Display method and system based on multi-dimension and multi-granularity public placement fund data

Abstract

The invention provides a method and system for displaying public funds based on multi-dimension and multi-granularity, and relates to the technical field of data visualization, and the method comprises the steps: obtaining multi-dimension data including fund positions, net value fluctuation and macroeconomic indicators, and carrying out the aggregation processing through a sliding window to generate asset distribution data. Extracting multi-dimensional features from the data to form an initial feature set, and organizing and constructing the initial feature set into a multi-granularity space by using a hierarchical structure; and reducing each layer of feature set based on a neighborhood relationship in the space to obtain key feature subsets, fusing the subsets, and inputting the subsets into a random forest model for processing to obtain an asset importance score. And finally, generating a three-dimensional display model by applying an instantiated grid technology based on the scoring data, and compressing model data through a spatial hash algorithm to realize overall visual display of the multi-dimensional and multi-granularity data of the public funds. According to the method, the visual display efficiency and definition of the hierarchical relationship of the complex fund data are improved.

Description

Technical Field

[0001] This application relates to the field of data visualization technology, and in particular to a method and system for displaying multi-dimensional and multi-granular public fund data. Background Technology

[0002] In the process of monitoring and analyzing the asset allocation of public funds, users need to simultaneously grasp the weight distribution, risk status, and correlations of the fund as a whole, different asset classes, and specific holdings within a single view. This scenario requires the display technology to effectively handle multi-source data with complex hierarchical structures and clearly present multi-granular information from macro overview to micro details within a limited screen space, avoiding visual confusion caused by data stacking.

[0003] In existing technologies, there is a two-dimensional visualization scheme based on a parallel coordinate system. This scheme maps fund holdings data to the vertical axis according to hierarchical relationships such as asset class and industry sector, expands the data along the horizontal axis according to dimensions such as time and weight, connects data points of the same asset at different levels with polylines, and uses color coding to distinguish risk status.

[0004] When presenting multi-level data, the connecting lines between different levels are prone to intersection and overlap, leading to visual clutter. Furthermore, as the time series lengthens and the number of asset categories increases, the information density within the two-dimensional plane reaches saturation, making it difficult for users to quickly focus on key data at critical levels. The hierarchical and relational relationships between levels also exhibit visual ambiguity. Summary of the Invention

[0005] This application provides a method and system for displaying public fund data based on multi-dimensional and multi-granularity data, in order to solve the problems of low efficiency and poor clarity in intuitively displaying the hierarchical relationship of complex fund data in the prior art.

[0006] To address the aforementioned technical issues, firstly, this application provides a method for displaying public fund data based on multi-dimensional and multi-granularity data, including:

[0007] Acquire multi-dimensional data of public funds, including fund holdings data, net asset value fluctuation data, and macroeconomic indicator data;

[0008] The multi-dimensional data is aggregated using a sliding window to generate asset distribution data that reflects changes in asset class weights at different time granularities.

[0009] Multidimensional features are extracted from the asset distribution data to form an initial feature set. The initial feature set is then organized using a hierarchical structure to construct a multi-granular space. The hierarchical structure includes an overall portfolio hierarchy, an asset class hierarchy, and an individual security hierarchy.

[0010] Within the multi-granularity space, a feature reduction algorithm based on neighborhood relationships reduces the initial feature set at each level to obtain key feature subsets corresponding to each level. The key feature subsets from different levels are then fused, and the fusion result is input into a random forest model for processing to obtain asset importance score data.

[0011] Based on the asset importance score data, a three-dimensional display model is generated using instantiated grid technology, and the geometric and attribute data of the three-dimensional display model are compressed using a spatial hashing algorithm to achieve a holistic visualization of multi-dimensional and multi-granular public fund data.

[0012] Optionally, within the multi-granularity space, a feature reduction algorithm based on neighborhood relationships reduces the initial feature set at each level to obtain key feature subsets corresponding to each level. These key feature subsets from different levels are then fused, and the fusion result is input into a random forest model for processing to obtain asset importance score data, including:

[0013] Within each level of the multi-granularity space, the similarity value between each feature in the initial feature set is calculated, and the level is either the top layer, the middle layer, or the bottom layer.

[0014] Based on the similarity value, a corresponding neighborhood range is determined for each feature;

[0015] Within the neighborhood, redundant features with similarity values ​​greater than a preset similarity threshold are removed, and representative features are retained to form a subset of key features;

[0016] The key feature subsets from the top-level feature set, the middle-level feature set, and the bottom-level feature set are merged to form a fused feature set, which is the fusion result;

[0017] The fused feature set is input into a random forest model, and asset importance score data is output through the joint decision-making of multiple decision trees in the random forest model.

[0018] Optionally, the step of inputting the fused feature set into a random forest model and outputting asset importance score data through the joint decision-making of multiple decision trees in the random forest model includes:

[0019] The fused feature set is simultaneously input into multiple decision trees contained in the random forest model, wherein each decision tree receives the fused feature set;

[0020] Within each decision tree, based on preset feature selection rules, a subset of feature components are selected from the fused feature set as the decision basis for the current decision tree;

[0021] According to the judgment conditions of each node within the decision tree, some feature components are numerically compared, and the branch path is determined based on the comparison results until the branch path reaches the leaf node.

[0022] At each leaf node, a scoring baseline value is generated, and the scoring baseline values ​​output by the leaf nodes of all decision trees are counted. A majority voting mechanism is used to integrate all the scoring baseline values ​​to obtain a preliminary scoring result.

[0023] The preliminary scoring results are smoothed to generate asset importance score data.

[0024] Optionally, within each decision tree, based on preset feature selection rules, a subset of feature components are selected from the fused feature set as the decision basis for the current decision tree, including:

[0025] Based on the preset feature selection rules and the total number of feature components in the fused feature set, calculate the number of feature components that the current decision tree needs to select.

[0026] From the fused feature set, feature components corresponding to the number of feature components are randomly selected to form the candidate feature set of the current decision tree;

[0027] Assign a corresponding identification identifier to each feature component in the candidate feature set;

[0028] The set of candidate features assigned with identification labels is loaded into the decision-making process of the current decision tree for branch judgment, so as to form the decision basis of the current decision tree.

[0029] Optionally, the step of performing sliding window aggregation on the multi-dimensional data to generate asset distribution data reflecting changes in asset class weights at different time granularities includes:

[0030] A first sliding window with a first length and a first sliding step size is provided, and a second sliding window with a second length and a second sliding step size is provided, wherein the first length is greater than the second length, and the first sliding step size is greater than the second sliding step size;

[0031] The first sliding window is used to perform a first aggregation on the multi-dimensional data to obtain a first aggregated data set, which corresponds to a first-level time granularity.

[0032] The second sliding window is used to perform a second aggregation on the multi-dimensional data to obtain a second aggregated data set, which corresponds to the second-level time granularity.

[0033] Weight distribution calculations are performed on the first aggregated data set and the second aggregated data set respectively to determine the relative weight value of each asset class in each sliding window;

[0034] Based on the relative weight values, asset distribution data is constructed that includes time, asset class, and weight dimensions.

[0035] Optionally, the step of organizing the initial feature set using a hierarchical structure to construct a multi-granularity space includes:

[0036] The basic features in the initial feature set are classified according to the hierarchical structure to form a top-level feature set corresponding to the overall combination level, a middle-level feature set corresponding to the asset class level, and a bottom-level feature set corresponding to the individual bond level.

