Data story interpretation method and device based on large language model and related equipment

Through large language models, we obtain and analyze the insight direction in the data story and generate analysis reports, solving the problem of low intelligence in the data story interpretation, and achieving more efficient data analysis and interpretation.

CN118964643APending Publication Date: 2024-11-15DATACANVAS LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410980688.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

In the existing technology, the degree of intelligent interpretation of data stories is relatively low, and it depends on the analysis level of technicians, and lacks automation and intelligent processing.

Method used

Obtain information about the original data set, topics and data facts through large language models, determine multiple insights, and generate analysis reports to interpret data stories, including filtering, scoring, and reconstructing analysis reports to improve interpretation intelligence.

Benefits of technology

It improves the intelligent interpretation ability of data stories, provides more interpretation methods, reduces dependence on technicians, and enhances the automation and efficiency of data analysis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118964643A_ABST
    Figure CN118964643A_ABST
Patent Text Reader

Abstract

The invention provides a data story interpretation method and device based on a large language model and related equipment, and relates to the technical field of artificial intelligence large language models and cue engineering, and the method comprises the steps: S1, obtaining an original data set and a theme corresponding to a to-be-analyzed data story, and a plurality of data facts included in the to-be-analyzed data story, and related information corresponding to each data fact, wherein the related information is used for describing attributes of the data facts; s2, determining a plurality of insight directions based on the theme, the plurality of data facts and the related information corresponding to each data fact through the large language model; and S3, through the large language model, based on each insight direction in the plurality of insight directions, an analysis report of each insight direction is generated for the original data set and the plurality of data facts, and the analysis report is used for interpreting the to-be-analyzed data story. The intelligent degree of data story interpretation is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence large language models and prompt word engineering technology, and in particular to a data story interpretation method, device and related equipment based on a large language model. Background Art

[0002] "Large language model" refers to a large-scale language model (LLM), which is a language model with a large parameter scale. It is designed to understand and generate human language. It is trained with a large amount of text data and can perform a wide range of tasks including text summarization, translation, sentiment analysis, etc. The rapid development of large language model technology has brought new opportunities for economic and social development. How to realize the practical application of large language models is an issue that needs to be studied urgently.

[0003] At present, in the scenario of processing and analyzing large amounts of data, large language models are usually used to expand the data to be analyzed, generate new data facts, and combine the new data facts into a complete data story to show the overall situation of the data to be analyzed through the data story.

[0004] However, data stories can only directly display the expanded data facts. Further interpretation of the data stories needs to rely on the analysis level of technical personnel, and the degree of intelligence of the interpretation of data stories is low. Summary of the invention

[0005] The embodiments of the present invention provide a data story interpretation method, apparatus and related equipment based on a large language model to solve the problem of low intelligence level of data story interpretation in the prior art.

[0006] To solve the above problems, the present invention is achieved as follows:

[0007] In a first aspect, an embodiment of the present invention provides a data story interpretation method based on a large language model, comprising:

[0008] Step S1, obtaining an original data set corresponding to a data story to be analyzed, a theme, a plurality of data facts included in the data story to be analyzed, and related information corresponding to each data fact, wherein the related information is used to describe the attributes of the data fact;

[0009] Step S2: determining multiple insight directions based on the topic, the multiple data facts, and the relevant information corresponding to each data fact by using the large language model;

[0010] Step S3: Generate an analysis report for each insight direction based on the original data set and the multiple data facts through the large language model, wherein the analysis report is used to interpret the data story to be analyzed.

[0011] In one embodiment, the multiple insight directions include a first insight direction, which is determined by the large language model based on the topic, one or more data facts among the multiple data facts, related information corresponding to the data facts, and a determined insight direction, and the determined multiple insight directions include an insight direction determined by the large language model based on the topic, the multiple data facts and the related information corresponding to each data fact.

[0012] In one embodiment, the multiple insight directions include a second insight direction, which is determined by the large language model based on the multiple determined insight directions, and the multiple determined insight directions include insight directions determined by the large language model based on the topic, the multiple data facts and the related information corresponding to each data fact.

[0013] In one embodiment, after step S3, the method further includes:

[0014] Step S4: screening multiple analysis reports associated with a target data fact from the analysis reports of the multiple insight directions by using the large language model, wherein the target data fact is one of the multiple data facts;

[0015] Step S5: Screening the multiple analysis reports using the large language model to obtain N target analysis reports, where N is a positive integer greater than 1;

[0016] Step S6: reconstruct the N target analysis reports through the large language model to obtain target analysis results, and the target analysis results are used to interpret the data stories to be analyzed based on the target data facts.

[0017] In one embodiment, step S5 includes:

[0018] Step S51: scoring the multiple analysis reports using the large language model to obtain scores corresponding to the multiple analysis reports;

[0019] Step S52: Using the large language model, set the N analysis reports with the highest scores among the multiple analysis reports as the N target analysis reports.

