the spatial feature matrix, temporal feature matrix, and scientific and cultural feature matrix to obtain feature-regionalized data.

[0076] The four-dimensional feature tagging system accurately identifies pain points in communication through cross-analysis of various dimensions, including geographical features such as population density and urban-rural differences, temporal features such as festival cycles, cultural features such as dialect vocabulary distribution, and scientific relevance such as industry-technology matching. This allows for the quantification of regional communication's "advantageous conditions" and "constraining factors." For example, in the second region, the system identifies the problem of "high demand for water resource science popularization but low media contact rate during the summer busy farming season," guiding communication organizations to select suitable channels such as radio broadcasts and field explanations, thus addressing the problem of blind placement in traditional channels.

[0077] 103. Language Deep Analysis Module: This module is used to perform sentiment analysis on feature regionalized data based on the RoBERTa fine-tuning model, and generate regional language feature maps by combining dialect sensitive word databases to obtain deep regionalized data.

[0078] Specifically, this embodiment also includes an emotion recognition submodule, which uses a pre-trained RoBERTa fine-tuning model to score the positive and negative emotions of social media texts, identify the public's emotional attitudes toward scientific topics, calculate the emotional intensity value, distinguish between mild concern and strong emotional expression, and obtain emotional data.

[0079] The language recognition submodule is used to build a regional dialect speech recognition model. It uses an LSTM deep neural network to convert dialect audio into text, thus obtaining deep regionalized data.

[0080] Affective computing captures the public’s implicit needs. By analyzing the emotional intensity and tendency of social media texts, it can identify the potential concerns or interests of the regional public regarding specific scientific topics.

[0081] Dialect recognition technology promotes culturally adaptable dissemination by enabling dialect speech transcription and semantic analysis, supporting the transformation of professional scientific terms into dialectal expressions (interpreting "blockchain" as "ledger sharing technology"), and identifying scientific cognitive misconceptions in dialects (dialectal colloquialisms that incorrectly attribute natural phenomena).

[0082] 104. Evaluation model building module, used to build a regionalized hybrid evaluation model based on GWR geographic weighted regression, XGBoost and GNN graph neural network, introduces spatiotemporal Transformer model to process sequence data with strong spatiotemporal dependence, and obtains target regionalized hybrid evaluation model.

[0083] Specifically, this embodiment also includes a weight determination submodule, which is used to construct a local regression model using geographic coordinates as weights and generate a spatial weight matrix;

[0084] The data input submodule is used to input preprocessed multidimensional features, uses XGBoost's tree model to handle high-order interactions between features, and outputs a ranking of feature importance.

[0085] The model building submodule treats each geographic unit as a graph node. Node attributes include spatiotemporal and cultural features, while edge attributes are defined as the degree of correlation between regions. The propagation dependencies between regions are learned through GCN graph convolution operations to obtain a regionalized hybrid evaluation model.

[0086] GWR captures the spatial heterogeneity of propagation effects by constructing a local model based on geographic weighted regression, which can quantify the differences in the effects of the same influencing factor in different regions.

[0087] GNN modeling of regional correlations and propagation paths: By analyzing the correlations between regions such as population flow and transportation connections through graph neural networks, the risk of cross-regional spread of scientific controversies can be predicted.

[0088] Core objective: To directly and efficiently model and learn dynamic sequence data with strong spatiotemporal dependencies.

[0089] Key technology: Transformer's self-attention mechanism:

[0090] Capturing Long-Term Dependencies: The core advantage of Transformers is their ability to model the relationship between any two elements in a sequence (regardless of distance) through a self-attention mechanism. In a spatiotemporal sequence, this means it can: learn the association between the current mood of a region and its own long-past mood; and learn the association between the current mood of a region and the mood of another region at a specific point in the past.

[0091] Parallel computing efficiency: Compared to recurrent networks such as RNN / LSTM that require sequential processing, Transformer can process the entire sequence in parallel, resulting in faster training speed.