[0037] A hierarchical link relationship is established between the top-level feature set, the middle-level feature set, and the bottom-level feature set, so that each upper-level feature is associated with the corresponding lower-level feature set.

[0038] Based on the hierarchical linking relationship, the top-level feature set, the middle-level feature set, and the bottom-level feature set are integrated into a feature organization framework with a hierarchical structure, and the feature organization framework is used as a multi-granularity space.

[0039] Optionally, generating a 3D display model based on the asset importance score data using instantiated mesh technology includes:

[0040] Based on the asset importance score data, the distribution location and hierarchical relationship of each asset element in three-dimensional space are determined;

[0041] Based on the distribution location and the hierarchical relationship, basic three-dimensional geometric elements are generated by instantiating mesh technology, and the basic three-dimensional geometric elements are combined and assembled according to the hierarchical relationship to form the framework structure of the three-dimensional display model;

[0042] Configure corresponding visualization attributes for each basic three-dimensional geometric element in the framework structure of the three-dimensional display model to form the three-dimensional display model.

[0043] Secondly, this application provides a system for displaying public fund data based on multi-dimensional and multi-granularity data, including:

[0044] The acquisition module is used to acquire multi-dimensional data of public funds, including fund holdings data, net asset value fluctuation data, and macroeconomic indicator data.

[0045] The aggregation module is used to perform sliding window aggregation processing on the multi-dimensional data to generate asset distribution data that reflects the changes in asset class weights at different time granularities.

[0046] The extraction module is used to extract multi-dimensional features from the asset distribution data to form an initial feature set, and to organize the initial feature set using a hierarchical structure to construct a multi-granularity space. The hierarchical structure includes an overall portfolio hierarchy, an asset category hierarchy, and an individual bond hierarchy.

[0047] The input module is used to reduce the initial feature set at each level in the multi-granularity space using a feature reduction algorithm based on neighborhood relations, to obtain the key feature subsets corresponding to each level, to fuse the key feature subsets from different levels, and to input the fusion result into the random forest model for processing to obtain asset importance score data.

[0048] The generation module is used to generate a three-dimensional display model based on the asset importance score data using instantiated mesh technology, and to compress the geometric and attribute data of the three-dimensional display model using a spatial hashing algorithm, so as to realize the overall visualization display of multi-dimensional and multi-granular public fund data.

[0049] Thirdly, this application provides an electronic device, comprising:

[0050] Memory, used to store computer programs;

[0051] A processor, used to execute the computer program to implement the steps of the method for displaying multi-dimensional and multi-granularity public fund data as described in the first aspect above.

[0052] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the method for displaying public fund data based on multi-dimensional and multi-granularity data as described in the first aspect above.

[0053] This application provides a method for displaying multi-dimensional and multi-granularity public fund data. The method includes: acquiring multi-dimensional data of public funds, including fund holdings data, net asset value fluctuation data, and macroeconomic indicator data; performing sliding window aggregation processing on the multi-dimensional data to generate asset distribution data reflecting changes in asset class weights at different time granularities; extracting multi-dimensional features from the asset distribution data to form an initial feature set; and organizing the initial feature set using a hierarchical structure to construct a multi-granularity space, the hierarchical structure including an overall portfolio hierarchy, an asset class hierarchy, and... Individual fund tiers; within the multi-granularity space, a feature reduction algorithm based on neighborhood relationships is used to reduce the initial feature set under each tier to obtain key feature subsets corresponding to each tier. The key feature subsets from different tiers are fused, and the fusion result is input into a random forest model for processing to obtain asset importance score data. Based on the asset importance score data, a three-dimensional display model is generated using instantiated grid technology, and the geometric and attribute data of the three-dimensional display model are compressed using a spatial hashing algorithm to achieve a holistic visualization of multi-dimensional and multi-granular public fund data.

[0054] The technical solution provided in this application has the following beneficial effects:

[0055] This application provides a comprehensive and diverse data foundation for subsequent analysis, ensuring the richness of information presented. It forms a data sequence that reflects the dynamic changes in asset weights over time, capturing evolutionary patterns at different time scales. The application structures and organizes the disorganized data features according to business logic, laying the framework for hierarchical and focused analysis and presentation. It refines the most representative information at each level and comprehensively assesses asset importance, providing core and differentiated driving data for visualization. It transforms abstract asset data into intuitive, hierarchical 3D graphics, and uses compression technology to ensure smooth presentation in complex scenarios.

[0056] Furthermore, within each level of the multi-granularity space, redundant features are identified and removed by calculating the similarity between features and setting the neighborhood range, thereby extracting key feature subsets for each level. Subsequently, these key feature subsets from different levels are merged to form a unified fused feature set. Finally, the fused feature set is input into a random forest model, and through its internal multi-decision tree joint decision-making mechanism, asset importance score data is output to drive visualization.

[0057] Furthermore, by refining and integrating multi-granularity features layer by layer, and using an ensemble learning model for comprehensive decision-making, the accuracy and robustness of asset importance assessment are effectively improved, thus providing high-quality and highly reliable core driving data for subsequent 3D visualization.

[0058] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 A flowchart illustrating a method for displaying multi-dimensional and multi-granularity public fund data, provided as an embodiment of this application;

[0061] Figure 2 A schematic diagram illustrating a specific implementation of a method for displaying public fund data based on multi-dimensional and multi-granularity data, provided in this application embodiment;

[0062] Figure 3 This is a schematic diagram of the structure of a system for displaying public fund data based on multi-dimensional and multi-granularity, as provided in an embodiment of this application. Detailed Implementation

[0063] Based on the above background technical analysis, existing two-dimensional visualization solutions face fundamental challenges when processing complex, multi-level data from mutual funds. When asset classes and time dimensions are superimposed and unfolded in a planar coordinate system, data connection lines at different levels inevitably intersect and overlap, causing visual confusion. As the amount of data and the time span increase, the limited information-carrying capacity of a two-dimensional plane tends to saturate, making it difficult for users to quickly locate key information. The hierarchical relationships and correlations between levels become ambiguous in their expression, affecting the efficiency and accuracy of data analysis.

[0064] To address the aforementioned issues, this application proposes a method for displaying public fund data based on multi-dimensional and multi-granularity data. This method constructs a multi-granular space to structurally organize fund data, and uses feature reduction and model fusion techniques to extract the core elements driving the display. Finally, it utilizes 3D instantiated grid technology to generate a hierarchical visualization model. This method maps the overall asset portfolio, major category distribution, and individual security details to different levels in a 3D space, clearly expressing the complex hierarchical structure and internal relationships through the natural distinction of spatial location and visual attributes. This approach effectively overcomes the inherent problems of information overlap and hierarchical confusion in a two-dimensional plane, enabling users to intuitively and smoothly grasp panoramic information from macro-allocation to micro-holdings, improving the efficiency of interpreting complex fund data and enhancing decision support capabilities.

[0065] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0066] The core of this application is to provide a method for displaying public fund data based on multi-dimensional and multi-granularity data. A flowchart of one specific implementation method is shown below. Figure 1 As shown, the method includes:

[0067] Step 101: Obtain multi-dimensional data of public funds, including fund holdings data, net asset value fluctuation data, and macroeconomic indicator data.

