AI-driven multi-type point map visualization intelligent point selection method and system

Abstract

The application relates to the technical field of digital media advertisement, in particular to an AI-driven multi-type point map visualization intelligent point selection method and system, which comprises the following steps: through space-semantic joint mapping, heterogeneous attributes are calculated to a unified semantic representation system, and the problem that multi-type points cannot be uniformly compared is solved; through joint determination of space compatibility and semantic matching, preliminary screening of candidate points is realized, and it is ensured that the points simultaneously satisfy spatial layout rationality and semantic demand matching; through the introduction of context semantic association strength and competition resolution constraints, secondary screening is carried out, and the deep matching degree of points and scenes and the competition conflict between points are fully considered; through the adoption of layered visualization identification projection to a map interface, users can intuitively perceive the characteristics of each candidate point in multiple dimensions; through the dynamic updating of point selection constraint conditions in response to the selection adjustment operation of the user on the visualization identification, the preference is continuously learned and the screening strategy is adjusted.

Description

Technical Field

[0001] This invention relates to the field of digital media advertising technology, and in particular to an AI-driven intelligent point selection method and system for visualizing multi-type location maps. Background Technology

[0002] Map point selection is a core decision-making process in urban planning, commercial site selection, and public service facility layout. It typically requires selecting the optimal combination of heterogeneous points within the same map space, including commercial outlets, transportation hubs, and public service facilities. Existing point selection methods generally employ single-dimensional evaluation strategies: either geometrically filtering based on spatial distance thresholds or semantically matching based on attribute keywords. Some solutions introduce simple weighted scoring to integrate multi-dimensional features. However, this approach lacks a spatial-semantic joint evaluation framework for cross-type points, making it impossible to align and compare heterogeneous attribute labels within a unified system.

[0003] However, when selecting points involves multiple types of heterogeneous points and simultaneously requires satisfying both spatial layout rationality and semantic matching, existing methods suffer from a high false positive rate in the initial screening due to the lack of a unified semantic representation system. Heterogeneous labels such as physical and functional attributes cannot be jointly judged across types, and spatial compatibility rules and semantic matching rules are disconnected. Furthermore, the absence of contextual semantic association analysis and inter-site competition resolution mechanisms prevents the automatic selection of candidate sites with overlapping functions or spatiotemporal conflicts. Additionally, the map interface only displays points with flat icons, lacking hierarchical information such as compatibility levels, making it difficult for users to quickly assess the overall merits of each candidate point, resulting in low decision-making efficiency and high subjectivity. Moreover, existing technologies lack an interactive feedback loop; constraints cannot be adaptively updated after users adjust their point selection preferences, failing to continuously learn user decision-making intentions, and thus exhibiting severely insufficient applicability in real-world business scenarios requiring iterative optimization. Therefore, this invention proposes an AI-driven intelligent point selection method and system for multi-type point map visualization. Summary of the Invention

[0004] The purpose of this invention is to solve the problems in the background art, and to propose an AI-driven intelligent point selection method and system for visualization of multi-type point maps.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of this invention provides an AI-driven intelligent point selection method for visualizing multi-type point maps, including: S1. Obtain spatial location data of multiple types of sites and heterogeneous attribute label sets corresponding to each type of site. Through spatial-semantic joint mapping, reduce the heterogeneous attribute label sets to a unified semantic representation system and construct a visual candidate map. S2. Obtain the point selection constraints of the target scene, and perform joint compatibility judgment between the point selection constraints and each candidate site in the visualized candidate map to select the first set of sites that simultaneously satisfies the spatial compatibility rules and semantic matching rules; wherein, the point selection constraints include spatial compatibility rules and semantic matching rules. S3. Based on the contextual semantic association strength between each point in the first point set and the target scene, and combined with the competition resolution constraints between each point in the first point set, the first point set is filtered a second time to generate the second point set. S4. Project each point in the second point set onto the map interface with a hierarchical visualization label, wherein the hierarchical visualization label includes the spatial location label, semantic type label label, and compatibility judgment level label of the point. S5. In response to the selection or adjustment of layered visual markers on the map interface, update the point selection constraints.

[0006] Further, S1 includes: Extract the spatial coordinates of each type of site, and construct a spatial distribution topology network based on the proximity relationships between the spatial coordinates; Simultaneously, heterogeneous attribute tag sets for each type of site are extracted, which include physical attribute tags, functional attribute tags, and scene attribute tags for the site. By mapping heterogeneous attribute tag sets to the same semantic representation system, determining the semantic affinity between different types of sites in the semantic representation system, and generating a semantic affinity relationship network; By overlaying and fusing spatially distributed topological networks with semantic affinity networks, a visual candidate map is generated.

[0007] Furthermore, mapping heterogeneous attribute tag sets to the same semantic representation system includes: Semantic encoding is performed on each type of attribute tag in the heterogeneous attribute tag set to obtain the semantic representation value of the corresponding attribute tag; Perform cross-type semantic alignment on semantic representation values ​​and classify semantic representation values ​​of different attribute types into the same semantic metric interval; Within the same semantic metric interval, the combination of semantic representation values ​​of multiple attribute labels belonging to each type of site is used as the unified semantic representation of that site in the semantic representation system.

[0008] Further, S2 includes: Based on the positional relationship between the spatial coordinates of the candidate site and the spatial range of the target scene, and combined with spatial compatibility rules, a spatial compatibility determination result is generated. Based on the degree of semantic matching between the unified semantic representation of candidate sites and the semantic requirement description of the target scene, and combined with semantic matching rules, a semantic compatibility judgment result is generated. The spatial compatibility determination result is combined with the semantic compatibility determination result. When both satisfy their respective compatibility conditions, the corresponding candidate site is determined to pass the joint compatibility determination.

[0009] Furthermore, the method for establishing the spatial compatibility rules and semantic matching rules in the point selection constraints includes: Obtain historical site selection decision records, and extract the spatial feature distribution range and semantic feature distribution range of adopted sites from the historical site selection decision records; The criteria for determining spatial compatibility rules are generated based on the spatial feature distribution range of the adopted sites, and the criteria for determining semantic matching rules are generated based on the semantic feature distribution range of the adopted sites. Two decision criteria are used as the initial settings for point selection constraints, and the decision criteria are iteratively corrected based on the subsequent point selection decision results.

[0010] Further, S3 includes: Obtain the contextual semantic description of the target scenario, which includes the functional requirements, audience characteristics, and timeliness requirements of the target scenario; The unified semantic representation of each point in the first point set is semantically matched with the context semantic description to generate the context semantic association strength level corresponding to each point. Obtain the degree of functional overlap and spatiotemporal conflict among the points in the first point set, and generate competition resolution constraints based on the degree of functional overlap and spatiotemporal conflict. Based on the contextual semantic association strength level and competition resolution constraints, the first point set is further filtered to retain the points whose contextual semantic association strength level reaches the preset association level and whose competition resolution constraints meet the preset resolution conditions, thus generating the second point set.

