Big data visual analysis display method

Abstract

The application relates to the technical field of big data analysis and visualization, in particular to a big data visualization analysis display method, which comprises the following steps: constructing and maintaining a unified knowledge graph for fusing and semantically representing multi-source heterogeneous data; based on the knowledge graph, a predefined visualization description language is used to generate a visualization configuration specification defining a plurality of views and linkage rules between the views; in response to a user's interactive operation on any view, other views affected according to the linkage rules are determined, and information is distributed to related views based on triggered data state changes to realize synchronous rendering update; the user's interactive history is collected and analyzed to construct and update a user portrait reflecting the user's visualization preferences; during the interactive process, based on the user portrait, a recommendation model is used to adjust the linkage rules or the display form of the views in real time. The application realizes intelligent linkage and personalized adaptive optimization of multiple views, and improves the efficiency and experience of the visualization analysis.

Description

Technical Field

[0001] This invention relates to the field of big data analysis and visualization technology, and in particular to a method for big data visualization analysis and display. Background Technology

[0002] As enterprises deepen their digital transformation, the data generated by business systems exhibits typical characteristics of massive scale, heterogeneous sources, and diverse structures. To gain insights from this data, big data visualization and analysis technologies have become key tools. Traditional solutions typically integrate data into a data warehouse through ETL processes and utilize business intelligence tools to build static reports or interactive dashboards to achieve graphical presentation of the data. Existing technologies can already achieve multi-view linkage based on predefined rules; for example, performing a filtering operation on one view will refresh the data in related views accordingly.

[0003] However, existing technologies still have significant limitations in practical applications, especially when facing complex analysis scenarios. The system cannot understand the user's analytical intent and preferences, leading to rigid interaction between multiple views and a lack of personalized analysis experience. Specifically, the interaction rules between views are statically predefined by developers during the design phase. Once the business analysis logic changes or the user's exploration path deviates from the preset scenario, the system cannot intelligently adjust the relationships between views or the view's own presentation to adapt to the current analysis context. This makes the analysis process highly dependent on repeated manual configuration, limiting the efficiency and depth of exploratory data analysis, and making it difficult for users to obtain a truly intelligent, guided visual analysis experience. Summary of the Invention

[0004] To address the above shortcomings, this invention provides a big data visualization and analysis method, which aims to improve the problems in traditional big data visualization and analysis systems, such as rigid interaction between multiple views and the inability to adaptively adjust in real time according to user preferences and the current analysis context, resulting in low efficiency and a lack of personalized experience in exploratory data analysis.

[0005] In a first aspect, the present invention provides the following technical solution: a big data visualization analysis and display method, comprising the following steps:

[0006] S1. Construct and maintain a unified knowledge graph for fusing and semantically representing multi-source heterogeneous data to be analyzed;

[0007] S2. Based on the knowledge graph, a visualization configuration specification is generated using a predefined visualization description language. The visualization configuration specification defines at least multiple views and the linkage rules between views based on data state.

[0008] S3. In response to the user's interaction with any of the multiple views, determine at least one other view affected by the interaction according to the linkage rules, and distribute information containing the data state change to the view and the at least one other view based on the data state change triggered by the interaction, so that they can perform synchronous rendering updates.

[0009] S4. Collect and analyze the user's interaction history and corresponding view update results in step S3, and build and update the user profile that reflects the user's visual preferences based on the analysis results.

[0010] S5. During the interaction process in step S3, based on the user profile, the linkage rules or the display format of the multiple views are adjusted in real time through the recommendation model.

[0011] Preferably, in step S1, the step of constructing and maintaining a unified knowledge graph for fusing and semantically representing multi-source heterogeneous data to be analyzed specifically includes:

[0012] Data to be analyzed is extracted from multiple heterogeneous data sources, and the extracted data to be analyzed is preprocessed.

[0013] Based on the preprocessed data to be analyzed, data entities, entity attributes, and relationships between entities are identified and labeled to form initial semantic data units.

[0014] The semantic data units are imported into a graph database, and a unified knowledge graph with entities as nodes and relationships as edges is constructed based on a preset graph ontology model.

[0015] In response to updates to the source data, incremental updates and consistency maintenance are performed on the corresponding nodes, attributes, and relationships in the unified knowledge graph.

[0016] Preferably, the step of incrementally updating and maintaining consistency of the corresponding nodes, attributes, and relationships in the unified knowledge graph in response to updates to the source data specifically includes:

[0017] Monitor the multiple heterogeneous data sources, identify and capture the updated source data and its changes;

[0018] The updated source data is cleaned, transformed, and its features are extracted. Based on the extraction results, entities, attributes, and relationships are identified and labeled to obtain incremental semantic data units.

[0019] The incremental semantic data unit is compared and matched with the existing nodes, attributes and relationships in the unified knowledge graph;

[0020] Based on the comparison and matching results, the unified knowledge graph is updated accordingly to ensure that the unified knowledge graph is consistent with the latest data state reflected by the incremental semantic data unit.

[0021] After the update operation is performed, the unified knowledge graph is logically verified and conflict resolved according to the preset consistency rules.

