Business process configuration method based on Activiti and AI decision

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a business process configuration method based on Activiti and AI decision, and the method comprises the steps: collecting the historical execution data of a business process through a process instance monitoring module, generating a process behavior feature data set, carrying out the training processing of an AI decision model based on the process behavior feature data set, and carrying out the analysis of the AI decision model. A business process decision tree model is constructed, AI decision routing labels are configured for decision nodes in an Activiti process engine, non-intrusive intelligent decision integration is achieved, and when a process instance is executed to a key decision node, a system automatically calls the pre-trained business process decision tree model to conduct real-time path analysis and generate an optimal dynamic routing instruction. The problem of decision stiffness caused by the fact that a traditional workflow depends on artificial experience configuration is solved, the accuracy and adaptability of flow branch selection in a complex service scene are improved, and meanwhile the compatibility and stability of an original flow engine are guaranteed.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a business process configuration method based on Activiti and AI decision-making. Background Technology

[0002] Artificial intelligence is an interdisciplinary and emerging field based on computer science, integrating computer science, psychology, philosophy, and other disciplines. It studies and develops theories, methods, technologies, and application systems to simulate, extend, and expand human intelligence. It attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems.

[0003] The business process configuration method based on Activiti and AI decision-making is an innovative system that integrates process engine and intelligent decision-making technology. It aims to improve the dynamic adaptability and decision accuracy of enterprise business processes. This is one of the core supporting technologies for enterprises to achieve intelligent automated management. As an open-source business process management framework, Activiti supports the design, deployment and execution of business processes through its standardized process engine, while the AI ​​decision-making model gives process nodes intelligent routing capabilities, together building a dynamic process configuration system with self-learning capabilities.

[0004] Currently, in actual enterprise application scenarios, the traditional Activiti process configuration has limitations: process decision nodes rely on preset fixed rules and manual experience configuration, and cannot be dynamically optimized based on historical execution data, resulting in low configuration efficiency and easy errors in complex business scenarios. When the budget approval process needs to divide different approval paths according to the amount threshold, it is necessary to manually maintain the exclusive gateway expression, which makes it difficult to respond to changes in business rules in real time.

[0005] Therefore, a business process configuration method based on Activiti and AI decision-making is proposed to solve the above problems. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a business process configuration method based on Activiti and AI decision-making. This method solves the problem mentioned in the background that process decision nodes rely on preset fixed rules and manual experience for configuration, making it impossible to dynamically optimize based on historical execution data. This results in low configuration efficiency and a high risk of errors in complex business scenarios.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a business process configuration method based on Activiti and AI decision-making, the method comprising the following steps: S1. Collect historical execution data of business processes through the process instance monitoring module to generate a process behavior feature dataset; S2. Based on the process behavior feature dataset, perform AI decision model training and processing to construct a business process decision tree model; S3. Configure business process nodes in the Activiti process engine and add AI decision routing tags to decision-making nodes; S4. When the process instance executes to a node with an AI decision routing label, the business process decision tree model is invoked to perform real-time decision path analysis and generate dynamic routing instructions. S5. Activate the target process branch according to the dynamic routing instruction, and at the same time update the process behavior feature dataset by collecting execution result data through the feedback learning module; S6. Incrementally train the decision model based on the updated process behavior feature dataset to generate an optimized business process decision tree model. S7. Dynamically adjust the AI ​​decision routing label binding relationship in the Activiti process definition through the visual configuration interface to generate a versioned process decision configuration scheme. S8. Perform root cause analysis on the abnormal process nodes and output suggestions for correcting the decision logic. S9. Based on the version history, the execution process will determine the canary release and rollback operations of the configuration scheme; S10. Generate a process execution analysis report that includes AI decision-making paths.

[0008] Preferably, step S1 includes the following steps: S11. Real-time capture of task processing time, path selection frequency, and abnormal interruption event parameters of process instances through the data collection device. S12. Perform feature vectorization transformation on the collected raw parameters to construct a multi-dimensional process behavior feature matrix; S13. Remove duplicate and invalid feature vectors through the data cleaning module to generate a standardized process behavior feature dataset.

