A data story generation method and system oriented to predictive analysis results

Abstract

The present application relates to a kind of data story generation method and system for predictive analysis result.The method comprises the following steps: predictive analysis system is handled as black box, story parameter is generated based on the input parameter of user t;Story character is generated based on story parameter;Story event is generated for story character according to story parameter;According to story event, story plot is generated according to multiple story stages;According to story plot, visual data story curve and data story report are generated.The technical scheme of the present application provides a new solution for the test of the three key problems of reliability, fairness and solvability in current big data application, supports What-if analysis and Why-not analysis two analysis tasks, has strong practicability, helps to solve the contradiction between model usability and explainability in big data era.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for generating data stories based on predictive analytics results. Background Technology

[0002] The interpretability and trustworthiness of predictive analytics results are a key focus for future society. With the widespread implementation of automated decision-making applications such as personalized recommendations, autonomous driving, smart healthcare, and machine translation, people are increasingly concerned about the interpretability of the underlying ethical, moral, and legal issues. The difficulty in interpreting predictive analytics results lies in the inappropriateness of explaining too much about the theoretical and technical aspects of the model to non-experts, and even more importantly, the need to avoid revealing the implementation details of the model itself and the trade secrets behind it.

[0003] Existing research primarily focuses on the interpretability of predictive algorithms and models. The interpretability of algorithms and models has become a hot topic in related fields and is gradually developing into a new field—Interpretable Machine Learning. Currently, Interpretable Machine Learning has made significant progress in methodology, key technologies, and application development. Model-independent, locally interpretable techniques, represented by the LIME algorithm, can provide a theoretical foundation for the narrative description of predictive analytics results. At the same time, "interpretation of algorithms or models" and "interpretation of analytics results" are related but distinct terms. The field of Interpretable Machine Learning mainly focuses on how to solve the interpretability problem of machine learning algorithms and the models they train.

[0004] Therefore, how to provide data storytelling solutions for predictive analytics results has become a technical problem that needs to be solved. Summary of the Invention

[0005] To address the aforementioned problems, the purpose of this invention is to provide a data story generation method and system for predictive analytics results. This method provides a data story generation scheme for the predictive analytics results of a predictive analytics system, without relying on the predictive analytics system's model, thus resolving the contradiction between usability and interpretability.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] One aspect of the present invention provides a data story generation method for predictive analytics results, comprising the following steps:

[0008] The predictive analytics system is treated as a black box, generating story parameters based on the input parameters of user t;

[0009] Generate story characters based on story parameters;

[0010] Generate story events for the story characters based on story parameters;

[0011] Generate storylines based on story events and multiple story stages;

[0012] Generate visual data story curves and data story reports based on the storyline.

[0013] Furthermore, the process of treating the predictive analytics system as a black box to generate story parameters specifically includes:

[0014] Obtain the input information t[X] of user t and the prediction result y expected by user t. target And the actual prediction result t[y] returned by the predictive analytics system to user t, where X is the feature set and y is the target vector; obtain the feature subset X of the feature set X from which bias needs to be detected. * The corresponding target vector is y * ; Obtain feature subset X * The complement of the feature subset is used as a feature subset that does not require detection bias. The corresponding target vector is

[0015] Obtain the immutable feature subset X" from the feature set X, corresponding to the target vector y"; obtain the complement of the feature subset X" as the mutable feature subset. The corresponding target vector is

[0016] The input information t[X], the prediction result t[y], and the prediction result expected by user t are given. target Feature subset X * and its target vector y * Feature subset and its target vector Feature subset X " and its target vector y", feature subset and its target vector Used as a story parameter to generate story characters.

[0017] Furthermore, the generation of story characters based on story parameters specifically includes:

[0018] Take user t as the protagonist t, and determine the protagonist t's input information t[X] and actual prediction result t[y];

[0019] Identify a character t of the same type as the protagonist t. ＝ / Different types of characters ≠ The criterion is based on the feature subset X that needs to detect bias. *If a character is identical to the protagonist (t), then the character is considered "the same type as the protagonist (t)". ＝ Otherwise, it is called "a character of a different type from the protagonist". ≠ ”;

[0020] Identify the positive character t relative to the protagonist t. + / Villain t - The criterion for judgment is whether the predicted result is the same as the actual predicted result for the protagonist t. If they are the same, then t is called a "positive character". + Otherwise, they are called "villains". - ".