[0022] Preferably, the specific steps for updating the unified knowledge graph based on the comparison and matching results include:

[0023] Analyzing the comparison and matching results, the difference types of the incremental semantic data units relative to the unified knowledge graph are identified, and the difference types include addition, modification, or invalidation;

[0024] If the identified difference type is new, then a new node, attribute, or relationship corresponding to the incremental semantic data unit is created in the unified knowledge graph;

[0025] If the identified difference type is a change, then the value of the existing node or attribute corresponding to the incremental semantic data unit is searched and modified in the unified knowledge graph;

[0026] If the identified difference type is failure, then the failure node, attribute, or relationship corresponding to the incremental semantic data unit is searched and deleted in the unified knowledge graph.

[0027] Preferably, in step S2, generating a visual configuration specification based on the knowledge graph using a predefined visual description language specifically includes:

[0028] Based on the analysis requirements, the target entities, attributes, and relationships to be analyzed are queried and determined from the knowledge graph, serving as the data foundation for visualization analysis;

[0029] Based on the data foundation, using the visualization description language, multiple views are defined to display the data foundation from different dimensions, and the corresponding data mapping rules and visual encoding methods are specified for each view;

[0030] Analyze the data dimensions corresponding to the multiple views and the logic of data state changes in the data foundation, and use the visualization description language to define the linkage rules between views based on data state changes.

[0031] Preferably, in S3, the step of responding to a user's interaction with any of the plurality of views, determining at least one other view affected by the interaction according to the linkage rules, and distributing information containing the data state change to the view and the at least one other view based on the data state change triggered by the interaction specifically includes:

[0032] Capture user interaction events on any of the views and parse the intent indicated by the interaction events;

[0033] Based on the parsed intent, calculate and determine the data state changes triggered by the interactive operation;

[0034] Based on the linkage rules in the visualization configuration specification, query at least one other view associated with the data dimension where the data state change has occurred;

[0035] The data state change is encapsulated into a unified message format, and the message is simultaneously distributed to any of the views that initiated the interaction and to at least one other view that was queried.

[0036] Upon receiving the message, each of the views and at least one other view re-queries the data and performs a rendering update based on the data state changes contained therein.

[0037] Preferably, the step of querying at least one other view associated with the data dimension where the data state change has occurred, according to the linkage rules in the visualization configuration specification, specifically includes:

[0038] Extract one or more data dimensions that have changed from the data state changes;

[0039] Using one or more data dimensions that have changed as query keys, a matching search is performed in the linkage rule base of the visualization configuration specification;

[0040] Retrieve one or more associated view identifiers bound in the matched linkage rules;

[0041] Based on the associated view identifier, at least one other view is determined from the plurality of views.

[0042] Preferably, in step S4, the step of collecting and analyzing the user's interaction history and corresponding view update results in step S3, and constructing and updating a user profile reflecting the user's visual preferences based on the analysis results, specifically includes:

[0043] The user interaction log is formed by continuously recording the interactive operation events that occur in step S3, the source view that triggers the interactive operation, and the view update results caused by the interactive operation.

[0044] Perform pattern analysis on the user interaction logs to extract user interaction behavior characteristics;

[0045] The extracted interactive behavior features are quantized into feature vectors, and visual preference tags for the corresponding users are created or updated based on the feature vectors.

[0046] Integrate all the user's visual preference tags to generate or update the user profile that reflects the user's visual preferences.

[0047] Preferably, in step S5, the step of adjusting the linkage rules or the display format of the multiple views in real time based on the user profile using a recommendation model specifically includes:

[0048] During the interaction in step S3, based on the visual preference tags reflected in the user profile, the user's potential interaction intentions or focus of attention in the current analysis context are predicted.

[0049] The predicted potential interaction intentions or focus of attention, as well as the features of the subset of data in the knowledge graph related to the current view, are input into the recommendation model;

[0050] Obtain adjustment suggestions from the recommendation model output, and modify the currently effective linkage rules according to the adjustment suggestions, or adjust the visual encoding, layout, or data filtering conditions of the corresponding views in the multiple views in real time.

[0051] Preferably, in the interaction process of step S3, the step of predicting the user's potential interaction intentions or focus of attention in the current analysis context based on the visual preference tags reflected in the user profile specifically includes:

[0052] From the user profile, read the set of visual preference tags corresponding to the current user;

[0053] Obtain the current analysis context information;

[0054] By performing correlation analysis between the set of visual preference tags and the current analysis context information, the user's next possible interaction or the data subset currently being focused on in the current analysis context can be inferred.

[0055] The next possible interactive action or data subset inferred is determined as the potential interactive intent or focus of attention.

[0056] The present invention has the following beneficial effects:

[0057] 1. In this invention, a unified knowledge graph is constructed to fuse and semantically represent structured and unstructured data from heterogeneous data sources such as databases, text, images, and time-series signals. This breaks through the limitations of traditional visualization solutions, which rely on single data models and make it difficult for unstructured data to be directly used in analysis. It provides a unified, rich, and semantically rich data foundation for panoramic data analysis in complex business scenarios.

