CN116340752A - Predictive analysis result-oriented data story generation method and system - Google Patents

Abstract

The invention relates to a data story generation method and system for predictive analysis results. The method comprises the following steps: treating the predictive analysis system as a black box, generating story parameters based on user t's input parameters; generating story characters based on the story parameters; generating story events for the story characters according to the story parameters; The story stage generates a storyline; according to the storyline, a visual data story curve and a data story report are generated. The technical scheme of the present invention provides a new solution for the testing of the three key problems of reliability, fairness and solvability in current big data applications, supports two analysis tasks of What-if analysis and Why-not analysis, and has relatively Strong practicability helps to solve the contradiction between model usability and interpretability in the era of big data.

Description

A data story generation method and system for predictive analysis results

technical field

The present invention relates to the technical field of artificial intelligence, in particular to a data story generation method and system for predictive analysis results.

Background technique

The interpretability and trustworthiness of predictive analysis results will be the focus of future society. With the widespread implementation of automatic decision-making applications such as personalized recommendation, automatic driving, intelligent medical care, and machine translation, people are paying more and more attention to the explainability of moral, ethical, and legal issues behind them. The difficulty in interpreting the results of predictive analysis is that it is not appropriate to explain the theoretical and technical issues related to the model to non-professionals, let alone reveal the implementation details of the model itself and the business secrets behind it.

Existing related research mainly focuses on the interpretability of predictive algorithms and models. The interpretability of algorithms and models has become a hot topic of research in related fields, and has gradually developed into a new field - interpretable machine learning (Interpretable Machine Learning). At present, interpretable machine learning has made great progress in methodology, key technologies, and application development. The model-independent and local interpretation technology represented by the LIME algorithm can provide a theoretical basis for the story-oriented description of predictive analysis results. . At the same time, "interpretation of algorithms or models" and "interpretation of analysis results" are terms that are both related and different. The field of interpretable machine learning is mainly concerned with how to solve the interpretability of machine learning algorithms and the models they train. .

Therefore, how to provide data storytelling solutions for predictive analysis results has become a technical problem that needs to be solved at present.

Contents of the invention

In view of the above problems, the object of the present invention is to provide a data story generation method and system for predictive analysis results, which is used to provide data storytelling solutions for the predictive analysis results of the predictive analysis system without relying on predictive analysis A model of a system that resolves the tension between usability and interpretability.

To achieve the above object, the present invention takes the following technical solutions:

One aspect of the present invention provides a method for generating data stories for predictive analysis results, comprising the following steps:

treat the predictive analytics system as a black box, generating story parameters based on the input parameters of user t;

Generate story characters based on story parameters;

Generate story events for story characters based on story parameters;

Generate storylines in multiple story stages based on story events;

Generate visual data story curves and data story reports based on storylines.

Further, the predictive analysis system is treated as a black box to generate story parameters, specifically including:

对应目标向量为/>

Obtain the input information t[X] of user t, the expected prediction result y ^target of user t, and the actual prediction result t[y] returned by the predictive analysis system to user t, where X is the feature set and y is the target vector; The feature subset X ^* that needs to detect bias in the feature set X, the corresponding target vector is y ^* ; get the complement of the feature subset X ^* as the feature subset that does not need to detect bias

The corresponding target vector is />

对应目标向量为/>

Get the immutable feature subset X" in the feature set X, and the corresponding target vector is y"; get the complement of the feature subset X" as a variable feature subset

The corresponding target vector is />

及其目标向量/>

特征子集X"及其目标向量y"、特征子集/>

及其目标向量/>

作为故事参数，用于生成故事人物。Input information t[X], prediction result t[y], prediction result y ^target expected by user t, 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 />

As a story parameter, used to generate story characters.

Further, said generating story characters according to story parameters specifically includes:

Take the user t as the protagonist t, and determine the input information t[X] and the actual prediction result t[y] of the protagonist t;

Determine the character t of the same type as the protagonist t ⁼ /character of a different type t ^≠ , the basis for judging is whether the same character as the protagonist t on the feature subset X ^* that needs to detect bias is the same, if they are the same, it is called "with the protagonist Characters of the same type t ⁼ ", otherwise it is called "characters of different types from the protagonist t ^≠ ";

Determine the positive character t ⁺ / negative character t ^- relative to the protagonist t, the judgment is based on whether the prediction result is the same as the actual prediction result of the protagonist t, if they are the same, it is called "positive character t ⁺ ", otherwise it is called "negative character Character ^t- ".

