Visual output method and system of artificial intelligence model

Abstract

The invention relates to the technical field of data visualization, in particular to a visual output method and system for an artificial intelligence model, and the method comprises the following steps: analyzing the information of nodes and edges in a network through model structure network data, recognizing the connection number of a plurality of nodes, calculating the weight, partitioning the network according to the connection strength of the nodes, and storing the partitioned network in a database; and generating a model structure mapping graph. According to the method, the network data of the model structure is analyzed, the connection number and the weight of the nodes are calculated, the relevance between the data and the core structure of the network are revealed, the statistical attributes of the nodes are classified, the chart types are matched according to the data characteristics, the visual presentation effect of the data is optimized, and the user interaction behavior is monitored in real time. And the chart layout is adjusted according to the visual focus of the user and the reaction data, so that the cognitive load of information processing is reduced, a decision maker can quickly identify modes, trends and abnormities in the data, and the ability of understanding the output data of the artificial intelligence model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data visualization, and in particular to a visualization output method and system for an artificial intelligence model. Background Art

[0002] The field of data visualization technology focuses on converting and presenting complex data through visual expressions, making the interpretation of information more intuitive and easier to understand. It combines design, statistics and computing techniques, and uses charts, interactive graphics, and geographic information systems to help users analyze and reason about data, identify data change patterns, trends, and anomalies, and effectively understand the meaning of complex data sets. It realizes the presentation of complex data and provides decision support. It is applied in multiple fields such as business, scientific research, engineering, and education to help decision makers gain insight into trends, patterns, and anomalies.

[0003] Among them, the visualization output method of the artificial intelligence model focuses on graphically displaying the internal working mechanism, prediction results and decision-making process of the artificial intelligence model through visual expression technology, aiming to enhance the transparency of the model and the user's understanding. By converting abstract model operations into visual graphics, users can intuitively observe and analyze the model's behavior and decision-making path when processing data, understand the working principle of the model, monitor and verify the accuracy of the model output, and help researchers and actual users evaluate model performance, adjust model parameters, and optimize and diagnose errors of the model.

[0004] The visualization output technology of traditional artificial intelligence models lacks in-depth analysis and personalized adaptation capabilities for user interaction behaviors, and ignores changes in user behavior patterns and cognitive load in actual operations, resulting in inefficiency and insufficient understanding when users interpret complex data. The data display method does not match the actual needs of users or device characteristics, leading to data interpretation errors or cognitive overload in the information parsing process when processing batch and complex data. It is insufficient in flexibility and real-time performance when dynamically adjusting the visual display to match multiple user responses, limiting the application effect in fast-paced and high-demand environments. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a visualization output method and system for an artificial intelligence model.

[0006] In order to achieve the above object, the present invention adopts the following technical solution, a visualization output method of an artificial intelligence model, comprising the following steps:

[0007] S1: Analyze the information of nodes and edges in the network through the model structure network data, identify the number of connections of multiple nodes and calculate the weights, partition the network according to the connection strength of the nodes, and generate a model structure map;

[0008] S2: Based on the model structure map, classify multiple types of data according to statistical properties of nodes, including average number of connections and node degrees, and generate data type evaluation information;

[0009] S3: According to the data type evaluation information, combined with data application scenarios and user needs, matching chart types for multiple data according to data characteristics, and generating chart type matching parameters;

[0010] S4: According to the chart type matching parameters, the user's behavior data during data analysis is collected in real time, and the user's facial expression image is collected by using an image acquisition device to evaluate the user's visual focus and eye movement path, and generate a user behavior data set;

[0011] S5: Based on the user behavior data set, by analyzing the user's reaction time and facial expression changes, evaluating the user's cognitive load when processing visual information, and obtaining a cognitive load evaluation result;

[0012] S6: Based on the cognitive load assessment result, the layout of the chart and text is adjusted according to the user's device characteristics, the information density of the displayed content, and the user's cognitive load capacity, and a visualization output parameter is generated.

[0013] As a further solution of the present invention, the model structure mapping diagram includes node clustering partition information, node importance level, and edge connection strength index data; the data type evaluation information includes classified data labels, statistical feature information of multiple data, and a distribution diagram of the data type; the chart type matching parameters include a chart style list, chart design parameters, and chart preset configurations; the user behavior data set includes user operation frequency information, user visual attention area, and interface interaction mode; the cognitive load evaluation results include a cognitive load index, user attention distribution information, and emotional response analysis results; the visualization output parameters include layout visual element adjustment records, spatial optimization results of information layout, and visual priority of information presentation.

[0014] As a further solution of the present invention, the steps of analyzing the information of nodes and edges in the network through the model structure network data, identifying the number of connections of multiple nodes and calculating the weights, partitioning the network according to the connection strength of the nodes, and generating the model structure map are specifically as follows:

[0015] S101: Identify the types and weight parameters of multiple nodes and edges in the network through the model structure network data, classify and record the information, and obtain node and edge analysis data;

[0016] S102: Analyzing the data of the nodes and edges, taking into account the connection density and connection quality of the nodes, calculating the betweenness centrality of multiple nodes, identifying key nodes in the network, and obtaining node statistical feature information;

[0017] S103: Based on the node statistical feature information, by analyzing the connection strength and node importance, the network is logically partitioned to obtain a model structure mapping diagram.