[0009] Preferably, step S2 includes the following steps: S21. Divide the process behavior feature dataset into a training set and a validation set; S22. Use the gradient boosting decision tree algorithm for iterative training on the training set; S23. Verify the model accuracy using the validation set. If the accuracy falls below a threshold, adjust the feature weights and retrain.

[0010] in, This is a dynamic accuracy threshold. This represents the historical average accuracy. The standard deviation of historical accuracy; Threshold explanation: Retraining is triggered when the current model accuracy is lower than the historical average accuracy minus 2 standard deviations; S24. Deploy the validated decision tree model to the AI ​​service gateway.

[0011] Preferably, step S3 includes the following steps: S31. Parse the exclusive gateway node in the Activiti process definition XML file; S32. Add an AI decision routing label to the specified exclusive gateway node in the visual configuration interface; S33. Persistently store the tag binding information in the process decision configuration database.

[0012] Preferably, step S4 includes the following steps: S41, Listen to the event bus of the process engine and capture node arrival events containing AI decision routing labels; S42. Extract the context feature parameters of the current process instance; S43. Call the AI ​​service gateway interface to input feature parameters and obtain the route direction prediction value; S44. Activate the corresponding process branch based on the predicted value.

[0013] Preferably, step S5 includes the following steps: S51. Collect the execution result data and time consumption parameters of the process branch; S52. Compare the execution results with the decision prediction values ​​to generate a confidence index:

[0014] in, To normalize the confidence level, For the actual result vector, For the predicted result vector, Number of decision dimensions; Threshold explanation: Time-marked as low-confidence sample; S53. Label decision data with confidence levels below the threshold as reinforcement learning samples.

[0015] Preferably, step S7 includes the following steps: S71. Display the node topology of the process definition diagram in a visual interface; S72. Bind the AI ​​decision routing label to the newly added decision node by dragging and dropping; S73. Generate a version difference comparison chart and record configuration change logs; S74. Perform syntax validation on the revised process definition file.

[0016] Preferably, step S8 includes the following steps: S81. Capture execution timeout and routing error exceptions; S82. Extract the input features and decision tree path of the abnormal nodes; S83. Determine defective nodes through the decision path inversion localization model; S84, Suggestions for adjusting output feature weights and supplementing samples.

[0017] Preferably, step S9 includes the following steps: S91. Create an isolated test environment for the new process decision configuration scheme; S92. Distribute production environment process instances to the test environment according to a preset ratio; S93. Monitor the fluctuation of decision accuracy in the test environment; S94. Automatically trigger version rollback when accuracy drops below the tolerance threshold.

[0018] Preferably, step S10 includes the following steps: S101, AI decision-making path data that aggregates historical process instances; S102. Generate a heatmap of decision path distribution and a branch selection probability matrix; S103, Mark the root cause localization markers for high-frequency abnormal decision-making paths; S104. Outputs a dual-mode report in PDF and interactive visualization.

[0019] (III) Beneficial Effects Compared with existing technologies, this invention provides a business process configuration method based on Activiti and AI decision-making, which has the following beneficial effects: 1. In this invention, by constructing a dynamic coupling mechanism between the AI ​​decision engine and workflow nodes, AI decision routing labels are configured for decision-making nodes in the Activiti process engine to achieve non-intrusive intelligent decision integration. When a process instance executes to a key decision node, the system automatically calls a pre-trained business process decision tree model to perform real-time path analysis and generate the optimal dynamic routing instructions. This solves the problem of rigid decision-making caused by the reliance on manual experience in traditional workflows, improves the accuracy and adaptability of process branch selection in complex business scenarios, and ensures the compatibility and stability of the native process engine.