[0058] 2. In this invention, configuration specifications are generated using a predefined visual description language, decoupling core logic such as view definition, data mapping, and linkage rules from hard coding. This not only significantly reduces the development and maintenance costs of multi-view, highly interactive visual dashboards, but also enables business analysts to quickly build and iterate analysis scenarios through configuration rather than programming, improving the agility of data analysis.

[0059] 3. In this invention, based on the linkage rules defined in the configuration specification, the system can automatically perceive the user's interaction intent on a single view and accurately calculate the affected related views. A unified status message mechanism enables millisecond-level synchronous rendering of multiple views. This overcomes the problems of complex logic, error-proneness, and slow response when manually associating views in the traditional way, ensuring a consistent and smooth experience in cross-view analysis. Attached Figure Description

[0060] Figure 1 This is a flowchart illustrating the big data visualization and analysis method proposed in this invention. Detailed Implementation

[0061] The technical solutions in 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.

[0062] This invention provides a method for big data visualization and analysis, such as... Figure 1 As shown, it includes the following steps:

[0063] S1. Construct and maintain a unified knowledge graph for fusing and semantically representing multi-source heterogeneous data to be analyzed.

[0064] Furthermore, in S1, the steps for constructing and maintaining a unified knowledge graph for fusing and semantically representing multi-source heterogeneous data to be analyzed specifically include:

[0065] Data to be analyzed is extracted from multiple heterogeneous data sources, and the extracted data is preprocessed.

[0066] Based on the preprocessed data to be analyzed, data entities, entity attributes, and relationships between entities are identified and labeled to form initial semantic data units;

[0067] Import semantic data units into a graph database, and construct a unified knowledge graph with entities as nodes and relationships as edges based on a pre-defined graph ontology model.

[0068] In response to updates to the source data, incremental updates and consistency maintenance are performed on the corresponding nodes, attributes, and relationships in the unified knowledge graph.

[0069] Furthermore, in response to updates to the source data, the specific steps for incremental updates and consistency maintenance of the corresponding nodes, attributes, and relationships in the unified knowledge graph include:

[0070] Monitor multiple heterogeneous data sources, identify and capture updated source data and its changes;

[0071] The source data that has been updated is cleaned, transformed and feature extracted, and entities, attributes and relationships are identified and labeled based on the extraction results, thereby obtaining incremental semantic data units;

[0072] The incremental semantic data units are compared and matched with the existing nodes, attributes and relationships in the unified knowledge graph.

[0073] Based on the comparison and matching results, the unified knowledge graph is updated accordingly to ensure that the unified knowledge graph is consistent with the latest data state reflected by the incremental semantic data units.

[0074] After the update operation is performed, the unified knowledge graph is logically verified and conflict resolved according to the preset consistency rules.

[0075] Furthermore, based on the comparison and matching results, the specific steps for updating the unified knowledge graph include:

[0076] The analysis and matching results identify the types of differences between incremental semantic data units and the unified knowledge graph. These differences include additions, changes, or invalidations.

[0077] If the identified difference type is new, then a new node, attribute, or relationship corresponding to the incremental semantic data unit is created in the unified knowledge graph;

[0078] If the identified difference type is a change, then search for and modify the value of the existing node or attribute corresponding to the incremental semantic data unit in the unified knowledge graph;

[0079] If the identified difference type is failure, then the failure node, attribute, or relationship corresponding to the incremental semantic data unit is searched and deleted in the unified knowledge graph.

[0080] Specifically, heterogeneous data sources include, but are not limited to, relational databases, application programming interfaces (APIs), text log files, image repositories, and IoT time-series data streams. By adapting dedicated connectors or universal interface protocols to each data source, data extraction tasks are performed periodically or in real-time. The data to be analyzed is read into a unified buffer, and the extracted data is preprocessed. For structured data, data cleaning and transformation are performed, specifically including: filling in missing values, correcting outliers, standardizing date and numerical formats, and performing data normalization. For unstructured text data, feature extraction and entity recognition are performed: first, word segmentation is performed using a word segmentation tool; then, a pre-trained natural language processing model, such as the BERT model, is used to generate semantic vector representations of the text, and named entity recognition technology is used to identify entities such as names of people, places, organizations, and proper nouns in the text. For image data, a convolutional neural network model is used to extract visual feature vectors, and an object detection model is used to identify specific objects in the image. For time-series data, noise reduction and resampling are performed, and time-domain and frequency-domain features, such as mean, variance, and major frequency components, are extracted.

[0081] Based on the preprocessed data, data entities, entity attributes, and relationships between entities are identified and labeled. This step is performed according to a predefined domain ontology model. First, entities identified in the preprocessing stage, such as "Product A," "Customer B," and "Q4 2023," are mapped to classes defined in the ontology model, such as "Product," "Customer," and "Quarter," as entity nodes. Second, numerical or textual information describing these entities is extracted from the preprocessed data and labeled as entity attributes, such as "sales revenue" for products and "region" for customers. Finally, the context and logic in the data are analyzed to identify relationships between entities, such as customers "purchasing" products and products "belonging to" a certain category, and these relationships are labeled according to the types defined in the ontology model. This forms structured semantic data units, each containing an entity identifier, attribute key-value pairs, and associated relation triples.

