A natural language-based enterprise data visualization analysis method and system

Abstract

The application relates to the technical field of enterprise data analysis, and discloses an enterprise data visualization analysis method and system based on natural language, which comprises the following steps: collecting enterprise business data and natural language query text, and determining a target analysis dimension; dividing core, auxiliary and associated data subtypes according to an analysis intention, and constructing a to-be-analyzed dimension set; calculating the dimension correlation degree among the subtypes, generating an analysis correlation graph and determining an optimal generation tree; determining a visualization priority order based on the optimal generation tree, and generating visualization results in sequence. The system comprises a data acquisition unit, a dimension division unit, an associated graph construction unit, a priority determination unit and a visualization generation unit. Through natural language interaction and intelligent correlation analysis, the application improves the accuracy and efficiency of enterprise data visualization, and is suitable for data analysis requirements in complex business scenarios.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of enterprise data analysis, in particular to an enterprise data visualization analysis method and system based on natural language. BACKGROUND

[0002] With the rapid growth of enterprise data, how to efficiently extract valuable information from massive business data has become a key challenge for enterprise decision-making. Traditional data analysis methods usually rely on manually writing query statements or using fixed report tools. This way is not only inefficient, but also requires higher technical ability of users. Especially for non-technical business personnel, complex query syntax and data analysis process often become a bottleneck that hinders them from quickly obtaining insights.

[0003] There are some data analysis tools based on natural language processing (NLP) in the current market, which allow users to input query requirements through natural language. However, these tools still have significant limitations in practical application. Most tools can only handle simple query intentions and cannot deeply understand complex analysis requirements, such as multi-dimensional association analysis or dynamic data sub-type division. Existing methods lack intelligent priority ordering in the visualization generation link, resulting in generated charts that may be redundant or not highlight the key points, reducing user experience and analysis efficiency. Traditional systems usually ignore the relevance between data sub-types, making it difficult to build a globally optimized visualization display scheme.

[0004] In view of the above problems, an enterprise data visualization analysis method combining natural language processing and graph optimization technology is needed. The method needs to achieve the following goals: accurately identify the core analysis intention in the natural language query, dynamically divide the data sub-type; quantify the relevance between data sub-types, build an analysis association graph and determine the optimal spanning tree; generate visualization results based on priority order to ensure that the display logic is clear and the key points are highlighted. The present application effectively solves the problems of single analysis dimension and scattered visualization results in existing technologies by introducing multi-dimensional correlation degree calculation and spanning tree optimization algorithm, providing a more efficient and intelligent solution for enterprise data visualization analysis. SUMMARY

[0005] The purpose of the present application is to provide an enterprise data visualization analysis method and system based on natural language to solve the problems raised in the background.

[0006] To achieve the above purpose, the present application provides the following technical solution: an enterprise data visualization analysis method based on natural language, the method comprising:

[0007] Collecting current enterprise business data to be analyzed and corresponding natural language query text, and obtaining target analysis dimensions according to the enterprise business data and natural language query text;

[0008] Based on the analysis intention of the natural language query text, the data subtypes covered by the target analysis dimension are divided into core data subtypes and auxiliary data subtypes, and an associated data subtype is determined among the data subtypes not covered by the target analysis dimension, the core data subtypes, auxiliary data subtypes and associated data subtypes constitute a set of dimensions to be analyzed;

[0009] The dimension correlation degree between any two adjacent subtypes in the set of dimensions to be analyzed is determined, and an analysis correlation graph of the set of dimensions to be analyzed is constructed with the subtypes in the set of dimensions to be analyzed as vertices and the dimension correlation degree as the edges connecting the vertices.

[0010] The optimal spanning tree of the analysis correlation graph is determined, and the visualization priority order of each subtype in the set of dimensions to be analyzed is determined according to the optimal spanning tree;

[0011] The visualization results of each subtype in the set of dimensions to be analyzed are generated in the order of the visualization priority order.

[0012] Preferably, the target analysis dimension is obtained according to the enterprise business data and the natural language query text, including:

[0013] According to the natural language query text, the analysis dimensions matching the enterprise business data in the historical analysis tasks are counted.

[0014] According to the semantic features of the natural language query text in each analysis dimension, the counted analysis dimensions are classified to form one or more sets of analysis dimensions.

[0015] In the set of analysis dimensions, the target analysis dimension set matching the current analysis intention of the natural language query text is determined, and the dimensions covered by the target analysis dimension set are taken as the target analysis dimension.

[0016] Preferably, the classified analysis dimensions include:

[0017] The semantic feature standard interval predefined in the enterprise data management system is obtained.

[0018] For any one of the counted analysis dimensions, the semantic feature standard interval to which the semantic feature of the natural language query text in the analysis dimension belongs is identified, and the analysis dimension is divided into the analysis dimension set corresponding to the identified semantic feature standard interval.

[0019] Preferably, based on the analysis intention of the natural language query text, dividing the data subtypes covered by the target analysis dimension into core data subtypes and auxiliary data subtypes comprises:

[0020] Determining the core analysis object and the auxiliary analysis object represented by the analysis intention of the natural language query text;

[0021] Determining a first data coverage range of the core analysis object in the target analysis dimension, and determining a second data coverage range of the auxiliary analysis object in the target analysis dimension;

[0022] Determining the data subtypes covered by the first data coverage range in the enterprise business data range as core data subtypes, and determining the data subtypes covered by the second data coverage range in the enterprise business data range as auxiliary data subtypes.