[0018] As a further solution of the present invention, the specific formula for calculating the betweenness centrality of multiple nodes is:

[0019]

[0020] Among them, C B (v) represents the betweenness centrality of node v, which is used to measure the frequency of node v as a bridge connecting other node pairs in the network. s is the starting node in the network, which is the starting point of the path. v is the node currently analyzed, which is different from the starting node s and the terminal node t. t is the terminal node in the network, which is the end point of the path. V represents the set of all nodes in the graph. σ st is the total number of shortest paths from node s to node t, σ st (v) is the number of shortest paths from node s to node t that pass through node v.

[0021] As a further solution of the present invention, based on the model structure map, the steps of classifying multiple types of data according to the statistical attributes of the nodes, including the average number of connections and the node degree, and generating data type evaluation information are specifically as follows:

[0022] S201: Analyze the average number of connections and node degrees of multiple nodes according to the model structure mapping diagram, the centrality and network influence of the nodes, and obtain node attribute analysis results;

[0023] S202: Based on the node attribute analysis result, the nodes are classified into core nodes and edge nodes according to the statistical attributes and connection characteristics of the nodes and the influence degree, and the influence classification result is obtained;

[0024] S203: Based on the influence classification result, evaluate the clustering relationship of nodes in the network, identify the interaction patterns of multiple node groups, and obtain data type evaluation information.

[0025] As a further solution of the present invention, according to the data type evaluation information, combined with the data application scenario and user needs, the steps of matching chart types for multiple data according to data characteristics and generating chart type matching parameters are specifically as follows:

[0026] S301: Based on the data type evaluation information, according to the frequency and distribution characteristics of the various types of data, analyzing the visualization requirements of various types of data, including quantitative data, qualitative data, and time series data, and generating a data visualization requirement list;

[0027] S302: Based on the data visualization requirement list, matching chart types for multiple data, including selecting line charts and bar charts for quantitative data, selecting pie charts and radar charts for qualitative data, and selecting timeline charts for time series data, and generating requirement type matching records;

[0028] S303: Based on the requirement type matching record, various parameters of the chart, including the scale of the axis, the color and label of the chart, are adjusted according to the data application scenario and user requirements to generate chart type matching parameters.

[0029] As a further solution of the present invention, according to the chart type matching parameters, the user's behavior data during data analysis is collected in real time, and the user's facial expression image is collected by using an image acquisition device to evaluate the user's visual focus and eye movement path. The steps of generating a user behavior data set are specifically as follows:

[0030] S401: According to the chart type matching parameters, real-time monitoring and recording of user behavior data during data analysis, including page click frequency, scrolling behavior, and page dwell time, to generate user interaction behavior data;

[0031] S402: Based on the user interaction behavior data, using an image acquisition device, collect the user's facial expression when analyzing the data to generate user facial expression data;

[0032] S403: Analyze the user's visual focus position and eye movement path according to the user's facial expression data, and record corresponding timestamp information to generate a user behavior data set.

[0033] As a further solution of the present invention, based on the user behavior data set, the user's cognitive load when processing visual information is evaluated by analyzing the user's reaction time and facial expression changes. The steps of obtaining the cognitive load evaluation result are specifically as follows:

[0034] S501: Based on the user behavior data set, calculate the average reaction time of the user during the interaction process, compare it with the preset baseline reaction time, analyze the reaction speed of the user under various data display pages, and generate a reaction time analysis result;

[0035] S502: Based on the reaction time analysis result and in combination with the user's facial expression change data, identifying high-load expression features shown in data processing, and generating a facial expression load index;

[0036] S503: Analyze the degree of burden of the user when processing the visual information according to the facial expression load index and in combination with the reaction time data, calculate the cognitive load index, and generate a cognitive load evaluation result.

[0037] As a further solution of the present invention, based on the cognitive load evaluation result, the layout of the chart and text is adjusted according to the user's device characteristics, the information density of the displayed content, and the user's cognitive load capacity, and the steps of generating visualization output parameters are specifically as follows:

[0038] S601: Based on the cognitive load assessment result, identify the screen size and resolution parameters of the user device, adjust the size of the chart and text to match the device parameters, optimize the clarity and readability of the visual information, and generate a device adaptability adjustment record;

[0039] S602: In combination with the device adaptability adjustment record, according to the user's cognitive load capacity, adjusting the information density to match the user's processing capacity, including adjusting the interval of data points in the chart and adding or removing chart elements, and generating information density optimization parameters;

[0040] S603: Apply the information density optimization parameters to adjust the layout of charts and texts, including changing the layout direction, enhancing color contrast and text readability, optimizing the information display effect and the user's task processing efficiency, and generating visualization output parameters.

