Data story interaction construction method and system based on hierarchical table insight correlation

Abstract

The application discloses a data story interaction construction method and system based on hierarchical table insight association, establishes the association between data insights based on the table header hierarchical structure of hierarchical table data, does not destroy the original hierarchical structure of the table, and enables a user to create meaningful data stories more quickly while retaining the understanding of the original hierarchical table; a nested graph visualization form based on edge binding is adopted to display the data insight association relationship graph extracted from the hierarchical table data, user interaction exploration is supported, the data insight interested by the user is selected, and the data insight selected by the user is organized into a data story.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically relating to a method and system for constructing interactive data stories based on hierarchical table insights and relationships. Background Technology

[0002] Tabular data is an important data management method with applications spanning finance, investment, scientific research, and human resource management. Discovering data insights (such as trends, correlations, and outliers) from tabular data and constructing data stories is a common and crucial data analysis task. Existing research on data insight extraction and data story generation primarily focuses on ordinary flat tables without hierarchical structures. A common method for extracting data insights from ordinary flat tables is to traverse the data and divide it into subsets, calculating and finding potential data insights from different analytical perspectives within each subset. This work focuses on tables with hierarchical headers, where the headers consist of categorical or time-series data fields, while the table content consists of numerical data. Existing methods fail to consider the hierarchical relationships within the headers of hierarchical tabular data, resulting in incomplete data insight extraction. For hierarchical tabular data, the hierarchical relationship of the headers expands the search space for data insights, meaning more potential data insights can be analyzed and extracted. Furthermore, data stories need to connect data insights into a logically coherent narrative, and the attribute relationships within hierarchical tables facilitate the construction of data stories. Therefore, a method that organizes a large amount of data insights into visual data stories based on the hierarchical structure of hierarchical tabular data is of great significance in helping analysts understand the data.

[0003] Based on the aforementioned technical problems in existing technologies, this invention proposes a data story interaction construction method and system based on hierarchical table insight association. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a data story interactive construction method and system based on hierarchical table insights. This method constructs a data insight graph by constructing data insights based on the relationships between attributes in the hierarchical table headers and related data insights extracted from the hierarchical table data. It proposes a method for defining a search space for different hierarchical table data organization states and extracting data insights. Related data insights are used to guide user exploration and assist in the generation of data stories. The relationships between data insights in the hierarchical table are defined and categorized, and presented to users using a clear nested graph data visualization method, assisting users in efficiently constructing meaningful data narratives.

[0005] The present invention adopts the following technical solution:

[0006] On the one hand, a method for constructing data story interactions based on hierarchical table insights and relationships is provided, including:

[0007] Step 1: Parse the hierarchical table data uploaded by the user and build a model;

[0008] Step 2: Transform the hierarchical table data to generate different data organization methods;

[0009] Step 3: Extract data insights for tables with different data organization methods;

[0010] Step 4: Build relationships among all the extracted data insights;

[0011] Step 5: Users explore building data insight maps, select data insights of interest, and build data stories;

[0012] Step 6: Organize the data insights into data stories using a tree structure.

[0013] Further, step 1 includes:

[0014] The header data is defined as either time-series or categorical data based on its data type. Furthermore, based on its position within the table, it is divided into row headers and column headers. The header structure is then constructed as a tree, with the row headers and column headers defined as row subtrees and column subtrees, respectively.

[0015] Further, step 2 includes:

[0016] Step 2.1: Perform row and column swapping operations within the row headers and column headers of the hierarchical table uploaded by the user to generate a new table data organization state;

[0017] Step 2.2: Perform row and column swapping operations between rows and column headers on the hierarchical table uploaded by the user to generate a new table organization method;

[0018] In step 2.1, the row and column swapping operation can change the hierarchical structure inside the row header and column header, thereby changing the way the table data is organized and arranged.

[0019] In step 2.2, the row-column swapping operation can change the row header to the column header or vice versa, thereby changing the arrangement of complex table data.