[0020] In one embodiment, step S51 includes:

[0021] Step S511, calculating at least one of the repetition, data relevance and word order disorder corresponding to the plurality of analysis reports by using the large language model, wherein the data relevance is used to characterize the relevance between the analysis report and the topic;

[0022] Step S512: Obtain scores corresponding to the plurality of analysis reports by calculating, using the large language model, at least one of the corresponding repetition, data value, and word order disorder.

[0023] In one embodiment, after step S6, the method further includes:

[0024] Step S7: Generate a graphic report and / or language digital broadcast of the target analysis result through the large language model, and display the graphic report and / or language digital broadcast.

[0025] In a second aspect, an embodiment of the present invention further provides a data story interpretation device based on a large language model, comprising:

[0026] An acquisition module, used to acquire an original data set corresponding to a data story to be analyzed, a theme, a plurality of data facts included in the data story to be analyzed, and related information corresponding to each data fact, wherein the related information is used to describe the attributes of the data fact;

[0027] a determination module, configured to determine a plurality of insight directions based on the topic, the plurality of data facts, and relevant information corresponding to each of the data facts by using the large language model;

[0028] A generation module is used to generate an analysis report for each insight direction based on the original data set and the multiple data facts for each of the multiple insight directions through the large language model, wherein the analysis report is used to interpret the data story to be analyzed.

[0029] In a third aspect, an embodiment of the present invention further provides an electronic device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the steps of the data story interpretation method based on a large language model as described in the first aspect above are implemented.

[0030] In a fourth aspect, an embodiment of the present invention further provides a readable storage medium for storing a program, which, when executed by a processor, implements the steps in the data story interpretation method based on a large language model as described in the first aspect above.

[0031] In a fifth aspect, an embodiment of the present invention further provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the steps in the data story interpretation method based on a large language model as described in the first aspect above.

[0032] In an embodiment of the present invention, the original data set corresponding to the data story to be analyzed, the theme, the multiple data facts included in the data story to be analyzed, and the related information corresponding to each data fact are obtained, and the related information is used to describe the attributes of the data fact; multiple insight directions are determined based on the theme, the multiple data facts, and the related information corresponding to each data fact through a large language model; based on each of the multiple insight directions, the large language model generates an analysis report for each insight direction for the original data set and the multiple data facts, and the analysis report is used to interpret the data story to be analyzed. In this way, multiple insight directions are determined through a large language model, and an analysis report for each insight direction is generated for the original data set and the multiple data facts based on the multiple insight directions. The data story to be analyzed is interpreted through the analysis report, which provides more interpretation methods, thereby improving the intelligence level of data story interpretation. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0034] Figure 1 is a flow chart of a data story interpretation method based on a large language model provided by an embodiment of the present invention;

[0035] Figure 2 It is one of the schematic diagrams for determining the insight direction provided by the embodiment of the present invention;

[0036] Figure 3 This is the second schematic diagram of determining the insight direction provided by the embodiment of the present invention;

[0037] Figure 4 This is the third schematic diagram of determining the insight direction provided by the embodiment of the present invention;

[0038] Figure 5 is a flow chart of generating target analysis results provided by an embodiment of the present invention;

[0039] Figure 6 It is a structural diagram of a data story interpretation device based on a large language model provided by an embodiment of the present invention;

[0040] Figure 7 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] The "large language model" described in the present invention refers to a large language model (LLM), which is a language model with a large parameter scale, designed to understand and generate human language, trained with a large amount of text data, and can perform a wide range of tasks including text summarization, translation, sentiment analysis, etc. Among them, the large language model can be a large language model such as GPT4, GLM-4 or Llama-3.

[0043] The "data facts" mentioned in the present invention refer to the real data information contained in the data set of the data to be analyzed. The data facts mentioned in the present invention also include the visual display of this information, that is, the data facts are displayed through various charts and auxiliary information on the charts, such as labels, indicator marks, text descriptions, etc., to intuitively display the data facts. Among them, the data facts in the present invention can be expanded based on the data to be analyzed through a large language model, and the expanded data facts are data that conform to the actual situation.

[0044] The "data story" described in the present invention refers to an orderly connection of a series of logically related data facts around a theme. Among them, the data story can be obtained by coherently expressing multiple data facts through pre-set conjunctions or connecting sentences. Through the data story, it is possible to display the data facts obtained by expanding the large language model from multiple dimensions.

[0045] The “original dataset” described in the present invention is a dataset used to expand a large language model to obtain data facts and data stories. The multiple data facts included in the data story are all obtained by expanding the original dataset with a large language model.

[0046] The "title" mentioned in the present invention refers to the theme of multiple data facts included in the data story. The multiple data facts included in the data story can be obtained by expanding the original data set based on the theme by a large language model. In a data story, the theme of each data story is the same, and the data story is obtained by combining data facts of the same theme.

[0047] The "relevant information" described in the present invention is used to describe the attributes of data facts. It can be the descriptive information (desc) about the data facts generated while expanding the large language model to obtain the data facts, and / or the data field information (dataframe) in the original data set that the data facts depend on.