[0041] A visual output system for an artificial intelligence model, the visual output system for an artificial intelligence model is used to execute the visual output method for the artificial intelligence model, the system comprising:

[0042] The structural data analysis module analyzes the information of nodes and edges in the network based on the model structure network data, identifies the number of connections of multiple nodes and calculates the weights, partitions the network according to the connection strength of the nodes, and generates network mapping information;

[0043] The visual demand analysis module classifies various types of data based on the network mapping information and the statistical attributes of the nodes, matches chart types for various data in combination with data application scenarios and user needs, and generates chart configuration analysis results;

[0044] The user interaction analysis module collects the user's behavior data and user facial expression images in real time during data analysis based on the analysis results of the chart configuration, analyzes the user's visual focus and eye movement path, and generates interaction data collection records;

[0045] The cognitive ability analysis module evaluates the user's cognitive load when processing visual information by analyzing the user's reaction time and facial expression changes based on the interactive data collection record, and generates a processing ability analysis result;

[0046] The visualization layout optimization module adjusts the layout of visualization elements and generates visualization output parameters based on the processing capability analysis result and in combination with the user's device characteristics, information density, and user cognitive load capacity.

[0047] Compared with the prior art, the advantages and positive effects of the present invention are:

[0048] In the present invention, by analyzing the model structure network data and calculating the number of connections and weights of the nodes, the correlation between the data and the core structure of the network are revealed, the statistical attributes of the nodes are classified, and the chart type is matched according to the data characteristics to optimize the visual presentation effect of the data. On the basis of real-time monitoring of user interaction behavior, the chart layout is adjusted according to the user's visual focus and response data to reduce the cognitive load of information processing, enable decision makers to quickly identify patterns, trends and anomalies in the data, and improve their ability to understand the output data of the artificial intelligence model. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0050] Figure 2 This is a detailed flow chart of S1 of the present invention;

[0051] Figure 3 This is a detailed flow chart of S2 of the present invention;

[0052] Figure 4 This is a detailed flow chart of S3 of the present invention;

[0053] Figure 5 This is a detailed flow chart of S4 of the present invention;

[0054] Figure 6 This is a detailed flow chart of S5 of the present invention;

[0055] Figure 7 This is a detailed flow chart of S6 of the present invention;

[0056] Figure 8 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0058] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0059] See also Figure 1 The present invention provides a technical solution, a visualization output method of an artificial intelligence model, comprising the following steps:

[0060] S1: Analyze the information of nodes and edges in the network through the model structure network data, identify the number of connections of multiple nodes and calculate the weights, partition the network according to the connection strength of the nodes, and generate a model structure map;

[0061] S2: Based on the model structure map, multiple types of data are classified according to the statistical properties of the nodes, including the average number of connections and node degrees, to generate data type evaluation information;

[0062] S3: Based on the data type evaluation information, combined with the data application scenario and user needs, the chart type is matched for various data according to the data characteristics, and the chart type matching parameters are generated;

[0063] S4: According to the chart type matching parameters, the user's behavior data is collected in real time during data analysis, and the user's facial expression images are collected using image acquisition equipment to evaluate the user's visual focus and eye movement path, and generate a user behavior data set;

[0064] S5: Based on the user behavior dataset, the cognitive load of users when processing visual information is evaluated by analyzing the user's reaction time and facial expression changes, and the cognitive load evaluation results are obtained;

[0065] S6: Based on the cognitive load assessment results, adjust the layout of charts and texts according to the user's device characteristics, the information density of the displayed content, and the user's cognitive load capacity, and generate visualization output parameters.

[0066] The model structure mapping diagram includes node clustering partition information, node importance level, and edge connection strength index data. The data type evaluation information includes classified data labels, statistical feature information of multiple data, and distribution diagram of data types. The chart type matching parameters include chart style list, chart design parameters, and chart preset configuration. The user behavior data set includes user operation frequency information, user visual attention area, and interface interaction mode. The cognitive load evaluation results include cognitive load index, user attention distribution information, and emotional response analysis results. The visualization output parameters include layout visual element adjustment records, spatial optimization results of information layout, and visual priority of information presentation.

[0067] See also Figure 2 ,Through the model structure network data, the information of nodes and edges in the network is analyzed, the number of connections of multiple nodes is identified and the weights are calculated, the network is partitioned according to the connection strength of the nodes, and the steps of generating the model structure map are as follows:

[0068] S101: Identify the types and weight parameters of multiple nodes and edges in the network through the model structure network data, classify and record the information, and obtain node and edge analysis data;

[0069] In sub-step S101, the graphical network analysis tools Gephi and NetworkX are used to identify the types of nodes and edges in the model structure network to achieve parsing and visualization of large-scale network data. The types of nodes and edges are identified based on the structural position and connection mode in the network, including degree, clustering coefficient and path length. The decision tree classification algorithm is used to assign weights to nodes and edges. The weight parameters are set according to the influence and information transmission ability of the node or edge in the network. Nodes at the center of the network have higher weights to ensure that the data of each node and edge is classified and recorded, providing a basis for network behavior analysis and pattern recognition.