[0082] Graph databases such as Neo4j or JanusGraph are selected as storage and query engines. Based on the patterns defined by the predefined graph ontology model, entities in semantic data units are treated as nodes and relationships as edges, and are imported into the graph database in batches. Each node stores its type and attributes, and each edge stores its relationship type and attributes, thereby constructing a unified knowledge graph with entities as nodes and relationships as edges, capable of semantic querying.

[0083] By monitoring database logs, file change events, or streaming data in message queues, the source data experiencing updates is identified in real time, and its changes and types are captured. For the captured incremental update data, the same preprocessing, entity identification, and attribute and relation annotation processes as described above are performed to obtain incremental semantic data units reflecting the latest changes. The entities and relations in these incremental semantic data units are then matched against existing nodes and edges in the graph database for similarity. The matching process is performed by calculating vector similarity or based on unique identifiers. For entities, the cosine similarity of their attribute vectors is calculated. The cosine similarity calculation formula is:

[0084] ;

[0085] Here, A and B represent the attribute feature vectors of the entity to be matched and existing entities in the graph, respectively. If the similarity exceeds a preset threshold and the unique identifiers are the same, they are determined to be the same entity; if the identifier does not exist and the semantic features are new, it is determined to be a newly added entity. Relationship matching follows a similar logic;

[0086] Based on the comparison and matching results, corresponding graph database operations are performed. If a new node or edge is identified, a creation operation is performed to insert it into the graph. If a node or edge is identified as a change, an update operation is performed to modify the attribute values ​​of existing nodes or edges. If a node or edge is identified as invalid, a deletion operation is performed to remove it. This process ensures that the knowledge graph remains consistent with the latest data state reflected by the incremental semantic data units.

[0087] After performing the update operation, a consistency check is initiated. This check is performed according to preset consistency rules, such as verifying the validity of attribute values, the symmetry of relationships, or transitivity constraints. If a conflict violating the rules is found, it is automatically repaired or an alarm is logged according to a preset resolution strategy. For example, when two "belong to" relationships are detected to form a cycle, the earlier or lower-confidence relationship is marked and removed based on timestamps or confidence rules to maintain the logical consistency of the graph.

[0088] Through the above process, a unified knowledge graph that can integrate multi-source heterogeneous data and support semantic queries and real-time incremental updates was built and maintained, providing a high-quality and timely data foundation for subsequent visualization analysis and intelligent recommendation.

[0089] S2. Based on the knowledge graph, a visual configuration specification is generated using a predefined visual description language. The visual configuration specification defines at least multiple views and the linkage rules between views based on data state.

[0090] Furthermore, in S2, based on the knowledge graph, a visual configuration specification is generated using a predefined visual description language, specifically including:

[0091] Based on the analysis requirements, the target entities, attributes, and relationships to be analyzed are queried and determined from the knowledge graph, serving as the data foundation for visualization analysis;

[0092] Based on the data foundation, multiple views are defined using a visual description language to display the data foundation from different dimensions, and the corresponding data mapping rules and visual encoding methods are specified for each view.

[0093] Analyze the data dimensions and data state change logic corresponding to multiple views, and define the linkage rules between views based on data state changes using a visual description language.

[0094] Specifically, users or the system input or select analytical intent through the front-end interface, such as analyzing product sales in a specific region within a specific time frame. The system parses this intent into a structured query on the knowledge graph. The query is executed using a graph query language, which retrieves target entities, attributes, and relationships from the knowledge graph that match the analytical intent by matching specific types of nodes and relationships and constraining node attribute conditions. The query results constitute a structured dataset, serving as the data foundation for subsequent visualization analysis.

[0095] The system uses a predefined formatted visualization description language to define views. First, based on the analytical dimensions included in the data foundation, the type and number of views to be created are determined. For example, to show the trend of sales revenue over time, a line chart view is defined; to compare the sales revenue of different products, a bar chart view is defined. For each view, its data mapping rules and visual encoding method must be defined in the visualization configuration specification. The data mapping rules specify how the view is bound to specific fields in the data foundation. For the line chart view, the mapping rule maps the time dimension field to the horizontal axis of the chart and the sales revenue measure field to the vertical axis of the chart. The visual encoding method specifies the graphical representation attributes of the data, including the type, color, shape, and size of the graphic markers. For the line chart above, the visual encoding method specifies the use of continuous line segments to connect data points and specifies the color coding scheme for the line segments and the display shape of the data points.

[0096] The system analyzes the core data dimensions relied upon in each view definition. For example, the core dimensions of the line chart view are time and sales revenue, while the core dimensions of the bar chart view are product and sales revenue. Based on the logical relationships between data dimensions, linkage rules between views are defined. The core of the linkage rule is to declare how to pass on a change in the data state of one view due to user interaction to other views. A change in data state mainly refers to a change in the data filtering range, such as selecting a specific time range or data item in a view. A complete linkage rule definition includes four elements: the triggering view, the triggering event, the list of affected views, and the state propagation logic. The state propagation logic specifically describes which data dimension of the triggering view has changed, and how this change should affect which data dimension of the affected views. For example, a rule can be defined: when a user selects a product in the bar chart view, the selected product identifier should be used as a filter condition and automatically passed to the line chart view, so that the line chart only displays the sales revenue trend data of the selected product. This rule indicates that the product dimension is a shared dimension and linkage link between the two views. All view definitions and linkage rules together constitute a complete visualization configuration specification. The specification is stored in the form of a structured text file, which is then loaded, parsed, and executed by the subsequent visualization rendering engine.