[0048] The "insight direction" (InsightDir) mentioned in the present invention refers to a descriptive sentence used to instruct the big language model to analyze the data story and / or the data facts included in the data story. The big language model can analyze the data story and / or the data story based on the insight direction to obtain conclusions or suggestions that are consistent with the insight direction.

[0049] See also Figure 1 , Figure 1 is a flow chart of a data story interpretation method based on a large language model provided by an embodiment of the present invention. Figure 1 As shown, the following steps are included:

[0050] Step S1, obtaining an original data set corresponding to a data story to be analyzed, a theme, a plurality of data facts included in the data story to be analyzed, and related information corresponding to each data fact, wherein the related information is used to describe the attributes of the data fact;

[0051] Step S2: determining multiple insight directions based on the topic, the multiple data facts, and the relevant information corresponding to each data fact by using the large language model;

[0052] Step S3: Generate an analysis report for each insight direction based on the original data set and the multiple data facts through the large language model, wherein the analysis report is used to interpret the data story to be analyzed.

[0053] In an embodiment of the present invention, the original data set corresponding to the data story to be analyzed, the theme, the multiple data facts included in the data story to be analyzed, and the related information corresponding to each data fact are obtained, and the related information is used to describe the attributes of the data fact; multiple insight directions are determined based on the theme, the multiple data facts, and the related information corresponding to each data fact through a large language model; based on each of the multiple insight directions, the large language model generates an analysis report for each insight direction for the original data set and the multiple data facts, and the analysis report is used to interpret the data story to be analyzed. In this way, multiple insight directions are determined through a large language model, and an analysis report for each insight direction is generated for the original data set and the multiple data facts based on the multiple insight directions. The data story to be analyzed is interpreted through the analysis report, which provides more interpretation methods, thereby improving the intelligence level of data story interpretation.

[0054] In an embodiment of the present invention, an original data set corresponding to a data story to be analyzed, a theme, a plurality of data facts included in the data story to be analyzed, and related information corresponding to each data fact are obtained. Specifically, an acquisition code may be pre-set, and the original data set corresponding to the data story to be analyzed, a theme, a plurality of data facts included in the data story to be analyzed, and related information corresponding to each data fact are obtained through the acquisition code.

[0055] For example, a dataset related to bank customer information (Customer-Records) generates data stories by expanding data facts, and obtains three data facts about "customer churn analysis", which are expressed in JSON format as follows:

[0056]

[0057]

[0058] In an embodiment of the present invention, a large language model is used to generate an analysis report for each insight direction based on the original data set and multiple data facts for each insight direction in multiple insight directions. Specifically, analysis prompt words are constructed, and the analysis prompt words are used to instruct the large language model to generate an analysis report for each insight direction based on the original data set and multiple data facts for each insight direction. Among them, the analysis prompt words may include data fact insight direction (InsightDir), thought (Thought), code (Code) required to execute insight, and / or insight analysis report (Report) obtained based on the results after code execution.

[0059] Exemplarily, the analysis prompt words constructed based on the subject, multiple data facts, and the relevant information corresponding to each data fact are as follows:

[0060] as follows:

[0061]

[0062]

[0063] After obtaining the analysis prompt words, the analysis prompt words are used to instruct the large model to generate the following insight directions (Insight Directions 1 to 5):

[0064] Insight direction 1:

[0065]

[0066] Insight direction 2

[0067]

[0068]

[0069] Insight three:

[0070]

[0071] Insight direction 4:

[0072]

[0073] Insight direction five:

[0074]

[0075] After obtaining the insight direction through the big language model, the code sample content of generating the analysis report for each insight direction based on the original data set and multiple data facts through the big language model is as follows:

[0076]

[0077] The output of the code that provides insights by running the above code is as follows: The point-biserial correlation coefficient between credit score and customer churn is: -0.1068, and the P value is: 0.0074.

[0078] Then, the big language model generates insight prompt words based on the topic, multiple data facts, and the relevant information corresponding to each data fact. The insight prompt words are used to instruct the big model to generate a report. The following is an example of the report content: [

[0080] {

[0081] "Report":"The point biserial correlation coefficient between credit score and customer churn is -0.1068, which indicates that there is a certain degree of negative correlation between the two. That is, customers with lower credit scores are more likely to churn, while customers with higher credit scores are relatively less likely to churn.",

[0082] } ]

[0084] In an embodiment of the present invention, a large language model is used to determine multiple insight directions based on a topic, multiple data facts, and related information corresponding to each data fact. Specifically, an insight prompt word is generated based on a large language model based on a topic, multiple data facts, and related information corresponding to each data fact. The insight prompt word instructs the large language model to generate information including insight directions, so as to determine the insight directions.

[0085] The insight prompt words include descriptive information (desc) of the data facts and / or thoughts (thought) on the insight direction, etc.

[0086] In one embodiment, multiple insight directions are determined based on a topic, multiple data facts, and related information corresponding to each data fact through a large language model. The insight directions may be generated based on a topic, one or more data facts, and related information of the data facts through a large language model.