[0020] 2. In this invention, by establishing a closed-loop learning system of decision feedback and model self-optimization, after the execution of dynamic routing instructions to activate the target process branch, the system collects execution result data in real time and generates a normalized confidence index. Low-confidence decision data is automatically marked as reinforcement learning samples. Based on the incrementally updated process behavior feature dataset, the system continuously optimizes the decision model, enabling the system to autonomously evolve decision logic from historical execution anomalies. This effectively reduces the risk of process interruption caused by dynamic changes in business rules and enhances the resilience of business processes in the face of complex environmental changes.

[0021] 3. In this invention, by developing a visual decision diagnosis and flexible control system, it supports decision path inversion analysis for abnormal process nodes, accurately locates defective nodes in the model and outputs feature weight adjustment suggestions. Combined with the canary release mechanism of the versioned process decision configuration scheme, when the decision accuracy fluctuation in the test environment exceeds the tolerance threshold, a rollback operation is automatically triggered, realizing second-level circuit breaker control for high-risk changes. This mechanism fundamentally eliminates systemic process collapse caused by intelligent decision model iteration, ensuring the continuity and reliability of the enterprise's core business links.

[0022] 4. In this invention, by innovating the multi-dimensional decision analysis report generation mechanism, AI decision path data of historical process instances are aggregated to construct a decision path distribution heatmap and branch selection probability matrix, and the root cause localization markers of high-frequency abnormal paths are marked. The output adopts a dual mode of PDF and interactive visualization, enabling managers to penetrate and analyze process decision bottlenecks, providing data-driven decision support for business process reengineering, and improving the scientificity and foresight of enterprise strategic process optimization.

[0023] 5. In this invention, a dynamic threshold management system driven by design rules and data is constructed. Adaptive retraining trigger conditions are built based on the standard deviation of historical decision accuracy. The confidence alarm threshold is automatically calibrated according to the Euclidean distance of multidimensional feature vectors. This mechanism achieves a dynamic balance between anomaly detection sensitivity and system stability, avoids the problems of excessive alarms and missed alarms caused by traditional fixed thresholds, and improves the robustness of the intelligent decision-making system in fluctuating business environments. Attached Figure Description

[0024] Figure 1 This is a flowchart of the overall method of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Please see Figure 1 This is a business process configuration method based on Activiti and AI decision-making, which includes the following steps: S1. Collect historical execution data of business processes through the process instance monitoring module to generate a process behavior feature dataset; S2. Train and process the AI ​​decision-making model based on the process behavior feature dataset to build a business process decision tree model; S3. Configure business process nodes in the Activiti process engine and add AI decision routing tags to decision-making nodes; S4. When the process instance executes to a node with an AI decision routing label, the business process decision tree model is invoked to perform real-time decision path analysis and generate dynamic routing instructions. S5. Activate the target process branch according to the dynamic routing instruction, and at the same time, collect the execution result data through the feedback learning module to update the process behavior feature dataset; S6. Incrementally train the decision model based on the updated process behavior feature dataset to generate an optimized business process decision tree model. S7. Dynamically adjust the AI ​​decision routing label binding relationship in the Activiti process definition through the visual configuration interface to generate a versioned process decision configuration scheme. S8. Perform root cause analysis on the abnormal process nodes and output suggestions for correcting the decision logic. S9. Based on the version history, the execution process will determine the canary release and rollback operations of the configuration scheme; S10. Generate a process execution analysis report that includes AI decision-making paths; S1 includes the following steps: S11. Real-time capture of task processing time, path selection frequency, and abnormal interruption event parameters of process instances through the data collection device. S12. Perform feature vectorization transformation on the collected raw parameters to construct a multi-dimensional process behavior feature matrix; S13. The data cleaning module removes duplicate and invalid feature vectors to generate a standardized process behavior feature dataset. The data cleaning module also calculates the feature vector dispersion and filters out inefficient features.

[0027] in, The feature vector dispersion, The standard deviation of the eigenvectors, , These are the maximum and minimum values ​​of the eigenvectors; Threshold explanation: The feature vectors are considered low-discrimination features and are automatically filtered. S2 includes the following steps: S21. Divide the process behavior feature dataset into a training set and a validation set; S22. Use the gradient boosting decision tree algorithm for iterative training on the training set; S23. Verify the model accuracy using the validation set. If the accuracy falls below a threshold, adjust the feature weights and retrain.