[0097] S3. In response to a user's interaction with any of the multiple views, determine at least one other view affected by the interaction according to the linkage rules, and distribute information containing the data state change to any view and at least one other view based on the data state change triggered by the interaction, so that they can perform synchronous rendering updates.

[0098] Furthermore, in S3, in response to a user's interaction with any of the multiple views, the steps of determining at least one other view affected by the interaction based on the linkage rules, and distributing information containing the data state change to any view and at least one other view based on the data state change triggered by the interaction specifically include:

[0099] Capture user interaction events on any view and parse the intent indicated by the interaction events;

[0100] Based on the parsed intent, calculate and determine the data state changes triggered by the interactive operation;

[0101] Based on the linkage rules in the visualization configuration specifications, query at least one other view associated with the data dimension whose data state has changed;

[0102] Encapsulate data state changes into a unified message format and simultaneously distribute the message to any view that initiated the interactive operation and at least one other queried view;

[0103] Upon receiving the message, any view and at least one other view will re-query the data and perform a rendering update based on the changes in the data state contained therein.

[0104] Furthermore, according to the linkage rules in the visualization configuration specification, the steps of querying at least one other view associated with the data dimension whose data state has changed specifically include:

[0105] Extract one or more data dimensions that have changed from the changes in data state;

[0106] Use one or more data dimensions that have changed as the query key to perform a matching search in the linked rule library of the visual configuration specification;

[0107] Retrieve one or more associated view identifiers bound in the matched linkage rules;

[0108] Based on the associated view identifier, identify at least one other view from a plurality of views.

[0109] Specifically, when a user interacts with any view component in the visual interface, the browser or client application triggers a corresponding event. The system captures this raw interaction event through event listeners. The captured event object contains the event type, the view identifier that triggered the event, and information about the data point or graphical region related to the event. The system parses the captured raw event to determine the user's intention. The parsing process is based on the event type. For example, a mouse selection operation is parsed as the user's intention to filter a range of data dimensions; a click on a data point is parsed as the user's intention to focus on or highlight the data item represented by that data point; a double-click operation may be parsed as the user's intention to drill down to a more detailed data level. The parsing result is output as a structured intent description object, which at least contains information about the intent type and the target data dimension.

[0110] Based on the parsed user intent, the system calculates the specific data state change corresponding to that intent. The data state change describes how the user's action alters the filtering, focus, or detail status of the data. The calculation process follows a preset mapping logic. If the user intent is range filtering, the data state change adds a set of filtering conditions to that data dimension, with values ​​representing the range selected by the user. If the user intent is data item focusing, the data state change sets a focus identifier on that data dimension, with values ​​representing the dimension value corresponding to the clicked data item. If the user intent is drill-down, the data state change switches to a finer-grained level of data detail. The calculated data state change is an object containing the following elements: the name of the data dimension that changed, the type of state change, and the specific value of the change.

[0111] Based on the aforementioned determined data state changes, the system queries the visualization configuration specifications to determine which other views will be affected by these changes. First, it extracts the names of one or more data dimensions that have changed from the data state change object. Then, using the extracted data dimension names as the query key, it performs a matching search in the predefined linkage rule library of the visualization configuration specifications. The linkage rule library is a data structure that stores multiple rules, each recording the associated view identifiers affected when a specific data dimension changes. The matching search finds rules whose triggering conditions include the currently changed data dimension. Then, from all matched linkage rules, it retrieves one or more associated view identifiers bound to the rules. Finally, based on the retrieved associated view identifiers, it finds the corresponding view component objects in the multiple view sets currently loaded by the system. These views are at least one other view affected by the current interaction operation.

[0112] The system encapsulates the calculated data state change objects into a message in a unified format. This message includes at least the message type, the data state change content, and a timestamp. After encapsulation, the system distributes the message simultaneously to two destinations via a message bus or event publish / subscribe mechanism: the original view that initiated this interaction, and all other related views queried in the third step. This distribution is concurrent to ensure that all relevant views receive the state update notification almost simultaneously.

[0113] Upon receiving the distributed data state change message, the original view and all related views initiate their respective update processes. First, each view parses the message and extracts the data state change content. Then, each view, based on its own data query definition and the newly received data state change, generates a new data query request with the latest filtering or focus conditions. This query request is sent to the backend data service or directly queries the local knowledge graph. After obtaining the new query result dataset, each view re-executes its graphics rendering logic according to its visual encoding method, drawing the new data onto the canvas. Finally, the interface content of all related views is synchronously updated based on the same data state change, providing users with a consistent analytical context.

[0114] S4. Collect and analyze the user's interaction history and corresponding view update results in step S3, and build and update the user profile that reflects the user's visualization preferences based on the analysis results.