[0068] In step 101, multi-dimensional data refers to a collection of fund-related information gathered from different sources. Fund holdings data records the specific composition of the various assets held by the fund. Net asset value fluctuation data reflects the changes in the fund's unit net asset value over time. Macroeconomic indicator data includes economic data related to the fund's investment environment.

[0069] In this embodiment of the application, publicly disclosed data from fund companies and macroeconomic data released by economic statistics agencies are obtained through data interfaces. The daily stock and bond holdings of the fund, the daily rise and fall of the fund's net asset value, and indicators such as the GDP growth rate are collected to form a structured raw data set, providing a complete data foundation for subsequent processing.

[0070] For example, taking Fund A as an example, we can obtain its holdings details for the most recent 30 trading days, including 100,000 shares of Company B stock and 50,000 bonds of Company C. We can also obtain the daily net asset value change data of the fund during this period, and collect the consumer price index and producer price index released in the same period. We can integrate these data into a multi-dimensional data set.

[0071] Step 102: Perform sliding window aggregation processing on the multi-dimensional data to generate asset distribution data that reflects the changes in asset class weights at different time granularities.

[0072] In step 102, sliding window aggregation is a data processing method that performs rolling calculations over fixed time intervals. Asset distribution data is structured data describing the changes in the proportion of each asset class within different time intervals.

[0073] In this embodiment of the application, time windows of different lengths are set to perform rolling calculations on the original data. First, aggregated data reflecting medium- and long-term trends is calculated using a longer window, and then aggregated data reflecting short-term changes is calculated using a shorter window. Finally, asset distribution data is generated based on the average market value percentage of each asset within the window period.

[0074] For example, a window of ten trading days is set for the data of Fund A. The window slides every two trading days to calculate the average weight of stocks and bonds within each window. The stock weight is calculated by dividing the total market value of stocks by the total assets of the fund within the window period. By continuously sliding, asset distribution data for fifteen windows is obtained.

[0075] Step 103: Extract multidimensional features from the asset distribution data to form an initial feature set, and use a hierarchical structure to organize the initial feature set to construct a multi-granularity space. The hierarchical structure includes an overall portfolio hierarchy, an asset category hierarchy, and an individual bond hierarchy.

[0076] In step 103, multidimensional features refer to multiple quantitative indicators extracted from asset distribution data, specifically including volatility features that measure the magnitude of price changes, correlation features that reflect the interrelationship between different assets, concentration features that reflect investment concentration, risk-return ratio features that describe the relationship between returns and risks, and liquidity features that assess the liquidity of assets. The initial feature set is a collection of these features. The multi-granularity space is a feature storage structure organized according to fund hierarchical relationships.

[0077] In this embodiment of the application, firstly, characteristic indicators such as volatility and correlation are calculated from asset distribution data. Then, the characteristics are classified and organized according to three levels: fund as a whole, asset class, and individual securities. The relationship between the levels is established, and finally a characteristic organization framework with a clear hierarchical structure is formed.

[0078] For example, twenty features can be extracted from the asset distribution data of Fund A, including the overall volatility of the fund and the correlation between stocks and bonds. Five of these features are assigned to the overall portfolio level, eight to the asset class level, and seven to the individual bond level, thus constructing a feature organization framework with a three-tier structure.

[0079] Step 104: Within the multi-granularity space, a feature reduction algorithm based on neighborhood relationships is used to reduce the initial feature set at each level to obtain the key feature subsets corresponding to each level. The key feature subsets from different levels are then fused, and the fusion result is input into a random forest model for processing to obtain asset importance score data.

[0080] In step 104, neighborhood relationship refers to the spatial range defined based on the degree of similarity between features. When the similarity value between two features reaches a preset similarity threshold, the two features are considered to be within the same neighborhood. This relationship is used to identify and distinguish similar feature groups in the feature set. The key feature subset is a representative feature set retained after screening. Asset importance score data is the asset value assessment result obtained through model calculation. The formation process of the initial feature set under each level is as follows: during the construction of the multi-granularity space, the overall initial feature set is classified and divided according to a predefined hierarchical structure. Specifically, features belonging to the overall portfolio level are assigned to the initial feature set of the overall portfolio level, features describing the attributes of major asset categories are assigned to the initial feature set of the major asset category level, and features involving the details of specific individual bonds are assigned to the initial feature set of the individual bond level.

[0081] In this embodiment, the similarity between features within each level is first calculated, and the neighborhood range is divided according to the similarity. The most representative features in each neighborhood are retained to form a key feature subset. Then, the feature subsets of each level are merged and input into a random forest model. An importance score is generated through the joint decision-making of multiple decision trees.

[0082] For example, within the multi-granularity space of Fund A, the similarity between features is calculated, features with a similarity greater than 0.8 are grouped into the same neighborhood, each neighborhood retains a representative feature, and finally twelve key features are selected from twenty features. After merging, these features are input into a random forest model containing one hundred decision trees, and the importance score of each asset is output.

[0083] Step 105: Based on the asset importance score data, a three-dimensional display model is generated using instantiated mesh technology, and the geometric and attribute data of the three-dimensional display model are compressed using a spatial hash algorithm to achieve a comprehensive visualization of multi-dimensional and multi-granular public fund data.

[0084] In step 105, mesh instantiation is a method of mapping data to three-dimensional geometric elements. Spatial hashing is a data compression technique based on spatial location.

[0085] In this embodiment, the position and size of each element in three-dimensional space are determined based on asset importance scores. Basic geometric units are generated using mesh instantiation technology and assembled according to hierarchical relationships. Then, a spatial hashing algorithm is applied to group and compress adjacent geometric units, ultimately generating a three-dimensional visualization scene that can be rendered in real time. By compressing the geometric and attribute data of the three-dimensional display model, the amount of data that needs to be transmitted and rendered is reduced, thereby improving graphics rendering efficiency and system response speed. This enables smooth display of complex three-dimensional scenes even on ordinary computing devices, which is a key technical guarantee for achieving efficient visualization.

[0086] For example, the importance score of Fund A is converted into three-dimensional coordinates, with highly important assets placed in the center. Cube geometric elements are generated and assembled hierarchically. A spatial hashing algorithm is applied to compress thirty adjacent geometric units into five storage groups, enabling a smooth 3D display on a regular computer.

[0087] This method achieves three-dimensional visualization through multi-dimensional data collection, multi-level feature extraction, and intelligent scoring, effectively improving the efficiency of interpreting complex fund data and providing a clear and intuitive presentation of the relationships and importance between assets at all levels.

[0088] To address the issue of low model efficiency caused by feature redundancy in multi-granularity spaces, in some embodiments, step 104 involves: within the multi-granularity space, a feature reduction algorithm based on neighborhood relationships is used to reduce the initial feature set at each level, obtaining key feature subsets corresponding to each level. These key feature subsets from different levels are then fused, and the fusion result is input into a random forest model for processing to obtain asset importance scoring data, such as... Figure 2 As shown, it includes:

[0089] Step 201: Within each level of the multi-granularity space, calculate the similarity value between each feature in the initial feature set, where the level is the top, middle, or bottom layer.

[0090] In step 201, the similarity value is a numerical indicator used to quantify the consistency of the changing trends between two features, and is obtained by calculating the statistical correlation of the feature data sequence.