[0070] S102: Analyze the data based on nodes and edges, consider the connection density and connection quality of the nodes, calculate the betweenness centrality of multiple nodes, identify the key nodes in the network, and obtain the node statistical feature information;

[0071] The specific formula for calculating the betweenness centrality of multiple nodes is:

[0072]

[0073] Among them, C B(v) represents the betweenness centrality of node v, which is used to measure the frequency of node v as a bridge connecting other node pairs in the network. s is the starting node in the network, which is the starting point of the path. v is the node currently analyzed, which is different from the starting node s and the terminal node t. t is the terminal node in the network, which is the end point of the path. V represents the set of all nodes in the graph. σ st is the total number of shortest paths from node s to node t, σ st (v) is the number of shortest paths from node s to node t that pass through node v.

[0074] formula:

[0075]

[0076] Detailed explanation of the formula and the process of formula calculation and derivation:

[0077] The formula is used to calculate the betweenness centrality of node v, that is, the frequency with which node v acts as an intermediary on the shortest paths between all pairs of nodes s and t. A high value of betweenness centrality indicates that the node has a high influence in the network and the ability to control information flow.

[0078] Parameter meaning and setting value:

[0079] V is the set of all nodes in the graph, assuming that the total number of nodes in the network is set to 10;

[0080] s and t are any two nodes in the network that are not v, used to calculate the shortest path from s to t;

[0081] σ st is the number of all shortest paths from node s to node t, σ st (v) is the number of shortest paths from node s to node t that pass through node v. Assume that in the network analysis, there are three paths from node s to node t, and v appears on two of the three paths. st =3,σ st (v) = 2;

[0082] Substitute the parameters into the formula for calculation:

[0083]

[0084]

[0085] result It shows that about 66.7% of all the shortest paths between nodes s and t pass through node v, indicating that node v plays an important role in the information flow between these two nodes. The calculation results are used to help understand the structure of information control in the network and optimize the network structure.

[0086] S103: Based on the node statistical feature information, by analyzing the connection strength and node importance, the network is logically partitioned to obtain a model structure mapping diagram;

[0087] In sub-step S103, based on the node statistical feature information, a hierarchical clustering algorithm is used to logically partition the network nodes. By calculating the similarity measure between nodes and merging nodes step by step according to the set threshold, a predetermined number of partitions are formed. The betweenness centrality and closeness centrality of each node are used to evaluate the activity and connection strength in the network. Through the hierarchical clustering method, the network is divided into different areas, each of which contains nodes with similar functional properties or close interactions, and the logical layout of the network structure is optimized. The partition information is displayed in the model structure mapping diagram, which indicates the node configuration of each logical partition and its mutual connection relationship, providing an important reference for network design and optimization.

[0088] See also Figure 3 ,Based on the model structure mapping diagram, according to the statistical attributes of the nodes, including the average number of connections and node degrees,,the steps of classifying various types of data and generating data type evaluation information are as follows:

[0089] S201: According to the model structure mapping diagram, the average number of connections and node degrees of multiple nodes are analyzed according to the centrality and network influence of the nodes to obtain the node attribute analysis results;

[0090] In sub-step S201, according to the model structure mapping diagram, the network analysis tool NetworkX is used to analyze the average number of connections and node degrees of multiple nodes to calculate the degree of each node in the network, that is, the number of edges directly connected to the node, and clarify the structural importance of each node in the network. The average number of connections is obtained by calculating the sum of the number of connections of all nodes in the network and dividing it by the total number of nodes. It reflects the overall connection density of the network and reveals the centrality of the node, that is, the connection strength between the node and other nodes in the network, and obtains the node attribute analysis results. The results help understand the structural characteristics of the network and the importance of the nodes, and provide data support for subsequent network optimization and decision-making.

[0091] S202: Based on the node attribute analysis results, the nodes are classified into core and edge nodes according to the statistical attributes and connection characteristics of the nodes and the influence degree, and the influence classification results are obtained;

[0092] In the above content, according to the node attribute analysis results, according to the formula Calculate the degree centrality C of node i i ;

[0093] In the formula, C i represents the degree centrality of node i, d irepresents the number of edges connected to node i, and N represents the total number of nodes in the network;

[0094] Detailed explanation of the formula and the process of formula calculation and derivation:

[0095] Assume that there are 5 nodes in the network, and node i is directly connected to the other two nodes, namely, d i =2, N is 5, calculate degree centrality:

[0096]

[0097] Result C i =0.5 indicates that the centrality of node i in the network accounts for 50%. The degree centrality index reflects the relative importance of node i in the network. The calculation process is used to analyze the network influence of the node and classify the nodes according to this, including core and edge nodes, to obtain the influence classification results.

[0098] S203: Based on the influence classification results, evaluate the clustering relationship of nodes in the network, identify the interaction patterns of multiple node groups, and obtain data type evaluation information;

[0099] In sub-step S203, based on the influence classification results, a hierarchical clustering algorithm is used to evaluate the node clustering relationship in the network, and the interaction patterns between different node groups are identified. The node data is processed using a hierarchical clustering algorithm. By gradually aggregating node groups that are close to each other, the interaction and connection patterns of each group in the network are displayed. By analyzing the interaction patterns of various types of nodes, the key node groups and relatively independent groups in the network are discovered. The statistical properties and influence data of the nodes are combined with the clustering analysis results to obtain data type evaluation information.

