Convolutional neural network-based intelligent text and creation block scenarized content generation method

Abstract

The invention relates to an intelligent text and creation block scenarized content generation method based on a convolutional neural network. The method comprises the following steps: acquiring a two-dimensional plane map, culture point locations and culture assets of a corresponding block to obtain a block culture map database; based on the block culture map database, calculating a culture value density score of each space grid, and generating a culture density network map; based on the culture density network diagram, analyzing the spatial aggregation degree of the culture assets to obtain an aggregation degree thermodynamic diagram, and based on the block culture map database, calculating the association degree between the culture point locations to generate a culture point location theme association degree matrix; generating a narrative touring path based on the aggregation degree thermodynamic diagram and the cultural point location theme association degree matrix; and based on the narrative touring path, determining the content needing to be generated at each culture point location, and obtaining the digital scene content. By adopting the method, the short-distance strong correlation effect can be captured, the long-distance weak correlation diffusion path can be analyzed, and the scene content with guidance can be generated.

Description

Technical Field

[0001] This invention belongs to the field of intelligent scene generation, and in particular relates to a method for generating scene-based content for smart cultural and creative blocks based on convolutional neural networks. Background Technology

[0002] With the deep integration of spatial information technology and artificial intelligence, spatial pattern recognition technology, with convolutional neural networks at its core, has made significant progress. This technology can automatically extract deep features and distribution patterns from complex spatial data, providing a new paradigm for the value assessment and intelligent integration of cultural resources, and also prompting a revolution in the planning of traditional cultural and creative district tour content. Traditionally, the planning of tour routes and content in cultural and creative districts relies mainly on human experience and static resource listing. Planners typically use scattered historical information and subjective aesthetics to simply linearly connect and fill in cultural points within the district. This approach focuses on the independent presentation of individual attractions, lacking a systematic analysis of the inherent network structure of the district's cultural resources. However, current traditional planning methods have significant problems. First, they cannot scientifically quantify the intrinsic connections between different cultural themes, making it difficult to uncover the deep cultural threads between scattered points, resulting in a fragmented tour experience and a lack of coherent cultural narrative. Secondly, traditional methods cannot accurately identify the spatial clustering characteristics and radiation effects of cultural hotspots. They cannot effectively distinguish between primary and secondary hotspots, nor can they generate paths that can scientifically guide tourists to move efficiently between multiple hotspots. This results in blind planning of tour routes and a lack of hierarchy and logic in the overall generated scenario-based content. Summary of the Invention

[0003] Therefore, it is necessary to address the aforementioned technical issues by providing a method for generating contextualized content for smart cultural and creative blocks based on convolutional neural networks. This method can accurately identify the spatial clustering characteristics and radiation effects of cultural hotspots, effectively distinguish between primary and secondary hotspots, and generate guiding contextualized content.

[0004] Firstly, this application provides a method for generating contextualized content for smart cultural and creative blocks based on convolutional neural networks, including:

[0005] Collect two-dimensional planar maps, cultural sites, and cultural assets of the corresponding blocks, and bind each cultural site with its corresponding cultural asset to obtain a block cultural map database;

[0006] The two-dimensional planar map is divided into spatial grids, and based on the neighborhood cultural map database, the cultural value density score of each spatial grid is calculated to generate a cultural density network map.

[0007] Based on the cultural density network diagram, the spatial clustering of cultural assets is analyzed to obtain a clustering heat map. Based on the neighborhood cultural map database, the correlation between cultural sites is calculated to generate a cultural site theme correlation matrix.

[0008] A narrative tour route is generated based on the clustering heatmap and the theme relevance matrix of cultural sites.

[0009] Based on the narrative tour path, the content to be generated at each cultural site is determined, resulting in digital scene content.

[0010] Furthermore, based on the cultural density network diagram, the spatial clustering of cultural assets is analyzed to obtain a clustering heatmap, including:

[0011] Define multiple multi-scale convolution kernels, and perform sliding window convolution on the cultural value density grid map to calculate the response value of each of the multi-scale convolution kernels, thereby obtaining multiple cultural space response maps;

[0012] Based on the response maps of all cultural spaces, the corresponding response values ​​are weighted and fused to calculate the comprehensive clustering degree, thus obtaining the comprehensive clustering degree distribution map;

[0013] Normalize the overall clustering degree in the overall clustering degree distribution map to obtain the normalized clustering degree distribution map;

[0014] Based on preset color mapping rules, the normalized clustering distribution map is mapped to the corresponding color to obtain the clustering heatmap.

[0015] Furthermore, based on the neighborhood cultural map database, the correlation between cultural sites is calculated, generating a cultural site theme correlation matrix, including:

[0016] Based on the word embedding model, the text descriptions of cultural points in the corresponding street cultural map database are transformed into feature vectors to obtain topic semantic vectors;

[0017] The aggregated topic vector is obtained by averaging all thematic semantic vectors corresponding to cultural locations.

[0018] Based on the aggregated topic vectors, the topic vector similarity between cultural locations is calculated using the following formula to obtain a list of topic vector similarities:

[0019]

[0020] in, Let represent the similarity of topic vectors between cultural point i and cultural point j, where n is the dimension of the topic vectors and k is the vector index. Let be the k-dimensional component value of the aggregated topic vector of cultural point i. Let be the component of the aggregated topic vector of cultural point j in the k-dimensional dimension;

[0021] Based on the list of topic vector similarity, a topic association matrix of cultural sites is generated; each element in the topic association matrix of cultural sites represents the degree of association between cultural sites.

[0022] Furthermore, based on the clustering heatmap and the cultural site theme correlation matrix, a narrative tour route is generated, including:

[0023] Based on the clustering heatmap and the theme correlation matrix of cultural sites, an optimization proposition is defined with the order of visiting cultural sites as the decision variable.

[0024] Solve the optimization problem to obtain a solution set, and map the solution set to the corresponding cultural points to obtain candidate paths;

[0025] Based on natural language generation, names and descriptions are generated for candidate paths to obtain narrative tour paths.

[0026] Furthermore, based on the clustering heatmap and the theme correlation matrix of cultural sites, and using the order of visiting cultural sites as the decision variable, optimization propositions are defined, including:

[0027] Based on the clustering heatmap, the sum of the spatial value of the path is defined as the spatial value objective function, and based on the neighborhood cultural map database, the total distance of the path is defined as the travel efficiency objective function; the path is the order in which cultural points are visited.

[0028] Based on the theme relevance matrix of cultural sites, calculate the objective function of local theme coherence and the objective function of global theme consistency;

[0029] Based on the spatial value objective function, the travel efficiency objective function, the local theme coherence objective function, and the global theme consistency objective function, a comprehensive objective function is constructed using the following formula:

[0030]

[0031] in, Let P be the overall objective function, and P be the path sequence. For spatial value objective function, For the objective function of local topic coherence, For the objective function of local topic coherence, Let the objective function be the travel efficiency. , , , These are the weighting coefficients;

[0032] Based on the neighborhood cultural graph database, the constraints in path planning are transformed into functional forms to obtain the constraint conditions; the constraint conditions include the starting point constraint, path length constraint, unique access constraint, and connectivity constraint.

[0033] By integrating the objective function and constraints, an optimization proposition is obtained.