[0091] In the embodiments of this application, within the constructed multi-granularity space, the top, middle and bottom layers are processed separately, and the similarity values ​​between all pairs of features within each layer are calculated to form a numerical set describing the similarity relationship of features within the layer.

[0092] Step 202: Based on the similarity value, determine the corresponding neighborhood range for each feature.

[0093] In step 202, the neighborhood range refers to the set of all other features that are similar to a certain feature and meet a certain standard.

[0094] In this embodiment of the application, based on the similarity values ​​calculated in step 201, the neighborhood range to which each feature belongs in the multi-granularity space is defined. Specifically, features with similarity values ​​exceeding a preset benchmark are grouped into the same neighborhood.

[0095] Step 203: Within the neighborhood, remove redundant features whose similarity values ​​are greater than a preset similarity threshold, retain representative features, and form a subset of key features.

[0096] In step 203, redundant features refer to other features that highly repeat the change pattern of the central feature within the neighborhood, and representative features are the features selected from each neighborhood that best reflect the common change pattern of the group of features.

[0097] In this embodiment of the application, within each neighborhood defined in step 202, features with similarity values ​​exceeding a preset similarity threshold are identified as redundant features and removed. Then, the most representative feature is retained from each neighborhood, and finally, a subset of key features is formed at each level.

[0098] Step 204: Merge the key feature subsets from the top-level feature set, the middle-level feature set, and the bottom-level feature set to form a fused feature set, which is the fusion result.

[0099] In step 204, the fused feature set is a unified feature set formed by merging key feature subsets from different levels.

[0100] In this embodiment of the application, the key feature subsets of the top-level feature set, the key feature subsets of the middle-level feature set, and the key feature subsets of the bottom-level feature set obtained in step 203 are simply merged, and duplicates are removed to form a fused feature set.

[0101] Step 205: Input the fused feature set into the random forest model, and output asset importance score data through the joint decision of multiple decision trees in the random forest model.

[0102] In this embodiment of the application, the fused feature set obtained in step 204 is input into a pre-trained random forest model. The model contains multiple decision trees, each of which independently analyzes and judges the input features. Finally, by combining the judgment results of all decision trees, the importance score data of each asset is output.

[0103] Here is a specific example:

[0104] Following the example of Fund A mentioned above, within the constructed multi-granularity space containing 20 features, the similarity values ​​between features in the initial feature set are first calculated at each level. Specifically, the Pearson correlation coefficient formula is used for calculation. ],in Indicates the similarity value. and Each represents one of the two features in the first place. The value of each time window. and These represent the mean values ​​of the two features across all time windows. The total number of time windows is used to calculate the similarity value between the overall fund volatility and the correlation between stocks and bonds. Based on the calculated similarity values, a corresponding neighborhood range is determined for each feature. Features with a similarity value greater than or equal to 0.8 are grouped into the same neighborhood. Therefore, overall fund volatility, stock-bond correlation, and another feature D with a similarity value of 0.85 are grouped into the same neighborhood. Within this neighborhood, a preset similarity threshold of 0.9 is set, and redundant features with similarity values ​​greater than this threshold are removed. Since the similarity value of overall fund volatility and stock-bond correlation is 0.92, which is greater than 0.9, the redundant feature of stock-bond correlation is removed, and overall fund volatility is retained as the representative feature. The formula forms key feature subsets at each level, and finally selects 12 key features from 20 features. The 3 key features from the top feature set, the 5 key features from the middle feature set, and the 4 key features from the bottom feature set are merged to form a fused feature set containing 12 features. The fused feature set is then input into a random forest model, which contains 100 decision trees. Through the joint decision of these 100 decision trees, the importance score data of each asset is output. Among them, the importance score of Company B's stock is 85 points, and the importance score of Company C's bond is 76 points. This completes the entire process from feature reduction to score generation.

[0105] In this embodiment, feature reduction and fusion processing in a multi-granularity space effectively removes redundant information and highlights representative features at each level, enabling subsequent models to more accurately assess asset importance and improving the efficiency and reliability of the entire analysis process.

[0106] To further improve the accuracy and stability of asset importance scoring, in some embodiments, step 205: inputting the fused feature set into a random forest model, and outputting asset importance score data through the joint decision-making of multiple decision trees in the random forest model, includes:

[0107] Step 301: Input the fused feature set into multiple decision trees contained in the random forest model simultaneously, wherein each decision tree receives the fused feature set.

[0108] In step 301, decision trees are the basic building blocks of the random forest model, and each decision tree is an independent classifier.

[0109] In this embodiment, the fused feature set is simultaneously distributed to all decision trees contained in the random forest model, ensuring that each decision tree can obtain complete feature information for independent analysis.

[0110] Step 302: Within each decision tree, based on preset feature selection rules, select a portion of feature components from the fused feature set as the decision basis for the current decision tree.

[0111] In step 302, a preset feature selection rule stipulates that each decision tree randomly selects a fixed number of feature components from the fused feature set as its decision basis. The specific number of selected features is determined by a mathematical formula, namely, the number of feature components is equal to the square root of the total number of features in the fused feature set. For example, when the fused feature set contains thirty-six features, each tree randomly selects six features. A partial feature component refers to a subset of features randomly selected for the current decision tree from all features contained in the fused feature set according to the preset rule. This subset contains only a part of the overall features, not all of them. Each decision tree uses different feature combinations for learning and decision-making based on different random selections. The decision basis refers to the role played by the selected partial feature components in the decision-making process of the current decision tree. As the comparison objects for node judgment conditions, they directly determine the flow path of data samples in the decision tree and are the data foundation upon which the decision tree relies for reasoning and judgment.

[0112] In this embodiment of the application, before each decision tree starts working, a certain number of feature components are randomly selected from the fusion feature set according to preset rules. These selected features will serve as the sole basis for the decision tree in the entire decision-making process.

[0113] Step 303: According to the judgment conditions of each node within the decision tree, perform numerical comparisons on some feature components, determine the branch path based on the comparison results, until the branch path reaches the leaf node.

[0114] In step 303, the judgment condition is a numerical comparison logic pre-set on each node inside the decision tree. Its content is to compare the specific value of the selected feature component with a preset threshold, such as judging whether "the overall volatility of the fund is greater than 5%" or "the correlation between stocks and bonds is less than 0.3". The subsequent branch path is determined according to the comparison result of "yes" or "no".

[0115] In this embodiment of the application, the decision tree starts from the root node, performs numerical comparisons on the feature components selected in step 302 according to the preset judgment conditions of each internal node, selects the corresponding branch path according to the comparison results, and passes through the subsequent nodes along the branch path until it reaches the leaf node that no longer branches.

[0116] Step 304: At each leaf node, generate a scoring baseline value, and count the scoring baseline values ​​output by the leaf nodes of all decision trees. Use a majority voting mechanism to integrate all the scoring baseline values ​​to obtain preliminary scoring results.

[0117] In step 304, the scoring baseline value is the pre-set base score of the leaf node. The majority voting mechanism is a method to determine the final conclusion by statistically analyzing the results of all decision trees. The preliminary scoring result refers to the intermediate score obtained after integrating the scoring baseline values ​​output by all decision tree leaf nodes through the majority voting mechanism. It reflects the collective decision-making tendency of the random forest model, but has not yet undergone subsequent smoothing processing and may still contain some random fluctuations.