[0011] Furthermore, the generation of competition resolution constraints based on the degree of functional overlap and the degree of spatiotemporal conflict includes: Analyze the functional attribute overlap between any two sites in the first point set and calculate the functional attribute overlap degree; identify site pairs whose functional attribute overlap degree exceeds a preset overlap threshold as functional attribute overlapping site pairs. Analyze the spatiotemporal coverage crossover between any two points in the first point set, and calculate the degree of spatiotemporal coverage crossover; identify point pairs whose spatiotemporal coverage crossover degree exceeds a preset crossover threshold as spatiotemporal coverage crossover point pairs; Pairs of sites with overlapping functional attributes and pairs of sites with overlapping spatiotemporal coverage are jointly labeled as competing site pairs. Among the competing site pairs, sites with high levels of contextual semantic association are retained, while sites with low levels of contextual semantic association are eliminated, which serves as the competition resolution constraint.

[0012] Further, S4 includes: Generate spatial location markers at the spatial coordinates corresponding to each second point on the map interface; The semantic type label for the site is displayed in an additional area besides the spatial location identifier; The compatibility assessment level of the site is displayed in the associated information area of ​​the spatial location identifier. The compatibility assessment level is presented in a hierarchical and visual manner to show the comprehensive evaluation results of the site in the joint compatibility assessment and secondary screening.

[0013] Further, S5 includes: The system detects the selection operation of any hierarchical visual marker on the map interface and identifies the site corresponding to the selection operation as the user's preferred site. Extract the spatial feature attributes of user preference sites, and adaptively shrink or expand the spatial compatibility rules based on the spatial feature attributes of user preference sites; Extract semantic feature attributes of user preference sites, and adjust the semantic matching rules according to the semantic feature attributes of user preference sites to match preferences; The spatial compatibility rules after adaptive shrinkage or expansion and the semantic matching rules after matching preference adjustment are used as the updated point selection constraints.

[0014] The second aspect of this invention provides an AI-driven intelligent point selection system for visualization of multi-type point maps, comprising: Visual candidate map construction module: acquires spatial location data of multiple types of sites and heterogeneous attribute label sets corresponding to each type of site, and reduces the heterogeneous attribute label sets to a unified semantic representation system through spatial-semantic joint mapping to construct a visual candidate map; The first site set filtering module: obtains the site selection constraints of the target scene, performs joint compatibility judgment between the site selection constraints and each candidate site in the visualized candidate map, and filters out the first site set that simultaneously satisfies the spatial compatibility rules and semantic matching rules; wherein, the site selection constraints include spatial compatibility rules and semantic matching rules; The second point set generation module: Based on the contextual semantic association strength between each point in the first point set and the target scene, and combined with the competition resolution constraints between each point in the first point set, the first point set is filtered a second time to generate the second point set. Layered visualization label projection module: Projects each point in the second point set onto the map interface with layered visualization labels, where the layered visualization labels include the spatial location label of the point, the semantic type label label, and the compatibility judgment level label; Interactive Response Constraint Update Module: Updates the selected point constraints in response to the selection or adjustment of hierarchical visual markers on the map interface.

[0015] Compared with existing technologies, the beneficial effects of this invention in providing an AI-driven intelligent point selection method and system for visualizing multiple types of point maps are as follows: 1) By using spatial-semantic joint mapping, heterogeneous attribute labels of multiple types of sites are reduced to a unified semantic representation system, effectively solving the problem of inconsistent comparison caused by differences in attribute dimensions of different types of sites. The constructed visual candidate map takes into account both spatial distribution and semantic association, providing a comprehensive, structured and computable candidate basis for subsequent site selection, and improving the scientificity and coverage of site selection. Through joint compatibility judgment of spatial compatibility rules and semantic matching rules, the preliminary screening of candidate sites is achieved, ensuring that the selected sites simultaneously meet the requirements of spatial layout rationality and semantic matching, effectively improving the reliability and accuracy of the first point set. 2) By introducing contextual semantic association strength and competition resolution constraints for secondary screening, the deep semantic fit between the site and the target scene, as well as the functional overlap and spatiotemporal conflict between sites, are fully considered. This effectively eliminates redundant competition, making the final site selection results more in line with actual application needs and improving the optimality and practicality of the solution. By using hierarchical visualization labels to synchronously project three-dimensional information such as spatial location, semantic type, and compatibility level onto the map interface, users can intuitively and multidimensionally perceive the characteristics and advantages and disadvantages of each candidate site, significantly reducing the cognitive load of decision-making and improving the efficiency of site selection interaction and user experience. By responding to users' selection or adjustment of visualization labels, the site selection constraints are dynamically updated, realizing adaptive iterative optimization through human-machine collaboration. This allows for continuous learning of user preferences and adjustment of screening strategies, improving the personalization and long-term practicality of the site selection results. Attached Figure Description

[0016] Figure 1 This is a flowchart of the AI-driven intelligent point selection method for visualizing multi-type point maps proposed in this invention.

[0017] Figure 2 This is a module diagram of the AI-driven intelligent point selection system for visualization of multi-type point maps proposed in this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 This invention provides an AI-driven intelligent point selection method for visualizing multiple types of point maps, including: S1. Obtain spatial location data of multiple types of sites and heterogeneous attribute label sets corresponding to each type of site. Through spatial-semantic joint mapping, reduce the heterogeneous attribute label sets to a unified semantic representation system and construct a visual candidate map containing the spatial distribution topology and semantic affinity of sites. S2. Obtain the point selection constraints of the target scene, and perform joint compatibility judgment between the point selection constraints and each candidate site in the visualized candidate map to select the first set of sites that simultaneously satisfies the spatial compatibility rules and semantic matching rules; wherein, the point selection constraints include spatial compatibility rules and semantic matching rules. S3. Based on the contextual semantic association strength between each point in the first point set and the target scene, and combined with the competition resolution constraints between each point in the first point set, the first point set is filtered a second time to generate the second point set. S4. Project each point in the second point set onto the map interface with a hierarchical visualization label, wherein the hierarchical visualization label includes the spatial location label, semantic type label label, and compatibility judgment level label of the point. S5. In response to the selection or adjustment operation of the layered visualization markers on the map interface, update the point selection constraints and return to steps S2 to S4; wherein, updating the point selection constraints and returning to steps S2 to S4 includes: Re-execute the joint compatibility judgment and secondary screening with the updated point selection constraints to generate an updated second point set; re-project the updated second point set onto the map interface with updated hierarchical visualization labels; continuously respond to interactive operations on the map interface until a point selection confirmation instruction is received or the updated second point set tends to be consistent with the user's expected point selection result.