[0020] Furthermore, step 3 includes:

[0021] Step 3.1: For each table state, the hierarchical table is divided into different data insight extraction blocks by traversal, and the data insight extraction blocks are used as the basic unit.

[0022] Step 3.2: Calculate data insights using each data region as the basic unit, process the data regions, and preserve the logical relationships between data items;

[0023] Step 3.3: Calculate data insights for the processed data area. During the calculation of data insights, design a scoring method to evaluate different types of data insights and record the quality of the data insights.

[0024] Furthermore, in step 3.3, data insights include: data insights for individual objects, data insights based on distribution and trends, and composite data insights.

[0025] Furthermore, in step 3.2, the data area processing includes data aggregation operations, which aggregate data items within the data area according to the hierarchical structure of the table header to obtain data overview information, thereby expanding the calculation scope of data insights.

[0026] Furthermore, in step 3.2, the data area processing includes data grouping operations, which group the data items according to the last header value of the row header and list header sequence used to locate each data item, and then perform inter-group comparison calculations on the data items after the grouping operation.

[0027] On the other hand, a data story interaction construction system based on hierarchical table insights and relationships is provided, including:

[0028] The parsing and modeling module is used to parse the hierarchical table data uploaded by users and build a model.

[0029] The data transformation module is used to transform hierarchical table data to generate different data organization methods;

[0030] The data insights module extracts data insights from tables organized in different ways.

[0031] Build a correlation module to establish correlations between all extracted data insights;

[0032] The data story building module allows users to explore and build data insight maps, select data insights of interest, and construct data stories.

[0033] The Data Story Organization module organizes data insights into data stories using a tree structure.

[0034] On another front, a computer device is provided, comprising: a processor and a memory, wherein the memory stores at least one program, which is loaded and executed by the processor to implement a data story interaction construction method based on hierarchical table insight associations.

[0035] The beneficial effects of this invention are:

[0036] The data story interaction construction method and system based on hierarchical table insight association described in this invention considers different table data organization states after data transformation of hierarchical tables. Using state as the basic unit, the original data is organized into different forms, supporting comprehensive calculation of data insights. For example, for an original data table where the first-level header is year information and the second-level header is month information, in the state without data transformation, data from different months of the same year are adjacent, meaning that only insights existing in certain data belonging to different months of the same year can be calculated. If the positions of the two levels of headers are swapped through data transformation, making the month the first-level header and the year the second-level header, data from different years of the same month are adjacent in this table organization state. Therefore, insights existing in data belonging to different years of the same month can be further calculated. Thus, generating different table data organization states through data transformation makes the calculation of data insights more comprehensive.

[0037] The data story interaction construction method and system based on hierarchical table insight and correlation described in this invention provides users with an understanding of various aspects of table data, promoting the construction of meaningful data stories;

[0038] The data story interaction construction method and system based on hierarchical table insight association described in this invention establishes the association between data insights based on the header hierarchy of hierarchical table data, without destroying the original hierarchical structure of the table, enabling users to create meaningful data stories more quickly while retaining their understanding of the original hierarchical table.

[0039] The data story interactive construction method and system based on hierarchical table insight association described in this invention uses a nested graph visualization form based on edge binding to display the data insight association graph extracted from hierarchical table data, and supports user interactive exploration, allowing users to select data insights of interest, and organize the data insights selected by the user into data stories. Attached Figure Description

[0040] Figure 1 This is a schematic diagram illustrating the construction of an abstract model for tabular data in an embodiment of the present invention;

[0041] Figure 2 This is a flowchart illustrating the creation of a data narrative from raw, complex tabular data in an embodiment of the present invention.

[0042] Figure 3 This is a schematic diagram illustrating different table states generated by data transformation in an embodiment of the present invention;

[0043] Figure 4 This is a visual illustration of different types of data insights in an embodiment of the present invention;

[0044] Figure 5 This is a schematic diagram illustrating the relationship between different data regions in an embodiment of the present invention;

[0045] Figure 6 This is a schematic diagram of an interactive visual analysis interface in an embodiment of the present invention. Detailed Implementation

[0046] To better understand the above-mentioned objectives, features and advantages of the present invention, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present application can be combined with each other.

