Automated data cross-checking system and automated data cross-checking method

Abstract

An automated data cross-checking method includes executing through a processor: receiving a question string and confirming key field contents of the question string; reading and executing a plurality of agent programs to parse the key field contents; determining whether to perform data cross-check based on key column contents; in response to performing data cross-checking is needed according to the key field contents, searching and reading a plurality of data sets corresponding to the key field contents, executing the agent program to parse the fields of each data set, generating cross-checking instructions and outputting cross-checking prompts accordingly, wherein each data set includes multiple fields; receiving a confirmation string related to the cross-checking prompt; executing the agent program to generate a program code based on the cross-checking instruction and the confirmation string; executing the program code to generate a cross-checking result file; and outputting a cross-checking result string corresponding to the question string based on the cross-checking result file.

Description

Technical Field

[0001] This invention relates to a cross-referencing technique, and more particularly to an automated data cross-referencing system and an automated data cross-referencing method. Prior Technology

[0002] Data analysis tools require professional statistical and data processing experience, demanding a strong background in statistics and precise data processing procedures, making them challenging to implement. Currently, many tools like chatGPT lack data cleaning capabilities, leading to errors. Furthermore, as data volumes increase, businesses face difficulties in cross-referencing data from multiple sources, especially datasets with inconsistent column names or data formats, making cross-referencing via column names extremely difficult.

[0003] Therefore, how to enable users to perform data analysis through dialogue with databases using natural language, automate task breakdown and planning, generate high-quality, low-error-rate code, and complete statistical analysis and chart creation are areas that urgently need improvement. Summary of the Invention

[0004] This invention provides an automated data reconciliation system, including memory and a processor, with the processor coupled to the memory. The memory stores multiple agent programs. The processor receives a query string and confirms the key field content of the query string; reads and executes the agent programs to parse the key field content; determines whether data reconciliation needs to be performed based on the key field content; in response to the need to perform data reconciliation based on the key field content, searches and reads multiple datasets related to the key field content, executes the agent programs to parse the fields of each dataset, generates reconciliation instructions and outputs reconciliation prompts accordingly, wherein each dataset includes multiple fields; receives a confirmation string related to the reconciliation prompts; executes the agent programs to generate code based on the reconciliation instructions and the confirmation string; executes the code to generate a reconciliation result file; and outputs a reconciliation result string corresponding to the query string based on the reconciliation result file.

[0005] The present invention also provides an automated data reconciliation method, comprising, through a processor, performing: receiving a query string and confirming the key field content of the query string; reading and executing multiple agent programs to parse the key field content; determining whether data reconciliation needs to be performed based on the key field content; in response to the need to perform data reconciliation based on the key field content, searching and reading multiple datasets related to the key field content, executing agent programs to parse the fields of each dataset, generating reconciliation instructions and outputting reconciliation prompts accordingly, wherein each dataset includes multiple fields; receiving a confirmation string related to the reconciliation prompts; executing agent programs to generate code based on the reconciliation instructions and the confirmation string; executing the code to generate a reconciliation result file; and outputting a reconciliation result string corresponding to the query string based on the reconciliation result file.

[0006] Based on the above, the automated data reconciliation system and automated data reconciliation method provided by the present invention can enable users to perform data analysis by communicating with databases in natural language, automate task decomposition and planning, generate high-quality, low-error-rate code, and complete statistical analysis and chart drawing. Simple Explanation of the Diagram

[0007] Figure 1 is a schematic diagram of an automated data reconciliation system according to an embodiment of the present invention. Figure 2 is a flowchart illustrating an automated data reconciliation method according to an embodiment of the present invention. Figure 3 is a schematic diagram of a reconciliation result file according to an embodiment of the present invention. Figure 4 is a schematic diagram of a reconciliation result file according to another embodiment of the present invention. Implementation

[0008] The following description will detail some exemplary embodiments of the present invention in conjunction with the accompanying drawings. Component symbols used in the following description are considered identical or similar when appearing in different drawings. These exemplary embodiments are only a part of the present invention and do not disclose all possible implementations disclosed herein. More precisely, these exemplary embodiments are merely examples of the methods and systems within the scope of the present invention's patent application.