[0034] Furthermore, based on the neighborhood cultural map database, the cultural value density score of each spatial grid is calculated to generate a cultural density network map, including:

[0035] Based on the neighborhood cultural map database, each cultural location is spatially matched with a spatial grid to obtain an attribution mapping table;

[0036] Based on the neighborhood cultural map database, weight coefficients are assigned to different types of cultural assets to obtain a weight comparison table;

[0037] Based on the attribution mapping table and weight comparison table, the cultural value density of each spatial grid is calculated to obtain the initial matrix of cultural value density.

[0038] The initial matrix of cultural value density is normalized to obtain the cultural density network diagram.

[0039] Furthermore, based on the attribution mapping table and weight lookup table, the cultural value density of each spatial grid is calculated to obtain an initial cultural value density matrix, including:

[0040] Based on the attribution mapping table, query the cultural assets bound to the cultural points corresponding to the spatial grid to obtain a list of cultural assets;

[0041] Iterate through each cultural asset in the cultural asset list, query the weight comparison table, and obtain the corresponding weight coefficient;

[0042] The cultural value density is obtained by summing all the weighting coefficients arithmetically.

[0043] By integrating the cultural value densities corresponding to all spatial grids, an initial matrix of cultural value densities is obtained.

[0044] Secondly, this application also provides a smart cultural and creative block scene-based content generation device based on convolutional neural networks, including:

[0045] The data acquisition module is used to collect two-dimensional planar maps, cultural sites, and cultural assets of the corresponding blocks, and bind each cultural site with its corresponding cultural asset to obtain a block cultural map database.

[0046] The density module is used to divide a two-dimensional planar map into spatial grids and, based on a neighborhood cultural map database, calculate the cultural value density score of each spatial grid to generate a cultural density network map.

[0047] The clustering module is used to analyze the spatial clustering of cultural assets based on the cultural density network diagram, obtain a clustering heat map, and calculate the correlation between cultural sites based on the neighborhood cultural map database, generating a cultural site theme correlation matrix.

[0048] The path module is used to generate narrative tour paths based on clustering heatmaps and cultural site theme correlation matrices.

[0049] The scene module is used to determine the content to be generated at each cultural site based on the narrative tour path, thus obtaining digital scene content.

[0050] Thirdly, this application also provides a computer device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement any step of the method provided in the first aspect of this application.

[0051] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any step of the method provided in the first aspect of this application.

[0052] The aforementioned method for generating scenario-based content for smart cultural and creative blocks based on convolutional neural networks involves: collecting a two-dimensional planar map, cultural sites, and cultural assets of the corresponding block; binding each cultural site with its corresponding cultural asset to obtain a block cultural map database; dividing the two-dimensional planar map into spatial grids; calculating the cultural value density score of each spatial grid based on the block cultural map database to generate a cultural density network map; analyzing the spatial clustering of cultural assets based on the cultural density network map to obtain a clustering heatmap; calculating the correlation between cultural sites based on the block cultural map database to generate a cultural site theme correlation matrix; generating a narrative tour path based on the clustering heatmap and the cultural site theme correlation matrix; and determining the content to be generated at each cultural site based on the narrative tour path to obtain digital scene content. It can effectively capture and model the dynamic impact diffusion path of a hotspot area on the surrounding areas, quantify the intrinsic correlation between different cultural themes, and avoid the problem of the lack of coherent cultural narrative in the generated scenario-based content. It can capture the strong correlation effect at close range, and analyze the weak correlation diffusion path at long distance. It can accurately identify the spatial clustering characteristics and radiation effect of cultural hotspots, and scientifically guide tourists to move efficiently between multiple hotspots in scenario-based content, so as to achieve the sense of hierarchy and logic of scenario-based content. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a schematic diagram illustrating the process of a method for generating scenario-based content in a smart cultural and creative block based on a convolutional neural network, according to an embodiment of the present invention.

[0055] Figure 2 A schematic diagram of the structure of a smart cultural and creative block scene-based content generation device based on a convolutional neural network provided in an embodiment of the present invention; Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0057] In one embodiment, such as Figure 1 As shown, a method for generating contextualized content for smart cultural and creative blocks based on convolutional neural networks is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and furthermore, to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0058] Step 101: Collect the two-dimensional planar map, cultural points and cultural assets of the corresponding block, and bind each cultural point and the corresponding cultural asset to obtain the block cultural map database.

[0059] The two-dimensional planar map refers to a standard two-dimensional electronic map of the target block, containing basic geographic information such as roads, buildings, and rivers, serving as the reference coordinate system for subsequent spatial analysis. Cultural sites refer to specific geographical locations within the block that possess cultural value, such as a historical building, a museum, a memorial square, or an archaeological site; each site has clearly defined geographic coordinates. Cultural assets refer to specific cultural content entities or abstract concepts associated with cultural sites. Binding refers to establishing associations between cultural sites and one or more corresponding cultural assets at the data level. The block cultural map database is a structured database that not only stores basic information about cultural sites and cultural assets but also stores binding relationships, forming a knowledge graph describing the block's cultural resources and their interrelationships. The terminal collects two-dimensional electronic maps of the target street area through various methods such as geographic information system data, public map platforms, field surveys, and literature and archives. It identifies and records all important cultural sites and their geographical locations. For each identified cultural site, it conducts a detailed investigation and lists all related cultural assets, and collects detailed descriptions, historical backgrounds, multimedia materials, etc. of these assets. In database or knowledge graph modeling tools, each cultural site is treated as the subject, and its corresponding cultural assets are treated as related objects. Clear ownership, association, and other relationship links are established, thereby connecting all discrete data into an organic network.

[0060] Step 102: Divide the two-dimensional planar map into spatial grids, and calculate the cultural value density score of each spatial grid based on the neighborhood cultural map database to generate a cultural density network map.

[0061] Specifically, a spatial grid refers to dividing a two-dimensional planar map into multiple regular squares according to a certain resolution. Each grid is the basic spatial unit for subsequent calculations. The cultural value density score is a numerical value used to quantify the level of cultural value within each spatial grid; a higher score indicates a higher concentration of cultural value within that grid. The cultural density network map is a matrix or raster map where the value of each grid cell is its cultural value density score, which can be visualized as a grid map where different shades of color represent varying densities. Based on a selected grid size, the terminal divides the entire street map into a uniform array of spatial grids. It iterates through each grid, finding all cultural points geographically located within that grid according to the street's cultural map database. Different weights are assigned to these cultural assets based on their importance and scarcity. For each grid, the weights of all cultural assets associated with its cultural points are summed to obtain the cultural value density score for that grid. The density scores of all grids are then integrated to form a complete cultural density network map. The scores are normalized to a range between 0 and 1 for easier comparison and subsequent calculations.

[0062] Step 103: Based on the cultural density network diagram, analyze the spatial clustering of cultural assets to obtain a clustering heat map, and based on the neighborhood cultural map database, calculate the correlation between cultural sites to generate a cultural site theme correlation matrix.