[0047] Example

[0048] The method for constructing data story interactions based on hierarchical table insights includes:

[0049] Step 1: The system parses the hierarchical table data uploaded by the user and builds a model;

[0050] Step 2: The system transforms the hierarchical table data to generate different data organization methods;

[0051] Step 3: The system extracts data insights for tables with different data organization methods;

[0052] Step 4: The system builds connections between all the extracted data insights;

[0053] Step 5: The system supports users in exploring the constructed data insight maps, selecting data insights of interest, and building data stories;

[0054] Step 6: The system organizes the data insights into data stories using a tree structure.

[0055] As a specific implementation method, in step 1:

[0056] like Figure 1 As shown, the table header is modeled as a row subtree and a column subtree. The first-level nodes of the row subtree represent continents, and the second-level nodes represent countries. The first-level nodes of the column subtree represent years, and the second-level nodes represent seasons. The node sequences of the row and column subtrees can represent data items, i.e., data regions. For example, the data items represented by the shaded area in the figure can be represented by the path "Europe, France" in the row subtree and the path "2022, Autumn" in the column subtree. In the data structure of this application embodiment, the paths in the corresponding two subtrees can be obtained through each data item, and a data item can be obtained through each pair of row and column subtree paths.

[0057] A hierarchical table consists of a header and data items. The header data is defined as either time-series or categorical data by determining its data type. Based on its position in the table, the header is divided into row headers and column headers. The header can be structured into a tree, with row headers and column headers defined as row subtrees and column subtrees, respectively. A hierarchical attribute is defined for each header data item, representing the depth of the header within the subtree.

[0058] Based on the relationship between the header and data items in a complex table, a data item is located by determining two paths in the row subtree and column subtree formed by the header. Each path is represented by a header sequence, where each header corresponds to a node in the row or column subtree. Therefore, a data item can be represented by a header sequence of a row subtree and a header sequence of a column subtree.

[0059] First, blocks in hierarchical table data are defined, which are composed of multiple adjacent table data items. Similar to each data item, a block is also represented by a set of header sequences. To simplify the representation, only different header sequences are retained when representing blocks, and headers belonging to the same subtree in the sequence are represented by wildcards (*). Based on this, a special type of block is defined, in which the set of header sequences contains only one header sequence, or the simplified result is a header sequence containing wildcards. This special type of block is called a data insight extraction block and serves as the basic unit for subsequent data insight extraction and data insight association construction.

[0060] As a specific implementation method, such as Figure 2 As shown, after inputting the table, the data is transformed to generate different states, including swapping levels within rows or list headers and swapping levels between rows and list headers. Then, the partitioning results of all data regions are traversed, and data insights are extracted for each data region, including data insights based on single objects, data insights based on distribution and trends, and composite data insights. Afterwards, relationships are established between all extracted data insights to generate a data insight map. Users can explore different data insights in the table under the guidance of the data insight map, and finally, a meaningful data narrative is generated, such as... Figure 3 As shown in the figure, (a) represents the original table data state, (b) represents the table data organization state obtained by swapping the two levels of list headers during the row and column swapping operation, and (c) represents the table data organization state that transforms the first level of list headers to the last level of row headers during the row and column swapping operation. Specifically, step 2 includes:

[0061] Step 201: The system performs row and column swapping operations on the user-uploaded hierarchical table, within the rows and column headers, to generate a new table data organization state. The above swapping operations can change the hierarchical structure within the rows and column headers, thereby changing the way the table data is organized and arranged.

[0062] Step 202: The system performs row and column swapping operations between rows and column headers on the hierarchical table uploaded by the user to generate a new table organization. This swapping operation can change row headers into column headers or vice versa, thereby changing the arrangement of complex table data.

[0063] As a specific implementation method, step 3 includes:

[0064] Step 301: For each table state, divide the hierarchical table into different data insight extraction blocks by traversing it, and use them as the basic unit;

[0065] Step 302: For each data region, since it retains the structure of a complex table and combines related data items together, when calculating data insights using it as the basic unit, we should focus on the inherent hierarchical structure between the data, and not treat all data items in the data region as equivalent. Therefore, the data region is processed to preserve the logical relationship between data items.