[0092] 105. Feasibility assessment module, used to input in-depth regional data into the target regional hybrid assessment model to generate a heat map of propagation potential and a conflict early warning index.

[0093] Specifically, this embodiment also includes a data partitioning submodule, which is used for stratified sampling according to regional geographical location to divide the training set, validation set and test set;

[0094] The model optimization submodule is used to adjust the bandwidth parameter of WR, the learning rate and tree depth of XGBoost, and the number of layers and embedding dimension of GNN using Bayesian optimization to obtain a regionalized hybrid evaluation model.

[0095] Its beneficial effects lie in the following: First, by acquiring research databases, social media, open data, and dialect data from specific regions, regionalized data is obtained. This regionalized data undergoes preprocessing to produce initial regionalized data. Second, a four-dimensional feature label is constructed, encompassing geographical, temporal, cultural characteristics, and scientific relevance. Based on this four-dimensional feature label, multi-dimensional feature extraction is performed on the initial regionalized data to obtain characteristic regionalized data. Third, sentiment analysis is conducted on the characteristic regionalized data using a RoBERTa fine-tuning model. Finally, a regional language feature map is generated by combining this with a dialect-sensitive word database, resulting in deep regionalized data. Fourth, a regionalized hybrid evaluation model is established based on GWR geographical weighted regression, XGBoost, and GNN graph neural networks. Finally, the deep regionalized data is input into the regionalized hybrid evaluation model to generate a heatmap of dissemination potential and a conflict warning index. 1. It fully considers the cultural dissemination possibilities of regional characteristics such as geographical environment, cultural context, and temporal patterns, improving the feasibility and accuracy of cultural dissemination. 2. Combining dialect and language analysis can avoid scientific cognitive biases behind emotional metaphors, focusing on emotional fluctuation characteristics and ensuring that the dissemination content aligns with the regional cultural and psychological needs. 3. It can conduct specific scientific and cultural dissemination based on the differences in acceptance levels in regions with different economic levels, thereby improving dissemination efficiency.

[0096] Please see Figure 2 In the feasibility assessment system for the dissemination of science and culture based on regional analysis, the regional data acquisition module includes the following sub-modules:

[0097] The scientific research data acquisition submodule is used to acquire publicly available experimental data from databases and scientific research institutions to obtain regional scientific research subset data.

[0098] The media data acquisition submodule is used to collect text, image, and video content data from social media platforms within the region to obtain social media subset data.

[0099] The open data acquisition submodule is used to connect to the public data platform to acquire data on population distribution, education level and cultural facility layout in the region, and obtain open subset data.

[0100] The dialect data acquisition submodule is used to acquire dialect speech data and language data from dialect databases in the region to obtain dialect subset data.

[0101] The data integration submodule is used to integrate regional scientific research subset data, social media subset data, open subset data, and dialect subset data to obtain regionalized data.

[0102] Please seeFigure 3 In the feasibility assessment system for the dissemination of science and culture based on regional analysis, the multi-dimensional feature extraction module includes the following sub-modules:

[0103] The spatial feature extraction submodule is used to construct four-dimensional feature labels including geographical, temporal, cultural features and scientific relevance. Based on the GIS geographic information system, the initial regionalized data is divided into grids, the statistical values ​​of geographical attributes in each grid are calculated, and a spatial feature matrix is ​​generated.

[0104] The time feature extraction submodule is used to perform sliding window analysis on the time series data in the initial regionalized data to extract trend terms, periodic terms and outliers, and obtain a time feature matrix.

[0105] The cultural feature extraction submodule is used to extract cultural keywords and scientific keywords from the initial regionalized data based on the LDA topic model, calculate the word frequency co-occurrence frequency and the region specificity index, and obtain the scientific and cultural feature matrix.