[0021] Furthermore, the generation of story events for story characters based on story parameters includes generating reliability test events, fairness test events, and solvability test events; wherein, generating reliability test events refers to judging whether the prediction labels t[y]' corresponding to the multiple output actual prediction results t[y] are the same or fluctuate within a range less than the reliability threshold by submitting the protagonist t's input information t[X] to the predictive analysis system multiple times; generating fairness .... ＝ The absolute value of the difference between the probability that the expression of the bias detection rule is true in one user and the probability that the expression of the bias detection rule is true in all users is less than the bias threshold.

[0022] The expression for the bias rule is:

[0023] Bias={|P((y * ==y target )|(X * ==t[X ＝ ]))-P(y==y target )|<ε1}

[0024] Where Bias indicates whether there may be bias in detecting human characteristics, t[X ＝ [t] represents a group of people of the same type as the protagonist who wants to detect bias. ＝ In the feature, ε1 represents the acceptable range threshold of bias, P((y * ==y target )|(X * ==t[X ＝ ])) represents a group of people of the same type t ＝ The probability that the bias expression Bias holds true in the expression is P(y==y). target ) represents the probability that the bias expression Bias appears among all characters;

[0025] Generating solvability test events refers to generating solvability test events by analyzing a variable subset of features of the protagonist t. The desired prediction result y is achieved by minimizing the changes in multiple feature attribute values. target .

[0026] Furthermore, the generation of storylines based on multiple story stages according to story events refers to generating storylines according to the introductory stage, the escalation stage, the climax stage, the descent stage, and the ending stage, specifically including:

[0027] In the initial stage, the story parameters are set, including the input information t[X] of the protagonist t and the actual prediction result t[y].

[0028] During the ascent phase, the reliability test event is set as an "inflammatory event" at the starting point, and the fairness test results are plotted at the fairness test event position; based on the protagonist t finding y for the first time. target Under the premise of multiple what-if analysis events are sorted by y target After sorting by distance, set the median event Q. b 2. Upper quartile event Q b 1 and lower quartile event Q b 3;

[0029] During the climax phase, a climax event is set, which is the first occurrence of a predicted outcome y that matches the protagonist t's expectations in the what-if analysis. target The event or predictive analytics system recommends events for the protagonist t;

[0030] During the descent phase, based on the protagonist T's first discovery of Y... target Under the premise of this, perform Why-not analysis, and categorize multiple Why-not analysis events according to y target Set median event Q after sorting by distance a 2. Upper quartile event Q a 1 and lower quartile event Q a 3;

[0031] In the final stage, suggested events are set, which include suggestions given to the protagonist by the predictive analytics system.

[0032] Furthermore, the generation of visualized data story curves based on the storyline specifically includes:

[0033] Using the timeline of the storyline as the horizontal axis, and the expected prediction result y of the first discovering user t as the vertical axis. target The similarity value is represented by the ordinate, with different types of points indicating whether the protagonist t is a similar character or a positive character, and the expected prediction result y of the first discovered user t is used. target The point is the apex of the pyramid, and the curve is drawn according to the pyramid model.

[0034] Furthermore, the columns of the data story report include reliability, fairness, and solvability, and the rows of the data story report include input samples, prediction results, analysis methods, analysis results, analysis conclusions, and suggestions from the predictive analytics system to user t.

[0035] Another aspect of the present invention provides a data story generation system for predictive analytics results, comprising:

[0036] The parameter generation module is used to treat the predictive analytics system as a black box and generate story parameters based on the input parameters of user t.

[0037] The character generation module is used to generate story characters based on story parameters;

[0038] The event generation module is used to generate story events for story characters based on story parameters;

[0039] The plot generation module is used to generate story plots based on story events and multiple story stages;

[0040] The view generation module is used to generate visual data story curves based on the storyline.

[0041] The report generation module is used to generate data story reports based on the storyline.

[0042] Another aspect of the present invention provides a processing apparatus, the processing apparatus including at least a processor and a memory, the memory storing a computer program, wherein the processor, when running the computer program, performs steps to implement a data story generation method for predictive analytics results.

[0043] Another aspect of the present invention provides a computer storage medium having computer-readable instructions stored thereon, the computer-readable instructions being executable by a processor to implement the steps of a data story generation method oriented towards predictive analytics results.