[0063] Specifically, spatial clustering refers to the degree to which cultural assets are not evenly distributed in space, but rather tend to cluster together; high-cluster areas are cultural hotspots. A clustering heatmap is a data visualization that uses color to visually display the degree of spatial clustering on a two-dimensional map, with red areas representing core areas where cultural assets are highly concentrated. Correlation refers to the strength of similarity or relevance between different cultural sites in terms of cultural theme, historical background, and semantic content. A cultural site theme correlation matrix is ​​a square matrix where rows and columns represent all cultural sites. The element in the i-th row and j-th column is a numerical value representing the degree of correlation between cultural site i and cultural site j. The terminal uses multiple windows of different sizes to slide and scan on the cultural density network map, calculating the cluster response value of each location at different scales, identifying cultural clusters of different sizes, and weighting and fusing the response values ​​calculated at multiple scales to obtain a comprehensive clustering distribution map. This map is then normalized and mapped onto a predefined color band to generate a clustering heatmap. Using natural language processing technology, the text descriptions of each cultural location and its assets in the neighborhood cultural map database are converted into a series of numerical feature vectors. These vectors capture the semantic information of the text. All relevant text feature vectors for a cultural location are averaged or combined to form an aggregated theme vector representing the overall cultural theme of that location. The cosine similarity between the aggregated theme vectors of every two cultural locations is calculated; a similarity value between 0 and 1 indicates the correlation between the locations. All correlation values ​​between point pairs are filled into a matrix to form a cultural location theme correlation matrix.

[0064] Step 104: Generate a narrative tour route based on the clustering heatmap and the cultural site theme correlation matrix.

[0065] The narrative tour route is an optimized tour sequence that considers the route's cultural value, thematic coherence, and tour efficiency, and possesses a storyline or theme name, making the tour experience feel like reading a story. The terminal models the route generation problem as a multi-objective optimization problem, with the decision variable being the order of visits to cultural points. For example, the objective functions include: maximizing spatial value (the route should pass through high-value areas on the clustering heatmap as much as possible); maximizing thematic coherence (the correlation between adjacent points in the route should be as high as possible, and the theme of the entire route should be consistent); and maximizing tour efficiency (the total length of the route should be as short as possible). Constraints are added, and optimization algorithms are used to solve the above problems, resulting in one or more optimized point-of-visit sequences. Natural language generation technology is then used to assign an attractive name and brief description to the final selected route, elucidating the route's thematic storyline, thus forming the narrative tour route.

[0066] Step 105: Based on the narrative tour path, determine the content to be generated at each cultural point to obtain digital scene content.

[0067] In this context, digital scene content refers to customized multimedia displays generated for each cultural point along the narrative tour route. Examples include historical information overlaid on the real-world environment using augmented reality technology, 3D model reconstruction animations, personalized voice narration, and interactive Q&A. The terminal analyzes the content to be displayed to tourists at each cultural point based on the determined narrative tour route and its theme, aiming to enhance the route's narrative. Based on detailed asset data associated with each point in the neighborhood cultural map database, it automatically or semi-automatically generates corresponding text, audio, video, or interactive content using content generation technology. All content generated for the points along the route is then integrated to form a complete digital scene content package usable by tourists.

[0068] This embodiment provides a method for generating scenario-based content for smart cultural and creative blocks based on convolutional neural networks. The method collects a two-dimensional planar map, cultural sites, and cultural assets of the corresponding block, and binds each cultural site with its corresponding cultural asset to obtain a block cultural map database. The two-dimensional planar map is divided into spatial grids, and based on the block cultural map database, the cultural value density score of each spatial grid is calculated to generate a cultural density network map. Based on the cultural density network map, the spatial clustering of cultural assets is analyzed to obtain a clustering heatmap. Based on the block cultural map database, the correlation between cultural sites is calculated to generate a cultural site theme correlation matrix. Based on the clustering heatmap and the cultural site theme correlation matrix, a narrative tour path is generated. Based on the narrative tour path, the content to be generated at each cultural site is determined to obtain digital scene content. By employing the above methods, we can effectively capture the dynamic impact diffusion path of hotspot areas on surrounding areas, quantify the intrinsic correlation between different cultural themes, capture strong correlation effects at close range, analyze weak correlation diffusion paths at long distances, accurately identify the spatial clustering characteristics and radiation effects of cultural hotspots, and scientifically guide tourists to move efficiently between multiple hotspots in contextualized content, thereby achieving a sense of hierarchy and logic in contextualized content.

[0069] In one embodiment, based on a cultural density network diagram, the spatial clustering of cultural assets is analyzed to obtain a clustering heatmap, including:

[0070] Step 201: Define multiple multi-scale convolution kernels, and perform sliding window convolution on the cultural value density grid map to calculate the response value of each of the multi-scale convolution kernels, thereby obtaining multiple cultural space response maps.

[0071] Multi-scale convolutional kernels are a set of two-dimensional matrices of different sizes, each representing a convolutional kernel whose size indicates the observation scale. Examples include small-scale kernels (e.g., 3x3 grid), medium-scale kernels (e.g., 7x7 grid), and large-scale kernels (e.g., 15x15 grid). Multi-scale convolutional kernels are used to simultaneously detect cultural clusters of different sizes. The cultural value density grid map is a two-dimensional matrix where each cell's value represents the cultural value density score of the corresponding spatial grid, reflecting the original spatial distribution of cultural values. It can also be used to directly generate a cultural spatial response map through a convolutional neural network. Sliding window convolution is a standard technique in image processing and signal analysis. In operation, a convolutional kernel is moved like a window, starting from the top left corner of the cultural value density grid map and moving one step at a time, covering the entire grid map. The response value is calculated by multiplying the pixel values ​​of the cultural value density grid map covered within the window at each sliding window position with the corresponding weight values ​​within the convolutional kernel window, and then summing the results. This summation reflects the weighted aggregation degree of cultural values ​​within the window region. A cultural space response map is a new two-dimensional grid map obtained after performing sliding window convolution calculations on the entire cultural value density grid map for each specific scale of convolution kernel used. The value of each point on the new map is the response value of the original map at that location and corresponding window scale. Therefore, using N convolution kernels of different scales will yield N cultural space response maps. The terminal prepares a set of preset convolution kernels of different sizes. One convolution kernel is selected and placed at the starting position of the cultural value density grid map, usually the upper left corner. The sum of the products of the corresponding grid values ​​of the original map and the weights of the convolution kernel under the coverage of the kernel is calculated to obtain the response value at the current position, which is recorded at the corresponding position in the new grid map. The convolution kernel is moved one step to the right, and the calculation is repeated until the entire row is traversed. Then, the kernel is moved down one row, and the process continues from left to right until the entire cultural value density grid map is traversed. After traversal, a cultural space response map corresponding to the current convolution kernel scale is generated. The process is repeated with a convolution kernel of another scale until all scales of convolution kernels have been processed.

[0072] Step 202: Based on the response maps of all cultural spaces, the corresponding response values ​​are weighted and fused to calculate the comprehensive clustering degree, and a comprehensive clustering degree distribution map is obtained.