Further, said generating story events for story characters according to story parameters includes generating reliability test events, fairness test events and solvability test events; wherein, generating reliability test events refers to the multi-direction predictive analysis system Submit the input information t[X] of the protagonist t to judge whether the prediction label t[y]' corresponding to the actual prediction result t[y] output multiple times is the same or whether it floats within the range of less than the reliability threshold; generate a fairness test event It means that the absolute value of the difference between the probability of the establishment of the expression of the bias detection rule in the character t ⁼ of the same type as the protagonist and the probability of the establishment of the expression of the bias detection rule in all users is within the range of less than the deviation threshold;

The bias rule expression Bias is:

Bias＝{|P((y ^* ＝＝y ^target )|(X ^* ＝＝t[X＝]))-P(y＝＝y ^target )|<ε ₁ }

Among them, Bias represents whether there may be a bias in the detection of character features, t[X ⁼ ] represents the characteristics of the same type of character group t ⁼ in the same type of character as the protagonist wants to detect bias, _ε1 represents the acceptable range threshold of bias, P(( y ^* ＝＝y ^target )|(X ^* ＝＝t[X ^＝ ])) indicates the probability that the bias expression Bias appears in the same type of characters t ^＝ , and P(y＝＝y ^target ) indicates the probability that Bias appears in all characters The probability of bias expression Bias;

的多个特征属性值进行最小变化处理实现用户的期待预测结果y^target。Generating solvability test events refers to a variable subset of features for the protagonist t

The minimum change processing is performed on multiple characteristic attribute values of the user to achieve the user's expected prediction result y ^target .

Further, said generating storylines according to multiple story stages according to story events refers to generating storylines according to the initial stage, rising stage, climax stage, descending stage and ending stage, specifically including:

In the initial stage, set the story parameters. The set story parameters include the input information t[X] of the protagonist t and the actual prediction result t[y];

In the rising stage, set the reliability test event as an "inciting event" at the starting point, and draw the fairness test result at the fairness test event position; based on the premise that the protagonist t finds y ^target for the first time, the If the analysis events are sorted according to the distance from the y ^target , set the median event Q ^b ₂ , the upper quartile event Q ^b ₁ and the lower quartile event Q ^b ₃ ;

In the climax stage, a climax event is set, and the climax event is the first event found in the what-if analysis that meets the expected prediction result y ^target of the protagonist t or an event recommended by the predictive analysis system for the protagonist t;

In the descending stage, based on the premise that the protagonist t finds the y ^target for the first time, the Why-not analysis is performed, and the multiple Why-not analysis events are sorted according to the distance from the y ^target , and then the median event Q ^a ₂ and the upper quartile are set event Q ^a ₁ and lower quartile event Q ^a ₃ ;

In the ending stage, a suggestion event is set, which includes a suggestion given by the predictive analysis system for the protagonist.

Further, the generating a visual data story curve according to the storyline specifically includes:

Take the development time of the storyline as the abscissa, and take the similarity value of the expected prediction result y ^target of the first discovery of user t as the ordinate, and use different types of points to indicate whether the protagonist t is a similar character and whether he is a positive person. It is found that the point of user t's expected prediction result y ^target is the curve drawn according to the pyramid model.

Further, 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, predictive analysis system for users t's suggestion.

Another aspect of the present invention also provides a data story generation system for predictive analysis results, including:

a parameter generation module for treating the predictive analytics system as a black box and generating story parameters based on the input parameters of the user t;

A character generation module, configured to generate story characters based on story parameters;

An event generation module, configured to generate story events for story characters according to story parameters;

A plot generation module for generating storylines in multiple story stages based on story events;

The view generation module is used to generate a visual data story curve according to the storyline;

The report generation module is used to generate a data story report according to the storyline.