[0100] See also Figure 4 , according to the data type evaluation information, combined with the data application scenario and user needs, and matching chart types for multiple data according to data characteristics, the steps for generating chart type matching parameters are as follows:

[0101] S301: Based on the data type assessment information, according to the frequency and distribution characteristics of various types of data, analyze the visualization requirements of various types of data, including quantitative data, qualitative data, and time series data, and generate a data visualization requirement list;

[0102] In sub-step S301, based on the data type assessment information, statistical analysis software R language and Python's Pandas library are used to perform data analysis to process and analyze batch data sets. The frequency and distribution characteristics of the data are determined through statistical analysis, including calculating the central tendency and dispersion of each type of data, such as the mean, median, standard deviation, and coefficient of variation. According to the attributes of the data, the needs of data visualization are identified, including quantitative data suitable for line graphs and bar graphs to show data change trends, qualitative data using pie charts and radar charts to show proportions and classifications, and time series data using timeline graphs to express. The analysis results are summarized into a data visualization requirements list, which describes the visualization methods and reasons for each type of data, providing a basis for chart design and implementation.

[0103] S302: Based on the data visualization requirement list, matching chart types for multiple data, including selecting line charts and bar charts for quantitative data, selecting pie charts and radar charts for qualitative data, and selecting timeline charts for time series data, and generating requirement type matching records;

[0104] In sub-step S302, according to the data visualization requirements list, the visualization design tools Tableau and Microsoft Excel are used to match chart types for various data, including selecting line graphs and bar graphs for quantitative data to accurately reflect changes in data volume, selecting pie charts and radar charts for qualitative data to intuitively display data proportions and structures, and selecting timeline graphs for time series data to clearly show data flow in the time dimension. Chart templates are selected and chart settings are adjusted to adapt to data types. After completing the chart matching, the generated requirement type matching record records the specific information and selection reasons for the selected charts for each data type, ensuring that the visualization of the data conforms to both data attributes and analysis requirements.

[0105] S303: Based on the demand type matching record, various parameters of the chart are adjusted according to the data application scenario and user needs, including the scale of the axis, the color and label of the chart, and the chart type matching parameters are generated;

[0106] In sub-step S303, based on the requirement type matching record, the parameters of the chart are adjusted to match the data application scenario and user needs. Adobe Illustrator and Photoshop are used to beautify the chart and adjust the parameters, including adjusting the axis scale of the chart to ensure the clarity and accuracy of the data display, selecting the color of the chart, using color theory to ensure that the visual effect of the chart is both beautiful and can effectively convey information, setting data labels and legends, and improving the readability of the data content of the chart. The adjustment of parameters is based on the data display requirements and the preferences of the target audience to achieve the best visual communication effect. The generated chart type matching parameters record the design details of all charts to ensure that the chart can meet both functional requirements and aesthetic standards in actual applications.

[0107] See also Figure 5 ,According to the chart type matching parameters, the user's behavior data is collected in real time during data analysis, and the user's facial expression images are collected using image acquisition equipment to evaluate the user's visual focus and eye movement path. The specific steps for generating the user behavior data set are as follows:

[0108] S401: According to the chart type matching parameters, real-time monitoring and recording of user behavior data during data analysis, including page click frequency, scrolling behavior, and page dwell time, to generate user interaction behavior data;

[0109] In the above content, according to the user interaction behavior data, according to the formula Calculate the click frequency P of the page;

[0110] In the formula, P represents the click frequency of the page, N c represents the total number of clicks, and T represents the total duration of the time period;

[0111] Detailed explanation of the formula and the process of formula calculation and derivation:

[0112] Assume that in a 1-minute user session, the user clicked 30 times in total. Calculate the click frequency:

[0113]

[0114] The result P=30 indicates that the user clicks 30 times per minute on average during this time period. The indicator is used to measure the interactive activity between the user and the page. The calculation process is used to monitor and record the user's interactive behavior in real time and generate user interactive behavior data.

[0115] S402: Based on the user interaction behavior data, the facial expression of the user when analyzing the data is collected by an image collection device to generate user facial expression data;

[0116] In sub-step S402, the facial recognition and emotion analysis technology Microsoft Face API is used to collect the user's facial expressions during the data analysis process. The user's expression changes are captured in real time and the emotional state is analyzed through image acquisition equipment, including configuring the camera in an appropriate position on the user's device to ensure clear capture of facial information. At the same time, the image data obtained by analyzing the facial recognition software is used to identify the user's basic emotions, including happiness, surprise, and confusion, to help reveal the user's emotional response when using the data analysis tool, and evaluate the tool's intuitiveness and ease of use. The generated user facial expression data includes expression types and changing trends, which are used to understand the user's emotional response and optimize user interaction design.