[0066] Data region processing includes aggregation operations. When a data region spans multiple subtrees or leaf nodes in the row or column subtree of the table header, the set of table header sequences used to locate this data region contains wildcards. Therefore, the system aggregates data items according to the level of the wildcards. There are various aggregation data processing strategies, such as minimum, maximum, average, and summation. The system uses summation as the default operation. The system aggregates data items within the data region based on the hierarchical structure of the table header to obtain data overview information, thereby expanding the calculation scope of data insights. The method described in this application embodiment automatically identifies the structural information of the table header, performs hierarchical aggregation of data according to the table header level, and executes a series of built-in aggregation calculations. At the same time, it supports users to customize specific aggregation algorithms according to their needs. For example, for a table where the first-level table header is the country name and the second-level table header is the province (state) name, all data belonging to the same country are aggregated into a whole. Calculations include common statistics such as summation, average, variance, maximum, and minimum. It is also supported for users to input a custom function as the aggregation calculation formula.

[0067] Data region processing includes grouping operations, which are similar to aggregation operations. For data regions where the header sequence set contains wildcards, the difference between the two types of operations is that the purpose of grouping operations is to divide the data region into fine-grained subsets and does not execute aggregation functions. The system groups data items based on the row used to locate each data item and the last header value of the header sequence. After grouping operations, the system can perform inter-group comparison calculations on data items, promoting the calculation of complex data insights. In some analysis scenarios, users do not want to calculate aggregate values. For example, for a table where the first-level header is year information and the second-level header is month information, the method described in this application divides all data belonging to the same year into one group, resulting in several data groups based on years. When calculating data insights, the data distribution pattern within each group can be calculated and compared between groups to obtain overview-level data insight information.

[0068] Step 303: The system calculates data insights for the processed data area, such as... Figure 4 As shown, this application embodiment mainly considers three types of data insights: data insights oriented towards a single object, data insights based on distribution and trends, and composite data insights; in the process of calculating data insights, the system designs a scoring method to evaluate different types of data insights and record the quality of data insights;

[0069] Among them, data insights for a single object describe the characteristics of significant data values ​​within a data region, including maximum dominance, top two dominance, and outliers. The calculation of dominant data insights requires that all entries within the data region are non-negative.

[0070] Distribution and trend-based data insights describe the trends or distribution patterns of data within a data area, including trend, kurtosis, skewness, and evenness. This type of data insight is only calculated when the header of the data item is time-series data.

[0071] Composite data insights involve comparing data items in different groups within a data region, focusing primarily on data correlations. These insights help reveal the relationships and dependencies between various subsets of data within the data region. Since the calculation of such data insights involves comparisons within the data region, these insights are calculated for data regions that have undergone grouping operations.

[0072] As a specific implementation method, in step 4:

[0073] The relationships between data insights are defined based on the relationships between the data areas corresponding to the data insights, including the relationships within the same table data organization state and the relationships between different table data organization states;

[0074] When constructing the relationships between data blocks within the same table's data organization state, the relationships between sets of header sequences used to locate data blocks are first defined. The length attribute of the header sequence is defined, i.e., the total number of headers contained in the sequence (excluding wildcards). For two identical header sequences, their relationship is defined as identical. Figure 5 As shown, (a) and (b) represent the same name relationship, (c) and (d) represent the sibling relationship, and (e) and (f) represent the parent-child relationship. For header sequences with the same length and the same last header (excluding wildcards) in the sequence, their relationship is defined as the same name relationship. For header sequences with the same length but different last headers (excluding wildcards) in the sequence, their relationship is defined as the sibling relationship. For two header sequences with an inclusion relationship, their relationship is defined as the parent-child relationship. Based on the above relationship of header sequences, it is stipulated that two data regions are related if and only if one of their header sequences is exactly the same. More specifically, the relationship between the other two header sequences besides the two identical header sequences represents the relationship between the two data regions.