[0115] Furthermore, in S4, the steps of collecting and analyzing the user's interaction history and corresponding view update results in step S3, and constructing and updating a user profile reflecting the user's visual preferences based on the analysis results, specifically include:

[0116] Continuously record the interactive events that occur in step S3, the source view that triggers the interactive operation, and the view update results caused by the interactive operation to form a user interaction log;

[0117] Perform pattern analysis on user interaction logs to extract user interaction behavior characteristics;

[0118] The extracted interactive behavior features are quantified into feature vectors, and corresponding user visual preference labels are created or updated based on the feature vectors;

[0119] Integrate all of the user's visual preference tags to generate or update user profiles that reflect the user's visual preferences.

[0120] Specifically, during the execution of step S3, the system synchronously initiates a logging mechanism. Whenever an interaction is completed and triggers a view update, the system records the complete context information of this interaction in the user interaction log. Each log entry contains the following fields: a unique user identifier, a timestamp of the interaction, the source view identifier that triggered the interaction, the specific type of the interaction, the target data dimension and value affected by the interaction, and a summary of the view update result ultimately caused by the interaction. The view update result summary includes a list of affected view identifiers and a summary of statistics for the updated dataset. These log entries are sequentially appended to persistent storage, forming a time-ordered sequence of user interaction logs.

[0121] The system performs batch analysis on user interaction logs periodically or after accumulating a certain amount of data. The goal of pattern analysis is to identify habitual and meaningful user operation patterns from discrete log records. The analysis process first segments the logs according to sessions, with each session representing a single, continuous analysis task for the user. Within a single session, frequent patterns in the operation sequence are analyzed. For example, the most frequently used interaction types are statistically analyzed, and their frequency distribution is calculated. The frequency and duration of user visits to different view types are analyzed. User-preferred analysis paths are identified, i.e., starting with a specific operation in one view and followed by a frequently occurring specific operation in another view, forming an operation transition probability matrix. The elements in the matrix represent the probability of transitioning from operation type i to operation type j, calculated using the following formula:

[0122] ;

[0123] in, This represents the number of times operation type i is followed immediately by operation type j in the log, with the denominator being the total number of times operation type i appears. (Probability value) The higher the value, the more likely the transfer path is to be the user's preferred path;

[0124] Furthermore, the analysis examines data dimensions that users frequently focus on or filter. For example, users often perform range filtering on the "time" dimension or single-selection focusing on the "product category" dimension. Based on the above analysis, a set of interactive behavior features representing user preferences is extracted. These features include, but are not limited to: high-frequency operation types, commonly used view types, habitual analysis paths, and key data dimensions.

[0125] The extracted interactive behavior features are quantified and vectorized. For high-frequency operation type features, they are transformed into a multi-dimensional vector, with each dimension corresponding to a possible operation type, and its value being the normalized frequency of that type of operation in the logs. For commonly used view type features, a similar method is used to generate a view type preference vector. For habitual analysis path features, the transition probability matrix P is flattened into a vector. For key data dimension features, a data dimension attention vector is generated. Then, all the above vectors are concatenated into a comprehensive user interaction feature vector. Based on this comprehensive feature vector, visual preference labels are created or updated for users using preset rules or clustering algorithms. For example, a rule can be defined: if the weight of a user's "drill-down" operation in the high-frequency operation type vector exceeds a threshold... If a user's data dimension attention vector shows the highest weight for the "time" dimension, then that user is labeled a "deep time trend analyst." Labels are stored as key-value pairs, where the key is the label name and the value is the confidence score.

[0126] The system maintains a user profile database, storing a user profile document for each user. The core content of each user profile document consists of all the user's visual preference tags and their confidence levels. When new preference tags are generated or the confidence levels of existing tags are updated, the system integrates this information into the corresponding user's profile document. Integration operations include adding new tags, updating the confidence levels of existing tags, or reducing the weight of old tags based on their timeliness. The final generated or updated user profile is a structured document that comprehensively reflects the user's visual analysis preferences at the current stage, providing a basis for subsequent real-time adaptive optimization.

[0127] S5. During the interaction process in step S3, based on the user profile, the linkage rules or the display format of multiple views are adjusted in real time through the recommendation model.

[0128] Furthermore, in S5, the specific steps for adjusting the linkage rules or the display format of multiple views in real time based on user profiles and recommendation models include:

[0129] During the interaction in step S3, based on the visual preference tags reflected in the user profile, the user's potential interaction intentions or focus of attention in the current analysis context are predicted.

[0130] The predicted potential interaction intentions or focus of attention, as well as the features of the subset of data in the knowledge graph that are related to the current view, are input into the recommendation model;

[0131] Obtain adjustment suggestions from the recommendation model output, modify the currently effective linkage rules based on the adjustment suggestions, or adjust the visual encoding, layout, or data filtering conditions of corresponding views in multiple views in real time.

[0132] Furthermore, in the interaction process of step S3, the step of predicting the user's potential interaction intentions or focus of attention in the current analysis context based on the visual preference tags reflected in the user profile specifically includes:

[0133] From the user profile, retrieve the set of visual preference tags corresponding to the current user;

[0134] Obtain the current analysis context information;

[0135] By correlating the set of visual preference labels with the current analysis context, we can infer the user's next possible interaction or the data subset currently being focused on in the current analysis context.