[0020] In this embodiment of the invention, the specific implementation of step S1, which acquires spatial location data of multiple types of sites and heterogeneous attribute label sets corresponding to each type of site, and constructs a visual candidate map by reducing the heterogeneous attribute label sets to a unified semantic representation system through spatial-semantic joint mapping, is as follows: S11. The process of extracting the spatial coordinates of each type of site and constructing a spatial distribution topology network based on the proximity relationships between the spatial coordinates is as follows: Retrieve the latitude and longitude coordinates of all candidate points from the point management database; where latitude and longitude coordinates are the absolute positions of the points in geographic space; take each point as a node, calculate the Euclidean distance between it and all other candidate points within the target area displayed on the current visualization map interface; for example, the Euclidean distance between point A and point B is calculated using the distance formula. An adaptive proximity threshold calculation method based on point density is adopted: First, the target area is divided into kilometer grids, and the number of points in each grid is counted to obtain the local point density; based on the ratio of the local point density to the baseline density, the initial proximity threshold (e.g., 500 meters) is scaled inversely, with the threshold shrinking in high-density areas and expanding in low-density areas, and the shrinkage and expansion not exceeding ±40% of the baseline threshold; the baseline density is preset according to the urban built-up area level of the target area, for example, the baseline density of the main urban area is 20 points / square kilometer, and the baseline density of the suburbs is 5 points / square kilometer; For any two points, if their calculated Euclidean distance is less than or equal to the proximity threshold, a connecting edge is established between the two points, indicating that they are spatially adjacent. After traversing all point pairs, an undirected graph structure consisting of point nodes and connecting edges is generated, i.e., a spatially distributed topological network. In this spatially distributed topological network, the weight of the connecting edge is initialized to the reciprocal of the Euclidean distance between the two points. The closer the Euclidean distance, the higher the weight, representing a tighter spatial association. At the same time, for point pairs that cross grids, a grid density correction coefficient is introduced to perform secondary calibration of the edge weights, eliminating weight deviations caused by density differences. For example, the distance between the gate of the east gate of community A and the express delivery locker inside community A is 80 meters, which is less than the proximity threshold, so a strong connecting edge is established between them. However, there is no direct connecting edge between the point of community A and the point of community B, which is 1 kilometer away, because the Euclidean distance exceeds the proximity threshold.

[0021] S12. Simultaneously, extract heterogeneous attribute tag sets for each type of site, where the heterogeneous attribute tag sets include physical attribute tags, functional attribute tags, and scene attribute tags for the sites. Specifically, attribute information of the locations is extracted from database tables of different dimensions and categorized and integrated; physical attribute tags are obtained directly from the equipment archive, including the physical size of the location, screen resolution, brightness, equipment model, installation height and power supply method; functional attribute tags are derived from operation records, including the advertising material formats supported by the location, whether interactive functions are supported, playable time periods, and the main audience types obtained from historical data statistics; scene attribute tags are obtained from the environment and business database, including the specific scene name where the location is located, the building age of the community, the average housing price range, and the industry advertising category with the best performance in the historical campaigns for that location.

[0022] S13. Map heterogeneous attribute tag sets to the same semantic representation system, determine the semantic affinity between different types of sites in the semantic representation system, and generate a semantic affinity network, including: Semantic encoding is performed on each type of attribute tag in the heterogeneous attribute tag set to obtain the semantic representation value of the corresponding attribute tag: An independent semantic encoder is used to process three types of attribute tags: physical, functional, and scene. For physical attribute tags, the semantic encoder maps their values ​​or categories to a preset specification level range. For example, a screen size of 1.5 square meters is quantized as a physical specification value of 0.75, and a resolution of 1920x1080 is quantized as a clarity value of 0.9. For functional attribute tags, the semantic encoder encodes them based on the matching degree between the keyword vector of the functional description and the functional vocabulary. For example, support for face recognition interaction is encoded as an interaction capability value of 0.95. For scene attribute tags, the semantic encoder determines their encoding based on the scene classification tree. For example, the scene of the main entrance of a high-end residential community is encoded as a scene value value of 0.88. All encoding results are normalized to values ​​between 0 and 1.

[0023] Perform cross-type semantic alignment on semantic representation values ​​and classify semantic representation values ​​of different attribute types into the same semantic metric interval; Understandably, semantic alignment is performed on physical, functional, and scene attributes due to their different encoding benchmarks; a semantic association matrix across attribute types is established through learning from a large amount of sample data. The construction process of the semantic association matrix involves collecting a large amount of historically successful location combination data and analyzing the co-occurrence patterns and statistical correlations between different attribute dimension values. For example, it involves analyzing the positive correlation between locations with large screen sizes and brand display function preferences, or the strong correlation between locations located in high-end communities and the characteristics of high-net-worth audiences. Simultaneously, domain expert knowledge can be introduced to manually calibrate and assign values ​​to the influence relationships between key dimensions. Based on statistical patterns and prior knowledge, an M×N matrix is ​​constructed, where M and N represent the number of dimensions of different attribute types, and each element value in the matrix represents the semantic influence weight of the row dimension on the column dimension. For example, the values ​​corresponding to the physical specifications row and the scene value column in the matrix quantify the weight of the physical specifications' contribution to the scene value. Using this semantic association matrix, the original semantic representation values ​​of various attributes are projected into a common high-dimensional semantic space, eliminating differences in dimensions and benchmarks; After projection, cross-domain semantic secondary calibration is performed, including: calculating the variance distribution of the projected vector of each point in each dimension, identifying anomalous dimensions whose variance deviates from the global mean by ±2 standard deviations; for anomalous dimensions, a dimension calibration coefficient is introduced for linear correction. The dimension calibration coefficient is determined by the distribution quantile of the anomalous dimension in the historical adopted points. After correction, the variance of all anomalous dimensions is converged to within ±0.5 standard deviations of the global mean, solving the technical problems of dimensional drift and partial dimension discrimination failure after heterogeneous attribute projection. Finally, the projected values ​​across all dimensions are linearly transformed and reduced to a dimensionless semantic metric range, such as 0 to 100. Specifically, the linear transformation involves pre-setting a target range (e.g., 0 to 100) for the reduced semantic metric range; calculating the minimum and maximum values ​​of each dimension across all point samples after projection using the semantic association matrix; and then linearly transforming the projected value of each dimension at each point using the formula (current value - minimum value) / (maximum value - minimum value) × 100 to map it to the target range of 0-100, thus normalizing and making the scores of all dimensions comparable. After the transformation, the physical specification score, functional adaptation score, and scene matching score of a point are all within the same score range, making them directly comparable.