[0106] The feature integration submodule is used to integrate the spatial feature matrix, temporal feature matrix, and scientific and cultural feature matrix to obtain feature-regionalized data.

[0107] Specifically, the present invention can also be implemented using the following methods:

[0108] I. Regionalized Data Acquisition and Preprocessing:

[0109] 1. Multi-source data acquisition:

[0110] Scientific Research Database: It selectively captures academic achievements (papers, patents, technical reports) from universities and research institutions within the region, integrates open data from local key laboratories, and synchronously accesses regional subset data from general databases such as CNKI and WebofScience.

[0111] Social media data: Collect text, image, and video content from mainstream social media platforms (local forums, government new media, and vertical community groups) within the region, with a focus on user-generated content (UGC) and officially released content (OGC) related to science topics.

[0112] Open data: Connect with government open data platforms (data from regional statistics bureaus, science and technology bureaus, and culture and tourism bureaus) to obtain structured data such as population distribution, education level, layout of cultural facilities, and science and technology policy documents.

[0113] Dialect data: Collect dialect speech data (local colloquialisms and popular science dialect expressions) through field surveys, integrate existing dialect databases (data from the China Language Resources Protection Project), and supplement materials on ethnic minority languages ​​or regional characteristic languages ​​in the region.

[0114] 2. Data preprocessing:

[0115] Cleaning and deduplication: Remove duplicate records and invalid data (garbled text, incorrect formatting), and filter out noise content (pure entertainment topics) that is irrelevant to the dissemination of science and culture through natural language processing (NLP).

[0116] Standardization processing: Unify data format (time field is unified to ISO standard, geographic coordinates are converted to WGS84 coordinate system), and perform vectorization encoding (TF-IDF, word embedding) on ​​unstructured data (text, audio).

[0117] Missing value imputation: Missing fields are filled by interpolating neighboring data, using time series forecasting models (ARIMA), or expert knowledge rules to ensure data integrity.

[0118] II. Four-dimensional feature label design and spatiotemporal feature extraction:

[0119] 1. Four-dimensional feature labeling system:

[0120] Geographical characteristics: Spatial attributes such as geographical coordinates (latitude and longitude), topography (plains / mountains / water), population density, and urban-rural differences (urbanization rate) are used to locate the physical carriers and audience base for communication.

[0121] Time characteristics: encompassing timestamps (accurate to the hour / day / month / year) and periodic patterns (busy farming seasons, summer and winter vacations, local festivals) to capture the timeliness and seasonality of dissemination.

[0122] Cultural characteristics: Extract dialect features (vocabulary differences, pronunciation tones), folk traditions, cultural identity (attention to local science and technology celebrities), and education level to assess the suitability of the disseminated content to the local cultural context.

[0123] 2. Spatiotemporal feature extraction methods:

[0124] Spatial dimension: The data is divided into grids (1km×1km grids) based on Geographic Information System (GIS), and the statistical values ​​of geographic attributes (average population density, cultural facility coverage) within each grid are calculated to generate a spatial feature matrix.

[0125] Time dimension: Perform sliding window analysis on time series data (aggregate by week / month) to extract trend items (annual growth in the volume of discussions on scientific topics) and periodic items (peak dissemination before and after holidays).

[0126] Cultural and scientific dimensions: Cultural keywords (dialect-specific vocabulary) and scientific keywords ("artificial intelligence" and "agricultural technology") are extracted from unstructured data using text mining techniques (topic modeling LDA), and word frequency, co-occurrence frequency, and region-specificity index are calculated.

[0127] III. Deep Language Analysis, Affective Computing, and Dialect Recognition:

[0128] 1. Affective computing module:

[0129] Sentiment Classification: Using a pre-trained sentiment analysis model (BERT sentiment classifier), positive and negative sentiment scores are assigned to social media texts to identify public sentiment attitudes toward scientific topics.