[0117] S403: Analyze the user's visual focus position and eye movement path according to the user's facial expression data, and record the corresponding timestamp information to generate a user behavior data set;

[0118] In sub-step S403, image analysis technology is used to analyze the user's eye position and identify the visual focus position and eye movement path through the image processing software OpenCV, including real-time capture of user facial images, detection of eye areas using image processing algorithms, eye positioning using Haar cascade classifiers and deep learning model YOLO, tracking the movement of the user's eyes on the screen, and analyzing the user's eye movement path and gaze point through continuous capture of the eye position, and recording the timestamp information corresponding to these data. The generated user behavior data set contains the coordinates, dwell time and timestamp of each gaze point. The data provides an important basis for understanding the user's attention allocation and information processing mode during the data analysis process.

[0119] See also Figure 6 Based on the user behavior dataset, the cognitive load of users when processing visual information is evaluated by analyzing the user's reaction time and facial expression changes. The specific steps for obtaining the cognitive load evaluation results are as follows:

[0120] S501: Based on the user behavior data set, calculate the average reaction time of the user during the interaction process, compare it with the preset baseline reaction time, analyze the reaction speed of the user under various data display pages, and generate reaction time analysis results;

[0121] In the above content, based on the user behavior data set, according to the formula Calculate the average reaction time of users during interaction

[0122] In the formula, represents the average reaction time, RT i represents the reaction time of each data point, and N represents the total number of data points;

[0123] Detailed explanation of the formula and the process of formula calculation and derivation:

[0124] Assume that during a session, the user's reaction time data points are 100ms, 120ms, 150ms, 130ms, and 110ms respectively. Calculate the average reaction time:

[0125]

[0126] The result 122ms indicates the average reaction speed of users under various data display pages. This indicator is used to compare with the preset baseline reaction time. Assuming the baseline is 150ms, the analysis shows that the user's reaction speed is faster than the baseline. The calculation method is used to analyze the user's reaction speed under various data display pages and generate reaction time analysis results.

[0127] S502: Based on the reaction time analysis result and in combination with the user's facial expression change data, identifying the high-load expression features shown in the data processing, and generating a facial expression load index;

[0128] In sub-step S502, based on the reaction time analysis results, the image processing and emotion analysis software OpenCV and AffectivaSDK are used to analyze the changes in the user's facial expressions during data processing, identify expression features, including frowning and wide eyes, extract facial feature points from facial images captured during user interaction, use a machine learning model to classify the features, determine whether they indicate stress or distress, and encode the target facial expression features into a facial expression load index to quantify the user's emotional load in the data processing task. The generated facial expression load index provides key emotional feedback data for optimizing interface design and user experience.

[0129] S503: Analyze the user's burden when processing visual information based on the facial expression load index and the reaction time data, calculate the cognitive load index, and generate a cognitive load evaluation result;

[0130] In sub-step S503, the facial expression load index and reaction time data are combined, and cognitive load theory and related algorithms, including G-index or NASA-TLX, are used to analyze the degree of burden on users when processing visual information, to achieve quantitative evaluation of cognitive load, and to calculate a cognitive load index, including aggregating reaction time data and facial expression load data, to calculate a comprehensive index representing the psychological and cognitive pressure that users endure in interface interaction. The generated cognitive load evaluation results help interface designers and data analysts understand the efficiency of existing interface designs in user cognitive processing, and provide a basis for designing more intuitive and lower-load user interfaces.

[0131] See also Figure 7Based on the cognitive load assessment results, the layout of charts and texts is adjusted according to the user's device characteristics, the information density of the displayed content, and the user's cognitive load capacity. The steps to generate visualization output parameters are as follows:

[0132] S601: Based on the cognitive load assessment result, identify the screen size and resolution parameters of the user's device, adjust the size of the chart and text to match the device parameters, optimize the clarity and readability of the visual information, and generate a device adaptability adjustment record;

[0133] In sub-step S601, based on the cognitive load assessment results, JavaScript and CSS media query functions are used to detect the screen size and resolution parameters of the user's device, so that the web page can dynamically adapt to different display devices, ensure the adaptability and readability of the content on various screen sizes, adjust the size of the chart and text according to the device parameters, modify the viewport and view box attributes of the SVG chart, adjust the font size and line spacing of the HTML text, and ensure the clarity and readability of the visual information. The generated device adaptability adjustment record lists the adjustments made for each device type, including size and resolution parameters, to provide a basis and reference for page adjustment and design.

[0134] S602: combining the device adaptability adjustment record, adjusting the information density to match the user's processing capacity according to the user's cognitive load capacity, including adjusting the interval of data points in the chart and adding or removing chart elements, and generating information density optimization parameters;

[0135] In sub-step S602, based on the device adaptability adjustment record, the data visualization tool D3.js is used to optimize the information density of the chart. According to the user's cognitive load capacity, the data point interval in the chart is adjusted, and the chart elements are reduced or increased to match the user's processing capacity, including using D3.js's .scale and .axis methods to adjust the scale and coordinate axis of the chart, and modify the layout properties of the chart, including the bar width of the bar chart and the point size of the scatter plot, to optimize the information density of the chart and ensure that the user can quickly and accurately interpret the chart data without feeling stressed. The generated information density optimization parameters record the value and application effect of each adjustment, providing data support for user interface optimization.