[0044] The present invention has the following advantages due to the adoption of the above technical solutions:

[0045] 1. This invention proposes a model-independent data storytelling method that can be applied to any predictive analytics algorithm and business, has strong versatility and flexibility, and can be used as a new functional module in commonly used data analysis software such as SPSS, SAS, and Excel.

[0046] 2. This invention provides a new solution for testing the reliability, fairness, and solvability of three key issues in current big data applications. It supports two analysis tasks: What-if analysis and Why-not analysis, and has strong practicality. It provides a new solution to the contradiction between model usability and interpretability in the big data era.

[0047] 3. This invention presents data in a storytelling manner, providing visual storyline curves and readable data story reports, which have high readability and comprehensibility, and are not limited by the knowledge level or professional field of the target audience.

[0048] 4. This invention supports the selection of different sources of story data, thereby supporting a variety of application purposes.

[0049] When story data is provided by the management of the business system (such as a merchant), the right of interpretation belongs to the management, and this invention better supports its business logic and business objectives; when it is generated by the user, this invention better supports user experience and interactive story generation methods; when it is generated by a third-party organization or uses a standard test dataset, this invention better supports third-party testing and evaluation scenarios.

[0050] 5. The data stories generated in this invention are visual data story curves and readable data story reports, which not only support reading by human users, but also by computer users, thereby supporting the development of new functions and the upgrading of functional modules, and have strong scalability.

[0051] 6. The final stage of the data story presented in this invention supports commercial advertising and algorithmic recommendations, and has strong commercial application prospects. Attached Figure Description

[0052] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. In the drawings:

[0053] Figure 1 This is a flowchart of a data story generation method for predictive analytics results according to an embodiment of the present invention;

[0054] Figure 2 This is a schematic diagram of the feature matrix and target vector in the predictive analysis results of an embodiment of the present invention;

[0055] Figure 3 This is a schematic diagram of the data story generation process for predictive analytics results according to an embodiment of the present invention;

[0056] Figure 4This is a schematic diagram of the visual data story curve in an embodiment of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0058] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0059] Data storytelling is a crucial method for interpreting predictive analytics results. From the perspective of how data is received, perception is the prerequisite for cognition, and cognition is a continuation of perception. Data visualization and data storytelling address the issues of data perception and cognition, respectively. Data visualization is characterized by its ease of understanding, perception, and insight, while data storytelling is characterized by its ease of memorization, cognition, and experience. Therefore, data storytelling will be widely used in scenarios where predictive analytics results are explained to non-experts, thereby gaining their trust in the predictive analytics findings.

[0060] One aspect of this invention is to provide a data story generation method for predictive analytics results. This method is independent of the model itself and can be applied to any predictive analytics algorithm and business system. In the data story generation method, the predictive analytics system is first treated as a black box, and story parameters are generated based on the input parameters of user t; story characters are generated based on the story parameters; story events are generated for the story characters based on the story parameters; and story plots are generated according to multiple story stages based on the story events.

[0061] Visualized data story curves and data story reports are generated based on the storyline. On the other hand, corresponding to this data story generation method, a data story generation system oriented towards predictive analytics results is also provided.

[0062] Example 1

[0063] This embodiment provides a data story generation method for predictive analytics results, such as... Figure 1 As shown, the method includes the following steps:

[0064] S1, Treat the predictive analytics system as a black box and generate story parameters based on the input parameters of user t;

[0065] S2, Generate story characters based on story parameters;

[0066] S3 generates story events for story characters based on story parameters;

[0067] S4 generates storylines based on story events across multiple story stages;

[0068] S5 generates visual data story curves and data story reports based on the storyline.

[0069] In step S1, story parameters are generated.

[0070] Treat the predictive analytics system S as a black box, and abstract (or map) its inputs and outputs into a new relational schema R(X,y), where X and y are the names of the feature set and the target vector, respectively. For example... Figure 2 The data format for X and y is shown below. Story parameters need to be provided by the story audience (or users of the predictive analytics system). User input data and their meanings are as follows:

[0071] (1) t[X]: Input information submitted by user t to predictive analytics system S;

[0072] (2) t[y]: The prediction result returned by the predictive analytics system S to user t;

[0073] (3)y target The prediction result that user t expects (or wants);

[0074] (4)X * : A subset of feature set X, representing the subset of features for which bias needs to be detected by user t.

[0075] For X * The complement, i.e. X * and The corresponding target vector is denoted as y. * and

[0076] (5)t[X ＝ ]: A group of similar characters whose characteristics are the same as the protagonist's desire to detect prejudice. ＝ The feature subset in.