[0075] When constructing relationships between data regions in different table data organization states, two strategies are adopted. The first strategy is more stringent, stipulating that two data regions with at least one identical header sequence are related. The first strategy yields stronger relationships, but the number of relationships that can be constructed is relatively small. The second strategy stipulates that two data regions are considered related as long as their header sequences have a common prefix (i.e., the values ​​of the first few headers are the same). This strategy can construct more relationships, but some of these relationships are relatively loose. Users can choose either strategy based on their analysis tasks and requirements.

[0076] As a specific implementation method, step 5 includes:

[0077] Step 501: Visualize the data insight map using a nested diagram format to represent different types of data insights and the different relationships between data insights (parent-child relationship, same name relationship, and sibling relationship). Users can click on the nodes in the data insight map to present detailed information about the insight. Each type of data insight is presented in a suitable data visualization format, and annotations related to attributes in the data insight are added to help users understand the specific meaning of the insight. The relationships guide users to explore related data insights, thereby building a meaningful data story.

[0078] Step 502: Support users to explore data insight maps and select data insights of interest. The selected data insights are saved in the user's exploration record and organized into data stories according to the order and relationship of the selected data insights. After the user finishes exploring, support the understanding of the explored data stories.

[0079] To verify the superior technical effects of the embodiments of this application, this embodiment presents a specific case study. This case study was conducted in collaboration with analysts in the game market, and mainly explored the sales data of game hardware. The analysts collected historical sales data of game hardware devices produced by major companies from data websites, recorded the quarterly sales of specific hardware by region, and created a complex table with a hierarchical structure. The header of this complex table represents each quarter from 2013 to 2020, and the row headers include the sales region, game platform, and their manufacturing company.

[0080] After applying the data to the method described in the embodiments of this application, the analyst can obtain an overview of the extracted insights and their relationships, and simplify the insight graph by performing filtering operations on the insights and relationships. In this case, the analyst decides to retain only the parent-child relationship and filter out all point insights in the insight graph.

[0081] Analysts paid particular attention to Microsoft's game console sales. They observed an outlier in Microsoft's total sales in 2014 and sought to uncover the reasons behind this anomaly. Specifically, they wanted to identify which game consoles, regions, and months significantly impacted the outlier. Based on the parent-child relationships shown in the insights chart, analysts clicked on two child nodes: one representing Xbox 360 sales and the other representing Xbox One sales. These two consoles' sales constituted Microsoft's total sales. Analyzing these two child nodes, analysts noted a downward trend in Xbox 360 sales, with no sudden increase in 2014. Therefore, they concluded that the Xbox 360 was not the primary cause of the outlier in Microsoft's total sales in 2014. Regarding the Xbox One, analysts found that its sales remained zero in 2013 but saw a significant increase in 2014. Analysts speculated that the 2014 launch of the Xbox One attracted a large number of users to purchase it. Furthermore, although Xbox... 360's annual sales are declining, but many users still made purchases in 2014. Therefore, analysts suggest that the above phenomenon is the reason for the outlier in 2014.

[0082] Next, the analyst wanted to conduct a more detailed analysis from a regional and monthly perspective. They decided to retain the point insights filtered in the previous steps and examine the sales insights for Xbox One and Xbox 360 in 2014. To compare the sales of these two consoles in 2014 with other years, the previously filtered sibling relationships were retained and used as a basis for examining relevant insights. Most of these points had two insights, one calculated from a regional perspective and the other from a monthly perspective, which the analyst could explore independently. Starting from the monthly perspective, the analyst found that, in most cases, March sales were significantly higher than other months. However, Xbox 360 sales in September 2014 were almost equivalent to March's. Therefore, the analyst believes that Xbox 360 sales in September 2014 may also be a factor contributing to outliers.

[0083] The analysts then switched the above insights to results calculated from a regional perspective for further analysis. Again, the analysts found that the sales distribution of Xbox One and Xbox 360 in different regions was mostly consistent, with North America being larger than Europe, followed by other regions. However, the distribution of Xbox 360 in 2014 was different, with a higher proportion in other regions. The analysts speculated that this may also be one of the reasons for the outlier in Microsoft's total sales in 2014.