[0136] S603: applying information density optimization parameters, adjusting chart and text layout, including changing layout direction, enhancing color contrast and text readability, optimizing information display effect and user task processing efficiency, and generating visualization output parameters;

[0137] In sub-step S603, the information density optimization parameters are applied, and the layout of charts and texts is adjusted through the design software Adobe Illustrator and Adobe XD, including changing the layout direction, changing the horizontal bar chart to a vertical one to adapt to the mobile device screen, enhancing the color contrast to improve the visual impact of the chart, and adjusting the text size and font style to enhance readability, including selecting matching color schemes and typesetting layouts, and applying color theory and visual design principles to ensure clear transmission and beauty of information. The generated visual output parameters record the content and goals of each adjustment, improving the effect of information display and the user's task processing efficiency.

[0138] See also Figure 8 , a visualization output system of an artificial intelligence model, the visualization output system of an artificial intelligence model is used to execute the visualization output method of the artificial intelligence model, and the system includes:

[0139] The structural data analysis module analyzes the information of nodes and edges in the network based on the model structure network data, identifies the number of connections of multiple nodes and calculates the weights, partitions the network according to the connection strength of the nodes, and generates network mapping information;

[0140] The visual demand analysis module classifies various types of data based on network mapping information and the statistical attributes of nodes, matches chart types for various data in combination with data application scenarios and user needs, and generates chart configuration analysis results;

[0141] The user interaction analysis module collects the user's behavior data and facial expression images in real time during data analysis based on the chart configuration analysis results, analyzes the user's visual focus and eye movement path, and generates interaction data collection records;

[0142] The cognitive ability analysis module is based on the interactive data collection records. By analyzing the user's reaction time and facial expression changes, it evaluates the user's cognitive load when processing visual information and generates processing ability analysis results;

[0143] The visualization layout optimization module adjusts the layout of visualization elements and generates visualization output parameters based on the processing power analysis results, combined with the user's device characteristics, information density, and user cognitive load capacity.

[0144] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for visual output of an artificial intelligence model, characterized in that: The following steps are involved: Through the model structure network data, the information of nodes and edges in the network is analyzed, the number of connections of multiple nodes is identified and the weights are calculated, the network is partitioned according to the connection strength of the nodes, and a model structure map is generated; Based on the model structure map, multiple types of data are classified according to statistical properties of nodes, including average number of connections and node degrees, to generate data type evaluation information; According to the data type evaluation information, combined with data application scenarios and user needs, matching chart types for multiple data according to data characteristics, and generating chart type matching parameters; According to the chart type matching parameters, the user's behavior data during data analysis is collected in real time, and the user's facial expression image is collected using an image acquisition device to evaluate the user's visual focus and eye movement path, and generate a user behavior data set; Based on the user behavior data set, by analyzing the user's reaction time and facial expression changes, the user's cognitive load when processing visual information is evaluated to obtain a cognitive load evaluation result; Based on the cognitive load assessment result, the layout of the chart and text is adjusted according to the user's device characteristics, the information density of the displayed content, and the user's cognitive load capacity, and the visualization output parameters are generated.

2. The visualization output method of the artificial intelligence model according to claim 1, characterized in that: The model structure mapping diagram includes node clustering partition information, node importance level, and edge connection strength index data; the data type evaluation information includes classified data labels, statistical feature information of multiple data, and distribution diagrams of data types; the chart type matching parameters include chart style lists, chart design parameters, and chart preset configurations; the user behavior data set includes user operation frequency information, user visual attention areas, and interface interaction modes; the cognitive load evaluation results include cognitive load index, user attention distribution information, and emotional response analysis results; the visualization output parameters include layout visual element adjustment records, spatial optimization results of information layout, and visual priority of information presentation.

3. The visualization output method of the artificial intelligence model according to claim 1, characterized in that: Through the model structure network data, the information of nodes and edges in the network is analyzed, the number of connections of multiple nodes is identified and the weights are calculated, and the network is partitioned according to the connection strength of the nodes. The specific steps for generating the model structure map are as follows: Through the model structure network data, the types and weight parameters of multiple nodes and edges in the network are identified, the information is classified and recorded, and the node and edge analysis data are obtained; According to the node and edge analysis data, considering the connection density and connection quality of the nodes, calculating the betweenness centrality of multiple nodes, identifying the key nodes in the network, and obtaining the node statistical feature information; Based on the node statistical feature information, the network is logically partitioned by analyzing the connection strength and node importance to obtain a model structure mapping diagram.

4. The method for visualizing an artificial intelligence model according to claim 3, characterized in that: The specific formula for calculating the betweenness centrality of multiple nodes is: Among them, C B (v) represents the betweenness centrality of node v, which is used to measure the frequency of node v as a bridge connecting other node pairs in the network. s is the starting node in the network, which is the starting point of the path. v is the node currently analyzed, which is different from the starting node s and the terminal node t. t is the terminal node in the network, which is the end point of the path. V represents the set of all nodes in the graph. σ st is the total number of shortest paths from node s to node t, σ st (v) is the number of shortest paths from node s to node t that pass through node v.