[0023] Preferably, determining the associated data subtype in the data subtypes not covered by the target analysis dimension comprises:

[0024] Setting a first extension range for the core data subtypes in the target analysis dimension, and setting a second extension range for the auxiliary data subtypes in the target analysis dimension;

[0025] For the data subtypes not covered by the target analysis dimension, if adjacent to the core data subtypes, selecting the data subtypes of the first extension range as the associated data subtypes; if adjacent to the auxiliary data subtypes, selecting the data subtypes of the second extension range as the associated data subtypes.

[0026] Preferably, determining the dimension correlation degree between any two adjacent subtypes in the set of dimensions to be analyzed comprises:

[0027] For the first subtype and the second subtype adjacent in the set of dimensions to be analyzed, respectively identifying the type category of the first subtype and the second subtype, the type category comprising one of core data subtypes, auxiliary data subtypes and associated data subtypes;

[0028] According to the identified type category, determining a first dimension correlation parameter of the first subtype, and determining a second dimension correlation parameter of the second subtype;

[0029] Calculating the difference value between the first dimension correlation parameter and the second dimension correlation parameter, and determining the dimension correlation degree between the first subtype and the second subtype according to the difference value.

[0030] Preferably, determining the dimension correlation degree between the first sub-type and the second sub-type according to the difference value comprises:

[0031] identifying a dimension correlation parameter of each sub-type in the set of dimensions to be analyzed, and determining an average value of the identified dimension correlation parameters;

[0032] determining a plurality of difference intervals according to the average value, and determining a target difference interval in which the difference value between the first dimension correlation parameter and the second dimension correlation parameter is calculated to be located;

[0033] determining a preset correlation degree corresponding to the target difference interval as the dimension correlation degree between the first sub-type and the second sub-type.

[0034] Preferably, generating the visualization results of each sub-type in the set of dimensions to be analyzed in the order of the visualization priority order comprises:

[0035] identifying a generation parameter of each generation node in the visualization priority order, and constructing an error feature sequence according to the identified generation parameter, the arrangement order of each parameter in the error feature sequence being consistent with the visualization priority order;

[0036] inputting the error feature sequence into an error correction model to generate a visualization action output for each generation node through the error correction model;

[0037] generating the visualization results of each sub-type according to each generated visualization action output.

[0038] Preferably, the present application further comprises an enterprise data visualization analysis system based on natural language, the system comprising:

[0039] a data acquisition unit configured to collect current enterprise business data to be analyzed and corresponding natural language query text, and to obtain target analysis dimensions according to the enterprise business data and the natural language query text;

[0040] a dimension division unit configured to divide data sub-types covered by the target analysis dimensions into core data sub-types and auxiliary data sub-types based on an analysis intention of the natural language query text, and to determine associated data sub-types in data sub-types not covered by the target analysis dimensions, the core data sub-types, the auxiliary data sub-types and the associated data sub-types constituting a set of dimensions to be analyzed;

[0041] An association graph construction unit is configured to determine a dimension association degree between any two adjacent subtypes in the set of dimensions to be analyzed, and construct an analysis association graph of the set of dimensions to be analyzed by taking the subtypes in the set of dimensions to be analyzed as vertices and the dimension association degrees as edges connecting the vertices.

[0042] A priority determination unit is configured to determine an optimal spanning tree of the analysis association graph, and determine a visualization priority order of each subtype in the set of dimensions to be analyzed according to the optimal spanning tree.

[0043] A visualization generation unit is configured to generate visualization results of each subtype in the set of dimensions to be analyzed in the visualization priority order.

[0044] Preferably, the visualization generation unit is specifically configured to identify generation parameters of each generated node in the visualization priority order, construct an error feature sequence according to the identified generation parameters, the arrangement order of each parameter in the error feature sequence being consistent with the visualization priority order, input the error feature sequence into an error correction model to generate a visualization action output for each generated node through the error correction model, and generate the visualization results of each subtype according to each generated visualization action output.

[0045] Compared with the prior art, the present application has the following advantages:

[0046] The present application improves the intelligent level and efficiency of enterprise data visualization analysis by introducing natural language processing technology and graph optimization algorithm. Based on the analysis intention of the natural language query text, the core data subtype, the auxiliary data subtype and the associated data subtype are dynamically divided, which can accurately capture user demand and avoid the information omission problem caused by single analysis dimension in traditional methods. By quantifying the association degree between data subtypes and constructing an analysis association graph, the system can optimize the data display logic from a global perspective to ensure that the visualization results are both comprehensive and highlight the key points.

[0047] The optimal spanning tree algorithm is used to determine the visualization priority order, effectively solving the problem of chart redundancy or disordered display logic in the prior art. The user does not need to manually adjust or filter, and the system can automatically generate a visualization result with clear hierarchy and clear main and secondary points. This feature is particularly suitable for multi-dimensional analysis requirements in complex business scenarios, such as comprehensive display of sales trends and regional correlation.

[0048] The error correction model is used to dynamically optimize the visualization generation process, further improving the accuracy and stability of the results. The introduction of the error feature sequence enables the system to automatically correct the deviation caused by data noise or query ambiguity, ensuring that the final output chart is highly consistent with the user's intention.