[0073] Specifically, weighted fusion is an information fusion technology that assigns a weight to each input and then combines them according to the weights. The overall clustering degree is a comprehensive metric that integrates information from cultural space response maps at different scales, used to characterize the overall degree of cultural value clustering at a given spatial location. The overall clustering degree distribution map is a new two-dimensional grid map. The value of each grid on this map is the overall clustering degree calculated by weighted fusion of the response values ​​of all cultural space response maps at the same geographical location. The terminal assigns a preset weight coefficient to each cultural space response map. The weight allocation can be based on prior knowledge; for example, if the clustering pattern at a medium scale is considered the most important, it is given a higher weight. For each grid location in the space, the response values ​​of all cultural space response maps at that location are collected. Each response value is multiplied by the weight of its corresponding response map, and then all the products are summed to obtain the final overall clustering degree value for that location. This calculation process is repeated for each grid location on the map. The overall clustering degree values ​​calculated for all grid locations are arranged in their original spatial order to form a new grid map, i.e., the overall clustering degree distribution map.

[0074] Step 203: Normalize the overall clustering degree in the overall clustering degree distribution map to obtain the normalized clustering degree distribution map.

[0075] Specifically, normalization is a data preprocessing method that aims to linearly scale a set of values ​​to a uniform, standardized range. A normalized clustering distribution map is a new grid map obtained by normalizing each clustering value in the comprehensive clustering distribution map. All grid values ​​in the map are between 0 and 1. The terminal finds the maximum and minimum values ​​of all comprehensive clustering values ​​in the comprehensive clustering distribution map. For each value in the distribution map, the maximum-minimum normalization formula is used for calculation. After calculation, the position of the minimum value in the original map becomes 0, the position of the maximum value becomes 1, and all other values ​​are linearly mapped to the interval [0,1]. The mapped values ​​constitute the normalized clustering distribution map.

[0076] Step 204: Based on the preset color mapping rules, the normalized clustering distribution map is mapped to the corresponding color to obtain the clustering heatmap.

[0077] Specifically, the preset color mapping rule is a predefined function or lookup table that maps numerical values ​​to colors. For example, the value 0 is mapped to blue, 0.5 to yellow, and 1.0 to red, forming a gradient color band from blue to yellow to red. The clustering heatmap is a data visualization graphic in which each value in the normalized clustering distribution map is converted into a specific color according to the color mapping rule. The entire map intuitively displays the distribution of spatial clustering of cultural assets through color. The terminal determines a clear and intuitive color mapping rule, optionally using cool colors to represent low values ​​and warm colors to represent high values. It iterates through each grid value in the normalized clustering distribution map. For each grid value, according to the color mapping rule, it finds the corresponding color and fills the corresponding grid position in the final output image with that color. After processing all grids, a colored clustering heatmap is generated, with a color bar legend illustrating the correspondence between colors and values.

[0078] This embodiment extracts spatial features at multiple scales. Small-scale convolutional kernels are sensitive to small-scale, fine-grained fluctuations in cultural values, which helps to discover small cultural clusters. Large-scale convolutional kernels can smooth details and capture large-scale, macro-level cultural value trend areas. This obtains information on the aggregation patterns of cultural values ​​at different observation granularities. The information from multiple scales is synthesized into a single, comprehensive indicator, overcoming the limitations of single-scale analysis. It considers both small-scale dense points and large-scale trend areas, and adjusts the contribution of information at different scales through weighting, thereby obtaining a more comprehensive and robust panoramic view of cultural spatial aggregation.

[0079] In one embodiment, based on a neighborhood cultural map database, the correlation between cultural sites is calculated to generate a cultural site theme correlation matrix, including:

[0080] Step 301: Based on the word embedding model, the text descriptions of cultural points in the corresponding street cultural map database are converted into feature vectors to obtain topic semantic vectors.

[0081] Word embedding models are a type of natural language processing technology. Their core function is to convert text into a series of numerical vectors. These vectors can represent the semantic information of the text to a certain extent; words with similar semantics have vectors that are closer together in space. Text descriptions refer to the textual information stored in the neighborhood cultural map database used to describe cultural sites and their cultural assets. For example, a cultural site named "Clock Tower" might have a text description containing keywords or paragraphs such as "ancient timekeeping building," "city landmark," "Ming Dynasty style," and "brick and stone structure." Feature vectors are numerical arrays obtained after processing text descriptions using word embedding models; they can be viewed as coordinate points of the text description in a multi-dimensional space. Thematic semantic vectors are the aforementioned feature vectors, called thematic semantic vectors, to emphasize that this vector represents the core theme and semantic information extracted from the text. Each cultural site may have one or more related text descriptions, thus multiple thematic semantic vectors can be generated. The terminal extracts all relevant text descriptions for each cultural location from the neighborhood cultural map database, selects a pre-trained word embedding model, inputs the text descriptions for each cultural location into the model, the model processes and analyzes the text, and outputs a fixed-length numerical array that reflects the overall semantics of the text, namely the topic semantic vector.

[0082] Step 302: Average and aggregate all thematic semantic vectors corresponding to the cultural points to obtain aggregated thematic vectors.

[0083] Averaging aggregation is a simple data fusion method that involves averaging the components at corresponding positions of multiple vectors to obtain a new, representative vector. Aggregated topic vectors, on the other hand, are used to represent a single, comprehensive topic vector for a cultural location if it is associated with multiple cultural assets and thus has multiple topic semantic vectors. For each specific cultural location, the terminal collects all topic semantic vectors with the same dimension. The values ​​of all vectors in the first dimension are added together and then divided by the total number of vectors to obtain the value of the first dimension of the new vector. The same operation is performed on the values ​​of all vectors in the second dimension to obtain the value of the second dimension of the new vector. This process is repeated multiple times until all dimensions are processed, resulting in the aggregated topic vector for that cultural location.

[0084] Step 303: Based on the aggregated topic vectors, calculate the topic vector similarity between cultural points using the following formula to obtain a topic vector similarity list:

[0085]

[0086] in, Let represent the similarity of topic vectors between cultural point i and cultural point j, where n is the dimension of the topic vectors and k is the vector index. Let be the k-dimensional component value of the aggregated topic vector of cultural point i. Let be the component of the aggregated topic vector of cultural point j in the k-dimensional dimension.

[0087] Specifically, topic vector similarity is a numerical metric used to measure how close the directions of two aggregated topic vectors are in the semantic space. The value ranges from 0 to 1; a higher value indicates that the cultural points represented by the two vectors are more similar in topic semantics and have a stronger correlation. The topic vector similarity list is a list containing the similarity calculation results for all pairwise combinations of cultural points. Each item in the list records which two points are paired and their similarity value. The terminal takes two points from all cultural points, denoted as point i and point j, obtains the corresponding aggregated topic vectors, and calculates the similarity using the provided cosine similarity formula. Essentially, the formula calculates the cosine of the angle between the two vectors; the smaller the angle, the closer the cosine value is to 1, indicating greater similarity.

[0088] Step 304: Based on the list of topic vector similarity, generate a topic association matrix for cultural sites; each element in the topic association matrix represents the association degree between cultural sites.

[0089] The cultural site theme association matrix is ​​a square, symmetrical matrix. The row and column labels are the names or numbers of all cultural sites. The element at the intersection of the i-th row and j-th column represents the theme vector similarity between cultural site i and cultural site j. The matrix presents the strength of the theme association between all pairs of sites in a compact and regular form. The terminal creates an N-row, N-column blank matrix, where N is the total number of cultural sites, and labels the rows and columns with site identifiers. It iterates through the theme vector similarity list, filling the corresponding position in the matrix for each item in the list. The diagonal of the matrix represents the similarity between the site and itself, typically with a value of 1. After all items in the list are filled into the matrix, the cultural site theme association matrix is ​​generated.