[0077] (6)X”: A subset of feature set X, representing a subset of features that cannot be changed for user t. The complement of X", i.e. X" and The corresponding target vectors are denoted as y" and

[0078] Input information t[X], prediction result t[y], and the prediction result y expected by user t. target Feature subset X * and its target vector y * Feature subset and its target vector Feature subset X" and its target vector y", feature subset and its target vector These parameters are used to generate story characters. In step S2, story characters are generated. Story characters are categorized into the protagonist, characters similar to / different from the protagonist, and characters who are positive / negative relative to the protagonist, among which:

[0079] (1) Protagonist t: A specific user t in the predictive analytics system S, whose feature information and prediction results are t[X] and t[y], respectively. The values ​​of t[X] and t[y] can be the user's current input values, historical input values, or default values ​​of the predictive analytics system S.

[0080] (2) Characters of the same type as the protagonist t ＝ / Different types of characters ≠ The judgment criterion is based on the feature subset X where bias may occur. * Are the characters who are the same as the protagonist considered the same? If they are the same, they are called "characters of the same type as the protagonist"; otherwise, they are called "characters of different types from the protagonist".

[0081] Among them, there are characters t who are the same type as the protagonist t. ＝ The generation method is: X * Take the same value as t, complement attribute Use random values ​​that conform to the corresponding attribute domain. (For characters of the same type...) ＝ :

[0082] t ＝ ={z|z[X * ] = t[X * ]∩z[X * [[is a random value]]

[0083] Among them, characters of a different type from the protagonist t ≠ The generation method is: X * Take a value different from t, complement property Use random values ​​that conform to the domain of the corresponding attribute. Different types of characters t ≠ :

[0084] t ≠ ={z|z[X * ]≠t[X * ]∩z[X * [[is a random value]]

[0085] (3) Positive characters t in relation to the protagonist t + / Villain t - The criterion is whether the predicted result (or classification result) is the same as the protagonist. If they are the same, the protagonist is called a "positive character". + Otherwise, it is "the villain". - ".

[0086] Among them, compared to the protagonist t, the positive character t + The definitions and generation methods specifically include:

[0087] If the value of y (or the classification result) is the same as t[y], randomly select k (k>0) nearest neighbor samples from the historical records as the positive person t. + If no positive characters can be found in the historical records + A positive character t is generated by fine-tuning the variable attributes of the protagonist t. + :

[0088]

[0089] In the formula, p is the variable attribute of the protagonist t. The number of; x j For user t, a set of variable attributes A member attribute; x j and x' j These are the j-th variable attribute value of the protagonist t and its fine-tuned value, respectively.

[0090] Among them, compared to the protagonist t, the antagonist t - The definitions and generation methods specifically include:

[0091] If the value of y (or the classification result) is different from t[y], select k (k>0) nearest neighbor samples from the historical records as the negative example t. - If no positive characters can be found in the historical record... - Generate a villain t using the following method. - :

[0092]

[0093] In the formula, p is the variable attribute of the protagonist t. The number of; x j For user t, a set of variable attributes A member attribute; x j and x' j These are the j-th variable attribute value of the protagonist t and its fine-tuned value, respectively.

[0094] In step S3, story events are generated, including reliability test events, fairness test events, and solvability test events.

[0095] Here, reliability refers to the confidence level of the predicted results corresponding to multiple identical actions of the same protagonist t; fairness refers to the confidence level of the predicted results among characters t of the same type as protagonist t. ＝ Does it contain discrimination or bias, i.e., does it satisfy the bias rule (Bias_benchmark)? Solvability refers to whether, for a specific user, it can be solved by modifying their variable feature set. The expected prediction result y was achieved. target .

[0096] (1) The specific method for generating reliability test events is as follows: Submit n (n≥2) feature information t[X] of user t to the predictive analysis system S, and check whether the corresponding label t[y]' is equal or its fluctuation is within the negligible reliability threshold range (ε2), that is:

[0097]

[0098] (2) The specific method for generating fairness test events is as follows:

[0099] User group t ＝ The absolute value of the difference between the probability of the bias rule expression "Bias" being true among all users and the probability of "Bias" being true among all users is within the acceptable range ε1:

[0100] Bias={|P((y * ==y target )|(X * ==t[X ＝ ]))-P(y==y target )|<ε1}

[0101] Where Bias indicates whether there may be bias in detecting human characteristics, t[X ＝ [t] represents a group of people of the same type as the protagonist who wants to detect bias. ＝ In the feature, ε1 represents the acceptable range threshold of bias, P((y * ==y target )|(X * ==t[X ＝ ])) represents a group of people of the same type t ＝The probability that the bias expression Bias holds true in the expression is P(y==y). target ) represents the probability that the bias expression Bias appears among all characters;

[0102] The meaning of the bias expression is: calculate the difference between the probability that the predicted result expected by user t can be obtained based on the features to be detected that may have bias and the probability that the predicted result expected by user t can be obtained based on all features. If this difference is less than the threshold ε1, then it is considered that these features to be detected do not have bias.

[0103] (3) The specific method for generating solvability test events is as follows:

[0104] For user t, by its variable characteristics To achieve the goal, make minimal changes to the above. target The specific method is as follows:

[0105]

[0106] In the formula, p is the variable attribute of the protagonist t. The number of; x j For the set of variable attributes of t A member attribute; x j and x' j These are the j-th variable attribute value of the protagonist t and its fine-tuned value, respectively.

[0107] In step S4, the storyline is generated. The storyline uses a pyramid model and is divided into five stages:

[0108] The initial stage, the rising stage, the climax stage, the decline stage, and the final stage, such as Figure 3 As shown.

[0109] (1) Initial stage: only one event, namely the setting of story parameters, including user feature information and prediction results of predictive analysis system S.

[0110] (2) Ascending phase: Set the user to find y target Events that occur under certain conditions include three types of events:

[0111] Set the reliability test event as an "inflammatory event" at the starting point;

[0112] Plot the fairness test results in the fairness test event field;

[0113] The remaining length of the rising phase is divided into four equal parts, and the user t is analyzed multiple times using what-if analysis events according to y. target Distance to the median event (Q) after sorting b2) Upper quartile event (Q) b 1) and lower quartile events (Q b 3), indicating that user t discovered y target All the attempts and efforts made before.

[0114] (3) Climax stage: This stage includes only one event, namely the event recommended by the predictive analytics system S or the first predicted result y that the user t expects in the what-if analysis. target The incident.

[0115] (4) Descent phase: includes at least three events, including user t finding y for the first time. target Under the premise of this, a Why-not analysis is performed to analyze the reasons for success for successful users. The length of the descent phase is divided into four parts, and the multiple Why-not analysis events of user t are compared with y. target Distance to the median event (Q) after sorting a 2) Upper quartile event (Q) a 1) and lower quartile events (Q a 3), indicating that user t discovered y target Various causal analyses were subsequently conducted.

[0116] (5) Ending stage: This stage includes only one event. The suggestions given by the predictive analytics system S to user t can be determined by the business logic of the predictive analytics system S, such as corresponding commercial advertisements or algorithmic recommendation information.

[0117] In step S5, data stories are generated. This involves two aspects: generating a visual data story curve and generating a readable data story report; these two processes complement each other.

[0118] The generated data story curves are as follows Figure 4 As shown, the specific drawing method is as follows:

[0119] ●Horizontal axis: Timeline of the storyline;

[0120] ● Vertical axis: Compared to the y-axis where the user first discovered it. target Similarity;

[0121] ●Curve type: Pyramid model, with the apex representing the y-axis initially observed by the user. target ;

[0122] ● The shape of the point: indicates whether it is of the same type as user t and whether it is a positive person.

[0123] pass Figure 4 It can be seen that the protagonist first conducts a reliability test in the data story, an event that becomes the inflammatory one. The character then continues to act to find y. targetThe events occurring at this stage are what-if analysis events, and the data story continues to unfold. When the story's protagonist first finds y... target At that point, the data story reached its climax. Then the characters obtained y target Based on this, further actions are taken. The events that occur in this stage are why-not analyzed events, and the y in this stage target The data story doesn't change; it reaches a decline phase. Finally, the characters in the story cease action, the data story concludes, and the user finds advice on making decisions or taking action.

[0124] A readable data story report consists of 3 columns and 6 rows, as shown in Table 1. The 3 columns represent reliability, fairness, and solvability; the 6 rows represent the input sample, prediction result, analysis method, analysis result, analysis conclusion, and the predictive analytics system S's recommendation to user t.