[0049] In practical applications, the present application greatly reduces the technical threshold of enterprise data analysis. Business personnel do not need to master professional query languages or data analysis tools, and can obtain high-quality visualization results by only inputting natural language descriptions. At the same time, the system has strong expansibility and can adapt to the data management needs of enterprises in different industries, such as multi-scene analysis tasks in the financial, retail or manufacturing industries. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 A working principle diagram of the natural language-based enterprise data visualization analysis method described in the present application;

[0051] Figure 2 A design diagram of the data sub-type division method;

[0052] Figure 3 A design diagram of the associated data sub-type determination method;

[0053] Figure 4 A design diagram of the dimension correlation degree determination method. DETAILED DESCRIPTION

[0054] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0055] Please refer to Figures 1-4 The natural language-based enterprise data visualization analysis method described in the present application has the following specific implementation steps:

[0056] Collect the current enterprise business data to be analyzed and the corresponding natural language query text, and obtain the target analysis dimension according to the enterprise business data and the natural language query text. The enterprise business data covers various data generated in the enterprise operation process, such as financial data, sales data, production data, etc. The natural language query text is the query requirement proposed by the user in the form of natural language.

[0057] Based on the analysis intention of the natural language query text, the data sub-types covered by the target analysis dimension are divided into core data sub-types and auxiliary data sub-types, and the associated data sub-types are determined in the data sub-types not covered by the target analysis dimension. The core data sub-types, auxiliary data sub-types and associated data sub-types constitute the set of dimensions to be analyzed. The analysis intention reflects the core purpose and focus of the user query.

[0058] A dimension correlation degree between any two adjacent subtypes in the set of dimensions to be analyzed is determined, and an analysis correlation graph of the set of dimensions to be analyzed is constructed with the subtypes in the set of dimensions to be analyzed as vertices and the dimension correlation degrees as edges connecting the vertices.

[0059] An optimal spanning tree of the analysis correlation graph is determined, and a visualization priority order of each subtype in the set of dimensions to be analyzed is determined according to the optimal spanning tree. The optimal spanning tree can optimize the connection relationship between the subtypes, thereby determining a reasonable visualization order.

[0060] The visualization results of the subtypes in the set of dimensions to be analyzed are generated in the visualization priority order. The visualization results present data information in the form of intuitive charts, graphs, etc.

[0061] Embodiment 1:

[0062] In obtaining the target analysis dimension, the current enterprise business data to be analyzed and the corresponding natural language query text are collected. The enterprise business data includes various data generated by the enterprise in each operation link, such as the details of income and expenditure, asset and liability situation in financial data, order information and customer purchase records in sales data, production capacity and raw material consumption in production data. The natural language query text is a query requirement proposed by the user in the form of daily language, such as "analyze the change trend of sales performance in each region in this quarter" and "view the cost composition of the company in the past year".

[0063] According to the natural language query text, the analysis dimensions matching the enterprise business data in the historical analysis tasks are counted. The historical analysis task database stores the similar queries and the corresponding analysis dimensions processed in the past. The system will perform semantic analysis on the natural language query text, extract the keywords, key phrases and core intent of the query, and then compare these features with the queries in the historical analysis tasks to find similar historical analysis tasks in terms of semantics and business logic, thereby counting the matching analysis dimensions. For example, when the natural language query text is "analyze the sales in the second quarter in the East China region", the system will retrieve the analysis tasks in the historical analysis tasks that involve the keywords "East China region", "second quarter" and "sales", and count the analysis dimensions used in these tasks, such as time dimension, region dimension and sales indicator.

[0064] According to the semantic features of the natural language query text in each analysis dimension, the statistical analysis dimensions are classified to form one or more analysis dimension sets. The analysis of semantic features covers multiple aspects, including keywords and their weights in the query text, the grammatical structure of the sentence, semantic relationships (such as cause-and-effect relationship, parallel relationship, subordinate relationship, etc.), and contextual context. In order to achieve accurate classification, the pre-defined semantic feature standard intervals in the enterprise data management system need to be obtained. These standard intervals are formed by classifying and defining common semantic features according to the business characteristics and data management needs of the enterprise. For example, the semantic features can be divided into different standard intervals such as "time-related", "region-related", "product-related", "financial indicator-related", etc., each interval has a clear feature description and range definition.

[0065] For any statistical analysis dimension, the system will identify the semantic features of the natural language query text in that analysis dimension and determine the specific semantic feature standard interval to which it belongs. During the identification process, the system will use natural language processing techniques such as morphological analysis, syntactic analysis, and semantic understanding to conduct in-depth analysis of the query text. Taking the query text "analyze the sales of product A in North China in 2024" corresponding to an analysis dimension as an example, the system will extract semantic features such as "North China" (region-related), "2024" (time-related), "product A" (product-related), and "sales" (indicator-related), and then match these features with the semantic feature standard intervals to determine that the analysis dimension should be classified into the analysis dimension set corresponding to the "region-time-product-indicator" related semantic feature standard interval.

[0066] After completing the classification of each analysis dimension, the target analysis dimension set that matches the current analysis intent of the natural language query text is determined in the analysis dimension set. The analysis intent is the core purpose and focus of the user's query, which needs to be understood through in-depth analysis of the natural language query text. The system will conduct more detailed analysis of the current natural language query text to determine the content the user wants to understand, the focus of attention, and the expected analysis angle, etc. For example, the user's query text is "compare the profit growth rate of different regions in the third quarter", and the analysis intent is to compare and analyze the profit growth rate from the perspective of region and time. Then, the system will compare the features of each analysis dimension set with the current analysis intent, and select the analysis dimension set that best meets the requirements in terms of semantics, business logic, and analysis angle, and determine it as the target analysis dimension set.