Another aspect of the present invention also provides a processing device, the processing device includes at least a processor and a memory, a computer program is stored on the memory, and the processor executes the computer program to implement predictive analysis The steps of the resulting data story generation method.

Another aspect of the present invention also provides a computer storage medium on which computer-readable instructions are stored, and the computer-readable instructions can be executed by a processor to implement the steps of the method for generating data stories oriented to predictive analysis results.

The present invention has the following advantages due to the adoption of the above technical scheme:

1. The present invention proposes a model-independent data storytelling method, which can be applied to any predictive analysis algorithm and business, has strong versatility and flexibility, and can be used as common data analysis software, such as SPSS, SAS, Excel New feature modules in .

2. The present invention provides a new solution for testing the reliability, fairness and solvability of the three key problems in current big data applications, and supports What-if analysis and Why-not analysis. Strong practicability provides a new solution to solve the contradiction between model usability and interpretability in the era of big data.

3. The present invention adopts data story-based presentation, provides visual story plot curves and readable data story reports, has high readability and comprehension, and is not limited to the knowledge level and professional field of the target audience.

4. The present invention supports selection of sources of different story data, and further supports multiple application purposes.

When story data is provided by the manager of the business system (such as a merchant), the right of interpretation belongs to the manager, and the present invention better supports its business logic and business purpose; when generated by the user himself, the present invention better supports the user Experiential and interactive story generation methods; the present invention better supports third-party testing and evaluation scenarios when generated by third-party institutions or using standard test data sets.

5. The data stories generated in the present invention are visual data story curves and readable data story reports, which not only support the reading of human users, but also the readability of computer users, and then support the development of new functions and the development of functional modules Upgrade, strong scalability.

6. The last stage of the data story provided by the present invention supports commercial advertisements and algorithm recommendations, and has strong commercial application prospects.

Description of drawings

Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiment. The drawings are only for the purpose of illustrating a preferred embodiment and are not to be considered as limiting the invention. Throughout the drawings, the same reference numerals are used to refer to the same parts. In the attached picture:

Fig. 1 is a flow chart of a data story generation method for predictive analysis results according to an embodiment of the present invention;

Fig. 2 is a schematic diagram of a feature matrix and a target vector in the predictive analysis results of an embodiment of the present invention;

FIG. 3 is a schematic diagram of a data story generation process for predictive analysis results according to an embodiment of the present invention;

Fig. 4 is a schematic diagram of a visual data story curve according to an embodiment of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention more clear, the following will clearly and completely describe the technical solutions of the embodiments of the present invention in conjunction with the drawings of the embodiments of the present invention. Apparently, the described embodiments are some, not all, embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention belong to the protection scope of the present invention.

It should be noted that the terminology used here is only for describing specific implementations, and is not intended to limit the exemplary implementations according to the present application. As used herein, unless the context clearly dictates otherwise, the singular is intended to include the plural, and it should also be understood that when the terms "comprising" and/or "comprising" are used in this specification, they mean There are features, steps, operations, means, components and/or combinations thereof.

Data storytelling is an important means of interpreting predictive analytics results. From the subject's data acceptance mode, perception is the premise of cognition, and cognition is the continuation of perception. Data visualization and data storytelling respectively solve the problems of data perception and cognition. Data visualization has the characteristics of easy understanding, perception and insight, while data storytelling has the characteristics of easy memory, cognition and experience. Therefore, data storytelling will be widely used in application scenarios where predictive analysis results are explained to non-professionals, and then non-professionals can gain trust in predictive analysis results.

One aspect of the embodiments of the present invention is to provide a predictive analysis result-oriented data story generation method, which has nothing to do with the model itself and can be applied to any predictive analysis algorithm and business system. In the data story generation method, firstly, the predictive analysis system is treated as a black box, and the story parameters are generated based on the input parameters of the user t; the story characters are generated based on the story parameters; the story events are generated for the story characters according to the story parameters; according to the story events, multiple The story stage generates the storyline;

Generate visual data story curves and data story reports based on storylines. On the other hand, corresponding to the data story generation method, a data story generation system oriented to predictive analysis results is also provided.