[0024] Within the same semantic metric interval, the combination of semantic representation values ​​of multiple attribute labels belonging to each type of site is used as the unified semantic representation of that site in the semantic representation system. After cross-type alignment and reduction, each location receives a set of multi-dimensional scores within the same semantic measurement interval. These multi-dimensional scores are then combined according to a fixed dimensional order to form a multi-dimensional semantic vector, which serves as the unified semantic representation of that location. For example, the unified semantic representation of a location can be a five-dimensional vector in the format of [physical specification score, functional interaction score, audience matching score, scene value score, time period adaptation score]. This vector comprehensively reflects the quantitative characteristics of the location across multiple core semantic dimensions. The unified semantic representations of all locations constitute a semantic feature matrix, providing a standardized and structured data foundation for subsequent semantic matching and comparison.

[0025] After mapping, each point is transformed into a set of standardized semantic feature vectors. The semantic affinity is determined by calculating the cosine similarity between the semantic feature vectors of two points. If the semantic similarity between two points exceeds a preset affinity threshold, a connection edge is established between the two points in the network. The weight of the connection edge is the similarity value, thus forming a network structure that reflects the semantic similarity between points, i.e., a semantic affinity network. The affinity threshold is a configurable parameter, such as 0.7, and its setting is based on historical data analysis and business experience.

[0026] S14. The specific steps for overlaying and fusing the spatially distributed topological network with the semantic affinity relation network to generate a visualized candidate map are as follows: The spatial distribution topology network and the semantic affinity relationship network are fused into a graph. For the same set of points, their connecting edges in the two networks are merged. If two points have connecting edges in both the spatial distribution topology network and the semantic affinity relationship network, the weight of the fused edge is the weighted sum of the weights of the two networks. For example, the spatial proximity weight accounts for 0.6 and the semantic affinity weight accounts for 0.4. Finally, a fused network is generated. The nodes of the fused network carry point information, and the edge weights comprehensively represent the strength of the dual spatial and semantic association between points. This fused network constitutes the underlying data structure of the visualization candidate graph.

[0027] In this embodiment of the invention, the detailed implementation steps of S2, which involves obtaining the selection constraints of the target scene and performing a joint compatibility determination between the selection constraints and each candidate site in the visualized candidate map, to select a first set of sites that simultaneously satisfies both spatial compatibility rules and semantic matching rules, include: S21. The method for generating spatial compatibility determination results based on the positional relationship between the spatial coordinates of candidate sites and the spatial range of the target scene, combined with spatial compatibility rules, is as follows:

[0028] The spatial scope of the target scene is defined in the form of a geofence, such as a polygonal area or a circular area with a certain radius around a center point. The shortest distance from the spatial coordinates of each candidate point to the geofence is calculated. Spatial compatibility rules pre-set various spatial relationship conditions and their corresponding compatibility levels. For example, the rules might stipulate: if a point is located inside a geofence, the spatial compatibility judgment result is fully compatible; if a point is located outside a geofence but within a preset buffer distance, such as 300 meters, it is determined to be edge compatible; if a point exceeds the buffer distance but is within the same administrative region, it is determined to be weakly compatible; if a point is located in completely different areas, it is determined to be incompatible. Based on the calculated distance and the positional relationship between the point and the geofence, the conditions in the rules are matched, and the corresponding compatibility judgment result and level score are output. For example, in an advertising campaign targeting a certain business district, points located inside the business district will be determined to be fully compatible.

[0029] S22. The specific process of generating a semantic compatibility determination result based on the degree of semantic matching between the unified semantic representation of candidate sites and the semantic requirement description of the target scene, combined with semantic matching rules, includes: The semantic requirements description of the target scenario is also encoded as a requirement vector with the same dimension as the unified semantic representation of the location; for example, for the new car community promotion scenario of a high-end car brand, its requirement vector may emphasize high scenario value score, high physical specification score and high net worth audience matching score; the semantic matching rules predefine the weight and matching threshold of each semantic dimension. Calculate the multi-dimensional semantic fit score between the candidate point semantic representation vector and the scene demand vector: Pre-define a corresponding fit calculation function for each dimension of the semantic representation, defined according to the business characteristics of that dimension; for example, for the scene value score, use the corresponding value in the demand vector / the actual value of the point to calculate the saturation fit; for the audience matching score, use 1 - the absolute difference between the two to calculate the proximity fit; assign a weight to each dimension, reflecting the importance of that dimension in the target of this point selection; finally, the semantic matching score is the weighted sum of the fit calculation results of each dimension and its weight. The semantic matching score is compared with the matching threshold in the semantic matching rule. For example, the semantic matching rule can be set as follows: a semantic matching score greater than or equal to 0.8 is a high match, generating a fully compatible result; a semantic matching score between 0.6 and 0.8 is a medium match, generating a partially compatible result; and a semantic matching score less than 0.6 is a low match, generating an incompatible result. The result and the corresponding semantic matching score are then output.

[0030] S23. Combine the spatial compatibility determination result with the semantic compatibility determination result. When both satisfy their respective compatibility conditions, the corresponding candidate site is determined to pass the joint compatibility determination. The compatibility conditions include spatial compatibility conditions and semantic compatibility conditions, which correspond to the spatial compatibility determination result and the semantic compatibility determination result, respectively. A flexible judgment logic with dynamic spatial-semantic coupling is adopted: a single extreme value exemption rule is pre-set. When the spatial compatibility of a candidate point is judged to be fully compatible and the spatial compatibility score ranks in the top 10% of the current candidate set, the matching threshold in the semantic matching rule is relaxed downward by 10%; when the semantic matching degree of a candidate point is judged to be highly matched and the score ranks in the top 10% of the current candidate set, the buffer distance of spatial compatibility is relaxed outward by 20%; at the same time, a fallback constraint is set, that is, if any dimension is judged to be incompatible, no matter how high the score of the other dimension is, the judgment shall not pass. The methods for establishing spatial compatibility rules and semantic matching rules in point selection constraints include: S231. The implementation method for obtaining historical site selection decision records and extracting the spatial feature distribution range and semantic feature distribution range of adopted sites from the historical site selection decision records is as follows: Retrieve all successfully executed site selection plans from the decision log database over a historical period; for each site selection plan record, extract the set of all sites that were ultimately adopted. The method for extracting the spatial feature distribution range is as follows: Calculate the latitude and longitude coordinates of all points in the set of adopted points, identify the main spatial clustering areas of the points using a clustering algorithm, such as using the DBSCAN algorithm to find the core clusters, and calculate the boundary or center and radius of each cluster to define the spatial preference distribution range of historical point selections. It is understandable that when identifying the main clustering areas (core clusters) from the latitude and longitude coordinates of historical adopted points using the DBSCAN algorithm, it is necessary to set key parameters of the algorithm; specifically, setting the neighborhood radius and the minimum number of points required to form a core cluster. In the community media point selection scenario of this invention, the spatial clustering of points typically corresponds to residential areas or business districts. The range has practical business significance; therefore, the neighborhood radius can be set to 200 meters, indicating that the algorithm considers points within 200 meters of each other as neighbors; the minimum number of points can be set to 5, indicating that at least 5 historically adopted points need to exist within a 200-meter neighborhood of a core point for the area to be judged as a statistically significant core cluster; the above two example parameter settings are based on the distribution density of typical urban community points and business logic, and are used to automatically and repeatedly extract the spatial preference distribution range from historical data; in practical applications, those skilled in the art can adaptively adjust the two parameters according to the actual distribution density of points in the target area (such as high density in the city center and low density in the suburbs); The process of extracting the semantic feature distribution range is as follows: obtain the semantic representation vector of all points in the set of adoption points, and calculate the numerical distribution of each semantic dimension (such as scene value score, audience matching score, etc.), such as calculating the mean, standard deviation, or determining the main distribution range by quantiles; for example, historical data shows that for the promotion of fast-moving consumer goods, the scene value score of adoption points is mostly concentrated between 40 and 70 points, and the audience matching score is mostly concentrated between 60 and 90 points.