[0009] Figure 1 is a schematic diagram of an automated data reconciliation system 1 according to an embodiment of the present invention. The automated data reconciliation system 1 of the present invention includes a memory 11 and a processor 12. The various components and their configuration relationships in the automated data reconciliation system 1 will be introduced first through Figure 1; detailed functions will be disclosed in conjunction with the subsequent embodiment.

[0010] Memory 11 is used to store multiple agent programs. In practice, memory 11 is, for example, static random-access memory (SRAM), dynamic random-access memory (DRAM), or other memory, but the present invention is not limited thereto.

[0011] The processor 12 is coupled to the memory 11 and is used to execute an automated data reconciliation method. In detail, the automated data reconciliation system 1 is coupled to the electronic device 21. The user of the electronic device 21 can input a query string to the automated data reconciliation system 1 through the electronic device 21. After receiving the query string, the automated data reconciliation system 1 executes the automated data reconciliation method through the processor 12 to output a reconciliation result string corresponding to the query string to the electronic device 21.

[0012] In practice, processor 12 can be, for example, a central processing unit (CPU), an application processor (AP), or other programmable general-purpose or special-purpose microprocessors, digital signal processors (DSPs), or other similar devices, integrated circuits, and combinations thereof. The automated data reconciliation system 1 can be implemented by a computer device, such as a desktop computer, laptop computer, tablet computer, workstation, or other computer device with computing capabilities. Memory 11 can be, for example, static random-access memory (SRAM), dynamic random-access memory (DRAM), or other memory types; the invention is not limited thereto.

[0013] Processor 12 is used to execute automated data reconciliation method 2. Figure 2 is a flowchart illustrating an automated data reconciliation method 2 according to an embodiment of the present invention. The automated data reconciliation method 2 of Figure 2 can be executed by processor 12 of the automated data reconciliation system 1 of Figure 1. The automated data reconciliation method 2 includes steps S21 to S29. The flowcharts of the automated data reconciliation system 1 and the automated data reconciliation method 2 will be described below. Please refer to Figures 1 and 2 simultaneously.

[0014] First, in step S21, the processor 12 receives the query string and confirms the content of at least one key field of the query string. Specifically, the user inputs the query string through the input unit of the electronic device 21, which then transmits the query string to the automated data reconciliation system 1, where the processor 12 receives the query string. The query string is in natural language.

[0015] In step S22, the processor 12 reads and executes multiple agent programs to parse the key field content. In one embodiment, each of the multiple agent programs includes a machine learning model that uses machine learning to process and parse the key field content of the query string.

[0016] In step S23, the processor 12 determines whether data cross-referencing needs to be performed based on the content of the key field. Specifically, if the content of the key field is distributed across different datasets, the processor 12 needs to perform data cross-referencing on multiple datasets related to the content of the key field and then generate cross-referencing instructions; conversely, if the content of the key field is distributed across the same dataset, the processor 12 can generate cross-referencing instructions without performing data cross-referencing on the content of the key field.

[0017] In response to the need to perform data verification based on the content of key fields, in step S24, the processor 12 searches for and reads multiple datasets related to the content of key fields, executes the agent program to parse multiple fields of each dataset, generates verification instructions and outputs verification prompts accordingly, wherein each dataset includes multiple fields.

[0018] In response to the fact that data reconciliation is not required based on the content of the key fields, and the content of the key fields is distributed in different datasets, in step S25, the processor 12 reads one dataset from among the multiple datasets that is related to the content of the key field, executes the agent program to parse the multiple fields of the dataset, generates reconciliation instructions and outputs reconciliation prompts accordingly.