[0118] In this embodiment of the application, when a data sample reaches a certain leaf node, the preset scoring benchmark value of that node is read. After all decision trees have completed their judgments, the scoring benchmark values ​​output by all leaf nodes are counted. A majority voting mechanism is used to integrate these benchmark values, and the range of values ​​with the highest frequency of occurrence is selected as the preliminary scoring result.

[0119] Step 305: Smooth the preliminary scoring results to generate asset importance score data.

[0120] In step 305, smoothing is a technique that uses mathematical methods to eliminate data fluctuations and make the results more stable.

[0121] In this embodiment of the application, the preliminary scoring results obtained in step 304 are subjected to data smoothing calculation to eliminate scoring fluctuations caused by random factors and generate stable and reliable asset importance scoring data.

[0122] Here is a specific example:

[0123] Following the example of Fund A mentioned above, after forming a fused feature set containing 12 features, this fused feature set is simultaneously input into the 100 decision trees in the random forest model, ensuring that each decision tree receives complete 12-dimensional feature data. Within the first decision tree, four feature components are randomly selected based on preset feature selection rules as the decision basis for the tree. Specifically, the selected features are the overall fund volatility, industry concentration, bond duration, and the percentage of the top five holdings. According to the judgment conditions of the nodes within the decision tree, the overall fund volatility feature value is first compared at the root node to see if it is greater than 5%. Since the fund volatility is 6%, it enters the right branch. At the second node, it is judged whether the industry concentration is less than 0.3. The measured value is 0.25, so it enters the left branch, and so on until the branch path reaches the seventh leaf node. A scoring benchmark value of 82 points is generated at this leaf node. After all 100 decision trees have been run, a total of [data missing] points are collected. Of the 100 benchmark scores, 62 fell within the 80-85 score range. A majority vote was used to calculate the arithmetic mean of all benchmark scores within this range, resulting in a preliminary score of 83. The formula is: the preliminary score equals the sum of the benchmark scores within the range divided by the number of benchmark scores within the range. The sum of the benchmark scores within the range is 5146, and there are 62 benchmark scores within the range, hence the score of 83. Finally, the preliminary scores of 81, 82, 83, 82, and 84 for five consecutive analysis periods were smoothed using a moving average. The formula is: the asset importance score equals the sum of the scores from periods one, two, three, four, and five, divided by 5. Substituting the values, we get 82.4. Therefore, the final asset importance score for Company B's stock is 82.4, completing the entire process of outputting the score data through a random forest model's collaborative decision-making.

[0124] In this embodiment of the application, the accuracy and stability of asset importance scoring are effectively improved by using a multi-decision tree independent decision-making and result integration mechanism, combined with data smoothing processing, so that the assessment results can better reflect the true value level of the assets.

[0125] To further improve the standardization and traceability of feature selection, in some embodiments, step 302: within each decision tree, based on preset feature selection rules, selecting a subset of feature components from the fused feature set as the decision basis for the current decision tree, includes:

[0126] Step 401: Based on the preset feature selection rules and the total number of feature components in the fused feature set, calculate the number of feature components that the current decision tree needs to select.

[0127] In step 401, the number of feature components refers to the number of features that each decision tree needs to randomly select from the fusion feature set.

[0128] In this embodiment of the application, based on the specific requirements for quantity calculation in the preset feature selection rules and combined with the total number of feature components contained in the known fusion feature set, the specific number of feature components that the current decision tree should select is determined through mathematical operations.

[0129] Step 402: Randomly select feature components corresponding to the number of feature components from the fused feature set to form the candidate feature set of the current decision tree.

[0130] In step 402, the candidate feature set is a subset of features formed by randomly selecting a specified number of features from the fused feature set.

[0131] In this embodiment of the application, after determining the number of feature components, a corresponding number of feature components are selected from the complete fusion feature set by random sampling. These selected feature components together constitute the candidate feature set dedicated to the current decision tree.

[0132] Step 403: Assign a corresponding identification identifier to each feature component in the candidate feature set.

[0133] In step 403, the identification identifier is a unique identity tag assigned to each feature component.

[0134] In this embodiment of the application, each feature component in the candidate feature set formed in step 402 is assigned an independent identification identifier, and a one-to-one correspondence between the feature component and the identifier is established.

[0135] Step 404: Load the set of candidate features assigned with identification labels into the decision process of the current decision tree for branch judgment, so as to form the decision basis of the current decision tree.

[0136] In step 404, the decision process is a logical sequence of branches within the decision tree that compares and judges based on features.

[0137] In this embodiment of the application, the candidate feature set that has completed the assignment of identification marks is loaded into the decision process of the current decision tree, so that the decision tree uses only these feature components with marks as the basis for comparison in all subsequent node judgments, thereby forming a complete decision basis.

[0138] Here is a specific example:

[0139] Following the example of Fund A mentioned above, given that decision tree #1 requires the selection of 4 feature components as the basis for decision-making, the number of feature components that the current decision tree needs to select is first calculated based on the preset feature selection rules and the total number of feature components in the fused feature set (12). The rule stipulates that the number of selections is the square root of the total number of feature components, and the calculation formula is as follows: ,in This indicates the number of feature components to be selected, expressed in units of 100. The total number of feature components in the fused feature set is expressed in units of 1, which is calculated by substituting the numerical values. Based on the rounding rule, three feature components need to be selected. Then, specific features corresponding to the three feature components are randomly selected from the fusion feature set containing 12 features. Three non-repeating random numbers are generated in the range of 1 to 12 using a random number generator. Assuming that the selected features are the overall volatility of the fund, industry concentration, and the percentage of the top five holdings, a candidate feature set for the current decision tree is formed. Then, each feature component in this candidate feature set is assigned a corresponding digital identification label, with the overall volatility of the fund identified as 1, industry concentration identified as 2, and the percentage of the top five holdings identified as 3. Finally, the candidate feature set with the assigned identification labels is loaded into the decision process of the first decision tree for branch judgment. In all subsequent node judgments, this decision tree only uses the three features identified as 1 (overall volatility of the fund), 2 (industry concentration), and 3 (the percentage of the top five holdings) for comparison and decision-making, thus forming the complete decision basis for the current decision tree and ensuring that the tree maintains consistency in the use of features in subsequent branch judgments.

[0140] In this embodiment, the consistency and manageability of the feature selection process for each decision tree are ensured through standardized quantity calculation, random selection, identifier allocation, and process loading, laying a reliable foundation for subsequent collaborative decision-making.

[0141] To further improve the comprehensiveness and hierarchy of multi-dimensional data processing, in some embodiments, step 102: performing sliding window aggregation on the multi-dimensional data to generate asset distribution data reflecting changes in asset class weights at different time granularities includes:

[0142] Step 501: Set a first sliding window with a first length and a first sliding step size, and a second sliding window with a second length and a second sliding step size, wherein the first length is greater than the second length, and the first sliding step size is greater than the second sliding step size.

[0143] In step 501, the first sliding window is an observation window used to capture long-term trends in the data, the first length is the number of time units covered by this window, and the first sliding step is the number of time units that this window moves each time; the second sliding window is an observation window used to capture short-term changes in the data, the second length is the number of time units covered by this window, and the second sliding step is the number of time units that this window moves each time.