[0067] The dimensions covered by the target analysis dimension set are taken as the target analysis dimensions. The target analysis dimension set contains multiple analysis dimensions that match the current analysis intent, which collectively form the framework for analyzing the user query. For example, the target analysis dimension set may contain the region dimension, the time dimension, the profit indicator dimension, etc. The target analysis dimensions are these specific dimensions, and the system will further analyze and process the enterprise business data based on these dimensions.

[0068] During the entire process of obtaining the target analysis dimensions, the system needs to continuously process and analyze the natural language query text and enterprise business data to ensure that the obtained target analysis dimensions accurately reflect the user's query requirements, laying a solid foundation for subsequent data visualization analysis work. At the same time, this process makes full use of the data of historical analysis tasks and the pre-defined semantic feature standard interval, improving the efficiency and accuracy of target analysis dimension acquisition, enabling the system to respond more quickly and accurately to user query requests.

[0069] Embodiment 2:

[0070] In dividing the core data sub-type and the auxiliary data sub-type based on the analysis intent of the natural language query text, it is necessary to clearly define the core analysis object and the auxiliary analysis object represented by the analysis intent of the natural language query text. The extraction process of the analysis intent relies on natural language processing technology, and the system will perform deep semantic analysis on the query text to identify entities, relationships, and the user's core demands. For example, when the query text is "analyze the R&D cost proportion of each department in the first half of 2024 and its impact on product iteration efficiency", through semantic analysis, it can be determined that "R&D cost" is one of the core analysis objects, and "product iteration efficiency" is the auxiliary analysis object, because the user's core purpose is to understand the proportion of R&D cost, and needs to combine its impact on product iteration efficiency for comprehensive analysis.

[0071] After determining the core analysis object and the auxiliary analysis object, it is necessary to determine their data coverage ranges in the target analysis dimensions. The target analysis dimensions are specific analysis dimensions obtained from the enterprise business data and the natural language query text before, such as time dimension, department dimension, cost indicator dimension, efficiency indicator dimension, etc. For the core analysis object "R&D cost", when determining its first data coverage range in the target analysis dimensions, it needs to consider the specific direction of the core analysis object in each dimension. In the time dimension, it covers "the first half of 2024"; in the department dimension, it covers all departments related to R&D in the enterprise; in the cost indicator dimension, it covers various constituent items of R&D cost, such as personnel salary, equipment procurement cost, material consumption cost, etc. For the auxiliary analysis object "product iteration efficiency", when determining its second data coverage range in the target analysis dimensions, it also needs to be defined in related dimensions. In the time dimension, it is consistent with the core analysis object, which is "the first half of 2024"; in the department dimension, it covers the departments with product iteration tasks; in the efficiency indicator dimension, it covers product iteration cycle, new version release frequency, demand response time, etc.

[0072] The data subtypes covered by the first data coverage range in the enterprise business data range are determined as core data subtypes. Enterprise business data contains data of multiple dimensions and types, and the first data coverage range is matched and filtered in these data to determine specific data subtypes. For example, in the enterprise's financial data, the data subtypes related to R&D cost may include personnel salary data, R&D equipment procurement data, R&D material consumption data, etc.; in the human resource data, it may include the number of R&D personnel, job distribution, working years, etc. These data subtypes are involved in the first data coverage range in the enterprise business data range, so they are determined as core data subtypes.

[0073] Similarly, the data subtypes covered by the second data coverage range in the enterprise business data range are determined as auxiliary data subtypes. For the auxiliary analysis object "product iteration efficiency", its second data coverage range in the enterprise business data may include product R&D project management data, such as iteration cycle records of each project; product version release data, such as the time and content of new version release; demand management data, such as user demand receiving, processing and response time, etc. These data subtypes constitute auxiliary data subtypes, which are not the core of user query, but play an important auxiliary role in the comprehensive analysis of the core analysis object.

[0074] In this process, attention should be paid to the division of data subtypes accurately reflecting the association between analysis intent and data. The system needs to have a clear understanding of the structure and content of enterprise business data, and can accurately identify the corresponding subtypes according to the data coverage range. For example, when determining the core data subtypes of R&D costs, it is necessary to distinguish R&D costs from other cost types of data, and avoid including non-R&D cost data subtypes; when determining the auxiliary data subtypes of product iteration efficiency, it is necessary to ensure that the selected data subtypes can indeed reflect various aspects of product iteration efficiency, rather than irrelevant or unrelated data.

[0075] The analysis intent may have complexity and multi-level, and the system needs to be able to handle it flexibly. For example, some query texts may involve multiple core analysis objects and auxiliary analysis objects, in which case the data coverage range and corresponding subtypes need to be determined for each object. For another example, the analysis intent may change with further interaction of the context or user, which requires a certain dynamic adjustment capability when dividing data subtypes, but in this embodiment, the analysis intent and data subtypes are mainly determined based on the current natural language query text.

[0076] Embodiment 3:

[0077] When determining the associated data subtypes in the data subtypes not covered by the target analysis dimension, first, the first extension range needs to be set for the core data subtypes in the target analysis dimension, and the second extension range needs to be set for the auxiliary data subtypes in the target analysis dimension. The setting of the extension range needs to be determined in combination with the internal logic of enterprise business data and the association between data. For the core data subtypes, the first extension range is usually the data range that is closely related to the core data subtypes in business logic and can further supplement the core data analysis. For example, if the core data subtype is "quarterly sales of product A", its first extension range may include product A's sales cost, inventory changes, marketing expenses, etc. These data have a direct association with sales in business and can help analyze core data more comprehensively.