[0084] Furthermore, the analysts wanted to further verify the existence of the anomaly by analyzing the absolute data values. They found that, in the current state, the sales data for Xbox One and Xbox 360 in the same region and the same month were not adjacent. Therefore, the analysts decided to switch the focus state from State 0 to State 2, where sales in the same region are adjacent. In State 2, the analysts found a node corresponding to the sales of all Microsoft game consoles in other regions and discovered an insight that indicated that the sales data in other regions in 2014 was an outlier. Similarly, the analysts then switched the focus state to State 3, where sales in the same month are adjacent. The analysts also discovered an insight that indicated that the sales data in September 2014 was an outlier.

[0085] In summary, the analyst used the method described in the embodiments of this application to construct a data narrative to aid the analysis, mainly pointing out that Microsoft's total sales data in 2014 was an outlier, which was closely related to the release of the Xbox One in 2014; in addition, growth in other regions and the sales surge in September also contributed to the occurrence of the sales outlier in 2014.

[0086] For example, in the field of financial investment analysis, by extracting insights and correlations from hierarchical table data, financial analysts can identify risks and opportunities in investment portfolios. For instance, by analyzing the correlations between different asset classes, potential investment opportunities or risk points can be discovered. In the field of market research, the methods described in the embodiments of this application can help researchers gain a deeper understanding of consumer purchasing behavior patterns. By analyzing consumer purchasing data at different time periods, consumption trends and seasonal changes can be revealed. In the field of human resource management, the methods in the embodiments of this application can be used to analyze employee performance data, identify common characteristics of high-performing employees, and patterns of team collaboration.

[0087] A data story interaction building system based on hierarchical table insights and relationships includes:

[0088] The parsing and modeling module is used to parse the hierarchical table data uploaded by users and build a model.

[0089] The data transformation module is used to transform hierarchical table data to generate different data organization methods;

[0090] The data insights module extracts data insights from tables organized in different ways.

[0091] Build a correlation module to establish correlations between all extracted data insights;

[0092] The data story building module allows users to explore and build data insight maps, select data insights of interest, and construct data stories.

[0093] The data story organization module organizes data insights into data stories using a tree structure.

[0094] For example, the interface of the system is as follows: Figure 6 As shown, it contains four subviews: (a) a data insight quantity distribution view, where (a1) shows the distribution of data insight relationships and (a2) shows the distribution of data insights; (b) displays the original hierarchical table data, where the color depth of each cell in the table maps to the number of data insights associated with that cell; (c) a data insight relationship graph shows the data insights extracted from the hierarchical table and the relationships between the insights; and (d) displays a list of related data insights within a specific data insight extraction block, allowing users to select other data insights of interest.

[0095] A computer device includes a processor and a memory, the memory storing at least one program, which is loaded and executed by the processor to implement a data story interaction construction method based on hierarchical table insight associations.

[0096] This invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the invention as claimed. The scope of protection of this invention is defined by the appended claims.