[0136] The next possible interactive action or data subset that is inferred is identified as the potential interactive intent or focus of attention.

[0137] Specifically, during the interaction in step S3, the system performs intent prediction in parallel. First, based on the unique identifier of the currently logged-in user, the system reads the user's set of visual preference tags from the user profile database. This set contains multiple tags, each with a confidence score. Simultaneously, the system obtains the current analysis context information. This information includes: the identifiers of all active views on the interface for the current user, the data dimensions currently loaded and displayed in these views, and the sequence of historical interaction operations performed by the user in this session. Subsequently, the system performs correlation analysis between the set of visual preference tags and the current analysis context information. The analysis process is based on predefined inference rules. For example, one rule is: if the user's preference tag contains "focus on outliers," and the data displayed in a view in the current context has statistical outliers on a specific dimension, then it is inferred that the user's potential focus is on this subset of outlier data. Another rule is: if the user's preference tag is "habit comparison analysis," and the user has already selected product A in a bar chart view in the current context, then it is inferred that the user's next possible interaction operation is to select product B in the same view for comparison. The inference results output by association analysis take two forms: one is a prediction of the user's next possible interaction, and the other is a prediction of the subset of data the user is most likely to be currently interested in. These inference results are identified as the user's potential interaction intentions or focus of attention.

[0138] The system takes the potential interaction intent or focus predicted in the first step as its first input. Simultaneously, based on the currently active views, the system queries the knowledge graph to retrieve a subset of data related to these views. Features are extracted from this subset to form the second input. These extracted features include statistical characteristics such as the mean, variance, and skewness of the subset; and semantic features such as the distribution of major entity types and the density of relationships. These two inputs—the predicted intent or focus and the feature vector of the data subset—are then fed into a pre-trained recommendation model. This recommendation model can be a machine learning-based classification or regression model. After receiving the input, the model outputs an adjustment suggestion. This suggestion specifically indicates which part of the system should be adjusted in what way. The suggestions are mainly divided into two categories: the first category is adjustment suggestions for linkage rules, such as suggesting dynamically enabling or disabling a specific view linkage rule, or modifying the threshold of a filtering condition in a rule. The second category is suggestions for adjusting the presentation of a view. For example, suggestions may be made to adjust the visual encoding method of a view, such as changing the color mapping scheme from "continuous" to "categorical"; or to adjust the layout position of the view; or to automatically apply a specific data filtering condition to a view.

[0139] The system executes the adjustment suggestions output by the recommendation model. If the suggestion is to adjust the linkage rules, the system accesses the linkage rule library stored in the visualization configuration specification, finds the target rule, and modifies its effective status or parameters according to the suggestion. The modification takes effect immediately, and subsequent user interactions will be based on the new rule. If the suggestion is to adjust the view display format, the system finds the target view component and calls its application programming interface (API), dynamically modifying its visual encoding, layout position in the dashboard, or internal data filtering conditions according to the suggestion. After the modification is completed, the view is immediately re-rendered to reflect the new display format. The entire process is completed in real time during user interaction intervals and is transparent to the user. Its effect is to enable the system's interactive behavior and view display to automatically adapt to the user's personalized preferences and the current analysis context, achieving intelligent analysis assistance.

[0140] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A big data visualization and analysis method, characterized in that, Includes the following steps: S1. Construct and maintain a unified knowledge graph for fusing and semantically representing multi-source heterogeneous data to be analyzed; S2. Based on the knowledge graph, a visualization configuration specification is generated using a predefined visualization description language. The visualization configuration specification defines at least multiple views and the linkage rules between views based on data state. S3. In response to the user's interaction with any of the multiple views, determine at least one other view affected by the interaction according to the linkage rules, and distribute information containing the data state change to the view and the at least one other view based on the data state change triggered by the interaction, so that they can perform synchronous rendering updates. S4. Collect and analyze the user's interaction history and corresponding view update results in step S3, and build and update the user profile that reflects the user's visual preferences based on the analysis results. S5. During the interaction process in step S3, based on the user profile, the linkage rules or the display format of the multiple views are adjusted in real time through the recommendation model.

2. The big data visualization analysis and display method according to claim 1, characterized in that, In S1, the steps of constructing and maintaining a unified knowledge graph for fusing and semantically representing multi-source heterogeneous data to be analyzed specifically include: Data to be analyzed is extracted from multiple heterogeneous data sources, and the extracted data to be analyzed is preprocessed. Based on the preprocessed data to be analyzed, data entities, entity attributes, and relationships between entities are identified and labeled to form initial semantic data units. The semantic data units are imported into a graph database, and a unified knowledge graph with entities as nodes and relationships as edges is constructed based on a preset graph ontology model. In response to updates to the source data, incremental updates and consistency maintenance are performed on the corresponding nodes, attributes, and relationships in the unified knowledge graph.