[0078] The second extension range of the auxiliary data subtypes is around the auxiliary data subtypes, selecting the data range associated with them. For example, the auxiliary data subtype is "customer satisfaction survey results", and its second extension range may include customer purchase frequency, complaint records, willingness to recommend, etc. These data can supplement the information of the auxiliary data subtypes from different angles and provide more support for overall analysis.

[0079] After setting the extension range, the system needs to determine the adjacent relationship between the data subtypes not covered by the target analysis dimension and the core data subtypes or auxiliary data subtypes, and select the data subtypes in the corresponding extension range as the associated data subtypes. Here, "adjacent" not only refers to the adjacent physical storage location, but more importantly refers to the close connection in business logic, data structure, or semantic relationship.

[0080] Specifically, for a data subtype not covered by the target analysis dimension, the system first analyzes its association with the core data subtypes. For example, assume that the data subtype not covered by the target analysis dimension is "material price fluctuation of product A", and the core data subtype is "production cost of product A". Since material price fluctuation directly affects production cost, the two have a close causal relationship in business logic, so the data subtype can be considered adjacent to the core data subtype. At this time, the system selects the corresponding data subtype from the first extension range set for the core data subtype as the associated data subtype. If the first extension range contains the data subtype related to "material price fluctuation", it is determined as the associated data subtype.

[0081] If the data subtype not covered by the target analysis dimension is adjacent to the auxiliary data subtype, the system selects the associated data subtype from the second extension range. For example, the auxiliary data subtype is "user's functional evaluation of product B", and the data subtype not covered by the target analysis dimension is "function iteration record of product B". Functional evaluation and function iteration record have a before-and-after relationship in business logic, and the user's evaluation may affect function iteration, while function iteration may change subsequent user evaluation, so they belong to an adjacent relationship. At this time, the system checks whether the second extension range set for the auxiliary data subtype contains a data subtype related to "function iteration record", and if it does, it is determined as the associated data subtype.

[0082] In determining the adjacent relationship between data subtypes, the system uses data correlation analysis techniques to comprehensively analyze the field attributes, semantic characteristics, and data flow in business processes of data subtypes. For example, by analyzing whether the key fields of data subtypes have correlations, such as product ID, timestamp, and other common fields, or by semantic analysis to determine whether the business concepts represented by data subtypes have logical connections, causal relationships, or supplementary relationships.

[0083] In addition, the extended range is not fixed and can be adjusted flexibly according to the business characteristics and data structure of the enterprise. For different industries or enterprises, the same data sub-type may have different degrees of business association, so it is necessary to reasonably define the first extended range and the second extended range according to the actual situation. For example, in manufacturing enterprises, the production man-hour data of products may have strong association with the core data sub-type of production cost and should be included in the first extended range. In service industry enterprises, similar data may have low association and do not need to be included in the extended range.

[0084] The process of determining the associated data sub-type is crucial for constructing a complete set of dimensions to be analyzed. These associated data sub-types, although not directly covered by the target analysis dimension, can enrich the depth and dimension of the analysis through association with core or auxiliary data sub-types, making the visual analysis results more comprehensive. For example, when analyzing sales performance, in addition to the core sales data and auxiliary customer satisfaction data, associated market competition data, promotion activity data and other associated data sub-types can help the enterprise better understand the reasons and trends of sales performance changes.

[0085] Embodiment 4:

[0086] In determining the dimension association degree between any two adjacent sub-types in the set of dimensions to be analyzed, the type category of the first sub-type and the second sub-type in the set of dimensions to be analyzed needs to be identified respectively. The type category includes one of the core data sub-type, the auxiliary data sub-type and the associated data sub-type. For example, assuming that there are adjacent “2024 third quarter R&D cost of each department” (core data sub-type) and “R&D personnel education distribution” (auxiliary data sub-type) in the set of dimensions to be analyzed, the system needs to first determine the type attribution of the two. The identification process is based on the previous dimension division logic, the core data sub-type corresponds to the core object coverage data of the analysis intent, the auxiliary data sub-type corresponds to the auxiliary object coverage data, and the associated data sub-type is determined by the extended range.

[0087] After identifying the type category, it is necessary to determine the first dimension correlation parameter of the first sub-type and the second dimension correlation parameter of the second sub-type according to the type category. The dimension correlation parameter is a characteristic value for measuring the degree of correlation between sub-types, and its value is related to the type category. For example, the core data sub-type may correspond to a higher basic correlation weight, the auxiliary data sub-type next, and the correlation data sub-type relatively low weight. Specific parameters can include the overlap degree of data fields, the correlation strength in business processes, the relevance at the semantic level, etc. Taking “R&D cost” (core) and “R&D personnel education distribution” (auxiliary) as examples, the first dimension correlation parameter can be set to reflect the correlation strength of cost data and personnel structure data in business logic, such as the historical correlation index of R&D personnel education and R&D investment; the second dimension correlation parameter corresponds to the correlation characteristic parameter of the education distribution data itself, such as the potential correlation degree index of different education levels and R&D efficiency.