[0019] After the reconciliation prompt is generated, the automated data reconciliation system 1 transmits the reconciliation prompt to the electronic device 21. After the user sees the reconciliation prompt through the output unit of the electronic device 21, he / she can respond to the reconciliation prompt by entering a confirmation string through the input unit of the electronic device 21.

[0020] Next, the electronic device 21 transmits the confirmation string to the automated data reconciliation system 1. In step S26, the processor 12 receives the confirmation string related to the reconciliation prompt.

[0021] In step S27, the agent program is executed by the processor 12 to generate code based on the collation instructions and the confirmation string.

[0022] In step S28, the processor 12 executes the program code to generate the reconciliation result file.

[0023] In step S29, the processor 12 outputs the cross-check result string corresponding to the query string based on the cross-check result file. The automated data cross-checking system 1 transmits the cross-check result string to the electronic device 21, and the user can see the cross-check result string corresponding to the query string through the output unit of the electronic device 21.

[0024] The cross-checking results file is stored in memory 11 via processor 12. The automated data cross-checking system 1 also transmits the cross-checking results file to electronic device 21 and stores it in the storage unit of electronic device 21. The cross-checking result string further includes the location of the cross-checking result file stored in the storage unit of electronic device 21.

[0025] Regarding the datasets mentioned in steps S24 and S25, the automated data reconciliation system 1 can obtain them from different locations via the processor 12. In one embodiment, the user can input at least one dataset through the input unit of the electronic device 21 before inputting a query string. The electronic device 21 then transmits the at least one dataset to the automated data reconciliation system 1. The automated data reconciliation system 1 can receive the at least one dataset through the processor 12 and store it in the memory 11. In another embodiment, multiple datasets are pre-stored in an external or internal database 3 connected to the automated data reconciliation system 1. The automated data reconciliation system 1 can connect to the database 3 and receive one or more datasets through the processor 12.

[0026] The following will further explain an implementation example of how the automated data reconciliation system 1 and the automated data reconciliation method 2 need to perform data reconciliation based on the content of key fields. Please refer to Figures 1 and 2 at the same time.

[0027] First, the user inputs the question string "Calculate the Pearson correlation coefficient between 'Humidity' and 'Visibility (km)'?" through the input unit of electronic device 21. Electronic device 21 transmits the question string to automated data reconciliation system 1, which receives the question string through processor 12 (step S21).

[0028] Processor 12 reads the question string "Calculate the Pearson correlation coefficient between 'Humidity' and 'Visibility (km)'?" and begins question parsing. Processor 12 first identifies the relevant fields in the question string and executes multiple agent programs to parse the question, i.e., parse the key field content (step S22). After the question string is parsed by processor 12, processor 12 obtains the key field content, which includes 'Humidity' and 'Visibility (km)'.

[0029] Next, the processor 12 determines whether data reconciliation needs to be performed based on the key field content 'Humidity' and 'Visibility (km)' (step S23).

[0030] Here, it is assumed that the key fields 'Humidity' and 'Visibility (km)' are distributed in different datasets. The processor 12 will determine that a cross-referencing instruction needs to be executed. In response to the need to perform data cross-referencing based on the key fields 'Humidity' and 'Visibility (km)', the processor 12 searches for and reads the datasets related to the key fields (step S24).

[0031] The key field 'Humidity' is related to the field name in the dataset Weather_a.csv, and the key field 'Visibility (km)' is related to the field name in the dataset Weather_b.csv. The contents of the datasets Weather_a.csv and Weather_b.csv are shown below.

[0032] Data set Weather_a.csv address Date … Humidity Yixin Street, Beitun District, Taichung City 2005-12-31 … 0.89 Zhongqing West 1st Street, Beitun District, Taichung City 2006-01-01 … 0.85 … … … …

[0033] Data set Weather_b.csv Address Time … Visibility (km) Yixin Street, Beitun District, Taichung City 2005 / 12 / 31 … 9.982 Zhongqing West 1st Street, Beitun District, Taichung City 2006 / 01 / 01 … 9.9015 … … … …

[0034] Comparing the datasets Weather_a.csv and Weather_b.csv reveals some differences in the format of the field names and data.