3. The big data visualization analysis and display method according to claim 2, characterized in that, The steps of incrementally updating and maintaining consistency of the corresponding nodes, attributes, and relationships in the unified knowledge graph in response to updates to the source data specifically include: Monitor the multiple heterogeneous data sources, identify and capture the updated source data and its changes; The updated source data is cleaned, transformed, and its features are extracted. Based on the extraction results, entities, attributes, and relationships are identified and labeled to obtain incremental semantic data units. The incremental semantic data unit is compared and matched with the existing nodes, attributes and relationships in the unified knowledge graph; Based on the comparison and matching results, the unified knowledge graph is updated accordingly to ensure that the unified knowledge graph is consistent with the latest data state reflected by the incremental semantic data unit. After the update operation is performed, the unified knowledge graph is logically verified and conflict resolved according to the preset consistency rules.

4. The big data visualization analysis and display method according to claim 3, characterized in that, The specific steps for updating the unified knowledge graph based on the comparison and matching results include: Analyzing the comparison and matching results, the difference types of the incremental semantic data units relative to the unified knowledge graph are identified, and the difference types include addition, modification, or invalidation; If the identified difference type is new, then a new node, attribute, or relationship corresponding to the incremental semantic data unit is created in the unified knowledge graph; If the identified difference type is a change, then the value of the existing node or attribute corresponding to the incremental semantic data unit is searched and modified in the unified knowledge graph; If the identified difference type is failure, then the failure node, attribute, or relationship corresponding to the incremental semantic data unit is searched and deleted in the unified knowledge graph.

5. The big data visualization analysis and display method according to claim 1, characterized in that, In step S2, generating a visual configuration specification based on the knowledge graph using a predefined visual description language specifically includes: Based on the analysis requirements, the target entities, attributes, and relationships to be analyzed are queried and determined from the knowledge graph, serving as the data foundation for visualization analysis; Based on the data foundation, using the visualization description language, multiple views are defined to display the data foundation from different dimensions, and the corresponding data mapping rules and visual encoding methods are specified for each view; Analyze the data dimensions corresponding to the multiple views and the logic of data state changes in the data foundation, and use the visualization description language to define the linkage rules between views based on data state changes.

6. The big data visualization analysis and display method according to claim 1, characterized in that, In S3, the step of responding to a user's interaction with any of the plurality of views, determining at least one other view affected by the interaction according to the linkage rules, and distributing information containing the data state change to the view and the at least one other view based on the data state change triggered by the interaction specifically includes: Capture user interaction events on any of the views and parse the intent indicated by the interaction events; Based on the parsed intent, calculate and determine the data state changes triggered by the interactive operation; Based on the linkage rules in the visualization configuration specification, query at least one other view associated with the data dimension where the data state change has occurred; The data state change is encapsulated into a unified message format, and the message is simultaneously distributed to any of the views that initiated the interaction and to at least one other view that was queried. Upon receiving the message, each of the views and at least one other view re-queries the data and performs a rendering update based on the data state changes contained therein.

7. The big data visualization analysis and display method according to claim 6, characterized in that, The step of querying at least one other view associated with the data dimension where the data state change has occurred, based on the linkage rules in the visualization configuration specification, specifically includes: Extract one or more data dimensions that have changed from the data state changes; Using one or more data dimensions that have changed as query keys, a matching search is performed in the linkage rule base of the visualization configuration specification; Retrieve one or more associated view identifiers bound in the matched linkage rules; Based on the associated view identifier, at least one other view is determined from the plurality of views.

8. The big data visualization analysis and display method according to claim 1, characterized in that, In step S4, the steps of collecting and analyzing the user's interaction history and corresponding view update results in step S3, and constructing and updating a user profile reflecting the user's visual preferences based on the analysis results, specifically include: The user interaction log is formed by continuously recording the interactive operation events that occur in step S3, the source view that triggers the interactive operation, and the view update results caused by the interactive operation. Perform pattern analysis on the user interaction logs to extract user interaction behavior characteristics; The extracted interactive behavior features are quantized into feature vectors, and visual preference tags for the corresponding users are created or updated based on the feature vectors. Integrate all the user's visual preference tags to generate or update the user profile that reflects the user's visual preferences.

9. The big data visualization analysis and display method according to claim 1, characterized in that, In step S5, the step of adjusting the linkage rules or the display format of the multiple views in real time based on the user profile using a recommendation model specifically includes: During the interaction in step S3, based on the visual preference tags reflected in the user profile, the user's potential interaction intentions or focus of attention in the current analysis context are predicted. The predicted potential interaction intentions or focus of attention, as well as the features of the subset of data in the knowledge graph related to the current view, are input into the recommendation model; Obtain adjustment suggestions from the recommendation model output, and modify the currently effective linkage rules according to the adjustment suggestions, or adjust the visual encoding, layout, or data filtering conditions of the corresponding views in the multiple views in real time.

10. The big data visualization analysis and display method according to claim 1, characterized in that, In the interaction process of step S3, the step of predicting the user's potential interaction intentions or focus of attention in the current analysis context based on the visual preference tags reflected in the user profile specifically includes: From the user profile, read the set of visual preference tags corresponding to the current user; Obtain the current analysis context information; By performing correlation analysis between the set of visual preference tags and the current analysis context information, the user's next possible interaction or the data subset currently being focused on in the current analysis context can be inferred. The next possible interactive action or data subset inferred is determined as the potential interactive intent or focus of attention.

Images