[0144] In this embodiment of the application, two sliding windows of different specifications are set according to the analysis requirements. One window has a larger length and step size for macro trend analysis, and the other window has a smaller length and step size for micro change analysis.

[0145] Step 502: Use the first sliding window to perform a first aggregation on the multi-dimensional data to obtain a first aggregated data set, which corresponds to the first level of time granularity.

[0146] In step 502, the first aggregated data set is the aggregated result set obtained by processing the original data through the first sliding window.

[0147] In this embodiment of the application, the first sliding window is moved sequentially along the time axis. After each movement, the multi-dimensional data in the window is statistically calculated, and the calculation results are saved in order to form a first aggregated data set with a corresponding longer time granularity.

[0148] Step 503: Use the second sliding window to perform a second aggregation on the multi-dimensional data to obtain a second aggregated data set, which corresponds to the second-level time granularity.

[0149] In step 503, the second aggregated data set is the aggregated result set obtained by processing the original data through the second sliding window.

[0150] In this embodiment, the second sliding window is moved along the time axis with smaller steps, and statistical calculations are performed on the multi-dimensional data in each window. The calculation results are saved in sequence to form a second aggregated data set with a corresponding shorter time granularity.

[0151] Step 504: Perform weight distribution calculations on the first aggregated data set and the second aggregated data set respectively to determine the relative weight value of each asset class in each sliding window.

[0152] In step 504, the relative weight value is the proportion of a certain asset class in the total assets within a specific sliding window.

[0153] In this embodiment of the application, for each window of data in the first aggregated data set and the second aggregated data set, the average market value of each asset class is calculated as the average proportion of the total assets of the fund during the window period, so as to obtain the relative weight value of each asset class in each window.

[0154] Step 505: Based on the relative weight values, construct asset distribution data that includes time dimension, asset class dimension and weight dimension.

[0155] In this embodiment of the application, all the relative weight values ​​calculated in step 504 are organized and arranged according to their corresponding time windows and asset categories to construct complete asset distribution data containing three dimensions.

[0156] Here is a specific example:

[0157] Following the example of Fund A mentioned above, based on the multi-dimensional data set obtained from its most recent 30 trading days, a first sliding window with a first length of 10 trading days and a first sliding step of 2 trading days is first set up, and a second sliding window with a second length of 5 trading days and a second sliding step of 1 trading day is seconded. The first length of 10 trading days is longer than the second length of 5 trading days, and the first sliding step of 2 trading days is longer than the second sliding step of 1 trading day. Then, the first sliding window is used to perform a first aggregation of the multi-dimensional data. Starting from the start date, data from trading days 1 to 10 are taken to calculate the average weight of each asset to form the first data point. Then, the window slides for 2 trading days to take data from trading days 3 to 12 to calculate the second data point. Two data points are used, and this process is repeated until the entire data period is traversed, resulting in a first aggregated data set of 11 windows corresponding to the first level of time granularity at the weekly level. Simultaneously, a second sliding window is used to perform a second aggregation of multi-dimensional data. Starting from the beginning of the period, data from the 1st to the 5th trading day is used to calculate the average weight of each asset to form the first data point. Then, the window slides one trading day, taking data from the 2nd to the 6th trading day to calculate the second data point, and so on, resulting in a second aggregated data set of 26 windows corresponding to the second level of time granularity at the daily level. Subsequently, weight distribution calculations are performed on the first and second aggregated data sets respectively to determine the relative weight value of each asset class within each sliding window. The formula for calculating the relative weight value of stocks is as follows: ,in The relative weighting of stocks is expressed as a percentage. This indicates the total market value of the stocks during the window period, expressed in ten thousand yuan. The total assets of the fund during the window period are expressed in ten thousand yuan. Taking the first window of the first aggregated data set as an example, the calculation is as follows: Ten thousand yuan, Ten thousand yuan, substituting into the formula, yields Finally, based on all the calculated relative weight values, an asset distribution dataset is constructed that includes specific window periods in the time dimension, stocks and bonds in the asset class dimension, and relative weight values ​​in the weight dimension, forming a complete structured dataset for subsequent analysis.

[0158] In this embodiment, by using sliding windows of different sizes in a coordinated manner, data aggregation analysis at multiple time granularities is achieved, and the constructed asset distribution data fully reflects the changing characteristics of asset weights at different time scales.

[0159] To further improve the structure and systematic nature of feature organization, in some embodiments, step 103: organizing the initial feature set using a hierarchical structure to construct a multi-granularity space includes:

[0160] Step 601: Classify the basic feature quantities in the initial feature set according to the hierarchical structure to form a top-level feature quantity group corresponding to the overall combination level, a middle-level feature quantity group corresponding to the asset category level, and a bottom-level feature quantity group corresponding to the individual bond level.

[0161] In step 601, the basic feature quantities are fundamental indicators extracted from asset distribution data to describe asset characteristics. The overall portfolio level refers to the dimension that analyzes the mutual fund as an indivisible whole; at this level, the analysis focuses on the comprehensive indicators of the fund as a single entity. The top-level feature quantity group is a set of features describing the overall performance of the fund. These features reflect the fund's overall risk-return characteristics and operating status from a global perspective, without distinguishing between specific asset classes or individual securities. The asset class level refers to the dimension that analyzes the mutual fund by dividing it into investment target categories (such as stocks, bonds, etc.). This level focuses on the overall performance and allocation of different asset classes within the fund. The middle-level feature quantity group is a set of features describing the characteristics of the fund's major asset class portfolios. These features focus on the overall performance and attributes of different asset classes such as stocks and bonds, without involving specific individual securities within each asset class. The individual security level refers to the dimension that analyzes the specific individual securities (such as a particular stock or bond) that constitute the mutual fund's investment portfolio. This level focuses on the individual characteristics of the bottom-level holdings. The underlying feature set refers to the set of features that describe the characteristics of specific individual securities (stocks or bonds) in the fund's holdings. These features focus on the attributes of each specific security that constitutes an asset class and its contribution to the portfolio.

[0162] In this embodiment of the application, based on the inherent hierarchical relationship in the fund business, all basic feature quantities in the initial feature set are grouped according to the different scale levels of the objects they describe, forming three feature quantity groups corresponding to the overall fund level, the asset class level, and the specific bond level, respectively.

[0163] Step 602: Establish a hierarchical link relationship between the top-level feature set, the middle-level feature set, and the bottom-level feature set, so that each upper-level feature is associated with the corresponding lower-level feature set.

[0164] In step 602, the hierarchical link relationship is a logical connection that describes the hierarchical association between different hierarchical feature quantities.

[0165] In this embodiment of the application, a clear directional association is established between the three grouped feature quantities, so that each top-level feature quantity describing the overall macroscopic situation can be associated with multiple middle-level feature quantities under it, and each middle-level feature quantity can be further associated with multiple bottom-level feature quantities contained therein.

[0166] Step 603: Based on the hierarchical linking relationship, integrate the top-level feature set, the middle-level feature set, and the bottom-level feature set into a feature organization framework with a hierarchical structure, and use the feature organization framework as a multi-granularity space.

[0167] In step 603, the feature organization framework is a structured data system formed by integrating all feature quantities according to hierarchical relationships.