Claims

1. A method for constructing interactive data stories based on hierarchical table insights, characterized in that, include: Step 1: Parse the hierarchical table data uploaded by the user and build a model: The header data is defined as either time-series data or categorical data by determining its data type. Based on its position in the table, the header is divided into row headers and column headers. The header is constructed as a tree structure, with the row headers and column headers defined as row subtrees and column subtrees, respectively. Based on the relationship between the header and data items in a complex table, a data item is located by determining two paths in the row subtree and column subtree formed by the header. Each path is represented by a header sequence, where each header corresponds to a node in the row or column subtree. A data item can be represented by a header sequence of a row subtree and a header sequence of a column subtree. Define a special type of block whose header sequence set contains only one header sequence, or whose simplified result is a header sequence containing wildcards. This special type of block is called a data insight extraction block and serves as the basic unit for subsequent data insight extraction and data insight association construction. Step 2: Transform the hierarchical table data to generate different data organization methods: Step 2.1: Perform row and column swapping operations within the row headers and column headers of the hierarchical table uploaded by the user to generate a new table data organization state; Step 2.2: Perform row and column swapping operations between row headers and column headers on the hierarchical table uploaded by the user to generate a new table organization method; Step 3: Extract data insights for tables with different data organization methods: Step 3.1: For each table state, the hierarchical table is divided into different data insight extraction blocks by traversal, and the data insight extraction blocks are used as the basic unit. Step 3.2: Calculate data insights using each data region as the basic unit, process the data regions, and preserve the logical relationships between data items: Data processing includes data aggregation operations, which aggregate data items within a data area based on the hierarchical structure of the table header to obtain data overview information, thereby expanding the scope of data insight calculations. Data processing includes data grouping operations, which group data items based on the last header value of the row header and list header sequence used to locate each data item, and then perform inter-group comparison calculations on the data items after the grouping operation. Step 3.3 involves calculating data insights from the processed data area. During this process, a scoring method is designed to evaluate different types of data insights and record their quality. Data insights include: data insights oriented towards a single object, data insights based on distribution and trends, and composite data insights; in the process of calculating data insights, the system is designed with a scoring method to evaluate different types of data insights and record the quality of the data insights; Step 4: Build relationships among all extracted data insights: The relationships between data insights are defined based on the relationships between the data areas corresponding to the data insights, including the relationships within the same table data organization state and the relationships between different table data organization states; When constructing the association relationship between data blocks within the same table data organization state, the association between the header sequence sets used to locate data blocks is first defined. The length attribute of the header sequence is defined, that is, the total number of headers contained in the sequence, excluding wildcards. For two identical header sequences, their relationship is defined as identical. a and b represent the same name relationship, c and d represent the sibling relationship, and e and f represent the parent-child relationship. For header sequences with the same length and the same last header in the sequence, their relationship is defined as the same name relationship. For header sequences with the same length but different last headers in the sequence, their relationship is defined as the sibling relationship. For two header sequences with an inclusion relationship, their relationship is defined as the parent-child relationship. Based on the above relationships of header sequences, it is stipulated that two data regions are related if and only if one of their header sequences is completely identical. More specifically, the relationship between the other two header sequences besides the two identical header sequences represents the relationship between the two data regions. When constructing the association between data regions in different table data organization states, two strategies are adopted. The first strategy is strict, which stipulates that two data regions with at least one identical header sequence are related. The first strategy results in a strong correlation, but the number of associations that can be constructed is small. The second strategy stipulates that two data regions are considered related as long as their header sequences have a common prefix. This strategy constructs more associations, but some of these relationships are loose. Users can choose either strategy according to their analysis tasks and requirements. Step 5: Users explore building data insight maps, selecting data insights of interest and constructing data stories: Step 501: Visualize the data insight map using a nested diagram format to represent different types of data insights and the different relationships between them, such as parent-child relationships, same-name relationships, and sibling relationships. Users can click on the nodes in the data insight map to display detailed information about the insight. Each type of data insight is presented in a suitable data visualization format, and annotations related to attributes in the data insight are added to help users understand the specific meaning of the insight. The relationships guide users to explore related data insights, thereby building a meaningful data story. Step 502: Support users to explore data insight maps and select data insights of interest. The selected data insights are saved in the user's exploration record and organized into data stories according to the order and relationship of the selected data insights. After the user finishes exploring, support the understanding of the explored data stories. Step 6: Organize the data insights into data stories using a tree structure.