[0130] Emotional intensity quantification: Combines emojis, interjections, and punctuation marks (frequency of exclamation marks) to calculate emotional intensity values, distinguishing between mild concern and strong emotional expressions (anger or excitement), and identifying sensitive content that may cause controversy.

[0131] Emotional dissemination path analysis: Tracking the phenomenon of emotional polarization (the spread of opposing viewpoints on scientific controversies) and locating key dissemination nodes (the influence of opinion leaders' emotional inclinations on the group).

[0132] 2. Dialect recognition technology:

[0133] Speech-to-text conversion: Construct a regional dialect speech recognition model and convert dialect audio into text using a deep neural network (LSTM-CTC), supporting semantic parsing of multimodal data (dialect explanations in short videos).

[0134] Dialect vocabulary detection: Based on a rule engine (dialect dictionary matching) and a statistical model (n-gram language model), dialect words in text are identified, their regional attributes are labeled, and mixed expressions of Mandarin and dialect are distinguished.

[0135] Cultural semantic analysis: Combining local knowledge bases to analyze the deeper cultural meanings of dialect texts, such as identifying localized metaphors for scientific concepts in dialects.

[0136] IV. Construction of a Regionalized Hybrid Assessment Model:

[0137] 1. Model Architecture Design:

[0138] Spatial Heterogeneity Modeling (GWR Geographically Weighted Regression): Using geographic coordinates as weights, a local regression model is constructed to capture the spatial differences in the effectiveness of scientific and cultural dissemination in different regions (differences in dissemination efficiency between mountainous and plain areas), generating a spatial weight matrix.

[0139] Nonlinear Feature Fusion (XGBoost): Input preprocessed multidimensional features (geographical, temporal, cultural, and scientific relevance), use XGBoost's tree model to process high-order interactions between features (the synergistic effect of "dialect popular science videos + off-season farming" on dissemination effectiveness), output feature importance ranking, and screen key influencing factors.

[0140] Regional association modeling (GNN graph neural network): Each geographical unit (township) is regarded as a graph node. The node attributes include spatiotemporal features and cultural features. The edge attributes are defined as the degree of association between regions (population flow, transportation connection). The propagation dependency between regions (the risk of scientific controversy events spreading in neighboring areas) is learned through graph convolution operation (GCN).

[0141] 2. Model Training and Optimization:

[0142] Data partitioning: Stratified sampling based on geographical location to divide the data into training set (70%), validation set (20%), and test set (10%) to avoid model bias caused by spatial autocorrelation.

[0143] Parameter tuning: Grid search or Bayesian optimization was used to adjust the bandwidth parameter of GWR, the learning rate and tree depth of XGBoost, and the number of layers and embedding dimension of GNN, with the propagation effect prediction accuracy (root mean square error RMSE) and conflict warning accuracy as objective functions.

[0144] Enhanced interpretability: The decision logic of the XGBoost model is analyzed through SHAP values, the spatial weight distribution of GWR is visualized, and a report on the contribution of each region's characteristics to the propagation potential is generated.

[0145] V. Results Output and Application:

[0146] 1. Heat map of dissemination potential:

[0147] Visualization format: Based on a GIS overlay geographic base map, the potential value for the dissemination of science and culture in each region is displayed through color gradients (red indicates high potential and blue indicates low potential). Layers such as population density and distribution of cultural facilities are overlaid to intuitively present key areas such as "high potential - high demand" and "low potential - high demand".

[0148] Application scenario: To assist in the formulation of regional dissemination strategies and identify "dissemination blind spots".

[0149] 2. Conflict Early Warning Index:

[0150] Indicator system: Based on the comprehensive emotional polarization level (difference between positive and negative emotions), the probability of cultural conflict, and the popularity of scientific controversies, an early warning index of 0-100 is constructed (the higher the value, the greater the risk).

[0151] Early warning mechanism: When a three-level early warning threshold is set, it automatically associates the conflict type (cultural misunderstanding, technical dispute) and the area where the conflict occurs, pushes a risk report to the communication management department, and prompts intervention measures.