[0028] in, This is a dynamic accuracy threshold. This represents the historical average accuracy. The standard deviation of historical accuracy; Threshold explanation: Retraining is triggered when the current model accuracy is lower than the historical average accuracy minus 2 standard deviations; S24. Deploy the validated decision tree model to the AI ​​service gateway; S3 includes the following steps: S31. Parse the exclusive gateway node in the Activiti process definition XML file; S32. Add an AI decision routing label to the specified exclusive gateway node in the visual configuration interface; S33. Persistently store the tag binding information in the process decision configuration database, and calculate the decision path complexity to optimize the tag binding logic:

[0029] in, To normalize the decision path complexity, This represents the number of conditional branches for the current node. This represents the total number of nodes in the process model. Threshold explanation: The nodes automatically enable AI decision routing labels; S4 includes the following steps: S41, Listen to the event bus of the process engine and capture node arrival events containing AI decision routing labels; S42. Extract the context feature parameters of the current process instance; S43. Call the AI ​​service gateway interface to input feature parameters and obtain the route direction prediction value; S44. Activate the corresponding process branch based on the predicted value, and calculate the process branch activation efficiency:

[0030] in, To normalize activation efficiency, For the theoretical shortest response time, This is the actual response time. This is the attenuation coefficient, with a default value of 0.5. , These are the encoding vectors for the actual path and the predicted path; Threshold explanation: An alarm is triggered at any time; S5 includes the following steps: S51. Collect the execution result data and time consumption parameters of the process branch; S52. Compare the execution results with the decision prediction values ​​to generate a confidence index:

[0031] in, To normalize the confidence level, For the actual result vector, For the predicted result vector, Number of decision dimensions; Threshold explanation: Time-marked as low-confidence sample; S53. Label decision data with confidence levels below the threshold as reinforcement learning samples; S7 includes the following steps: S71. Display the node topology of the process definition diagram in a visual interface; S72. Bind the AI ​​decision routing label to the newly added decision node by dragging and dropping; S73. Generate a version difference comparison chart and record configuration change logs; S74. Perform syntax validation on the revised process definition file; S8 includes the following steps: S81. Capture execution timeout and routing error exceptions; S82. Extract the input features and decision tree path of the abnormal nodes; S83. Determine defective nodes using the decision path inversion localization model:

[0032] in, To normalize defect weights, For the first The path weights of each decision node. The depth of the node in the decision tree. For the number of historical errors of a node, Total number of errors in the process; Threshold explanation: The node output is the critical defect node; S84. Suggestions for adjusting output feature weights and supplementing samples; S9 includes the following steps: S91. Create an isolated test environment for the new process decision configuration scheme; S92. Distribute production environment process instances to the test environment according to a preset ratio; S93. Monitor the fluctuation of decision accuracy in the test environment; S94. Automatically trigger version rollback when accuracy drops below the tolerance threshold. S10 includes the following steps: S101, AI decision-making path data that aggregates historical process instances; S102. Generate a heatmap of decision path distribution and a branch selection probability matrix:

[0033] in, For nodes Select branch The normalized probability, For nodes Select branch Historical frequency This is the smoothing coefficient, with a default value of 0.7. For nodes The total number of branches; Threshold explanation: The branches are marked as low-probability abnormal paths; S103, Mark the root cause localization markers for high-frequency abnormal decision-making paths; S104. Outputs a dual-mode report in PDF and interactive visualization.

[0034] The steps for configuring a business process based on Activiti and AI decision-making are as follows: I. Construction of the Integrated Architecture of Activiti and AI Decision Making 1. Process Engine Initialization The business process framework is built on the Activiti engine. The process model is defined according to the BPMN2.0 specification, and the process definition is deployed to the database using XML configuration files. The database-schema-update:true parameter is enabled during engine initialization to ensure that the database tables required for the process are created and updated automatically.

[0035] AI integration point: Embed AI decision routing labels in decision-making nodes. These labels are bound to pre-trained AI decision models to form a mapping relationship of "conditional branch - model call".