[0087] For example, Figure 2 and Figure 3 As shown, the data story includes data facts 1 to 5. In one embodiment, the large language model determines the insight direction based on the topic, a data fact, and the relevant information of the data fact. Specifically, the large language model generates an insight prompt word based on the topic, data fact 1, the description information of data fact 1, and the data field information of data fact 1, and the insight prompt word instructs the large language model to generate the insight direction 1 corresponding to data fact 1.

[0088] Furthermore, if Figure 4 As shown, the data facts include "the correlation between credit score and customer churn", "the distribution of customer churn in different regions", and "the impact of customer satisfaction on customer churn". Based on the data fact "the impact of customer satisfaction on customer churn", the big language model determines the new insight direction "whether customer satisfaction is the biggest factor affecting customer churn".

[0089] In another embodiment, the large language model determines the insight direction based on the topic, multiple data facts, and data fact-related information. Specifically, the large language model generates insight prompt words based on the topic, data fact two and data fact three, descriptive information of data fact two and data fact three, and data field information of data fact two and data fact three, and then the large language model generates insight direction 23 corresponding to data fact two and data fact three based on the insight prompt words.

[0090] In an embodiment of the present invention, a large language model is used based on a topic, multiple data facts, and related information corresponding to each data fact to achieve determination of multiple insight directions.

[0091] In one embodiment, the multiple insight directions include a first insight direction, which is determined by the large language model based on the topic, one or more data facts among the multiple data facts, related information corresponding to the data facts, and a determined insight direction, and the determined multiple insight directions include an insight direction determined by the large language model based on the topic, the multiple data facts and the related information corresponding to each data fact.

[0092] In an embodiment of the present invention, the insight direction can also be determined through existing insight directions and data facts, specifically, a first insight direction is determined through a large language model based on the topic, one or more data facts among multiple data facts, related information corresponding to the data facts, and an already determined insight direction, thereby realizing more ways to determine the insight direction.

[0093] Among them, the first insight direction is determined by the large language model based on the topic, one or more data facts among multiple data facts, relevant information corresponding to the data facts, and a determined insight direction. Specifically, a new insight prompt word is generated by the large language model based on the topic, one or more data facts among multiple data facts, relevant information corresponding to the data facts, and a determined insight direction, and the new insight prompt word instructs the large language model to generate information on the first insight direction.

[0094] It should be noted that in the embodiments of the present invention, it is necessary to determine a new insight direction through existing insight directions, and there is at least a first determined insight direction among the existing insight directions. This insight direction must be determined through a large language model based on the topic, one or more data facts, and related information corresponding to the data facts. On the basis of this insight direction, a first insight direction is determined through a large language model based on the topic, one or more data facts among multiple data facts, related information corresponding to the data facts, and an already determined insight direction.

[0095] For example, Figure 3 As shown, the data story includes data facts 1 to 5. After the big language model determines the insight direction 12 based on the topic, data fact 1 and data fact 2, and the related information of data fact 1 and data fact 2, the big language model generates a new insight prompt word based on the topic, data fact 3 and insight direction 12, and the new insight prompt word indicates that the big language model generates information of insight direction 123.

[0096] Furthermore, if Figure 4 As shown, the data facts include "the correlation between credit score and customer churn", "the distribution of customer churn in different regions" and "the impact of customer satisfaction on customer churn". Based on the data facts "the correlation between credit score and customer churn" and "the distribution of customer churn in different regions", the big language model has determined the insight direction of "exploring the differences in the impact of credit scores on customer churn in different regions". Based on the determined insight direction of "exploring the differences in the impact of credit scores on customer churn in different regions" and the data fact "the impact of customer satisfaction on customer churn", the big language model further determines a new insight direction of "which has a greater impact on customer churn, customer satisfaction or credit scores in different regions?".

[0097] In one embodiment, the multiple insight directions include a second insight direction, which is determined by the large language model based on the multiple determined insight directions, and the multiple determined insight directions include insight directions determined by the large language model based on the topic, the multiple data facts and the related information corresponding to each data fact.

[0098] In an embodiment of the present invention, the insight direction may also be determined only through existing insight directions, specifically by determining the second insight direction based on the topic and the determined multiple insight directions through a large language model, thereby achieving more ways to determine the insight direction.

[0099] Among them, the second insight direction is determined by the large language model based on the topic and multiple determined insight directions. Specifically, a new insight prompt word is generated by the large language model based on the topic, one or more data facts among multiple data facts, related information corresponding to the data facts, and a determined insight direction, and the new insight prompt word instructs the large language model to generate information on the first insight direction.

[0100] It should be noted that in the embodiments of the present invention, it is necessary to determine a new insight direction through existing insight directions, and there is at least a first determined insight direction among the existing insight directions. This insight direction must be determined through a large language model based on the topic, one or more data facts, and related information corresponding to the data facts. On the basis of this insight direction, a first insight direction is determined through a large language model based on the topic, one or more data facts among multiple data facts, related information corresponding to the data facts, and an already determined insight direction.