[0035] For example, the fields named "address" in both the Weather_a.csv and Weather_b.csv contain addresses, indicating they are identical. Similarly, the fields named "Date" in both Weather_a.csv and Weather_b.csv contain date information.

[0036] Furthermore, the date format in the dataset Weather_a.csv is 2005-12-31, while the date format in the dataset Weather_b.csv is 2005 / 12 / 31, which are different formats. The addresses in the fields of the Weather_a.csv dataset use "Taichung City," while the addresses in the fields of the Weather_b.csv dataset also use "Taichung City," which are also different formats.

[0037] The processor 12 executes an agent program to parse multiple fields in the datasets Weather_a.csv and Weather_b.csv, generating reconciliation instructions and outputting reconciliation prompts accordingly (step S24). It should be noted that the automated data reconciliation system 1 and automated data reconciliation method 2 of this invention can generate reconciliation instructions for fields with different names but similar information.

[0038] In one embodiment, the reconciliation prompt includes a prompt to modify the field name of each field in each dataset. In one embodiment, the reconciliation prompt includes a prompt to modify the field data of each field in each dataset. In one embodiment, the reconciliation prompt includes a prompt to merge at least one of the fields from multiple datasets.

[0039] The following is a verification prompt: question [Reconciliation Required] Suggested analysis workflow for the two datasets Weather_a.csv and Weather_b.csv: [1.] [Confirm field names match] - Check the field names in the data. The address in Weather_a.csv and Weather_b.csv is the same. They should be unified with the same name, Address_address. Both the Date in Weather_a.csv and the Time in Weather_b.csv contain date information, so they should be unified under the same name: Date_Time. [2.] [Standard Date Format] The date format of Weather_a.csv is 2005-12-31, but Weather_b.csv uses 2005 / 12 / 31. To unify them into YYYY-MM-DD format, you can use the pd.to_datetime function to convert them. [3.] [Find the merge key] - Use the Address_address and Date_Time fields as the merged fields. [4.] [Data Preprocessing] - After merging the data, check for any missing values.

[0040] In addition, processor 12 will perform data preprocessing on datasets Weather_a.csv and Weather_b.csv to merge the two datasets into a single dataset Weather_pre.csv, as shown below.

[0041] Data set Weather_pre.csv address Date Humidity … Visibility (km) Yixin Street, Beitun District, Taichung City 2005-12-31 0.89 … 9.982 Zhongqing West 1st Street, Beitun District, Taichung City 2006-01-01 0.85 … 9.9015 … … … … …

[0042] After the reconciliation prompt is generated, the automated data reconciliation system 1 transmits the reconciliation prompt to the electronic device 21. After the user sees the reconciliation prompt through the output unit of the electronic device 21, they can respond to the reconciliation prompt by entering the confirmation string "OK" through the input unit of the electronic device 21. The electronic device 21 transmits the confirmation string "OK" to the automated data reconciliation system 1, and the processor 12 receives the confirmation string "OK" (step S26).

[0043] After receiving the confirmation string "OK", the automated data reconciliation system 1 executes the agent program through the processor 12 to generate program code according to the reconciliation instructions and the confirmation string (step S27).

[0044] The processor 12 executes the code to generate the cross-referencing result file scatterplot_1.png (step S28). Figure 3 is a schematic diagram of the cross-referencing result file scatterplot.png_1 drawn according to an embodiment of the present invention.

[0045] Processor 12 outputs the cross-reference result string corresponding to the query string based on the cross-reference result file scatterplot_1.png (step S29).

[0046] The reconciliation result string is shown below: [Analysis Results:] The Pearson correlation coefficient between 'Humidity' and 'Visibility (km)' is approximately -0.542711. The two are negatively correlated. This result has been saved as a scatter plot in the file 'scatterplot_1.png' in the '. / results' directory.