[0125] Table 1. Structure of the Readable Data Story Report

[0126]

[0127] This invention proposes a model-independent, post-interpretive data storytelling method for predictive analytics results. The technical solution of this invention can be incorporated as a new functional module into commonly used data analysis software (such as SPSS, SAS, and Excel). Currently, data analysis software lacks data storytelling capabilities. The data model, surrogate analysis model training, formal script definition and validation, and Python toolkit in the data storytelling method for predictive analytics results proposed in this invention can provide theoretical basis and tool support for adding new data storytelling functional modules to such software.

[0128] Example 2

[0129] The above-described embodiment 1 provides a data story generation method for predictive analytics results. Correspondingly, this embodiment provides a data story generation system for predictive analytics results. The system provided in this embodiment can implement the data story generation method for predictive analytics results of embodiment 1. The system can be implemented through software, hardware, or a combination of both. For example, the system may include integrated or separate functional modules or units to execute the corresponding steps in the methods of embodiment 1. Since the system in this embodiment is basically similar to the method embodiment, the description process in this embodiment is relatively simple. Relevant details can be found in the description of embodiment 1. The system embodiment provided in this embodiment is merely illustrative.

[0130] This embodiment provides a data story generation system for predictive analytics results, including:

[0131] The parameter generation module is used to treat the predictive analytics system as a black box and generate story parameters based on the input parameters of user t.

[0132] The character generation module is used to generate story characters based on story parameters;

[0133] The event generation module is used to generate story events for story characters based on story parameters;

[0134] The plot generation module is used to generate story plots based on story events and multiple story stages;

[0135] The view generation module is used to generate visual data story curves based on the storyline.

[0136] The report generation module is used to generate data story reports based on the storyline.

[0137] Example 3

[0138] This embodiment provides a processing device corresponding to the data story generation method for predictive analytics results provided in Embodiment 1. The processing device can be a client-side processing device, such as a mobile phone, laptop, tablet, or desktop computer, to execute the method of Embodiment 1.

[0139] The processing device includes a processor, a memory, a communication interface, and a bus. The processor, memory, and communication interface are connected via the bus to enable communication between them. The memory stores a computer program that can run on the processor. When the processor runs the computer program, it executes the data story generation method for predictive analytics results provided in Embodiment 1.

[0140] In some embodiments, the memory may be high-speed random access memory (RAM), and may also include non-volatile memory, such as at least one disk storage device.

[0141] In other embodiments, the processor can be a general-purpose processor of various types, such as a central processing unit (CPU) or a digital signal processor (DSP), and is not limited thereto.

[0142] Example 4

[0143] The data story generation method for predictive analytics results in Embodiment 1 can be specifically implemented as a computer program product. The computer program product may include a computer-readable storage medium on which computer-readable program instructions for executing the data story generation method for predictive analytics results described in Embodiment 1 are loaded.

[0144] A computer-readable storage medium can be a tangible device that holds and stores instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof.

[0145] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A data story generation method for predictive analytics results, characterized in that, The method includes the following steps: The predictive analytics system is treated as a black box, generating story parameters based on the input parameters of user t; Generate story characters based on story parameters; Based on story parameters, story events are generated for the story characters, including reliability test events, fairness test events, and solvability test events. Reliability test events involve repeatedly submitting the protagonist's input information t[X] to the predictive analytics system and determining whether the prediction labels t[y]' corresponding to the multiple output prediction results t[y] are the same or fluctuate within a reliability threshold range. Fairness test events involve generating story events for characters of the same type as the protagonist. The absolute value of the difference between the probability that the expression of the bias detection rule is true in one user and the probability that the expression of the bias detection rule is true in all users is less than the bias threshold. The expression for the bias rule is: Here, Bias indicates whether there might be bias in detecting human characteristics. This refers to a group of people with similar traits to the one the protagonist wants to detect as bias. Features in The threshold representing the acceptable range of bias. Represents a group of people of the same type The probability that the bias expression "Bias" is true in the given context. This represents the probability that the bias expression "Bias" appears among all characters. Generating solvability test events refers to generating solvability test events by analyzing a variable subset of features of the protagonist t. The desired prediction result y is achieved by minimizing the changes in multiple feature attribute values. target ; Generate storylines based on story events and multiple story stages; Generate visual data story curves and data story reports based on the storyline.