[0101] For example, Figure 2 As shown, the data story includes data facts 1 to 5. After the big language model determines insight directions 12 and 23 based on the topic, data fact 1, data fact 2 and data fact 3, and the relevant information of data fact 1, data fact 2 and data fact 3, the big language model generates new insight prompt words based on the topic, insight direction 12 and insight direction 23, and the new insight prompt words instruct the big language model to generate information on insight direction 123.

[0102] Furthermore, if Figure 4As shown, the data facts include "the correlation between credit score and customer churn", "customer churn distribution in different regions" and "the impact of customer satisfaction on customer churn". The big language model has determined the insight direction "whether customer satisfaction is the biggest factor affecting customer churn" based on the data fact "the impact of customer satisfaction on customer churn", and the big language model has determined the insight direction "exploring the differences in the impact of credit scores on customer churn in different regions" based on the data facts "the correlation between credit score and customer churn" and "the distribution of customer churn in different regions". At this point, the big language model determines a new insight direction "whether customer satisfaction is the biggest factor affecting customer churn" based on the determined insight directions "exploring the differences in the impact of credit scores on customer churn in different regions" and "whether customer satisfaction is the biggest factor affecting customer churn".

[0103] In one embodiment, multiple insight directions can be determined based on different arrangements or combinations, such as jointly determining a new insight direction by combining multiple adjacent data facts or insight directions, or jointly determining a new insight direction by combining any multiple data facts or insight directions, or randomly sampling multiple data facts or insight directions to jointly determine a new insight direction.

[0104] Each form of combination can determine multiple insight directions, and the large language model can determine multiple random insight directions based on a combination. In the embodiment of the present invention, only one insight direction is selected as the insight direction corresponding to the combination. It can be understood that for a data fact, the large language model can determine multiple different insight directions related to it.

[0105] For example, Figure 2 and Figure 3 As shown, the data story includes data facts 1 to 5. At this time, the new insight direction is determined by combining adjacent data facts, and insight directions 12, 23, 34 and 45 can be obtained; if the new insight direction is determined by combining adjacent insight directions, insight directions 123, 234 and 345 can be obtained.

[0106] It should be noted that Figure 2 and Figure 3 It further demonstrates the possibility of partial combination. Technicians can set different combinations according to actual needs to obtain insights that meet their needs.

[0107] Step S4: screening multiple analysis reports associated with a target data fact from the analysis reports of the multiple insight directions by using the large language model, wherein the target data fact is one of the multiple data facts;

[0108] Step S5: Screening the multiple analysis reports using the large language model to obtain N target analysis reports, where N is a positive integer greater than 1;

[0109] Step S6: reconstruct the N target analysis reports through the large language model to obtain target analysis results, and the target analysis results are used to interpret the data stories to be analyzed based on the target data facts.

[0110] In an embodiment of the present invention, after obtaining analysis reports of multiple insight directions through a large language model, the analysis reports of the multiple insight directions are screened to obtain multiple analysis reports associated with the target data, and then screened and reconstructed to generate target analysis results, thereby achieving target analysis results that interpret data stories from multiple insight directions and improving the interpretation effect of the target analysis results.

[0111] In an embodiment of the present invention, a large language model is used to screen multiple analysis reports associated with target data facts from analysis reports of multiple insight directions. The insight directions corresponding to the multiple analysis reports are insight directions associated with the target data facts, which may be directly determined by the target data facts through the large language model, or indirectly determined by the target data facts through the large language model.

[0112] In an embodiment of the present invention, N target analysis reports are obtained by screening multiple analysis reports through a large language model. N target analysis reports can be randomly selected after deleting duplicate analysis reports, or the analysis reports can be scored according to preset rules, and the N analysis reports with the highest scores are selected as the target analysis reports to further improve the interpretation effect of the target analysis results.

[0113] In an embodiment of the present invention, N target analysis reports are reconstructed based on target data facts through a large language model to obtain target analysis results. Reconstruction prompt words may be set in advance, and the content to be output by the large language model may be set in the reconstruction prompt window, including, for example, thoughts on the data facts, whether the data facts can be analyzed for insight, insight analysis results on the data facts, suggestions on the insight analysis results of the data facts, etc.

[0114] Exemplarily, multiple analysis reports associated with target data facts are screened from analysis reports of multiple insight directions by using a large language model as follows: [

[0116] {"InsightDir":"Explore the impact of regional differences on customer satisfaction and customer churn",

[0117] "Report":"""The churn rates for different regions are as follows (expressed as a percentage): Exited NotChurnedChurned Geography

[0118] France 83.825289 16.174711 Germany 67.556796 32.443204 Spain83.32660516.673395, from the data, we can see that Germany has the highest customer churn rate, followed by France and Spain. In terms of satisfaction, the distribution between regions is relatively even, with no significant differences. In terms of customer churn rate, Germany's customer churn rate is significantly higher than France and Spain.