[0088] The difference value between the first dimension correlation parameter and the second dimension correlation parameter is calculated. The calculation method of the difference value needs to be combined with the specific form of the parameter. If the parameter is numerical, the absolute difference or relative difference can be directly calculated; if it is a semantic characteristic parameter, the difference degree can be calculated by a semantic similarity algorithm. For example, assuming that the first dimension correlation parameter is 0.8 (indicating strong correlation between R&D cost and personnel structure), and the second dimension correlation parameter is 0.5 (indicating moderate correlation between education distribution and R&D efficiency), the absolute difference value between them is 0.3.

[0089] In order to determine the dimension correlation degree according to the difference value, it is necessary to identify the dimension correlation parameters of each sub-type in the set of dimensions to be analyzed, and calculate the average value of these parameters. For example, the set of dimensions to be analyzed contains 5 sub-types, and their dimension correlation parameters are 0.6, 0.8, 0.5, 0.7, and 0.9, with an average value of (0.6+0.8+0.5+0.7+0.9) / 5=0.7.

[0090] Then, according to the average value, a plurality of difference intervals are determined. The division of difference intervals needs to cover the possible range of difference values and reflect the hierarchy of different correlation degrees. For example, taking the average value 0.7 as a reference, the difference interval can be set as: [0, 0.2) for “high correlation”, [0.2, 0.4) for “high-medium correlation”, [0.4, 0.6) for “moderate correlation”, and [0.6, 1.0] for “low correlation”. The specific interval division can be adjusted according to business needs and data characteristics to ensure that the interval can reasonably reflect the correlation strength difference between sub-types.

[0091] The target difference interval where the calculated difference value is located is determined. Taking the difference value 0.3 as an example, it is in the [0.2, 0.4) interval, which corresponds to the “high-medium correlation” interval.

[0092] The preset correlation degree corresponding to the target difference interval is determined as the dimension correlation degree between the first sub-type and the second sub-type. For example, the preset correlation degree corresponding to the "medium-high correlation" interval can be set to 0.75 (this value is only an example, and the actual preset value needs to be determined according to business logic and historical data), indicating that the correlation between the two adjacent sub-types is medium-high.

[0093] For another example, if the adjacent sub-types in the dimension set to be analyzed are "product A sales" (core data sub-type, first dimension correlation parameter 0.9) and "market competitor price" (correlation data sub-type, second dimension correlation parameter 0.3), the difference value is 0.6, which is in the interval [0.6, 1.0], corresponding to "low correlation", and the preset correlation degree is 0.3, indicating that the correlation between sales and competitor price is low.

[0094] In actual operation, the determination of the dimension correlation parameter needs to consider many factors. For example, for "customer complaint rate" (auxiliary data sub-type) and "product return rate" (correlation data sub-type), the dimension correlation parameter can be set based on indicators such as the co-occurrence frequency of complaint rate and return rate in historical data, lag correlation on time series, etc. When calculating the difference value, if the correlation parameter of the complaint rate is 0.6, the correlation parameter of the return rate is 0.4, and the difference value is 0.2, which is in the interval [0, 0.2), corresponding to "high correlation", the preset correlation degree is 0.9, indicating that the two are closely related in business.

[0095] The system needs to standardize the value logic of the dimension correlation parameter, to ensure that the parameters of different types of sub-types are comparable. For example, the parameter value range of the core data sub-type can be set to [0.7, 1.0], the auxiliary data sub-type to [0.4, 0.7), and the correlation data sub-type to [0.1, 0.4), to reflect the importance difference of different types in analysis, and thus affect the calculation result of the correlation degree.

[0096] The process of determining the dimension correlation degree is crucial for constructing the analysis correlation graph, as it directly determines the weight of the edges between the vertices (sub-types) in the graph, and thus affects the construction of the optimal spanning tree and the determination of the visualization priority. By scientifically and reasonably calculating the dimension correlation degree, the analysis correlation graph can accurately reflect the internal relationship between data sub-types, and the visualization process can be presented in order according to the closeness of data correlation, thereby providing users with logical and focused data analysis results. The entire process requires the system to have accurate recognition ability of data types, deep understanding of business logic, and reasonable design of parameter calculation, to ensure the accuracy and effectiveness of the dimension correlation degree.

[0097] Example 5:

[0098] In generating the visualization results of each sub-type in the set of dimensions to be analyzed in order of the visualization priority order, the generation parameters of each generation node in the visualization priority order need to be identified, and error feature sequences are constructed according to these parameters. The generation parameters cover various attributes of the data sub-type, such as data format (e.g., numerical, textual, time series), data precision (e.g., two decimal places), data dimension (e.g., time dimension, region dimension), data correlation (e.g., correlation strength with other sub-types), etc. Taking “R&D costs of each department in the third quarter of 2024” (core data sub-type) and “R&D personnel education distribution” (auxiliary data sub-type) as examples, the generation parameters of the former may include data time range (July-September 2024), department classification (technology department, product department, etc.), cost composition item (personnel salary, equipment procurement, etc.), and the generation parameters of the latter include education level (bachelor's degree, master's degree, doctorate), personnel quantity statistics method, etc.

[0099] After identifying the generation parameters, the system arranges these parameters in the order of the visualization priority order to construct the error feature sequence. The arrangement order of the parameters in the sequence is strictly consistent with the visualization priority order to ensure the logical coherence of subsequent processing. For example, if the visualization priority order is “R&D costs → education distribution → competitor prices”, the parameter arrangement of the error feature sequence is in turn the generation parameters of R&D costs, the generation parameters of education distribution, and the generation parameters of competitor prices. Each parameter corresponds to a position in the sequence, reflecting its order and importance in the visualization process.