[0168] In this embodiment of the application, based on the hierarchical link relationship established in step 602, the three feature groups are organized and integrated in a tree structure, with the top-level feature group as the root node, the middle-level feature group as the intermediate node, and the bottom-level feature group as the leaf node. Finally, this complete tree structure is defined as a multi-granularity space.

[0169] Here is a specific example:

[0170] Following the example of Fund A mentioned above, after extracting twenty features from the asset distribution data to form an initial feature set, the basic features in the initial feature set are first categorized according to a hierarchical structure. Five features describing the overall attributes of the fund—including annualized volatility, Sharpe ratio, maximum drawdown, fund size change rate, and asset turnover—are categorized into the top-level feature set corresponding to the overall portfolio level. Eight features describing the attributes of major asset classes—including stock portfolio volatility, bond portfolio volatility, stock-bond correlation, stock industry concentration, bond duration, bond credit rating distribution, stock portfolio turnover rate, and bond portfolio liquidity—are categorized into the middle-level feature set corresponding to the major asset class level. Seven features describing the attributes of specific individual bonds—including the weight of the largest holding, the concentration of the top five holdings, the weight of the largest bond, the average yield to maturity of the bond portfolio, the contribution of individual bond volatility, the correlation coefficient of individual bonds, and the liquidity indicator of individual bonds—are categorized into the bottom-level feature set corresponding to the individual bond level. Then, in the top... A hierarchical link relationship is established between the top-level feature set, the middle-level feature set, and the bottom-level feature set. Foreign key associations in the database are used to link each upper-level feature to its corresponding lower-level feature set. Specifically, the annualized volatility of funds in the top-level feature set is linked to the stock portfolio volatility and bond portfolio volatility in the middle-level feature set; the industry concentration of stocks in the middle-level feature set is linked to the concentration of the top five holdings in the bottom-level feature set, thus establishing a complete hierarchical relationship network. Finally, based on these hierarchical links, the three feature sets are integrated into a feature organization framework with a clear hierarchical structure. This framework uses a tree-like data model for storage. The top-level feature set, as the root node, contains five features; the middle-level feature set, as the branch nodes, contains eight features; and the bottom-level feature set, as the leaf nodes, contains seven features. The nodes are interconnected through the established links. This complete tree-like feature organization framework serves as the multi-granular space used in this scheme, providing a structured data foundation for subsequent feature reduction and model analysis.

[0171] In this embodiment of the application, a well-structured multi-granular space is constructed through hierarchical feature organization and association establishment, providing a good data foundation for subsequent feature analysis and visualization.

[0172] To further improve the accuracy and structure of the 3D display, in some embodiments, step 105: generating a 3D display model based on the asset importance score data using instantiated mesh technology includes:

[0173] Step 701: Based on the asset importance score data, determine the distribution location and hierarchical relationship of each asset element in three-dimensional space.

[0174] In step 701, asset elements refer to the basic units that need to be visualized after the aforementioned data processing flow. Their physical meaning is that they represent specific constituent objects at different levels within a public fund, including fund portfolio entities representing the entire fund, asset class entities representing major asset classes such as stocks and bonds, and individual bond entities representing single stocks or bonds. These elements originate from the targets to be displayed after aggregating multi-dimensional data, extracting features, and constructing a multi-granular space. Their specific identity and hierarchical affiliation are determined during the construction of the multi-granular space, and their positioning and visualization attributes in the three-dimensional space are obtained through asset importance scoring data. Distribution location refers to the specific spatial coordinates of each asset element in the three-dimensional coordinate system, and hierarchical relationship refers to the level and subordinate relationship of each asset element within the fund structure.

[0175] In this embodiment of the application, based on the differences in the asset importance score data and combined with the inherent organizational structure of the fund, the precise coordinate position of each asset element in three-dimensional space is calculated and determined, and its specific level of belonging to the overall portfolio, asset class or individual security is clarified.

[0176] Step 702: Based on the distribution location and the hierarchical relationship, generate basic three-dimensional geometric elements using instantiated mesh technology, and assemble the basic three-dimensional geometric elements according to the hierarchical relationship to form the framework structure of the three-dimensional display model.

[0177] In step 702, the basic three-dimensional geometric elements are basic graphic units generated by instantiating mesh technology, and the framework structure is a three-dimensional model skeleton formed by assembling these geometric elements according to specific rules.

[0178] In this embodiment of the application, based on the distribution location and hierarchical relationship determined in step 701, basic three-dimensional geometric elements of uniform specifications are created in batches using instantiated mesh technology, and these elements are spatially arranged and logically combined according to hierarchical relationships to construct the basic framework of the three-dimensional display model.

[0179] Step 703: Configure corresponding visualization attributes for each basic three-dimensional geometric element in the framework structure of the three-dimensional display model to form a three-dimensional display model.

[0180] In step 703, visualization attributes are visual feature parameters assigned to the 3D geometric elements. The 3D display model is the final display object that completes all visual configurations.

[0181] In this embodiment of the application, each basic three-dimensional geometric element in the three-dimensional display model framework structure formed in step 702 is configured with corresponding visual attributes such as color, size, and transparency according to the asset characteristics and importance score it represents, and finally a complete three-dimensional model that can be directly used for rendering and display is formed.

[0182] Here is a specific example:

[0183] In this embodiment of the application, the accurate conversion and intuitive display of asset data into a 3D model are achieved through a complete process from data mapping to spatial layout and then to visual configuration.

[0184] Figure 3 This application provides a schematic diagram of the structure of a system for displaying multi-dimensional and multi-granularity public fund data, and the specific implementation details are as follows:

[0185] The acquisition module 31 is used to acquire multi-dimensional data of public funds, including fund holdings data, net asset value fluctuation data and macroeconomic indicator data.

[0186] The aggregation module 32 is used to perform sliding window aggregation processing on the multi-dimensional data to generate asset distribution data that reflects the changes in asset class weights at different time granularities.

[0187] Extraction module 33 is used to extract multi-dimensional features from the asset distribution data to form an initial feature set, and to organize the initial feature set using a hierarchical structure to construct a multi-granularity space. The hierarchical structure includes an overall portfolio hierarchy, an asset category hierarchy, and an individual bond hierarchy.

[0188] The input module 34 is used to perform a feature reduction algorithm based on neighborhood relationships on the initial feature set at each level in the multi-granularity space to obtain the key feature subsets corresponding to each level, fuse the key feature subsets from different levels, and input the fusion result into the random forest model for processing to obtain asset importance score data.

[0189] The generation module 35 is used to generate a three-dimensional display model based on the asset importance score data using instantiated mesh technology, and to compress the geometric data and attribute data of the three-dimensional display model using a spatial hash algorithm, so as to realize the overall visualization display of multi-dimensional and multi-granular public fund data.

[0190] The system for displaying public fund data based on multi-dimensional and multi-granularity in this application is used to implement the aforementioned method for displaying public fund data based on multi-dimensional and multi-granularity. Therefore, the specific implementation of the system for displaying public fund data based on multi-dimensional and multi-granularity can be found in the embodiment section of the method for displaying public fund data based on multi-dimensional and multi-granularity above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0191] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the above-described method for displaying multi-dimensional and multi-granularity public fund data.

[0192] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described methods for displaying public fund data based on multi-dimensional and multi-granularity data.