[0047] The automated data cross-checking system 1 transmits the cross-checking result string to the electronic device 21. The user can view the cross-checking result string corresponding to the query string through the output unit of the electronic device 21. The user can obtain the result corresponding to the query string through the cross-checking result string and can find the cross-checking result file by following the cross-checking result string.

[0048] The following will further explain an embodiment of the automated data reconciliation system 1 and the automated data reconciliation method 2 that does not require data reconciliation based on the content of key fields. Please refer to Figures 1 and 2 at the same time.

[0049] First, the user inputs the query string "Based on existing files, analyze the humidity ('Humidity') in the dataset and find out the maximum humidity?" through the input unit of electronic device 21. Electronic device 21 transmits the query string to automated data reconciliation system 1, which receives the query string through processor 12 (step S21).

[0050] Processor 12 reads the query string "Based on existing files, analyze the humidity ('Humidity') in the dataset and find the maximum humidity?" and begins query parsing. Processor 12 first identifies the relevant fields in the query string and executes multiple agents to parse the query, i.e., parse the key field content (step S22). After being parsed by processor 12, the query string is found to contain the key field 'Humidity'.

[0051] Next, the processor 12 determines whether data reconciliation needs to be performed based on the key field content 'Humidity' (step S23).

[0052] Processor 12 reads Weather_a.csv, a dataset related to the key field 'Humidity', from multiple datasets and determines that no cross-checking instruction is required. An agent program is executed to parse multiple fields of the Weather_a.csv dataset, generate cross-checking instructions, and output cross-checking prompts accordingly (step S25).

[0053] The following is a verification prompt: 'Humidity' is in Weather_a.csv and does not require cross-referencing. The following is the complete process for analyzing the maximum humidity value: Data preprocessing: - Check if there are any missing values ​​in the Humidity field of Weather_a.csv and remove the missing data.

[0054] After the reconciliation prompt is generated, the automated data reconciliation system 1 transmits the reconciliation prompt to the electronic device 21. After the user sees the reconciliation prompt through the output unit of the electronic device 21, they can respond to the reconciliation prompt by entering the confirmation string "OK" through the input unit of the electronic device 21. The electronic device 21 transmits the confirmation string "OK" to the automated data reconciliation system 1, and the processor 12 receives the confirmation string "OK" (step S26).

[0055] After receiving the confirmation string "OK", the automated data reconciliation system 1 executes the agent program through the processor 12 to generate program code according to the reconciliation instructions and the confirmation string (step S27).

[0056] The processor 12 executes the code to generate the cross-referencing result file scatterplot_2.png (step S28). Figure 4 is a schematic diagram of the cross-referencing result file scatterplot_2.png drawn according to another embodiment of the present invention.

[0057] Processor 12 outputs the cross-reference result string corresponding to the query string based on the cross-reference result file scatterplot_2.png (step S29).

[0058] The reconciliation result string is shown below: [Analysis Results:] The Humidity column in Weather_a.csv will have 53 missing values ​​filled with 0. The maximum value for humidity is 1.0.

[0059] The automated data cross-checking system 1 transmits the cross-checking result string to the electronic device 21, and the user can see the cross-checking result string corresponding to the query string through the output unit of the electronic device 21.

[0060] The automated data cross-checking system 1 and automated data cross-checking method 2 provided by this invention can enhance the logic of program generation, possess knowledge of standardized data analysis processes, and improve accuracy. In quantitative evaluation, the comparison object is ChatGPT-4o, and each question is tested in a new window. Thirty data cross-checking test questions were generated based on weather datasets and then manually selected and adjusted. The evaluation indicators used below are the pass rate and the coverage rate. The pass rate is defined as the ability to answer questions correctly without error messages. The coverage rate is defined as the accuracy of answers as determined by human judgment.