[0100] The error feature sequence is input into the error correction model, and the visualization action output for each generation node is generated through the model. The error correction model is trained based on historical data and machine learning algorithms, which can identify potential errors in the feature sequence and generate corresponding correction actions. For example, when there are problems such as inconsistent data formats (e.g., R&D cost amounts have both ten-thousand-unit and yuan-unit) and missing dimensions (e.g., education distribution is not labeled with statistical time) in the error feature sequence, the model will generate standardized format conversion, supplementary time dimension, etc. according to the preset rules and training experience.

[0101] Taking the “R&D costs” generation node as an example, if there is an error “cost data is not aggregated by department classification” in its generation parameters, the error correction model will identify the deviation of this parameter from the preset standard (which requires department classification), and then generate the visualization action output “re-aggregate data by department dimension”; for the “education distribution” generation node, if the parameter “the number of personnel with master's degree is missing in a certain department”, the model will generate the action output “supplement the data of personnel with master's degree in XX department”.

[0102] After generating the visual action output, the system generates the visual results of each sub-type in sequence according to these actions. The presentation form of the visual results is determined according to the data characteristics and analysis requirements, including column chart, line chart, pie chart, heat map, etc. For example, the "R&D cost" data can generate a column chart divided by department to intuitively show the cost proportion of each department; the "education distribution" can generate a pie chart to show the proportion of personnel at different education levels; the "competitor price" can generate a time series line chart to reflect the price fluctuation trend.

[0103] In the process of generating visual results, the system will strictly follow the instructions of the visual action output to ensure that errors are corrected. For example, when processing R&D cost data, first process the data according to the "classify by department" action, then generate a column chart; when processing education distribution data, first supplement the missing department data, then generate a pie chart, to ensure the accuracy and completeness of the visual results.

[0104] For another example, if the sub-types in the set of dimensions to be analyzed are "Product A sales" (core, priority first), "customer satisfaction" (auxiliary, priority second), and "promotion investment" (associated, priority third). The generation parameters of "sales" include time interval (January-December 2024), regional distribution (East China, North China, etc.), and sales value, but there is an error of "missing data for South China region in the regional dimension"; the generation parameters of "customer satisfaction" include satisfaction score interval and customer group classification, but there is an error of "inconsistent score data format (mixing percentage and five-point system)"; the generation parameters of "promotion investment" include activity time and investment amount, but there is an error of "no product model marked for corresponding activities".

[0105] When constructing the error feature sequence, arrange the generation parameters of these three sub-types in order of priority to form a sequence containing error information such as missing regions, inconsistent formats, and unmarked models. After inputting this sequence into the error correction model, the model generates the action of "supplementing sales data for South China region" for "sales", the action of "unifying to percentage scores" for "satisfaction", and the action of "supplementing activity model annotations for Product A" for "promotion investment".

[0106] The system processes the data according to these actions: adds statistical data for South China region to sales data, unifies satisfaction data to percentage scores, and annotates activity models for Product A in promotion investment data, then generates visual results respectively. For example, sales generates a regional sales column chart with South China region, satisfaction generates a score distribution line chart with unified format, and promotion investment generates a heat map of investment and sales correlation with Product A annotations, so that the visual results can accurately reflect the business situation behind the data and provide users with clear and reliable analysis views.

[0107] In practical applications, the training of the error correction model requires a large amount of historical visualization task data, and through learning the error types and correction methods in the past processing process, the judgment and output ability of the model is continuously optimized. For example, when the error of "missing time dimension" occurs multiple times, the model will strengthen the detection logic of the time parameter, and improve the identification accuracy of similar errors.

[0108] The visualization action output also needs to consider the personalized needs of users and industry specifications. For example, the financial industry may require that the visualization results must contain data source annotations, and the retail industry may pay more attention to the real-time update frequency of sales data. When generating action output, the system will combine industry characteristics and user preset visualization rules to ensure that the results meet the requirements of the actual application scenario.

[0109] It should be noted that in this paper, relational terms such as first and second are used only to distinguish one entity or action from another, and do not necessarily require or imply that there is any such actual relationship or order between these entities or actions. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device.