[0152] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A feasibility assessment system for the dissemination of science and culture based on regional analysis, characterized in that: The feasibility assessment system for the dissemination of science and culture includes the following modules: The regional data acquisition module is used to acquire scientific research databases, social media, open data and dialect data in a specific region to obtain regional data, and to perform data preprocessing on the regional data to obtain initial regional data; A multi-dimensional feature extraction module is used to construct four-dimensional feature labels including geographical, temporal, cultural features and scientific relevance. Based on the four-dimensional feature labels, multi-dimensional feature extraction is performed on the initial regionalized data to obtain feature regionalized data. The language deep analysis module is used to perform sentiment analysis on the feature regional data based on the RoBERTa fine-tuning model, and generate regional language feature maps by combining the dialect sensitive word library to obtain deep regional data. The evaluation model building module is used to build a regionalized hybrid evaluation model based on GWR geographic weighted regression, XGBoost and GNN graph neural network. It introduces the spatiotemporal Transformer model to process sequence data with strong spatiotemporal dependence and obtains the target regionalized hybrid evaluation model. The feasibility assessment module is used to input the deep regionalized data into the target regionalized hybrid assessment model to generate a heat map of propagation potential and a conflict early warning index. The multi-dimensional feature extraction module Includes the following sub-modules: The spatial feature extraction submodule is used to construct four-dimensional feature labels including geographical, temporal, cultural features and scientific relevance. Based on the GIS geographic information system, the initial regionalized data is divided into grids, and the statistical values ​​of geographical attributes in each grid are calculated to generate a spatial feature matrix. The time feature extraction submodule is used to perform sliding window analysis on the time series data in the initial regionalized data to extract trend items, periodic items and outliers, and obtain a time feature matrix. The cultural feature extraction submodule is used to extract cultural keywords and scientific keywords from the initial regionalized data based on the LDA topic model, calculate the word frequency co-occurrence frequency and the region specificity index, and obtain the scientific and cultural feature matrix. The feature integration submodule is used to integrate the spatial feature matrix, temporal feature matrix, and scientific and cultural feature matrix to obtain feature regionalization data; The language deep analysis module includes the following sub-modules: The emotion recognition submodule is used to score the positive and negative emotions of social media texts using a pre-trained RoBERTa fine-tuning model, identify the public's emotional attitude towards scientific topics, calculate the emotion intensity value, distinguish between mild concern and strong emotional expression, and obtain emotion data. The language recognition submodule is used to build a regional dialect speech recognition model. It uses an LSTM deep neural network to convert dialect audio into text and obtain deep regionalized data. The evaluation model building module includes the following sub-modules: The weight determination submodule is used to construct a local regression model using geographic coordinates as weights and generate a spatial weight matrix. The data input submodule is used to input preprocessed multidimensional features, uses XGBoost's tree model to handle high-order interactions between features, and outputs a ranking of feature importance. The model building submodule treats each geographic unit as a graph node. Node attributes include spatiotemporal and cultural features, while edge attributes are defined as the degree of correlation between regions. The propagation dependencies between regions are learned through GCN graph convolution operations to obtain a regionalized hybrid evaluation model.

2. The feasibility assessment system for the dissemination of science and culture based on regionalization analysis as described in claim 1, characterized in that, The regionalized data acquisition module includes the following sub-modules: The scientific research data acquisition submodule is used to acquire publicly available experimental data from databases and scientific research institutions to obtain regional scientific research subset data. The media data acquisition submodule is used to collect text, image, and video content data from social media platforms within the region to obtain social media subset data. The open data acquisition submodule is used to connect to the public data platform to acquire data on population distribution, education level and cultural facility layout in the region, and obtain open subset data. The dialect data acquisition submodule is used to acquire dialect speech data and language data from dialect databases in the region to obtain dialect subset data. The data integration submodule is used to integrate the regional scientific research subset data, social media subset data, open subset data, and dialect subset data to obtain regionalized data.