[0031] S232. Generate a judgment criterion for spatial compatibility rules based on the spatial feature distribution range of the adopted sites, and generate a judgment criterion for semantic matching rules based on the semantic feature distribution range of the adopted sites. Specifically, based on the extracted historical spatial feature distribution range, a judgment criterion for spatial compatibility rules is generated; the spatial range of the main cluster area (core cluster) of historical adoption points is used as the initial preferred spatial range; the judgment criterion for spatial compatibility rules is set as follows: if a candidate point is located within any preferred spatial range or its buffer, its spatial compatibility judgment result is at least edge compatibility or higher; for semantic matching rules, based on the extracted historical semantic feature distribution range, preference intervals for each semantic dimension are generated; for example, for the dimension of audience matching score, if the score of the historical adoption point is mainly distributed in the [60,90] interval, then this interval is set as the benchmark matching interval for this dimension; the judgment criterion for semantic matching rules can be set as follows: in the semantic representation vector of the candidate point, the score of the key dimension (determined according to the weight of historical data) must fall within the corresponding benchmark matching interval to obtain a judgment result of medium matching or above.

[0032] S233. Use two judgment criteria as the initial settings for point selection constraints, and iteratively correct the judgment criteria based on subsequent point selection decision results: Use the generated spatial compatibility rule judgment criteria and semantic matching rule judgment criteria as the initial point selection constraints; After each subsequent point selection decision is completed, regardless of whether it is adopted or not, use the point set considered in this decision, the final adopted point set, and the user feedback results as new samples to re-execute the extraction and analysis of the spatial and semantic feature distribution range, and use this to fine-tune and update the judgment criteria such as the preferred spatial range and the benchmark matching interval, so that the rules can adapt to changes in business trends and user preferences.

[0033] Meanwhile, a sample confidence weighting mechanism is introduced during the iterative correction process: based on the actual deployment effect data of historical site selection schemes, a confidence weight is assigned to each historical sample. The sample weight of the scheme with a conversion rate higher than the industry average is set to 1.5, the sample weight of the scheme with a conversion rate lower than the industry average is set to 0.5, and the sample weight of the scheme with unverified effect is set to 1.0. Based on the weighted sample set, the spatial and semantic feature distribution range is recalculated so that the judgment benchmark converges to the high-performance scheme.

[0034] Understandably, the overall logic of the joint compatibility determination is an AND operation; it reads the spatial compatibility determination result and the semantic compatibility determination result respectively; the preset compatibility condition requires that neither determination result can be incompatible; specifically, only when the spatial compatibility determination result is not incompatible and the semantic compatibility determination result is not incompatible, is the candidate point determined to pass the joint compatibility determination; for example, even if a point is spatially edge-compatible, but has a high semantic matching score, it is determined to pass the overall determination; conversely, if a point is spatially fully compatible, but has a low semantic matching score, it is determined to fail the overall determination; points that pass the determination will be included in the first point set; the spatial compatibility level and semantic matching score of each passing point will be recorded as a reference for subsequent sorting or filtering.

[0035] In this embodiment of the invention, the specific implementation of step S3, which involves secondary filtering of the first point set based on the contextual semantic association strength between each point in the first point set and the target scene, and combined with the competition resolution constraints among the points in the first point set, to generate the second point set, includes: S31. Obtain the contextual semantic description of the target scene, wherein the contextual semantic description includes the functional requirements, audience characteristics and timeliness requirements of the target scene; Specifically, the contextual semantic description of the target scenario is entered by the user when creating the selection task or loaded from the project template; the functional requirements describe the specific goals that the advertising campaign hopes to achieve, such as brand image display, product promotion and traffic generation, or user interaction collection; the audience characteristics describe the attributes of the target audience that the advertisement hopes to reach, such as working people aged 25-40, families with cars, or families with children in the community; the timeliness requirements describe the time constraints or periodic characteristics of the advertisement, such as needing to be launched one week before the National Day holiday, only playing during weekday morning rush hours, or continuously launching for one month; these natural language descriptions are converted into structured contextual semantic description objects.

[0036] S32. The method for generating the context semantic association strength level corresponding to each point by performing semantic association matching between the unified semantic representation of each point in the first point set and the context semantic description is as follows: The structured contextual semantic description is also quantized into a multi-dimensional demand vector; the correlation strength between the semantic representation vector of each point in the first point set and the demand vector is calculated; The association strength is calculated using a matching model based on a multidimensional semantic attention network; It should be noted that the matching model based on a multidimensional semantic attention network is a lightweight neural network designed for community media location selection scenarios. Its necessary structure includes an input layer, an attention fusion layer, and an output scoring layer. The input layer receives two types of data: first, the semantic representation vector of the location (e.g., a multidimensional vector containing physical specifications, scene value, audience matching scores, etc.); second, the contextual semantic description vector encoded by natural language processing, which also structurally represents the functional requirements, audience characteristics, and timeliness requirements of the target scene. The attention fusion layer is the key to combining the model with domain knowledge. This layer has built-in trainable weight parameters, which dynamically calculate the attention weights of each element in the context description for different semantic dimensions of the location. For example, when the context description emphasizes brand image display, the model parameters will cause this layer to assign higher attention weights to the physical specifications and scene value scores of the location; if the description emphasizes holiday promotions, it will assign higher weights to the time period suitability score and pedestrian coverage score. Through this mechanism, the model weightedly fuses the original semantic vector of the point with the context attention weight to generate a context-aware point feature vector; the output scoring layer is a fully connected layer that maps the fused context-aware point feature vector to a scalar value, namely the context semantic association strength score; the score directly reflects the degree of fit of the point under specific context constraints. The training steps of the model are as follows: First, collect historical high-quality site selection scheme data as training samples. Each sample includes: 1) the semantic representation vector of the site; 2) the context description vector of the site selection task; 3) the label of the association strength. For example, the sites that are finally adopted are marked as high-intensity positive samples, and the sites that are explicitly excluded are marked as low-intensity negative samples. Second, using mean squared error or cross-entropy as the loss function, the backpropagation algorithm is used to optimize all parameters in the model, including the weight matrix of the attention layer and the weights and biases of the output layer. Through iterative training with a large number of samples, the model learns to accurately evaluate the association strength between the site and the context description.