[0090] This embodiment transforms the thematic correlation between cultural sites from a subjective perception into a measurable objective value. By calculating cosine similarity, it can accurately quantify the correlation between any two sites in terms of cultural content, providing data support for identifying clusters of sites with common themes and constructing thematically coherent tour routes, thereby improving the targeting of contextualized content generation.

[0091] In one embodiment, a narrative tour path is generated based on a clustering heatmap and a cultural site theme relevance matrix, including:

[0092] Step 401: Based on the clustering heatmap and the cultural site theme correlation matrix, define the optimization proposition with the order of visiting cultural sites as the decision variable.

[0093] In optimization problems, decision variables are the core unknowns that require decision-making. Their values ​​determine the solution. In this embodiment, the decision variable is the order of visiting cultural points, i.e., a permutation sequence containing cultural points, also known as a path. An optimization proposition is a precisely defined mathematical problem comprising two parts: an objective function and constraints. The objective function is the metric to be maximized or minimized, while the constraints are the rules that the solution must follow. Defining an optimization proposition transforms the fuzzy objective of generating a good path into a computable mathematical problem. The terminal constructs the objective function, quantifies multiple mathematical objectives, and combines them into a comprehensive objective function. Objectives typically include: the path should prioritize areas with high cultural concentration; the themes of adjacent points in the path should transition smoothly; and the total physical distance of the path should be as short as possible. The objective functions are combined using a weighted summation method to form the final comprehensive objective function. The hard rules that the path must satisfy are defined, and these rules are transformed into constraints on the decision variables. The comprehensive objective function and all constraints are formally expressed, forming a complete optimization proposition.

[0094] Step 402: Solve the optimization problem to obtain the solution set, and map the solution set to the corresponding cultural points to obtain candidate paths.

[0095] Specifically, solving optimization problems refers to the process of applying specialized mathematical algorithms to search for decision variables that maximize the value of the comprehensive objective function, while satisfying all constraints. The solution set is the output of the algorithm; in path optimization problems, it is one or more sequences representing the optimal order of visiting cultural sites. Candidate paths are one or more specific, selectable tour routes obtained by mapping the mathematical solution set back to the actual cultural sites. The terminal selects an appropriate optimization algorithm based on the complexity of the optimization problem. Optionally, this includes heuristic algorithms such as genetic algorithms, simulated annealing, and ant colony optimization. The defined objective function and constraints are input into the algorithm, which searches, iterates, and evaluates within a vast space of possible paths, continuously trying new path sequences and retaining those sequences that maximize the comprehensive objective function value, thus obtaining the solution set. When the algorithm reaches its termination condition, it outputs the best sequence or a set of high-quality sequences. The numerical sequences in the solution set are then used to reconstruct the actual names and geographical order of the cultural sites according to a table of cultural site numbers, thereby obtaining one or more specific candidate paths.

[0096] Step 403: Based on natural language generation, generate names and descriptions for candidate paths to obtain narrative tour paths.

[0097] Specifically, Natural Language Generation (NLP) is a branch of artificial intelligence that refers to the technology of automatically generating readable natural language text using structured data or knowledge bases. Narrative tour routes are candidate routes that are given a story-like package; they are storylines with names, themes, and brief introductions, aiming to enhance the visitor experience and immersion. The terminal analyzes the characteristics of the final selected candidate routes. Based on the analyzed characteristics, NLP technology generates an attractive and concise name. This can be achieved, for example, by filling in keywords using predefined templates, or by using more advanced large language models. A short text description is then generated, explaining the route's theme, highlights, and story outline. The introduction connects key points along the route, forming a simple narrative logic. The generated name and introduction are then linked to the geographical information of the candidate routes to create a narrative tour route.

[0098] This embodiment optimizes the problem-solving process, efficiently identifying the best balance between spatial value, thematic coherence, and travel efficiency from all possible path combinations, thereby improving the reliability and feasibility of scenario-based content generation.

[0099] In one embodiment, based on a clustering heatmap and a cultural site topic association matrix, and using the order of visiting cultural sites as a decision variable, an optimization proposition is defined, including:

[0100] Step 501: Based on the clustering heatmap, the sum of the spatial values ​​of the paths is defined as the spatial value objective function, and based on the neighborhood cultural map database, the total distance of the paths is defined as the travel efficiency objective function; the path is the order in which cultural points are visited.

[0101] The spatial value objective function is a mathematical function that quantifies the cultural value of a candidate path in terms of spatial dimension. Its value is the sum of the spatial cultural agglomeration values ​​corresponding to all cultural points along the path; a higher value indicates that the path passes through more cultural hotspots. The travel efficiency objective function is another mathematical function that quantifies the efficiency of a candidate path in terms of physical distance. Its value is the sum of the physical distances between adjacent cultural points in the path sequence; a lower value indicates a shorter total journey and higher travel efficiency. For any given path, the terminal iterates through each cultural point along the path. For each point, it queries the agglomeration value corresponding to its geographical location from the agglomeration heatmap, sums the agglomeration values ​​of all points, and the sum is the value of the spatial value objective function. It then calculates the physical distance between each pair of adjacent points sequentially, based on the latitude and longitude coordinates of each point stored in the street cultural map database. The sum of the distances between all adjacent point pairs is the value of the travel efficiency objective function.

[0102] Step 502: Based on the cultural site theme correlation matrix, calculate the local theme coherence objective function and the global theme consistency objective function.

[0103] The local theme coherence objective function is a mathematical function that quantifies the smoothness and naturalness of the theme transition between adjacent points on a candidate path. Its value is the sum of the theme relevance of all adjacent point pairs on the path; a larger value indicates better thematic connection between points visited consecutively. The global theme consistency objective function is also a mathematical function that quantifies whether a candidate path revolves around a clear and consistent theme. Its value is typically the average theme relevance of all points on the path, or the concentration of the overall theme distribution; a larger value indicates a more distinct and unified theme for the entire path. For each path, the terminal sequentially retrieves each pair of adjacent points. For each pair, it queries the cultural point theme relevance matrix to obtain the relevance values ​​between points. The sum of the relevance values ​​of all adjacent point pairs is the value of the local theme coherence objective function. The average relevance value between all points on the path is calculated, reflecting the overall theme consistency of the path. Specifically, this is done by calculating the average relevance value of all point pairs included in the path, including non-adjacent point pairs, and using this as the value of the global theme consistency objective function.

[0104] Step 503: Based on the spatial value objective function, the travel efficiency objective function, the local theme coherence objective function, and the global theme consistency objective function, a comprehensive objective function is constructed using the following formula:

[0105]

[0106] in, Let P be the overall objective function, and P be the path sequence. For spatial value objective function, For the objective function of local topic coherence, For the objective function of local topic coherence, Let the objective function be the travel efficiency. , , , These are the weighting coefficients.