[0193] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0194] The embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the method for displaying multi-dimensional and multi-granular public fund data.

[0195] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0196] The foregoing has provided a detailed description of a method, system, electronic device, and storage medium for displaying multi-dimensional and multi-granularity public fund data provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for displaying public fund data based on multi-dimensional and multi-granularity data, characterized in that, include: Acquire multi-dimensional data of public funds, including fund holdings data, net asset value fluctuation data, and macroeconomic indicator data; The multi-dimensional data is aggregated using a sliding window to generate asset distribution data that reflects changes in asset class weights at different time granularities. Multidimensional features are extracted from the asset distribution data to form an initial feature set. The initial feature set is then organized using a hierarchical structure to construct a multi-granular space. The hierarchical structure includes an overall portfolio hierarchy, an asset class hierarchy, and an individual security hierarchy. Within the multi-granularity space, a feature reduction algorithm based on neighborhood relationships reduces the initial feature set at each level to obtain key feature subsets corresponding to each level. The key feature subsets from different levels are then fused, and the fusion result is input into a random forest model for processing to obtain asset importance score data. Based on the asset importance score data, a three-dimensional display model is generated using instantiated grid technology, and the geometric and attribute data of the three-dimensional display model are compressed using a spatial hashing algorithm to achieve a holistic visualization of multi-dimensional and multi-granular public fund data.

2. The method according to claim 1, characterized in that, Within the multi-granularity space, a feature reduction algorithm based on neighborhood relationships reduces the initial feature set at each level to obtain key feature subsets corresponding to each level. These key feature subsets from different levels are then fused, and the fusion result is input into a random forest model for processing to obtain asset importance scoring data, including: Within each level of the multi-granularity space, the similarity value between each feature in the initial feature set is calculated, and the level is either the top layer, the middle layer, or the bottom layer. Based on the similarity value, a corresponding neighborhood range is determined for each feature; Within the neighborhood, redundant features with similarity values ​​greater than a preset similarity threshold are removed, and representative features are retained to form a subset of key features; The key feature subsets from the top-level feature set, the middle-level feature set, and the bottom-level feature set are merged to form a fused feature set, which is the fusion result; The fused feature set is input into a random forest model, and asset importance score data is output through the joint decision-making of multiple decision trees in the random forest model.

3. The method according to claim 2, characterized in that, The process of inputting the fused feature set into a random forest model, and outputting asset importance score data through the joint decision-making of multiple decision trees in the random forest model, includes: The fused feature set is simultaneously input into multiple decision trees contained in the random forest model, wherein each decision tree receives the fused feature set; Within each decision tree, based on preset feature selection rules, a subset of feature components are selected from the fused feature set as the decision basis for the current decision tree; According to the judgment conditions of each node within the decision tree, some feature components are numerically compared, and the branch path is determined based on the comparison results until the branch path reaches the leaf node. At each leaf node, a scoring baseline value is generated, and the scoring baseline values ​​output by the leaf nodes of all decision trees are counted. A majority voting mechanism is used to integrate all the scoring baseline values ​​to obtain a preliminary scoring result. The preliminary scoring results are smoothed to generate asset importance score data.

4. The method according to claim 3, characterized in that, Within each decision tree, based on preset feature selection rules, a subset of feature components are selected from the fused feature set as the decision basis for the current decision tree, including: Based on the preset feature selection rules and the total number of feature components in the fused feature set, calculate the number of feature components that the current decision tree needs to select. From the fused feature set, feature components corresponding to the number of feature components are randomly selected to form the candidate feature set of the current decision tree; Assign a corresponding identification identifier to each feature component in the candidate feature set; The set of candidate features assigned with identification labels is loaded into the decision-making process of the current decision tree for branch judgment, so as to form the decision basis of the current decision tree.

5. The method according to claim 1, characterized in that, The process of performing sliding window aggregation on the multi-dimensional data to generate asset distribution data reflecting changes in asset class weights at different time granularities includes: A first sliding window with a first length and a first sliding step size is provided, and a second sliding window with a second length and a second sliding step size is provided, wherein the first length is greater than the second length, and the first sliding step size is greater than the second sliding step size; The first sliding window is used to perform a first aggregation on the multi-dimensional data to obtain a first aggregated data set, which corresponds to a first-level time granularity. The second sliding window is used to perform a second aggregation on the multi-dimensional data to obtain a second aggregated data set, which corresponds to the second-level time granularity. Weight distribution calculations are performed on the first aggregated data set and the second aggregated data set respectively to determine the relative weight value of each asset class in each sliding window; Based on the relative weight values, asset distribution data is constructed that includes time, asset class, and weight dimensions.

6. The method according to claim 1, characterized in that, The method of organizing the initial feature set using a hierarchical structure to construct a multi-granularity space includes: The basic features in the initial feature set are classified according to the hierarchical structure to form a top-level feature set corresponding to the overall combination level, a middle-level feature set corresponding to the asset class level, and a bottom-level feature set corresponding to the individual bond level. A hierarchical link relationship is established between the top-level feature set, the middle-level feature set, and the bottom-level feature set, so that each upper-level feature is associated with the corresponding lower-level feature set. Based on the hierarchical linking relationship, the top-level feature set, the middle-level feature set, and the bottom-level feature set are integrated into a feature organization framework with a hierarchical structure, and the feature organization framework is used as a multi-granularity space.

7. The method according to claim 1, characterized in that, The process of generating a 3D display model based on the asset importance score data using instantiated mesh technology includes: Based on the asset importance score data, the distribution location and hierarchical relationship of each asset element in three-dimensional space are determined; Based on the distribution location and the hierarchical relationship, basic three-dimensional geometric elements are generated by instantiating mesh technology, and the basic three-dimensional geometric elements are combined and assembled according to the hierarchical relationship to form the framework structure of the three-dimensional display model; Configure corresponding visualization attributes for each basic three-dimensional geometric element in the framework structure of the three-dimensional display model to form the three-dimensional display model.

8. A system for displaying public fund data based on multi-dimensional and multi-granularity methods, characterized in that, include: The acquisition module is used to acquire multi-dimensional data of public funds, including fund holdings data, net asset value fluctuation data, and macroeconomic indicator data. The aggregation module is used to perform sliding window aggregation processing on the multi-dimensional data to generate asset distribution data that reflects the changes in asset class weights at different time granularities. The extraction module is used to extract multi-dimensional features from the asset distribution data to form an initial feature set, and to organize the initial feature set using a hierarchical structure to construct a multi-granularity space. The hierarchical structure includes an overall portfolio hierarchy, an asset category hierarchy, and an individual bond hierarchy. The input module is used to reduce the initial feature set at each level in the multi-granularity space using a feature reduction algorithm based on neighborhood relations, to obtain the key feature subsets corresponding to each level, to fuse the key feature subsets from different levels, and to input the fusion result into the random forest model for processing to obtain asset importance score data. The generation module is used to generate a three-dimensional display model based on the asset importance score data using instantiated mesh technology, and to compress the geometric and attribute data of the three-dimensional display model using a spatial hashing algorithm, so as to realize the overall visualization display of multi-dimensional and multi-granular public fund data.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the method for displaying multi-dimensional and multi-granularity public fund data as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the display method based on multi-dimensional and multi-granular public fund data as described in any one of claims 1 to 7.

Images