5. The method for visualizing an artificial intelligence model according to claim 1, characterized in that: Based on the model structure map, the steps of classifying various types of data according to the statistical properties of the nodes, including the average number of connections and the node degree, and generating data type evaluation information are as follows: According to the model structure mapping diagram, the average number of connections and node degrees of multiple nodes are analyzed according to the centrality and network influence of the nodes to obtain the node attribute analysis results; Based on the node attribute analysis results, the nodes are classified into core and edge nodes according to the statistical attributes and connection characteristics of the nodes and the degree of influence, and the influence classification results are obtained; Based on the influence classification results, the clustering relationship of nodes in the network is evaluated, the interaction patterns of multiple node groups are identified, and the data type evaluation information is obtained.

6. The method for visual output of an artificial intelligence model according to claim 1, characterized in that: According to the data type evaluation information, combined with the data application scenario and user needs, the steps of matching chart types for multiple data according to data characteristics and generating chart type matching parameters are specifically as follows: Based on the data type assessment information, according to the frequency and distribution characteristics of various types of data, the visualization requirements of various types of data are analyzed, including quantitative data, qualitative data, and time series data, and a data visualization requirement list is generated; Based on the data visualization requirement list, matching chart types for multiple data, including selecting line charts and bar charts for quantitative data, selecting pie charts and radar charts for qualitative data, and selecting timeline charts for time series data, and generating requirement type matching records; Based on the requirement type matching record, various parameters of the chart are adjusted according to the data application scenario and user needs, including the scale of the axis, the color and label of the chart, to generate chart type matching parameters.

7. The method for visual output of an artificial intelligence model according to claim 1, characterized in that: According to the chart type matching parameters, the user's behavior data during data analysis is collected in real time, and the user's facial expression images are collected using image acquisition equipment to evaluate the user's visual focus and eye movement path. The specific steps for generating a user behavior data set are as follows: According to the chart type matching parameters, real-time monitoring and recording of user behavior data during data analysis, including page click frequency, scrolling behavior, and page dwell time, generates user interaction behavior data; Based on the user interaction behavior data, the facial expression of the user when analyzing the data is collected by an image acquisition device to generate user facial expression data; According to the user's facial expression data, the user's visual focus position and eye movement path are analyzed, and the corresponding timestamp information is recorded to generate a user behavior data set.

8. The method for visualizing an artificial intelligence model according to claim 1, characterized in that: Based on the user behavior data set, the user's cognitive load when processing visual information is evaluated by analyzing the user's reaction time and facial expression changes. The steps for obtaining the cognitive load evaluation result are specifically as follows: Based on the user behavior data set, the average reaction time of the user during the interaction process is calculated, and compared with the preset baseline reaction time, the reaction speed of the user under various data display pages is analyzed, and the reaction time analysis results are generated; Based on the reaction time analysis results, combined with the user's facial expression change data, identifying high-load expression features shown in data processing, and generating a facial expression load index; According to the facial expression load index, combined with the reaction time data, the burden degree of the user when processing the visual information is analyzed, and the cognitive load index is calculated to generate a cognitive load evaluation result.

9. The method for visual output of an artificial intelligence model according to claim 1, characterized in that: Based on the cognitive load assessment result, the layout of the chart and text is adjusted according to the user's device characteristics, the information density of the displayed content, and the user's cognitive load capacity, and the steps of generating visualization output parameters are specifically as follows: Based on the cognitive load assessment result, identify the screen size and resolution parameters of the user's device, adjust the size of graphics and text to match the device parameters, optimize the clarity and readability of visual information, and generate a device adaptability adjustment record; In combination with the device adaptive adjustment record, according to the user's cognitive load capacity, the information density is adjusted to match the user's processing capacity, including adjusting the interval of data points in the chart and adding or removing chart elements, and generating information density optimization parameters; Apply the information density optimization parameters to adjust the layout of charts and texts, including changing the layout direction, enhancing color contrast and text readability, optimizing the information display effect and the user's task processing efficiency, and generating visualization output parameters.

10. A visual output system for an artificial intelligence model, characterized in that: According to the method for visual output of an artificial intelligence model according to any one of claims 1 to 9, the system comprises: The structural data analysis module analyzes the information of nodes and edges in the network based on the model structure network data, identifies the number of connections of multiple nodes and calculates the weights, partitions the network according to the connection strength of the nodes, and generates network mapping information; The visual demand analysis module classifies various types of data based on the network mapping information and the statistical attributes of the nodes, matches chart types for various data in combination with data application scenarios and user needs, and generates chart configuration analysis results; The user interaction analysis module collects the user's behavior data and user facial expression images in real time during data analysis based on the analysis results of the chart configuration, analyzes the user's visual focus and eye movement path, and generates interaction data collection records; The cognitive ability analysis module evaluates the user's cognitive load when processing visual information by analyzing the user's reaction time and facial expression changes based on the interactive data collection record, and generates a processing ability analysis result; The visualization layout optimization module adjusts the layout of visualization elements and generates visualization output parameters based on the processing capability analysis result and in combination with the user's device characteristics, information density, and user cognitive load capacity.

Images