[0107] Specifically, the comprehensive objective function is a final evaluation function that integrates all the aforementioned individual objective functions. The entire task of the optimization algorithm is to find the path that maximizes this function value, representing a complete mathematical definition of the optimal path. The weighting coefficients are a set of pre-defined positive numbers used to adjust the relative importance of different objectives in the comprehensive evaluation. The terminal linearly weights and combines the four objective functions according to a given formula. The smaller the value of the travel efficiency objective function, the better; therefore, a negative sign is used, meaning that the shorter the distance, the greater the contribution to the comprehensive objective function. The specific values ​​of the weighting coefficients need to be set and adjusted according to actual business requirements.

[0108] Step 504: Based on the neighborhood cultural graph database, the constraints in the path planning are transformed into function form to obtain the constraint conditions; the constraint conditions include the starting point constraint, path length constraint, unique access constraint, and connectivity constraint.

[0109] Constraints are mathematical expressions of hard rules that a set of path solutions must satisfy. Any path solution that does not meet the constraints is considered invalid, regardless of its comprehensive objective function value, thus limiting the search range of the optimization algorithm. The starting point constraint requires the path to begin from a specified cultural point; the unique access constraint means that each cultural point in the path sequence can only appear once; the path length constraint means that the number of points included in the path must be within a specified range; and the connectivity constraint means that the path sequence must be continuous, i.e., one point can reach the next. The terminal transforms the path planning rules into mathematical constraints, specifying through functions that the first point of the path must be a specified cultural point; the number of cultural points included in the path must be within a certain range; each cultural point in the path can only appear once, which translates to requiring the path to be a sequence without repeating elements; and ensuring that any two adjacent points in the path are reachable in the real world, which can be guaranteed through distance calculation or by pre-defining a reachability network in a database.

[0110] Step 505: Integrate the objective function and constraints to obtain the optimization proposition.

[0111] Specifically, an optimization proposition is a complete and formally formatted statement of a mathematical optimization problem, clearly indicating the objective to be optimized and the rules to be followed. The terminal integrates the constructed comprehensive objective function and all defined constraints to form a complete proposition description.

[0112] This embodiment ensures that the path has a core theme by introducing core theme consistency, avoiding the problem of loss of immersion caused by irrelevant cultural points, guaranteeing the narrative quality of the path, and improving the accuracy of scene-based content generation.

[0113] In one embodiment, based on a neighborhood cultural map database, the cultural value density score of each spatial grid is calculated to generate a cultural density network map, including:

[0114] Step 601: Based on the neighborhood cultural map database, spatial matching is performed between each cultural location and the spatial grid to obtain an attribution mapping table.

[0115] Spatial matching, an operation within a Geographic Information System (GIS), refers to the process of determining which spatial grid a cultural point falls into based on its geographic coordinates. The attribution mapping table is a list whose core function is to record the correspondence between each cultural point and its corresponding spatial grid. Each row in the table may contain two pieces of information: a unique identifier for the cultural point and a unique identifier for the spatial grid to which that point belongs. The terminal reads the precise geographic coordinates of each cultural point from the neighborhood cultural map database and obtains the boundary coordinate information of all spatial grids. Each grid has a clearly defined geographic range. For each cultural point, its coordinates are compared with the boundaries of all spatial grids, and geometric calculations are used to determine which grid the point is located within. Once it is determined that a point belongs to a specific grid, this relationship is recorded in the attribution mapping table. A grid can contain multiple cultural points, but a cultural point belongs to only one grid in a well-defined grid system.

[0116] Step 602: Based on the neighborhood cultural map database, assign weight coefficients to different types of cultural assets to obtain a weight comparison table.

[0117] Specifically, the weight coefficient is a numerical value used to characterize the relative importance of different types of cultural assets. The higher the weight, the greater the contribution of that type of asset in calculating cultural value. The weight lookup table is a data structure whose core function is to store the mapping relationship between cultural asset types and their corresponding weight coefficients. Each row in the table contains the asset type and its corresponding weight value. Based on domain knowledge or established standards, the terminal classifies all cultural assets in the neighborhood cultural map database to form a unified type system. According to factors such as cultural value, scarcity, and historical significance, a weight coefficient is assigned to each asset type. The weight can be determined based on expert scoring, policy regulations, or historical data statistics. The established correspondence between asset types and weight coefficients is compiled into a weight lookup table.

[0118] Step 603: Based on the attribution mapping table and weight comparison table, calculate the cultural value density of each spatial grid to obtain the initial matrix of cultural value density.

[0119] Specifically, cultural value density is a quantitative indicator for a single spatial grid, representing the total value of all cultural assets contained within that grid. It is obtained by summing the values ​​of all cultural assets within the grid. The initial cultural value density matrix is ​​a two-dimensional array whose rows and columns correspond perfectly to the spatial grid division method. The value of each element in the matrix is ​​the cultural value density of the corresponding spatial grid, digitizing the cultural value distribution of the entire neighborhood into a grid numerical map. The terminal processes each spatial grid sequentially. For the currently processed grid, it queries the attribution mapping table to find all cultural points located within that grid. Through the neighborhood cultural map database, it finds all cultural assets bound to those points, forming a list of cultural assets for that grid. It iterates through each asset in the list, and based on the asset type, queries the weight lookup table to obtain the corresponding weight coefficient. The weight coefficients of all assets within the grid are summed, and the sum is the cultural value density of that grid. The calculated density value is then filled into the corresponding position in a blank matrix that perfectly corresponds to the grid layout. This process continues until all grids are processed, generating the initial cultural value density matrix.

[0120] Step 604: Normalize the initial matrix of cultural value density to obtain the cultural density network diagram.

[0121] Normalization is a data standardization technique that aims to linearly scale a set of values ​​to a uniform, fixed interval to eliminate differences in the absolute size of the original data and make them comparable. The cultural density network diagram is the result of normalizing the initial cultural value density matrix. It is still a matrix, but each element's value is between 0 and 1, representing the relative cultural value density of its corresponding grid. The terminal scans the entire initial cultural value density matrix, finding the maximum and minimum values ​​among all density values. For each original density value in the matrix, the normalization formula is used for calculation. After calculation, the position of the minimum value in the original matrix becomes 0, the position of the maximum value becomes 1, and all other values ​​are linearly mapped to between 0 and 1, forming the cultural density network diagram.

[0122] This embodiment introduces a value assessment system, which makes the calculation of cultural value take into account asset quality, can truly reflect the distribution of cultural value, and thus improves the reliability of contextualized content generation.

[0123] In one embodiment, based on the attribution mapping table and weight lookup table, the cultural value density of each spatial grid is calculated to obtain an initial cultural value density matrix, including:

[0124] Step 701: Based on the attribution mapping table, query the cultural assets bound to the cultural points corresponding to the spatial grid to obtain a list of cultural assets.

[0125] The cultural asset list is a list specific to a particular spatial grid, listing every specific cultural asset associated with each cultural point belonging to that grid, and detailing all cultural asset items contained within that grid. The terminal selects a spatial grid to be processed, queries the attribution mapping table, and finds all cultural points mapped to that grid. For example, assuming the query results have three points: point A, point B, and point C, based on the point-asset binding relationships recorded in the neighborhood cultural map database, it identifies all cultural assets associated with point A, point B, and point C respectively, and aggregates these assets to form a list specific to the corresponding grid.