[0037] S33. Obtain the degree of functional overlap and spatiotemporal conflict among the points in the first point set, and generate competition resolution constraints based on the degree of functional overlap and spatiotemporal conflict: S331. Analyze the functional attribute overlap between any two sites in the first point set and calculate the functional attribute overlap degree; identify site pairs whose functional attribute overlap degree exceeds the preset overlap threshold as functional attribute overlapping site pairs. Specifically, the process involves iterating through all possible point pairs in the first point set; for each point pair, extracting the functionally relevant sub-vectors in the unified semantic representation, such as dimensions like industry fit, audience profile, and interaction capabilities; calculating the percentage of overlapping area between the two sub-vectors as the functional attribute overlap of the point pair; comparing the values ​​(or value ranges) of the two points for each functional dimension and calculating the proportion of their overlapping portion to their union; weighting the overlap ratios of all functional dimensions, with the weights set according to the importance of each dimension in distinguishing functions; and finally, obtaining the weighted average overlap ratio, which is the functional attribute overlap of the point pair. The overlap threshold is a preset empirical value, such as 0.7. This means that when the similarity of two locations in terms of functional attributes exceeds 70%, their target audience and advertising functions are considered to be highly homogeneous, which may lead to internal competition. All locations whose functional attribute overlap calculation results exceed this overlap threshold are marked to form a list of functional attribute overlap location pairs. For example, two locations that both target families with children and are good at playing educational and training advertisements may have a functional overlap of up to 0.9, which exceeds the overlap threshold, and are therefore identified as an overlap pair.

[0038] S332. Analyze the spatiotemporal coverage crossover between any two points in the first point set, and calculate the degree of spatiotemporal coverage crossover; identify point pairs whose spatiotemporal coverage crossover degree exceeds the preset crossover threshold as spatiotemporal coverage crossover point pairs. Specifically, for each pair of locations, the degree of spatiotemporal coverage overlap is calculated comprehensively from both time and space dimensions. In the time dimension, the overlap ratio of available advertising time slots between the two locations is calculated. In the spatial dimension, the physical distance between the two locations is calculated, and the overlap of their spatial coverage influence range is evaluated based on a distance decay model. For example, a distance threshold is set, and locations within the distance threshold are considered to have spatial coverage overlap. The degree of spatiotemporal coverage overlap is a weighted product of the time overlap ratio and the spatial coverage overlap. The overlap threshold is a preset comprehensive threshold, such as 0.5. The degree of spatiotemporal coverage overlap for all location pairs is calculated, and location pairs exceeding the overlap threshold are identified as spatiotemporal coverage overlap location pairs. For example, two locations that are only 30 meters apart and have 6 hours of completely overlapping broadcast time slots each day may have a spatiotemporal overlap degree of 0.8 and be identified as an overlap pair.

[0039] S333. Mark the functional attribute overlapping site pairs and the spatiotemporal coverage intersection site pairs as competing site pairs; among the competing site pairs, retain the sites with high contextual semantic association strength level and eliminate the sites with low contextual semantic association strength level as the competition resolution constraint condition; For site pairs with functional attribute overlap of less than 0.3 and complementary scene attribute coverage, their complementary gain value is calculated. The complementary gain value is calculated by multiplying the inverse of the audience overlap between the two sites by the proportion of the union of scene coverage. When the complementary gain value exceeds the preset gain threshold, even if there is some overlap in the spatiotemporal coverage of the two sites, they are not marked as competing site pairs, but rather as cooperative site pairs, and are retained in subsequent screening to achieve a balance between competition resolution and cooperative gain. Meanwhile, a multi-round iterative global competition resolution algorithm is adopted, including: in the first round, all competing site pairs are identified, sorted in descending order of association strength difference, and competing pairs with the largest differences are resolved first; after each set of competing pairs is resolved, the competition relationship network of the remaining sites is updated, and then the next round of resolution is performed; the iteration is repeated until there are no competing site pairs that meet the conditions; at the same time, a global coverage constraint is set to ensure that the spatial coverage of the final second site set is not less than 60% of the first site set, to prevent the coverage shrinkage caused by excessive resolution, and to achieve a balance between local competition optimization and global coverage effect; Understandably, for each pair of locations, the degree of spatiotemporal coverage overlap is calculated comprehensively from both time and space dimensions. In the time dimension, the overlap ratio of available advertising time slots between the two locations is calculated. In the space dimension, the physical distance between the two locations is calculated, and the overlap of their spatial coverage influence range is evaluated based on a distance decay model. For example, a distance threshold is set, and locations within the distance threshold are considered to have spatial coverage overlap. The degree of spatiotemporal coverage overlap is a weighted product of the time overlap ratio and the spatial coverage overlap. The overlap threshold is a preset comprehensive threshold, such as 0.5. The degree of spatiotemporal coverage overlap for all location pairs is calculated, and location pairs exceeding the overlap threshold are identified as spatiotemporal coverage overlap location pairs. For example, two locations that are only 30 meters apart and have 6 hours of completely overlapping broadcast time slots each day may have a spatiotemporal overlap of 0.8 and be identified as an overlap pair.

[0040] S34. Based on the contextual semantic association strength level and competition resolution constraints, perform secondary screening on the first point set, retaining the points whose contextual semantic association strength level reaches the preset association level and whose competition resolution constraints meet the preset resolution conditions, and generate the second point set. Specifically, the first set of points is traversed and filtered; a preset threshold for the association level is set, such as requiring the contextual semantic association strength level of the points to be strong or medium; points with weak association levels are directly eliminated; the competitive relationship between points that meet the association level requirements is handled; the resolution condition is preset as follows: in each pair of competing points identified due to functional overlap or spatiotemporal conflict, only points with higher association strength levels are retained; if the association strength levels are the same, points with higher original scores in spatial compatibility or semantic matching degree are retained; according to the resolution condition, points with competition are adjudicated; finally, all points that pass the association level screening and win in all the adjudication of the competing point pairs involved or are judged as non-competitive are set into the second set of points; this process effectively ensures high association and low internal friction within the final set of points.

[0041] In this embodiment of the invention, step S4 projects each point in the second point set onto the map interface using a hierarchical visualization identifier. The detailed implementation steps of this hierarchical visualization identifier, which includes spatial location identifiers, semantic type label identifiers, and compatibility judgment level identifiers, include: S41. Generate spatial location markers at the spatial location coordinates corresponding to each second point on the map interface. Specifically, based on the latitude and longitude coordinates of each point in the second point set, they are mapped to the corresponding pixel positions on the visualization map interface; at the pixel position, a visual graphic marker is rendered as a spatial location identifier; the spatial location identifier is usually an icon, such as a fixed-style dot or thumbtack; the center point of the icon is aligned with the coordinate position of the point.