[0036] 2. Dynamic Data Acquisition Layer The monitoring module captures multi-dimensional feature data of process instances in real time, including: Temporal characteristics: task processing time, node dwell time; Path characteristics: branch selection frequency, number of abnormal interruptions; Contextual features: process variables, executor roles.

[0037] The data cleaning stage employs a discrete filtering algorithm: in Features that are considered low-discrimination data are automatically removed to improve the quality of the dataset.

[0038] II. AI Decision Model Training and Dynamic Routing Mechanism 3. Decision Tree Model Construction Training a dynamic routing model based on gradient boosting decision tree algorithm: Feature engineering: converting process behavior data into a feature vector matrix; Dynamic threshold training: The threshold for triggering retraining based on validation set accuracy is adaptively adjusted according to historical performance. This represents the historical average accuracy. The standard deviation is denoted as .

[0039] 4. Real-time decision path generation When a process instance executes to a node containing an AI route tag: Extract the context feature vector of the current instance; Call the AI ​​service interface to obtain the route direction prediction value; Calculate branch activation efficiency to verify the effectiveness of the decision: An alarm is triggered at any time.

[0040] III. Closed-Loop Feedback and Model Optimization System 5. Confidence-driven incremental learning The execution results of the collected branches are compared with the predicted values ​​to generate normalized confidence scores. The samples are labeled as low-confidence data and input into the reinforcement learning sample pool.

[0041] Optimization mechanism: Incremental training: The decision tree model is updated daily with new data. Feature weight adjustment: Correct the node splitting rules based on backpropagation of sample errors.

[0042] 6. Root Cause Analysis of Decision Anomalies Perform inverse diagnostics on nodes with execution timeouts and routing errors: Extract the input features and decision tree paths of outlier nodes; Key issues in calculating defect node weights for localization: The node output is a high-risk defect.

[0043] IV. Visual Configuration and Flexible Control 7. Dynamic binding and version management Achieve this through a visual interface: Drag-and-drop binding: Dynamically associate AI route tags with newly added decision nodes; Complexity check: Nodes must meet path complexity conditions before enabling AI routing: It will be enabled automatically at that time.

[0044] 8. Gray-scale release and circuit breaker mechanism The new version's decision configuration allocates 10% of production environment instances to the testing environment. Monitoring the fluctuations in decision accuracy, when the decrease is greater than Then automatically roll back the version. The version difference comparison chart records the configuration change trajectory.

[0045] V. Generation of Multidimensional Decision Analysis Reports 9. Probabilistic Modeling of Decision Paths Generate a branch selection probability matrix by aggregating historical instance data: :node Select branch The normalized probability; The branch is marked as an abnormally low-frequency path.

[0046] 10. Root Cause Labeling and Dual-Mode Output Mark the defect location points along high-frequency anomaly paths on the heat map; Outputs PDF reports and interactive visualization dashboards, supporting drill-down analysis.

[0047] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0048] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A business process configuration method based on Activiti and AI decision-making, characterized in that, The method includes the following steps: S1. Collect historical execution data of business processes through the process instance monitoring module to generate a process behavior feature dataset; S2. Based on the process behavior feature dataset, perform AI decision model training and processing to construct a business process decision tree model; S3. Configure business process nodes in the Activiti process engine and add AI decision routing tags to decision-making nodes; S4. When the process instance executes to a node with an AI decision routing label, the business process decision tree model is invoked to perform real-time decision path analysis and generate dynamic routing instructions. S5. Activate the target process branch according to the dynamic routing instruction, and at the same time update the process behavior feature dataset by collecting execution result data through the feedback learning module; S6. Incrementally train the decision model based on the updated process behavior feature dataset to generate an optimized business process decision tree model. S7. Dynamically adjust the AI ​​decision routing label binding relationship in the Activiti process definition through the visual configuration interface to generate a versioned process decision configuration scheme. S8. Perform root cause analysis on the abnormal process nodes and output suggestions for correcting the decision logic. S9. Based on the version history, the execution process will determine the canary release and rollback operations of the configuration scheme; S10. Generate a process execution analysis report that includes AI decision-making paths.