2. A data story interaction construction system based on hierarchical table insights and relationships, characterized in that, include: The parsing and modeling module is used to parse the hierarchical table data uploaded by users and build a model: The header data is defined as either time-series data or categorical data by determining its data type. Based on its position in the table, the header is divided into row headers and column headers. The header is constructed as a tree structure, with the row headers and column headers defined as row subtrees and column subtrees, respectively. Based on the relationship between the header and data items in a complex table, a data item is located by determining two paths in the row subtree and column subtree formed by the header. Each path is represented by a header sequence, where each header corresponds to a node in the row or column subtree. A data item can be represented by a header sequence of a row subtree and a header sequence of a column subtree. Define a special type of block whose header sequence set contains only one header sequence, or whose simplified result is a header sequence containing wildcards. This special type of block is called a data insight extraction block and serves as the basic unit for subsequent data insight extraction and data insight association construction. The data transformation module is used to transform hierarchical table data to generate different data organization methods: The system performs row and column swapping operations within the row headers and column headers of user-uploaded hierarchical tables to generate a new table data organization state. Perform row and column swapping operations between row headers and column headers on the hierarchical table uploaded by the user to generate a new table organization method; The data insights module extracts data insights from tables organized in different ways: For each table state, the hierarchical table is divided into different data insight extraction blocks by traversal, and the data insight extraction blocks are used as the basic unit. Data insights are calculated using each data region as the basic unit, and the data regions are processed while preserving the logical relationships between data items: Data processing includes data aggregation operations, which aggregate data items within a data area based on the hierarchical structure of the table header to obtain data overview information, thereby expanding the scope of data insight calculations. Data processing includes data grouping operations, which group data items based on the last header value of the row header and list header sequence used to locate each data item, and then perform inter-group comparison calculations on the data items after the grouping operation. Data insights are calculated on the processed data areas. During the calculation of data insights, a scoring method is designed to evaluate different types of data insights and record the quality of the data insights: Data insights include: data insights oriented towards a single object, data insights based on distribution and trends, and composite data insights; in the process of calculating data insights, the system is designed with a scoring method to evaluate different types of data insights and record the quality of the data insights; The module for building associations establishes relationships between all extracted data insights: The relationships between data insights are defined based on the relationships between the data areas corresponding to the data insights, including the relationships within the same table data organization state and the relationships between different table data organization states; When constructing the association relationship between data blocks within the same table data organization state, the association between the header sequence sets used to locate data blocks is first defined. The length attribute of the header sequence is defined, that is, the total number of headers contained in the sequence, excluding wildcards. For two identical header sequences, their relationship is defined as identical. a and b represent the same name relationship, c and d represent the sibling relationship, and e and f represent the parent-child relationship. For header sequences with the same length and the same last header in the sequence, their relationship is defined as the same name relationship. For header sequences with the same length but different last headers in the sequence, their relationship is defined as the sibling relationship. For two header sequences with an inclusion relationship, their relationship is defined as the parent-child relationship. Based on the above relationships of header sequences, it is stipulated that two data regions are related if and only if one of their header sequences is completely identical. More specifically, the relationship between the other two header sequences besides the two identical header sequences represents the relationship between the two data regions. When constructing the association between data regions in different table data organization states, two strategies are adopted. The first strategy is strict, which stipulates that two data regions with at least one identical header sequence are related. The first strategy results in a strong correlation, but the number of associations that can be constructed is small. The second strategy stipulates that two data regions are considered related as long as their header sequences have a common prefix. This strategy constructs more associations, but some of these relationships are loose. Users can choose either strategy according to their analysis tasks and requirements. The data story building module allows users to explore and build data insight maps, select data insights of interest, and construct data stories. The data insight diagram is visualized using a nested diagram format to represent different types of data insights and the different relationships between them, such as parent-child relationships, same-name relationships, and sibling relationships. Users can click on nodes in the data insight diagram to view detailed information about the insight. Each type of data insight is presented in a suitable data visualization format, and annotations related to attributes in the data insight are added to help users understand the specific meaning of the insight. Relationships guide users to explore related data insights, thereby building meaningful data stories. Users can explore data insight maps and select data insights of interest. The selected data insights are saved in the user's exploration record and organized into data stories according to the order and relationship of the selected data insights. After the user finishes exploring, the user can understand the explored data stories. The Data Story Organization module organizes data insights into data stories using a tree structure.

3. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing at least one program, which is loaded and executed by the processor to implement the data story interaction construction method based on hierarchical table insight association as described in claim 1.

Images