2. The data story generation method for predictive analytics results according to claim 1, characterized in that, The process of treating the predictive analytics system as a black box to generate story parameters specifically includes: Obtain the input information t[X] of user t and the prediction result y expected by user t. target And the actual prediction result t[y] returned by the predictive analytics system to user t, where X is the feature set and y is the target vector; obtain the feature subset of the feature set X from which bias needs to be detected. The corresponding target vector is ; Obtain feature subset The complement of the feature subset is used as a feature subset that does not require detection bias. The corresponding target vector is ; Obtain an immutable subset of features from feature set X The corresponding target vector is ; Obtain feature subset The complement of the set as a variable feature subset The corresponding target vector is ; The input information t[X], the prediction result t[y], and the prediction result expected by user t are given. target Feature subset and its target vector Feature subset and its target vector Feature subset and its target vector Feature subset and its target vector Used as a story parameter to generate story characters.

3. The data story generation method for predictive analytics results according to claim 2, characterized in that, The process of generating story characters based on story parameters specifically includes: Take user t as the protagonist t, and determine the protagonist t's input information t[X] and actual prediction result t[y]; Identify a character t of the same type as the protagonist t. = / Different types of characters ≠ The criterion is based on the feature subset where bias needs to be detected. If a character is identical to the protagonist (t), then they are considered "characters of the same type as the protagonist". Otherwise, it is called "a character of a different type from the protagonist". ”; Identify the positive character t relative to the protagonist t. + / Villain t - The criterion is whether the predicted result is the same as the actual predicted result for the protagonist t. If they are the same, then t is called a "positive character". + Otherwise, they are called "villains". - ".

4. The data story generation method for predictive analytics results according to claim 1, characterized in that, The generation of storylines based on multiple story stages according to story events refers to generating storylines according to the introductory stage, the rising stage, the climax stage, the decline stage, and the ending stage, specifically including: In the initial stage, the story parameters are set, including the input information t[X] of the protagonist t and the actual prediction result t[y]. During the ascent phase, the reliability test event is set as an "inflammatory event" at the starting point, and the fairness test results are plotted at the fairness test event position; based on the protagonist t finding y for the first time. target Under the premise of multiple what-if analysis events are sorted by y target After sorting by distance, set the median event Q. b 2. Upper quartile event Q b 1 and lower quartile event Q b 3; During the climax phase, a climax event is set, which is the first occurrence of a predicted outcome y that matches the protagonist t's expectations in the what-if analysis. target The event or predictive analytics system recommends events for the protagonist t; During the descent phase, based on the protagonist T's first discovery of Y... target Under the premise of this, perform Why-not analysis, and categorize multiple Why-not analysis events according to y target Set median event Q after sorting by distance a 2. Upper quartile event Q a 1 and lower quartile event Q a 3; In the final stage, suggested events are set, which include suggestions given to the protagonist by the predictive analytics system.

5. The data story generation method for predictive analytics results according to claim 4, characterized in that, The generation of visualized data story curves based on the storyline specifically includes: Using the timeline of the storyline as the horizontal axis, and the expected prediction result y of the first discovering user t as the vertical axis. target The similarity value is represented by the ordinate, with different types of points indicating whether the protagonist t is a similar character or a positive character, and the expected prediction result y of the first discovered user t is used. target The point is the apex of the pyramid, and the curve is drawn according to the pyramid model.

6. The data story generation method for predictive analytics results according to claim 4, characterized in that, The columns of the data story report include reliability, fairness, and solvability, and the rows of the data story report include input samples, prediction results, analysis methods, analysis results, analysis conclusions, and suggestions from the predictive analytics system to user t.

7. A data story generation system for predictive analytics results, used to implement the data story generation method for predictive analytics results as described in any one of claims 1 to 6, characterized in that, include: The parameter generation module is used to treat the predictive analytics system as a black box and generate story parameters based on the input parameters of user t. The character generation module is used to generate story characters based on story parameters; The event generation module is used to generate story events for story characters based on story parameters; The plot generation module is used to generate story plots based on story events and multiple story stages; The view generation module is used to generate visual data story curves based on the storyline. The report generation module is used to generate data story reports based on the storyline.

8. A processing apparatus, the processing apparatus comprising at least a processor and a memory, the memory storing a computer program, characterized in that, When the processor runs the computer program, it performs steps to implement the data story generation method for predictive analytics results as described in any one of claims 1 to 6.

9. A computer storage medium, characterized in that, It stores computer-readable instructions that can be executed by a processor to implement the steps of the data story generation method for predictive analytics results according to any one of claims 1 to 6.

Images