[0061] The test example is as follows: Test Project question Answer Data verification and statistical analysis Based on existing records, calculate the average temperature (Temperature (C)) when visibility ('Visibility (km)') is greater than or equal to 10 km and the average temperature (Temperature (C)) when visibility ('Visibility (km)') is less than 10 km. Average temperature for visibility less than 10 km: 9.12 °C Average temperature for visibility of 10 km or more: 14.68 °C Data verification and statistical analysis Based on existing records, what are the average perceived temperature (°C) when the wind speed (km / h) is 20 km / h or higher, and the average perceived temperature (°C) when the wind speed is below 20 km / h? Average wind chill temperature when wind speed is below 20 km / h: 11.05 °C Average wind chill temperature at wind speeds of 20 km / h or higher: 9.32 °C Data verification, statistical analysis, statistical charts Based on existing records, calculate the average 'Wind Speed ​​(km / h)' when 'Temperature (C)' is below 10 degrees and the average 'Wind Speed ​​(km / h)' when 'Temperature (C)' is above 10 degrees and present the results in a graph. When the temperature (C) is below 10 degrees, the average wind speed (km / h) is 11.068 km / h. When the temperature (C) is above 10 degrees, the average wind speed (km / h) is 10.602 km / h.

[0062] Take the question "According to the existing files, calculate the average temperature ('Temperature (C)') when the visibility ('Visibility (km)') is greater than or equal to 10 km and the average temperature ('Temperature (C)') when the visibility ('Visibility (km)') is less than 10 km)" as an example. Since ChatGPT-4o does not have the function of data cleaning, the answer obtained is "According to the existing data, when the visibility ('Visibility (km)') is greater than or equal to 10 km, the average temperature ('Temperature (C)') is 14.67°C, and when the visibility is less than 10 km, the average temperature is 9.14°C." After comparing the obtained answer with the existing files, it can be found that although the successful execution rate of ChatGPT-4o is the same as that of the present invention, both reaching 100%, in terms of the logical accuracy rate, the technology of the present invention can reach 76%, while ChatGPT-4o only reaches 70%.

[0063] During the test, it was found that ChatGPT-4o could not automatically identify and convert data with similar meanings during data reconciliation, and would not actively perform data cleaning after reconciliation. The automated data reconciliation system 1 and automated data reconciliation method 2 provided by the present invention add a prompt engineering task planning tool, which enhances the data understanding ability, adds task decomposition, and knowledge prompts in the field of data analysis, having the ability of data cleaning, and constructs a complete problem analysis process.

[0064] Take the question "What is the Pearson correlation coefficient between 'Humidity' and 'Visibility (km)'?" as an example. Since ChatGPT-4o does not automatically identify and convert the format of field data (for example, it does not recognize that '臺' and '台' are only differences in format case and does not convert '臺' to '台'), the reconciliation was not successful. The answer obtained by ChatGPT-4o is "The calculated result of the Pearson correlation coefficient between Humidity and Visibility (km) is NaN, which may be because there is no matching data for these two fields during the merging of the two files", that is, no reconciliation result can be generated. Although the successful execution rate of ChatGPT-4o is the same as that of the present invention, both reaching 100%, in terms of the logical accuracy rate, ChatGPT-4o is not good.

[0065] In addition, the automated data reconciliation system and automated data reconciliation method provided by this invention are significantly different from ChatGPT-4o in that the automated data reconciliation system provided by this invention is a ground-based model, and the modules can be replaced according to the user's needs. This is a customization flexibility that ChatGPT-4o does not have.

[0066] In summary, the automated data cross-referencing system and method provided by this invention enable users to directly interact with the database using natural language to perform data analysis. It requires professional statistical and data processing knowledge, automates task decomposition and planning, generates high-quality, low-error-rate code, completes statistical analysis and chart generation, and returns the results to the user without requiring unnecessary manual intervention. It has a low barrier to entry and improves work efficiency. Furthermore, this invention supports cross-referencing of multiple datasets, automatically understanding and reconciling data based on field content, achieving full automation without human intervention.