[0119] "

[0120] },

[0121] {"InsightDir":"Is customer satisfaction the biggest factor affecting customer churn?"

[0122] "Report":"""The relationship between customer satisfaction score and customer churn rate is as follows:

[0123] Satisfaction score 1: Churn rate is about 20.03%

[0124] Satisfaction score 2: Churn rate is about 21.80%

[0125] Satisfaction score 3: Churn rate is about 19.64%

[0126] Satisfaction score 4: Churn rate is about 20.62%

[0127] Satisfaction score 5: Churn rate is about 19.81%

[0128] From these data, although the churn rates of customers with different satisfaction scores are different, these differences are not significant. This suggests that satisfaction scores may not be the main factor affecting customer churn.

[0129] Based on the analysis of different factors, here are some results that are significantly correlated with customer churn rate:

[0130] Age: The churn rate of customers in the age group of 41-50 years old increases significantly, especially the churn rate of customers between the ages of 51-60 years old is the highest, reaching about 56.21%. This shows that age is an important influencing factor.

[0131] Credit score: Customers with very low credit scores have the highest churn rate at 32.28%, while customers with higher credit scores have a relatively low churn rate. Credit score has some impact on customer churn, but not as significant as age.

[0132] Account balance: The churn rate of customers with a balance of 0 is relatively low (13.82%), while the churn rate of customers with a balance of more than 200,000 is as high as 54.55%. This shows that customers with higher balances have a higher churn rate.

[0133] Number of products: The churn rate of customers with 3 products reaches 82.71%, while the churn rate of customers with 4 products reaches 100%. In contrast, the churn rate of customers with 1 or 2 products is lower. This shows that the number of products is a significant factor affecting customer churn.

[0134] Active member: The churn rate of inactive members is 26.87%, while the churn rate of active members is lower, at 14.27%. Active status significantly affects customer churn.

[0135] In summary, while there is a correlation between customer satisfaction and churn rate, other factors such as age, account balance, number of products, and active membership status have a more significant impact on customer churn. """

[0136] }]

[0137] Exemplarily, the above target analysis report is reconstructed to refine the report content and obtain the target analysis results. Specifically, the following results are obtained:

[0138] {

[0139] "Thinking":"Our data analysis shows that multiple factors affect customer churn rate, and we need to pay special attention to those factors that are significantly correlated with high churn rate.",

[0140] "Data Facts Insight": "The customer churn rate in Germany is 32.44%, significantly higher than around 16% in France and Spain. In addition, the churn rate for customers with 3 products reaches 82.71%, and the churn rate for customers with 4 products is 100%. The churn rate for inactive members is 26.87%, and the churn rate for active members is 14.27%.

[0141] "

[0142] ,

[0143] "Conclusion":"Specific regions, number of products and customer activity are key factors affecting customer churn.

[0144] "

[0145] ,

[0146] "Suggestion": "Customize customer retention strategies for regions and groups with high churn rates, increase customer activity, and optimize product launch strategies."

[0147] }

[0148] In an embodiment of the present invention, N target analysis reports are reconstructed through a large language model to obtain a target analysis result. The N target analysis reports can be connected in sequence according to pre-set conjunctions to form a complete textual expression, or charts corresponding to the N target analysis reports can be generated according to preset chart types.

[0149] In one embodiment, step S5 includes:

[0150] Step S51: scoring the multiple analysis reports using the large language model to obtain scores corresponding to the multiple analysis reports;

[0151] Step S52: Using the large language model, set the N analysis reports with the highest scores among the multiple analysis reports as the N target analysis reports.

[0152] In an embodiment of the present invention, multiple analysis reports are scored using a large language model to obtain scores corresponding to the multiple analysis reports, and then the N analysis reports with the highest scores among the multiple analysis reports are set as N target analysis reports through the large language model to further improve the interpretation effect of the target analysis results.

[0153] In one embodiment, step S51 includes:

[0154] Step S511, calculating at least one of the repetition, data relevance and word order disorder corresponding to the plurality of analysis reports by using the large language model, wherein the data relevance is used to characterize the relevance between the analysis report and the topic;

[0155] Step S512: Obtain scores corresponding to the plurality of analysis reports by calculating, using the large language model, at least one of the corresponding repetition, data value, and word order disorder.

[0156] In an embodiment of the present invention, scoring is performed by using at least one of repetition, data correlation and word order disorder, so that the score corresponding to the analysis report can reflect the repetition, data correlation and / or word order disorder of the data report, and then N target analysis reports with low repetition, high data correlation and / or low word order disorder can be screened out.

[0157] In an embodiment of the present invention, at least one of the repetition, data association and word order disorder corresponding to the analysis report of each insight direction is calculated by a large language model. Specifically, a scoring prompt word is pre-set, and the scoring prompt word indicates that the large language model calculates at least one of the repetition, data association and word order disorder corresponding to the analysis report.