[0042] S42. Display the semantic type label of the site in an additional area of ​​the spatial location identifier, wherein the semantic type label distinguishes different types of sites with a differentiated graphic style; Next to the graphic of each spatial location, an additional label indicating the semantic type of that location is displayed; the label distinguishes the location type through different graphic styles; for example, the location representing a community gate may have a border with a railing pattern on the outside; the location representing a unit access control screen may be designed as a small house shape; and the location representing an elevator advertising screen may be designed as an upward arrow pattern.

[0043] S43. Display the compatibility assessment level indicator of the site in the associated information area of ​​the spatial location identifier. The compatibility assessment level indicator presents the comprehensive evaluation results of the site in joint compatibility assessment and secondary screening in a hierarchical visualization manner. Specifically, when a user hovers their mouse over or clicks on the spatial location marker of a point, an information card or tooltip pops up in its associated information area. This area clearly displays the compatibility rating of the point. This compatibility rating is presented in a hierarchical and visual manner; for example, three stars represent high compatibility, two stars represent moderate compatibility, and one star represents basic compatibility. The star rating combines the point's score ranking in the previous joint compatibility assessment with its contextual semantic relevance strength level in the secondary screening. At the same time, the information card will also briefly display its detailed scores for spatial compatibility and semantic matching in the form of text or a progress bar, allowing users to have a quantitative understanding of the point's advantages and disadvantages.

[0044] In this embodiment of the invention, the specific implementation of S5 updating the point selection constraints in response to the selection or adjustment operation of the layered visualization markers on the map interface is as follows: S51. The method for detecting the selection operation of any layered visual marker on the map interface and determining the site corresponding to the selection operation as the user's preferred site is as follows: The system monitors user interaction events on the map interface in real time. When a user selects a layered visual marker for a location via mouse or touchscreen click, the interaction event is captured. The system parses the interaction event to obtain the unique ID of the location bound to the selected marker. The location corresponding to this ID is extracted from the second location set and marked as the user's preferred location. The final scheduling instruction is generated based on the user's subsequent confirmation action. For example, a user may make a final selection and click the "Confirm Selection" button in a temporary list containing multiple preferred locations. At this time, the system packages the set of all user-confirmed location IDs in the list, along with the constraint context of this selection, to generate the final selection scheduling instruction, which is then sent to the downstream advertising scheduling or resource management system to execute subsequent advertising placement tasks.

[0045] S52. The steps for extracting the spatial feature attributes of user preference sites and adaptively shrinking or expanding the spatial compatibility rules based on the spatial feature attributes of user preference sites are as follows: Extract the spatial coordinates and spatial clustering characteristics of user-preferred locations. If a user selects multiple spatially clustered locations multiple times or in this round, it is determined that the user prefers this area. In this case, the judgment criteria of the spatial compatibility rule can be dynamically reduced. For example, the preferred spatial range can be converged towards this clustered area, or the spatial compatibility judgment threshold for locations outside this area can be increased. Conversely, if the locations selected by the user are very scattered in space, it may be inferred that the user is not sensitive to spatial constraints. Thus, the judgment criteria of the spatial compatibility rule can be appropriately expanded. For example, the buffer distance of the preferred spatial range can be increased so that more surrounding locations can pass the spatial compatibility judgment.

[0046] S53. The specific methods for extracting semantic feature attributes of user preference sites and adjusting semantic matching rules based on these semantic feature attributes include: The system analyzes the set of semantic representation vectors of user preference locations and calculates the statistical characteristics of all semantic representation vectors across various semantic dimensions, such as the mean or mode. If it finds that the locations selected by users exhibit significant consistency across one or more semantic dimensions, the semantic matching rules will be adjusted accordingly. For example, if the location selections by users generally have higher scene value scores, the weight of the scene value dimension will be increased in subsequent matching. Or, if the location selections by users all point to the automotive industry in the advertising industry dimension, the preference for automotive industry adaptation will be implicitly enhanced in the semantic demand description, thereby enabling locations with similar characteristics to obtain higher semantic matching scores in subsequent screening. This adjustment achieves adaptive adaptation between rules and user preferences.

[0047] S54. Use the spatial compatibility rules after adaptive shrinkage or expansion and the semantic matching rules after matching preference adjustment as the updated point selection constraints.

[0048] Please see Figure 2 This invention provides an AI-driven intelligent point selection system for visualization of multiple types of point maps, including: Visual candidate map construction module: acquires spatial location data of multiple types of sites and heterogeneous attribute label sets corresponding to each type of site, and reduces the heterogeneous attribute label sets to a unified semantic representation system through spatial-semantic joint mapping to construct a visual candidate map; The first site set filtering module: obtains the site selection constraints of the target scene, performs joint compatibility judgment between the site selection constraints and each candidate site in the visualized candidate map, and filters out the first site set that simultaneously satisfies the spatial compatibility rules and semantic matching rules; wherein, the site selection constraints include spatial compatibility rules and semantic matching rules; The second point set generation module: Based on the contextual semantic association strength between each point in the first point set and the target scene, and combined with the competition resolution constraints between each point in the first point set, the first point set is filtered a second time to generate the second point set. Layered visualization label projection module: Projects each point in the second point set onto the map interface with layered visualization labels, where the layered visualization labels include the spatial location label of the point, the semantic type label label, and the compatibility judgment level label; Interactive Response Constraint Update Module: Updates the selected point constraints in response to the selection or adjustment of hierarchical visual markers on the map interface.

[0049] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0050] It should be noted that all formulas in this manual are calculated by removing dimensions and taking their numerical values. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0051] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. An AI-driven intelligent point selection method for visualization of multi-type point maps, characterized in that, include: S1. Obtain spatial location data of multiple types of sites and heterogeneous attribute label sets corresponding to each type of site. Through spatial-semantic joint mapping, reduce the heterogeneous attribute label sets to a unified semantic representation system and construct a visual candidate map. S2. Obtain the point selection constraints of the target scene, and perform joint compatibility judgment between the point selection constraints and each candidate site in the visualized candidate map to select the first set of sites that simultaneously satisfies the spatial compatibility rules and semantic matching rules; wherein, the point selection constraints include spatial compatibility rules and semantic matching rules. S3. Based on the contextual semantic association strength between each point in the first point set and the target scene, and combined with the competition resolution constraints between each point in the first point set, the first point set is filtered a second time to generate the second point set. S4. Project each point in the second point set onto the map interface with a hierarchical visualization label, wherein the hierarchical visualization label includes the spatial location label, semantic type label label, and compatibility judgment level label of the point. S5. In response to the selection or adjustment of layered visual markers on the map interface, update the point selection constraints.