[0126] Step 702: Traverse each cultural asset in the cultural asset list, query the weight comparison table, and obtain the corresponding weight coefficient.

[0127] Specifically, obtaining the weight coefficient is a query operation, which refers to finding the corresponding value weight value of a cultural asset from a preset rule based on its type. The terminal processes each cultural asset in a grid's cultural asset list one by one. For the asset being processed, it identifies its type, queries the weight lookup table, finds the corresponding row in the table based on the asset type, reads its weight coefficient, and associates the weight coefficient with the asset. This process is repeated for each asset in the list.

[0128] Step 703: Summate all weight coefficients arithmetically to obtain the cultural value density.

[0129] Specifically, cultural value density is a single numerical result for the current spatial grid being processed, representing the total value of all cultural assets within that grid. It is calculated by simply summing the weight coefficients of all assets within the grid. The terminal simply adds up all the obtained weight coefficients to obtain the cultural value density; the more high-value assets a grid contains, the higher its overall cultural value density.

[0130] Step 704: Integrate the cultural value densities corresponding to all spatial grids to obtain the initial matrix of cultural value densities.

[0131] The initial cultural value density matrix is ​​a two-dimensional data structure whose rows and columns correspond exactly to the spatial grid division. The value of each cell in the matrix is ​​the cultural value density of the corresponding spatial grid. The terminal creates a blank matrix isomorphic to the spatial grid division. For example, if the map is divided into a 10x10 grid, a 10-row, 10-column blank matrix is ​​created. For each spatial grid in the map, its unique cultural value density value is strictly calculated. After calculating the density value of each grid, this value is filled into the cell in the blank matrix corresponding to the geographical location of that grid. When all grids have been processed, the matrix is ​​filled, and the matrix filled with values ​​is the initial cultural value density matrix.

[0132] This embodiment combines the discrete, individual cultural asset values ​​within a single grid into a single density score that represents the overall cultural value of that grid. This score intuitively reflects the total cultural value of the grid and accurately reflects the original spatial distribution of cultural value. It serves as the data foundation for subsequent path generation and enhances the targeting of scenario-based content generation.

[0133] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0134] Based on the same inventive concept, this application also provides a convolutional neural network-based smart cultural and creative street scene-based content generation device for implementing the above-mentioned method for generating scene-based content for smart cultural and creative streets based on convolutional neural networks. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the convolutional neural network-based smart cultural and creative street scene-based content generation device provided below can be found in the limitations of the convolutional neural network-based smart cultural and creative street scene-based content generation method described above, and will not be repeated here.

[0135] In one exemplary embodiment, such as Figure 2 As shown, a smart cultural and creative block scene-based content generation device 800 based on convolutional neural networks is provided, including:

[0136] The data acquisition module 801 is used to acquire two-dimensional planar maps, cultural sites, and cultural assets of the corresponding blocks, and bind each cultural site with its corresponding cultural asset to obtain a block cultural map database.

[0137] Density module 802 is used to divide a two-dimensional planar map into spatial grids and, based on a neighborhood cultural map database, calculate the cultural value density score of each spatial grid to generate a cultural density network map.

[0138] The aggregation module 803 is used to analyze the spatial aggregation degree of cultural assets based on the cultural density network diagram, obtain the aggregation degree heat map, and calculate the correlation degree between cultural sites based on the neighborhood cultural map database, generating a cultural site theme correlation degree matrix.

[0139] Path module 804 is used to generate narrative tour paths based on clustering heatmaps and cultural site theme correlation matrices;

[0140] Scene module 805 is used to determine the content to be generated at each cultural point based on the narrative tour path, and obtain digital scene content.

[0141] Furthermore, the aggregation module 803 is also used for:

[0142] Define multiple multi-scale convolution kernels, and perform sliding window convolution on the cultural value density grid map to calculate the response value of each of the multi-scale convolution kernels, thereby obtaining multiple cultural space response maps;

[0143] Based on the response maps of all cultural spaces, the corresponding response values ​​are weighted and fused to calculate the comprehensive clustering degree, thus obtaining the comprehensive clustering degree distribution map;

[0144] Normalize the overall clustering degree in the overall clustering degree distribution map to obtain the normalized clustering degree distribution map;

[0145] Based on preset color mapping rules, the normalized clustering distribution map is mapped to the corresponding color to obtain the clustering heatmap.

[0146] Furthermore, the aggregation module 803 is also used for:

[0147] Based on the word embedding model, the text descriptions of cultural points in the corresponding street cultural map database are transformed into feature vectors to obtain topic semantic vectors;

[0148] The aggregated topic vector is obtained by averaging all thematic semantic vectors corresponding to cultural locations.

[0149] Based on aggregated topic vectors, the topic vector similarity between cultural locations is calculated using the following formula, resulting in a list of topic vector similarities:

[0150]

[0151] in, Let represent the similarity of topic vectors between cultural point i and cultural point j, where n is the dimension of the topic vectors and k is the vector index. Let be the k-dimensional component value of the aggregated topic vector of cultural point i. Let be the component of the aggregated topic vector of cultural point j in the k-dimensional dimension;

[0152] Based on the list of topic vector similarity, a topic association matrix of cultural sites is generated; each element in the topic association matrix of cultural sites represents the degree of association between cultural sites.

[0153] Furthermore, path module 804 is also used for:

[0154] Based on the clustering heatmap and the theme correlation matrix of cultural sites, an optimization proposition is defined with the order of visiting cultural sites as the decision variable.

[0155] Solve the optimization problem to obtain a solution set, and map the solution set to the corresponding cultural points to obtain candidate paths;

[0156] Based on natural language generation, names and descriptions are generated for candidate paths to obtain narrative tour paths.

[0157] Furthermore, path module 804 is also used for:

[0158] Based on the clustering heatmap, the sum of the spatial value of the path is defined as the spatial value objective function, and based on the neighborhood cultural map database, the total distance of the path is defined as the travel efficiency objective function; the path is the order in which cultural points are visited.

[0159] Based on the theme relevance matrix of cultural sites, calculate the objective function of local theme coherence and the objective function of global theme consistency;

[0160] Based on the spatial value objective function, the travel efficiency objective function, the local theme coherence objective function, and the global theme consistency objective function, a comprehensive objective function is constructed using the following formula:

[0161]

[0162] in, Let P be the overall objective function, and P be the path sequence. For spatial value objective function, For the objective function of local topic coherence, For the objective function of local topic coherence, Let the objective function be the travel efficiency. , , , These are the weighting coefficients;

[0163] Based on the neighborhood cultural graph database, the constraints in path planning are transformed into functional forms to obtain the constraint conditions; the constraint conditions include the starting point constraint, path length constraint, unique access constraint, and connectivity constraint.

[0164] By integrating the objective function and constraints, an optimization proposition is obtained.

[0165] Furthermore, density module 802 is also used for:

[0166] Based on the neighborhood cultural map database, each cultural location is spatially matched with a spatial grid to obtain an attribution mapping table;

[0167] Based on the neighborhood cultural map database, weight coefficients are assigned to different types of cultural assets to obtain a weight comparison table;

[0168] Based on the attribution mapping table and weight comparison table, the cultural value density of each spatial grid is calculated to obtain the initial matrix of cultural value density.