[0067] 1: Automated data reconciliation system 11: Memory 12: Processor 2: Automated Data Reconciliation Method 21: Electronic devices 3: Database S21~S29: Steps scatterplot.png_1, scatterplot.png_2: Reconciliation results files

Claims

1. An automated data reconciliation system, comprising: Memory, used to store multiple agent programs; The processor, coupled to the memory, is configured to: receive a query string and confirm the key field content of the query string; read and execute the agent programs to parse the key field content; determine whether data cross-checking needs to be performed based on the key field content; in response to the need to perform data cross-checking, search and read multiple datasets related to the key field content, wherein each dataset includes multiple fields, execute the agent programs to parse the fields in each dataset, generate cross-checking instructions, and output cross-checking prompts accordingly; receive a confirmation string related to the cross-checking prompts; execute the agent programs to generate code based on the cross-checking instructions and the confirmation string; execute the code to generate a cross-checking result file; and output a cross-checking result string corresponding to the query string based on the cross-checking result file. In response to the need not to perform the data cross-check, the processor further searches and reads one of the datasets related to the content of the key field, executes the agents to parse the fields of the dataset, generates the cross-check instruction, and outputs the cross-check prompt accordingly.

2. The automated data reconciliation system as claimed in claim 1, wherein the processor is further configured to receive the data sets and store them in the memory.

3. The automated data reconciliation system as claimed in claim 1, wherein the processor is further configured to receive pre-stored data sets from a database.

4. The automated data reconciliation system as described in claim 1, wherein each of the fields includes a field name and field data.

5. The automated data reconciliation system as described in claim 4, wherein the reconciliation prompt includes a prompt to modify the name of the field for each of the fields in each of the datasets.

6. The automated data reconciliation system as described in claim 4, wherein the reconciliation prompts include a prompt for modification of the data in each field of each of the datasets.

7. The automated data reconciliation system as described in claim 4, wherein the reconciliation prompt includes a merge prompt for at least one of the fields of the datasets.

8. The automated data reconciliation system as described in claim 1, wherein the agents include machine learning models.

9. An automated data reconciliation method, comprising: Executed by the processor: Receive the query string and confirm the content of the key fields in the query string; The automated data reconciliation method further includes: reading and executing multiple agents to parse the content of the key field; determining whether data reconciliation needs to be performed based on the content of the key field; in response to the need to perform data reconciliation, searching and reading multiple datasets related to the content of the key field, wherein each dataset includes multiple fields, executing the agents to parse the fields of each dataset, generating reconciliation instructions and outputting reconciliation prompts accordingly; receiving a confirmation string related to the reconciliation prompts; executing the agents to generate code based on the reconciliation instructions and the confirmation string; executing the code to generate a reconciliation result file; and outputting a reconciliation result string corresponding to the query string based on the reconciliation result file. In response to the need not to perform data reconciliation, the automated data reconciliation method further includes: reading one of the datasets through the processor, executing the agents to parse the fields of the dataset, generating the reconciliation instructions and outputting the reconciliation prompts accordingly.

10. The automated data reconciliation method as described in claim 9 further includes: The processor receives these data sets and stores them in memory.

11. The automated data reconciliation method as described in claim 9 further includes: The processor receives these pre-stored data sets from the database.

12. The automated data reconciliation method as described in claim 9, wherein each of the fields includes a field name and field data.

13. The automated data cross-referencing method as described in claim 12, wherein the cross-referencing prompt includes a prompt to modify the field name of each of the fields in each of the datasets.

14. The automated data cross-referencing method as described in claim 12, wherein the cross-referencing prompts include a prompt for modification of the data in each of the fields of each of the datasets.

15. The automated data cross-referencing method as described in claim 12, wherein the cross-referencing prompt includes a merge prompt for at least one of the fields of the datasets.

16. The automated data reconciliation method as described in claim 9, wherein the agents include machine learning models.

Images