[0158] In the embodiment of the present invention, the score corresponding to the analysis report of each insight direction is calculated by the large language model based on at least one of the corresponding repetition, data value and word order disorder, specifically including:

[0159] Use the big language model to set the repetition, data value or word order disorder as the corresponding score of the analysis report for each insight direction;

[0160] The weighted sum of at least two of the repetition, data value and word order disorder is set as the score corresponding to the analysis report of each insight direction through the large language model.

[0161] In one embodiment, after step S6, the method further includes:

[0162] Step S7: Generate a graphic report and / or language digital broadcast of the target analysis result through the large language model, and display the graphic report and / or language digital broadcast.

[0163] In an embodiment of the present invention, a graphic report and / or language digital broadcast of the target analysis results is generated by a large language model, and the graphic report and / or language digital broadcast is displayed to achieve a visual display of the target analysis results, thereby facilitating personnel to understand the data story and further improving the interpretation effect of the data story.

[0164] The process of generating target analysis results in the embodiment of the present invention can be as follows: Figure 5 As shown in the figure, first obtain the original data set and topic corresponding to the data story, determine the insight direction through the big language model, and then generate an analysis report through the big language model. After obtaining the analysis report corresponding to each insight direction, filter and reconstruct it to generate analysis results in seconds, so as to interpret the data story from multiple insight directions.

[0165] See also Figure 6 , Figure 6 is a structural diagram of a data story interpretation device based on a large language model provided by an embodiment of the present invention, such as Figure 6 As shown, the data story interpretation device 600 based on the large language model includes:

[0166] An acquisition module 601 is used to acquire an original data set corresponding to a data story to be analyzed, a theme, a plurality of data facts included in the data story to be analyzed, and related information corresponding to each data fact, wherein the related information is used to describe the attributes of the data fact;

[0167] A determination module 602, configured to determine a plurality of insight directions based on the topic, the plurality of data facts, and the relevant information corresponding to each of the data facts by using the large language model;

[0168] The generation module 603 is used to generate an analysis report for each insight direction based on the original data set and the multiple data facts for each of the multiple insight directions through the large language model, wherein the analysis report is used to interpret the data story to be analyzed.

[0169] In one embodiment, the multiple insight directions include a first insight direction, which is determined by the large language model based on the topic, one or more data facts among the multiple data facts, related information corresponding to the data facts, and a determined insight direction, and the determined multiple insight directions include an insight direction determined by the large language model based on the topic, the multiple data facts and the related information corresponding to each data fact.

[0170] In one embodiment, the multiple insight directions include a second insight direction, which is determined by the large language model based on the multiple determined insight directions, and the multiple determined insight directions include insight directions determined by the large language model based on the topic, the multiple data facts and the related information corresponding to each data fact.

[0171] In one embodiment, after the generating module 603, the data story interpretation device 600 based on the large language model further includes:

[0172] A first screening module is used to screen multiple analysis reports associated with a target data fact from among the analysis reports of the multiple insight directions by using the large language model, where the target data fact is one of the multiple data facts;

[0173] A second screening module, configured to screen the plurality of analysis reports using the large language model to obtain N target analysis reports, where N is a positive integer greater than 1;

[0174] A reconstruction module is used to reconstruct the N target analysis reports through the large language model to obtain a target analysis result, and the target analysis result is used to interpret the data story to be analyzed based on the target data facts.

[0175] In one embodiment, the second screening module comprises:

[0176] a scoring unit, configured to score the plurality of analysis reports using the large language model to obtain scores corresponding to the plurality of analysis reports;

[0177] A determination unit is used to set the N analysis reports with the highest scores among the multiple analysis reports as the N target analysis reports through the large language model.

[0178] In one embodiment, the scoring unit includes:

[0179] A first calculation subunit is used to calculate at least one of the repetition degree, data relevance degree and word order disorder degree corresponding to the plurality of analysis reports by using the large language model, wherein the data relevance degree is used to characterize the relevance degree between the analysis report and the topic;

[0180] The second calculation subunit is used to calculate the scores corresponding to the multiple analysis reports based on at least one of the corresponding repetition, data value and word order disorder through the large language model.

[0181] In one embodiment, after the analysis module, the data story interpretation device 600 based on the large language model further includes:

[0182] A display module is used to generate a graphic report and / or a language digital broadcast of the target analysis result through the large language model, and to display the graphic report and / or the language digital broadcast.

[0183] The data story interpretation device based on a large language model provided in an embodiment of the present invention is capable of implementing each process of each embodiment of the above-mentioned data story interpretation method based on a large language model. The technical features correspond one to one and can achieve the same technical effect. To avoid repetition, they will not be described here.

[0184] It should be noted that the data story interpretation device based on the large language model in the embodiment of the present invention can be a device, or a component, integrated circuit, or chip in an electronic device.

[0185] The embodiment of the present invention further provides an electronic device, see Figure 7 , Figure 7 701, a processor 702, and a program or instruction stored in the memory 701 and executed by the processor 702. Figure 1 Any steps in the corresponding embodiment of the data story interpretation method based on a large language model and the same beneficial effects are achieved will not be repeated here.