[0110] Although embodiments of the present application have been shown and described, it will be understood by those having ordinary skill in the art that various changes, modifications, substitutions and alterations can be made hereto without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A natural language based enterprise data visualization analysis method, characterized in that, The method comprises: collecting current enterprise business data to be analyzed and corresponding natural language query text, and obtaining a target analysis dimension according to the enterprise business data and the natural language query text; based on an analysis intention of the natural language query text, dividing data subtypes covered by the target analysis dimension into a core data subtype and an auxiliary data subtype, and determining an associated data subtype in a data subtype not covered by the target analysis dimension, the core data subtype, the auxiliary data subtype, and the associated data subtype forming a set of dimensions to be analyzed; determining a dimension correlation degree between any two adjacent subtypes in the set of dimensions to be analyzed, and constructing an analysis correlation graph of the set of dimensions to be analyzed by taking the subtypes in the set of dimensions to be analyzed as vertices and the dimension correlation degrees as edges connecting the vertices; determining an optimal spanning tree of the analysis correlation graph, and determining a visualization priority order of each subtype in the set of dimensions to be analyzed according to the optimal spanning tree; generating visualization results of each subtype in the set of dimensions to be analyzed in the order of the visualization priority order; determining a dimension correlation degree between any two adjacent subtypes in the set of dimensions to be analyzed comprises: for a first subtype and a second subtype adjacent in the set of dimensions to be analyzed, respectively identifying a type category of the first subtype and the second subtype, the type category including one of the core data subtype, the auxiliary data subtype, and the associated data subtype; determining a first dimension correlation parameter of the first subtype and a second dimension correlation parameter of the second subtype according to the identified type category; calculating a difference value between the first dimension correlation parameter and the second dimension correlation parameter, and determining a dimension correlation degree between the first subtype and the second subtype according to the difference value; determining a dimension correlation degree between the first subtype and the second subtype according to the difference value comprises: identifying a dimension correlation parameter of each subtype in the set of dimensions to be analyzed, and determining an average value of the identified dimension correlation parameters; determining a plurality of difference intervals according to the average value, and determining a target difference interval in which the difference value between the first dimension correlation parameter and the second dimension correlation parameter is located; determining a preset correlation degree corresponding to the target difference interval as the dimension correlation degree between the first subtype and the second subtype. 2.The natural language based enterprise data visualization analysis method of claim 1, wherein, obtaining a target analysis dimension according to the enterprise business data and the natural language query text comprises: counting analysis dimensions in historical analysis tasks that match the enterprise business data according to the natural language query text; grouping each analysis dimension according to semantic features of the natural language query text to form one or more sets of analysis dimensions; determining a target set of analysis dimensions that matches a current analysis intention of the natural language query text in the set of analysis dimensions, and taking dimensions covered by the target set of analysis dimensions as the target analysis dimension. 3.The natural language based enterprise data visualization analysis method of claim 2, wherein, According to the semantic features of the natural language query text in each analysis dimension, the statistical analysis dimensions are classified, including: Obtaining predefined semantic feature standard intervals in the enterprise data management system; For any one of the statistical analysis dimensions, identifying the semantic feature standard interval to which the semantic features of the natural language query text in the analysis dimension belong, and dividing the analysis dimension into the analysis dimension set corresponding to the identified semantic feature standard interval.

4. The natural language based enterprise data visualization analysis method of claim 1, wherein, Based on the analysis intent of the natural language query text, the data subtypes covered by the target analysis dimension are divided into core data subtypes and auxiliary data subtypes, including: Determining the core analysis object and the auxiliary analysis object represented by the analysis intent of the natural language query text; Determining the first data coverage range of the core analysis object in the target analysis dimension, and determining the second data coverage range of the auxiliary analysis object in the target analysis dimension; Determining the data subtypes covered by the first data coverage range in the enterprise business data range as core data subtypes, and determining the data subtypes covered by the second data coverage range in the enterprise business data range as auxiliary data subtypes.

5. The natural language based enterprise data visualization analysis method of claim 1, wherein, Determining the associated data subtypes in the data subtypes not covered by the target analysis dimension, including: Setting a first extension range for the core data subtypes in the target analysis dimension, and setting a second extension range for the auxiliary data subtypes in the target analysis dimension; For the data subtypes not covered by the target analysis dimension, if adjacent to the core data subtypes, selecting the data subtypes of the first extension range as the associated data subtypes; if adjacent to the auxiliary data subtypes, selecting the data subtypes of the second extension range as the associated data subtypes.

6. The natural language based enterprise data visualization analysis method of claim 1, wherein, Generating the visualization results of each subtype in the set of analysis dimensions in the order of the visualization priority order, including: Identifying the generation parameters of each generation node in the visualization priority order, and constructing an error feature sequence according to the identified generation parameters, the arrangement order of each parameter in the error feature sequence being consistent with the visualization priority order; Inputting the error feature sequence into an error correction model to generate visualization action outputs for each generation node through the error correction model; Generating the visualization results of each corresponding subtype according to each generated visualization action output.

7. A natural language based enterprise data visualization analysis system for implementing a natural language based enterprise data visualization analysis method according to any one of claims 1-6, characterized in that, The system includes: A data acquisition unit for collecting current enterprise business data to be analyzed and corresponding natural language query text, and obtaining a target analysis dimension based on the enterprise business data and natural language query text; A dimension division unit for dividing the data subtypes covered by the target analysis dimension into core data subtypes and auxiliary data subtypes based on the analysis intent of the natural language query text, and determining associated data subtypes in the data subtypes not covered by the target analysis dimension, the core data subtypes, auxiliary data subtypes, and associated data subtypes forming a set of analysis dimensions to be analyzed; The association graph construction unit is configured to determine a dimension association degree between any two adjacent subtypes in the set of dimensions to be analyzed, and construct an analysis association graph of the set of dimensions to be analyzed, with the subtypes in the set of dimensions to be analyzed as vertices and the dimension association degrees as edges connecting the vertices. The priority determination unit is configured to determine an optimal spanning tree of the analysis association graph, and determine a visualization priority order of each subtype in the set of dimensions to be analyzed according to the optimal spanning tree. The visualization generation unit is configured to generate visualization results of each subtype in the set of dimensions to be analyzed in the visualization priority order.

8. The system of claim 7, wherein, The visualization generation unit is specifically configured to identify generation parameters of each generated node in the visualization priority order, and construct an error feature sequence according to the identified generation parameters, the arrangement order of each parameter in the error feature sequence being consistent with the visualization priority order. The error feature sequence is input into an error correction model, so as to generate a visualization action output for each generated node through the error correction model; and each subtype is visualized according to each generated visualization action output.

Images