2. The AI-driven intelligent point selection method for visualizing multiple types of point maps according to claim 1, characterized in that, S1 includes: Extract the spatial coordinates of each type of site, and construct a spatial distribution topology network based on the proximity relationships between the spatial coordinates; Simultaneously, heterogeneous attribute tag sets for each type of site are extracted, which include physical attribute tags, functional attribute tags, and scene attribute tags for the site. By mapping heterogeneous attribute tag sets to the same semantic representation system, determining the semantic affinity between different types of sites in the semantic representation system, and generating a semantic affinity relationship network; By overlaying and fusing spatially distributed topological networks with semantic affinity networks, a visual candidate map is generated.

3. The AI-driven intelligent point selection method for visualizing multi-type point maps according to claim 2, characterized in that, The mapping of heterogeneous attribute tag sets to the same semantic representation system includes: Semantic encoding is performed on each type of attribute tag in the heterogeneous attribute tag set to obtain the semantic representation value of the corresponding attribute tag; Perform cross-type semantic alignment on semantic representation values ​​and classify semantic representation values ​​of different attribute types into the same semantic metric interval; Within the same semantic metric interval, the combination of semantic representation values ​​of multiple attribute labels belonging to each type of site is used as the unified semantic representation of that site in the semantic representation system.

4. The AI-driven intelligent point selection method for visualizing multiple types of point maps according to claim 1, characterized in that, S2 includes: Based on the positional relationship between the spatial coordinates of the candidate site and the spatial range of the target scene, and combined with spatial compatibility rules, a spatial compatibility determination result is generated. Based on the degree of semantic matching between the unified semantic representation of candidate sites and the semantic requirement description of the target scene, and combined with semantic matching rules, a semantic compatibility judgment result is generated. The spatial compatibility determination result is combined with the semantic compatibility determination result. When both satisfy their respective compatibility conditions, the corresponding candidate site is determined to pass the joint compatibility determination.

5. The AI-driven intelligent point selection method for visualizing multi-type point maps according to claim 4, characterized in that, The methods for establishing the spatial compatibility rules and semantic matching rules in the point selection constraints include: Obtain historical site selection decision records, and extract the spatial feature distribution range and semantic feature distribution range of adopted sites from the historical site selection decision records; The criteria for determining spatial compatibility rules are generated based on the spatial feature distribution range of the adopted sites, and the criteria for determining semantic matching rules are generated based on the semantic feature distribution range of the adopted sites. Two decision criteria are used as the initial settings for point selection constraints, and the decision criteria are iteratively corrected based on the subsequent point selection decision results.

6. The AI-driven intelligent point selection method for visualizing multi-type point maps according to claim 1, characterized in that, S3 includes: Obtain the contextual semantic description of the target scenario, which includes the functional requirements, audience characteristics, and timeliness requirements of the target scenario; The unified semantic representation of each point in the first point set is semantically matched with the context semantic description to generate the context semantic association strength level corresponding to each point. Obtain the degree of functional overlap and spatiotemporal conflict among the points in the first point set, and generate competition resolution constraints based on the degree of functional overlap and spatiotemporal conflict. Based on the contextual semantic association strength level and competition resolution constraints, the first point set is further filtered to retain the points whose contextual semantic association strength level reaches the preset association level and whose competition resolution constraints meet the preset resolution conditions, thus generating the second point set.

7. The AI-driven intelligent point selection method for visualizing multi-type point maps according to claim 6, characterized in that, The generation of competition resolution constraints based on the degree of functional overlap and spatiotemporal conflict includes: Analyze the functional attribute overlap between any two sites in the first point set and calculate the functional attribute overlap degree; identify site pairs whose functional attribute overlap degree exceeds a preset overlap threshold as functional attribute overlapping site pairs. Analyze the spatiotemporal coverage crossover between any two points in the first point set, and calculate the degree of spatiotemporal coverage crossover; identify point pairs whose spatiotemporal coverage crossover degree exceeds a preset crossover threshold as spatiotemporal coverage crossover point pairs; Pairs of sites with overlapping functional attributes and pairs of sites with overlapping spatiotemporal coverage are jointly labeled as competing site pairs. Among the competing site pairs, sites with high levels of contextual semantic association are retained, while sites with low levels of contextual semantic association are eliminated, which serves as the competition resolution constraint.

8. The AI-driven intelligent point selection method for visualizing multi-type point maps according to claim 1, characterized in that, S4 includes: Generate spatial location markers at the spatial coordinates corresponding to each second point on the map interface; The semantic type label for the site is displayed in an additional area besides the spatial location identifier; The compatibility assessment level of the site is displayed in the associated information area of ​​the spatial location identifier. The compatibility assessment level is presented in a hierarchical and visual manner to show the comprehensive evaluation results of the site in the joint compatibility assessment and secondary screening.

9. The AI-driven intelligent point selection method for visualizing multi-type point maps according to claim 1, characterized in that, S5 includes: The system detects the selection operation of any hierarchical visual marker on the map interface and identifies the site corresponding to the selection operation as the user's preferred site. Extract the spatial feature attributes of user preference sites, and adaptively shrink or expand the spatial compatibility rules based on the spatial feature attributes of user preference sites; Extract semantic feature attributes of user preference sites, and adjust the semantic matching rules according to the semantic feature attributes of user preference sites to match preferences; The spatial compatibility rules after adaptive shrinkage or expansion and the semantic matching rules after matching preference adjustment are used as the updated point selection constraints.

10. An AI-driven intelligent point selection system for multi-type point map visualization, characterized in that: The system, applied to the AI-driven intelligent point selection method for visualization of multi-type point maps as described in any one of claims 1-9, comprises: Visual candidate map construction module: acquires spatial location data of multiple types of sites and heterogeneous attribute label sets corresponding to each type of site, and reduces the heterogeneous attribute label sets to a unified semantic representation system through spatial-semantic joint mapping to construct a visual candidate map; The first site set filtering module: obtains the site selection constraints of the target scene, performs joint compatibility judgment between the site selection constraints and each candidate site in the visualized candidate map, and filters out the first site set that simultaneously satisfies the spatial compatibility rules and semantic matching rules; wherein, the site selection constraints include spatial compatibility rules and semantic matching rules; The second point set generation module: Based on the contextual semantic association strength between each point in the first point set and the target scene, and combined with the competition resolution constraints between each point in the first point set, the first point set is filtered a second time to generate the second point set. Layered visualization label projection module: Projects each point in the second point set onto the map interface with layered visualization labels, where the layered visualization labels include the spatial location label of the point, the semantic type label label, and the compatibility judgment level label; Interactive Response Constraint Update Module: Updates the selected point constraints in response to the selection or adjustment of hierarchical visual markers on the map interface.

Images