[0186] The processor 702 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or a complex programmable logic device (CPLD).

[0187] A person skilled in the art can understand that all or part of the steps of implementing the above-mentioned data story interpretation method based on a large language model can be completed by hardware related to program instructions, and the program can be stored in a readable medium.

[0188] The embodiment of the present invention further provides a readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the above Figure 1 The corresponding steps in the data story interpretation method embodiment based on a large language model can achieve the same technical effect, and will not be repeated here to avoid repetition. The storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0189] The embodiment of the present invention also provides a computer program product, including computer instructions, which, when executed by a processor, implement the following Figure 1 The steps in the data story interpretation method based on the large language model can achieve the same technical effect. To avoid repetition, they will not be repeated here.

[0190] The terms "first", "second" etc. in the embodiments of the present invention are used to distinguish similar objects, and need not be used to describe a specific order or sequential order. In addition, the terms "include" and "have" and any variation thereof are intended to cover non-exclusive inclusions, for example, the process, method, system, product or equipment comprising a series of steps or units need not be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment. In addition, "and / or" is used in the present application to represent at least one of the connected objects, such as A and / or B and / or C, representing the 6 situations that include separate A, separate B, separate C, and A and B all exist, B and C all exist, A and C all exist, and A, B and C all exist.

[0191] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0192] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a second terminal device, etc.) to execute the methods of each embodiment of the present application.

[0193] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A data story interpretation method based on a large language model, characterized in that: include: Step S1, obtaining an original data set corresponding to a data story to be analyzed, a theme, a plurality of data facts included in the data story to be analyzed, and related information corresponding to each data fact, wherein the related information is used to describe the attributes of the data fact; Step S2: determining multiple insight directions based on the topic, the multiple data facts, and the relevant information corresponding to each data fact by using the large language model; Step S3: Generate an analysis report for each insight direction based on the original data set and the multiple data facts through the large language model, wherein the analysis report is used to interpret the data story to be analyzed.

2. The method according to claim 1, characterized in that The multiple insight directions include a first insight direction, which is determined by the large language model based on the topic, one or more data facts among the multiple data facts, related information corresponding to the data facts, and a determined insight direction, and the determined multiple insight directions include an insight direction determined by the large language model based on the topic, the multiple data facts and the related information corresponding to each data fact.

3. The method according to claim 1, characterized in that The multiple insight directions include a second insight direction, which is determined by the large language model based on the multiple determined insight directions, and the multiple determined insight directions include insight directions determined by the large language model based on the topic, the multiple data facts and the related information corresponding to each data fact.

4. The method according to any one of claims 1 to 3, characterized in that After step S3, the method further includes: Step S4: screening multiple analysis reports associated with a target data fact from the analysis reports of the multiple insight directions by using the large language model, wherein the target data fact is one of the multiple data facts; Step S5: Screening the multiple analysis reports using the large language model to obtain N target analysis reports, where N is a positive integer greater than 1; Step S6: reconstruct the N target analysis reports through the large language model to obtain target analysis results, and the target analysis results are used to interpret the data stories to be analyzed based on the target data facts.

5. The method according to claim 4, characterized in that The step S5 comprises: Step S51: scoring the multiple analysis reports using the large language model to obtain scores corresponding to the multiple analysis reports; Step S52: Using the large language model, set the N analysis reports with the highest scores among the multiple analysis reports as the N target analysis reports.

6. The method according to claim 5, characterized in that The step S51 comprises: Step S511, calculating at least one of the repetition, data relevance and word order disorder corresponding to the plurality of analysis reports by using the large language model, wherein the data relevance is used to characterize the relevance between the analysis report and the topic; Step S512: Obtain scores corresponding to the plurality of analysis reports by calculating, using the large language model, at least one of the corresponding repetition, data value, and word order disorder.

7. The method according to claim 4, characterized in that After step S6, the method further includes: Step S7: Generate a graphic report and / or language digital broadcast of the target analysis result through the large language model, and display the graphic report and / or language digital broadcast.

8. A data story interpretation device based on a large language model, characterized in that: include: An acquisition module, used to acquire an original data set corresponding to a data story to be analyzed, a theme, a plurality of data facts included in the data story to be analyzed, and related information corresponding to each data fact, wherein the related information is used to describe the attributes of the data fact; a determination module, configured to determine a plurality of insight directions based on the topic, the plurality of data facts, and relevant information corresponding to each of the data facts by using the large language model; A generation module is used to generate an analysis report for each insight direction based on the original data set and the multiple data facts for each of the multiple insight directions through the large language model, wherein the analysis report is used to interpret the data story to be analyzed.

9. An electronic device, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps of the data story interpretation method based on a large language model as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the data story interpretation method based on a large language model as described in any one of claims 1 to 7.

11. A computer program product, characterized in that The method comprises computer instructions, which, when executed by a processor, implement the steps of the data story interpretation method based on a large language model as described in any one of claims 1 to 7.