3. The feasibility assessment system for the dissemination of science and culture based on regionalization analysis as described in claim 2, characterized in that, The regionalized data acquisition module also includes the following sub-modules: The cleaning and deduplication submodule is used to acquire regionalized data, remove duplicate and invalid data from the regionalized data, and filter noise content unrelated to the dissemination of science and culture in the regionalized data through an NLP natural language model to obtain cleaned regionalized data. The standardization processing submodule is used to unify the time field in the cleaning regional data to the ISO standard, convert the geographic coordinates to the WGS84 coordinate system, and perform vectorization encoding on the unstructured data in the cleaning regional data to obtain standard regional data. The missing value filling submodule is used to fill in the missing fields in the standard regionalized data by interpolating neighboring regional data and using the ARIMA time series prediction model to obtain the initial regionalized data.

4. The feasibility assessment system for the dissemination of science and culture based on regionalization analysis as described in claim 1, characterized in that, The evaluation model building module also includes the following sub-modules: The data partitioning submodule is used for stratified sampling based on geographical location, dividing the data into training, validation, and test sets. The model optimization submodule is used to adjust the bandwidth parameter of WR, the learning rate and tree depth of XGBoost, and the number of layers and embedding dimension of GNN using Bayesian optimization to obtain a regionalized hybrid evaluation model.

5. A method for implementing the feasibility assessment system for the dissemination of science and culture based on regional analysis as described in claim 1, characterized in that, The method includes the following steps: Acquire scientific research databases, social media, open data, and dialect data from a specific region to obtain regionalized data; perform data preprocessing on the regionalized data to obtain initial regionalized data. A four-dimensional feature label including geographical, temporal, cultural features and scientific relevance is constructed. Based on the four-dimensional feature label, multi-dimensional feature extraction is performed on the initial regionalized data to obtain feature regionalized data. Sentiment analysis is performed on the feature regionalized data based on the RoBERTa fine-tuning model, and regional language feature maps are generated by combining the dialect sensitive word library to obtain deep regionalized data. A regionalized hybrid evaluation model is established based on GWR geographic weighted regression, XGBoost and GNN graph neural network. A spatiotemporal Transformer model is introduced to process sequence data with strong spatiotemporal dependence, and the target regionalized hybrid evaluation model is obtained. The deep regionalized data is input into the target regionalized hybrid assessment model to generate a heat map of propagation potential and a conflict early warning index.

6. A method for implementing the feasibility assessment system for the dissemination of science and culture based on regionalization analysis as described in claim 1, characterized in that, The method further includes the following steps: Obtain publicly available experimental data from databases and research institutions to obtain regional research subset data; Collect text, image, and video content data from social media platforms within the region to obtain social media subset data; By connecting to publicly available data platforms, data on population distribution, education level, and cultural facility layout in the region can be obtained, resulting in open subset data. Obtain dialect speech data and language data from dialect databases in the region to obtain dialect subset data; The regionalized data is obtained by integrating the regional scientific research subset data, social media subset data, open subset data, and dialect subset data.

7. A method for implementing the feasibility assessment system for the dissemination of science and culture based on regional analysis as described in claim 1, characterized in that, The method further includes the following steps: Acquire regionalized data, remove duplicate and invalid data from the regionalized data, and filter out noise content unrelated to the dissemination of science and culture from the regionalized data using an NLP natural language model to obtain cleaned regionalized data; The time field in the cleaned regional data is standardized to ISO standard, the geographic coordinates are converted to WGS84 coordinate system, and the unstructured data in the cleaned regional data is vectorized to obtain standard regional data. The missing fields in the standard regionalized data are filled by interpolating neighboring regional data and using the ARIMA time series prediction model to obtain the initial regionalized data.

Images