[0169] The initial matrix of cultural value density is normalized to obtain the cultural density network diagram.

[0170] Furthermore, density module 802 is also used for:

[0171] Based on the attribution mapping table, query the cultural assets bound to the cultural points corresponding to the spatial grid to obtain a list of cultural assets;

[0172] Iterate through each cultural asset in the cultural asset list, query the weight comparison table, and obtain the corresponding weight coefficient;

[0173] The cultural value density is obtained by summing all the weighting coefficients arithmetically.

[0174] By integrating the cultural value densities corresponding to all spatial grids, an initial matrix of cultural value densities is obtained.

[0175] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the above-described method for generating scenario-based content for a smart cultural and creative block based on a convolutional neural network.

[0176] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0177] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0178] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for generating scenario-based content for smart cultural and creative blocks based on convolutional neural networks, characterized in that, The method includes: Collect two-dimensional planar maps, cultural sites, and cultural assets of the corresponding blocks, and bind each cultural site with its corresponding cultural asset to obtain a block cultural map database; The two-dimensional planar map is divided into spatial grids, and based on the neighborhood cultural map database, the cultural value density score of each spatial grid is calculated to generate a cultural density network map. Based on the cultural density network diagram, the spatial clustering of the cultural assets is analyzed to obtain a clustering heat map. Based on the neighborhood cultural map database, the correlation between the cultural sites is calculated to generate a cultural site theme correlation matrix. Based on the clustering heatmap and the cultural site theme correlation matrix, a narrative tour path is generated; Based on the narrative tour path, the content to be generated at each of the cultural points is determined, resulting in digital scene content.

2. The method according to claim 1, characterized in that, The process of analyzing the spatial clustering of cultural assets based on the cultural density network diagram to obtain a clustering heatmap includes: Multiple multi-scale convolution kernels are defined, and each of the multi-scale convolution kernels is subjected to sliding window convolution on the cultural value density grid map to calculate the response value, thereby obtaining multiple cultural space response maps; Based on all the cultural space response maps, the corresponding response values ​​are weighted and fused to calculate the comprehensive clustering degree, and a comprehensive clustering degree distribution map is obtained. The overall clustering degree in the aforementioned overall clustering degree distribution map is normalized to obtain a normalized clustering degree distribution map; Based on preset color mapping rules, the normalized clustering distribution map is mapped onto the corresponding color to obtain the clustering heatmap.

3. The method according to claim 1, characterized in that, The step of calculating the correlation between cultural sites based on the street cultural map database and generating a cultural site theme correlation matrix includes: Based on the word embedding model, the text descriptions of the cultural locations corresponding to the street cultural map database are transformed into feature vectors to obtain topic semantic vectors; The aggregated topic vector is obtained by averaging all the topic semantic vectors corresponding to the cultural locations. Based on the aggregated topic vectors, the topic vector similarity between the cultural locations is calculated using the following formula to obtain a list of topic vector similarities: in, Let represent the similarity of topic vectors between cultural point i and cultural point j, where n is the dimension of the topic vectors and k is the vector index. Let be the k-dimensional component value of the aggregated topic vector of cultural point i. Let be the component of the aggregated topic vector of cultural point j in the k-dimensional dimension; Based on the topic vector similarity list, the cultural site topic association degree matrix is ​​generated; each element in the cultural site topic association degree matrix represents the association degree between the cultural sites.

4. The method according to claim 1, characterized in that, The generation of a narrative tour path based on the clustering heatmap and the cultural site theme correlation matrix includes: Based on the clustering heatmap and the cultural site theme correlation matrix, an optimization proposition is defined with the order of visiting the cultural sites as the decision variable. Solve the optimization problem to obtain a solution set, and map the solution set to the corresponding cultural points to obtain candidate paths; Based on natural language generation, names and descriptions are generated for the candidate paths to obtain the narrative tour path.

5. The method according to claim 4, characterized in that, Based on the clustering heatmap and the cultural site theme correlation matrix, and using the order of visiting the cultural sites as the decision variable, an optimization proposition is defined, including: Based on the clustering heatmap, the sum of the spatial values ​​of the paths is defined as the spatial value objective function, and based on the neighborhood cultural map database, the total distance of the paths is defined as the travel efficiency objective function; the paths are the order in which the cultural sites are visited. Based on the cultural site theme correlation matrix, calculate the local theme coherence objective function and the global theme consistency objective function; Based on the spatial value objective function, the travel efficiency objective function, the local theme coherence objective function, and the global theme consistency objective function, a comprehensive objective function is constructed using the following formula: in, Let P be the overall objective function, and P be the path sequence. For spatial value objective function, For the objective function of local topic coherence, For the objective function of local topic coherence, Let the objective function be the travel efficiency. , , , These are the weighting coefficients; Based on the street culture map database, the constraints in path planning are transformed into function form to obtain the constraint conditions; the constraint conditions include the starting point constraint, path length constraint, unique access constraint, and connectivity constraint. By integrating the objective function and the constraints, the optimization proposition is obtained.

6. The method according to claim 1, characterized in that, The process of calculating the cultural value density score for each spatial grid based on the neighborhood cultural map database and generating a cultural density network map includes: Based on the street culture map database, each cultural location is spatially matched with the spatial grid to obtain an attribution mapping table; Based on the neighborhood cultural map database, weight coefficients are assigned to different types of cultural assets to obtain a weight comparison table; Based on the attribution mapping table and the weight lookup table, the cultural value density of each spatial grid is calculated to obtain the initial matrix of cultural value density. The initial matrix of cultural value density is normalized to obtain the cultural density network graph.

7. The method according to claim 6, characterized in that, The process of calculating the cultural value density of each spatial grid based on the attribution mapping table and the weight lookup table to obtain an initial cultural value density matrix includes: Based on the attribution mapping table, query the cultural assets bound to the cultural points corresponding to the spatial grid to obtain a list of cultural assets; Iterate through each cultural asset in the cultural asset list, query the weight lookup table, and obtain the corresponding weight coefficient; The cultural value density is obtained by summing all the weight coefficients arithmetically. By integrating the cultural value densities corresponding to all the spatial grids, an initial matrix of cultural value densities is obtained.

8. A device for generating scenario-based content for a smart cultural and creative street based on a convolutional neural network, characterized in that, The device includes: The data acquisition module is used to collect two-dimensional planar maps, cultural sites, and cultural assets of the corresponding blocks, and bind each cultural site with its corresponding cultural asset to obtain a block cultural map database. The density module is used to divide the two-dimensional planar map into spatial grids and, based on the neighborhood cultural map database, calculate the cultural value density score of each spatial grid to generate a cultural density network map. The aggregation module is used to analyze the spatial aggregation degree of the cultural assets based on the cultural density network diagram, obtain an aggregation degree heat map, and calculate the correlation degree between the cultural sites based on the neighborhood cultural map database, and generate a cultural site theme correlation degree matrix. The path module is used to generate a narrative tour path based on the clustering heatmap and the cultural site theme correlation matrix. The scene module is used to determine the content to be generated at each of the cultural points based on the narrative tour path, thereby obtaining digital scene content.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images