2. The business process configuration method based on Activiti and AI decision-making according to claim 1, characterized in that, S1 includes the following steps: S11. Real-time capture of task processing time, path selection frequency, and abnormal interruption event parameters of process instances through the data collection device. S12. Perform feature vectorization transformation on the collected raw parameters to construct a multi-dimensional process behavior feature matrix; S13. Remove duplicate and invalid feature vectors through the data cleaning module to generate a standardized process behavior feature dataset.

3. The business process configuration method based on Activiti and AI decision-making according to claim 1, characterized in that, S2 includes the following steps: S21. Divide the process behavior feature dataset into a training set and a validation set; S22. Use the gradient boosting decision tree algorithm for iterative training on the training set; S23. Verify the model accuracy using the validation set. If the accuracy falls below a threshold, adjust the feature weights and retrain. ； in, This is a dynamic accuracy threshold. This represents the historical average accuracy. The standard deviation of historical accuracy; Threshold explanation: Retraining is triggered when the current model accuracy is lower than the historical average accuracy minus 2 standard deviations; S24. Deploy the validated decision tree model to the AI ​​service gateway.

4. The business process configuration method based on Activiti and AI decision-making according to claim 1, characterized in that, S3 includes the following steps: S31. Parse the exclusive gateway node in the Activiti process definition XML file; S32. Add an AI decision routing label to the specified exclusive gateway node in the visual configuration interface; S33. Persistently store the tag binding information in the process decision configuration database.

5. A business process configuration method based on Activiti and AI decision-making according to claim 1, characterized in that, S4 includes the following steps: S41, Listen to the event bus of the process engine and capture node arrival events containing AI decision routing labels; S42. Extract the context feature parameters of the current process instance; S43. Call the AI ​​service gateway interface to input feature parameters and obtain the route direction prediction value; S44. Activate the corresponding process branch based on the predicted value.

6. The business process configuration method based on Activiti and AI decision-making according to claim 1, characterized in that, S5 includes the following steps: S51. Collect the execution result data and time consumption parameters of the process branch; S52. Compare the execution results with the decision prediction values ​​to generate a confidence index: ； in, To normalize the confidence level, For the actual result vector, For the predicted result vector, Number of decision dimensions; Threshold explanation: Time-marked as low-confidence sample; S53. Label decision data with confidence levels below the threshold as reinforcement learning samples.

7. A business process configuration method based on Activiti and AI decision-making according to claim 1, characterized in that, S7 includes the following steps: S71. Display the node topology of the process definition diagram in a visual interface; S72. Bind the AI ​​decision routing label to the newly added decision node by dragging and dropping; S73. Generate a version difference comparison chart and record configuration change logs; S74. Perform syntax validation on the revised process definition file.

8. A business process configuration method based on Activiti and AI decision-making according to claim 1, characterized in that, S8 includes the following steps: S81. Capture execution timeout and routing error exceptions; S82. Extract the input features and decision tree path of the abnormal nodes; S83. Determine defective nodes through the decision path inversion localization model; S84, Suggestions for adjusting output feature weights and supplementing samples.

9. A business process configuration method based on Activiti and AI decision-making according to claim 1, characterized in that, S9 includes the following steps: S91. Create an isolated test environment for the new process decision configuration scheme; S92. Distribute production environment process instances to the test environment according to a preset ratio; S93. Monitor the fluctuation of decision accuracy in the test environment; S94. Automatically trigger version rollback when accuracy drops below the tolerance threshold.

10. A business process configuration method based on Activiti and AI decision-making according to claim 1, characterized in that, S10 includes the following steps: S101, AI decision-making path data that aggregates historical process instances; S102. Generate a heatmap of decision path distribution and a branch selection probability matrix; S103, Mark the root cause localization markers for high-frequency abnormal decision-making paths; S104. Outputs a dual-mode report in PDF and interactive visualization.

Images