Example 1

This embodiment provides a data story generation method for predictive analysis results, as shown in Figure 1, the method includes the following steps:

S1, treat the predictive analysis system as a black box, and generate story parameters based on the input parameters of user t;

S2, generating story characters based on story parameters;

S3, generating story events for story characters according to story parameters;

S4, generating a storyline according to a plurality of story stages according to a story event;

S5, generating a visualized data story curve and a data story report according to the storyline.

In step S1, story parameters are generated.

Treat the predictive analytics system S as a black box, and abstract (or map) its input and output into a new relational schema R(X,y), where X and y are the names of feature sets and target vectors, respectively. As shown in Figure 2, the data form of X and y is given. Story parameters need to be provided by the story audience (or predictive analytics system user). The user input data and their meaning are as follows:

(1) t[X]: input information submitted by user t to predictive analysis system S;

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

(3) y ^target : the predicted result that user t expects (or wants);

(4) X ^* : A subset of the feature set X, representing the subset of features that user t needs to detect bias,

为X^*的补集，即/>

X^*和/>

对应的目标向量记为y^*和/>

is the complement of X ^* , ie />

X ^* and />

The corresponding target vectors are denoted as y ^* and />

(5) t[X ^＝ ]: A subset of features in the group t ^＝ of the same type of characters as the protagonist wants to detect bias.

为X"的补集，即/>

X"和/>

对应的目标向量记为y"和/>

(6) X": A subset of the feature set X, which represents a subset of features that cannot be changed for user t, such as gender, skin color, belief, and ethnicity.

is the complement of X", ie />

X" and />

The corresponding target vectors are denoted as y" and />

及其目标向量/>

特征子集X"及其目标向量y"、特征子集/>

及其目标向量/>

作为故事参数，用于生成故事人物。在步骤S2中，生成故事人物。故事人物分为主人公、与主人公同类/异类人物、相对于主人公而言的正面/反面人物，其中：Input information t[X], prediction result t[y], prediction result y ^target expected by user t, 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 />

As a story parameter, used to generate story characters. In step S2, story characters are generated. The characters in the story are divided into the protagonist, similar/different characters to the protagonist, and positive/negative characters relative to the protagonist, among which:

(1) Protagonist t: a specific user t of the predictive analysis 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 not only the current input values by the user, but also the historical input values or the default values of the predictive analysis system S.

(2) Character t of the same type as the protagonist t ⁼ /character of a different type t ^≠ : the basis for judging is whether the same character as the protagonist is the same in the characteristic subset X ^* that may appear biased. If they are the same, it is called "the same type of character as the protagonist", otherwise it is called "a different type of character from the protagonist".

采用符合相应属性定义域的随机值。相同类型的人物t^＝：Among them, the generation method of the character t ⁼ of the same type as the protagonist t is: X ^* takes the same value as t, and the complement attribute

Takes a random value that fits the domain of the corresponding property. People of the same type t ⁼ :

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

采用符合相应属性定义域的随机值。不同类型的人物t^≠：Among them, the generation method of a character t ^≠ different from the protagonist t is: X ^* takes a value different from t, and the complement attribute

Takes a random value that fits the domain of the corresponding property. Different types of characters t ^≠ :

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

(3) Positive character t ⁺ /villain t ⁻ relative to the protagonist t: the basis for judging is whether the prediction result (or classification result) is the same as the protagonist. If they are the same, it is called “positive character t ⁺ ”, otherwise it is called “negative character t ⁻ ”.

Among them, relative to the protagonist t, the definition and generation method of the positive character t ⁺ specifically include:

If the value of y (or 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 the positive person t ⁺ cannot be found in the historical record , generate a positive character t ⁺ by fine-tuning the variable attributes of the protagonist t:

的个数；x_j为用户t的可变属性集/>

的一个成员属性；x_j和x’_j分别为主人公t的第j个可变属性值及其微调后的取值。In the formula, p is the variable attribute of the protagonist t

The number of; x _j is the variable attribute set of user t/>

A member attribute of ; x _j and x' _j are respectively the jth variable attribute value of the protagonist t and its fine-tuned value.

Among them, relative to the protagonist t, the definition and generation method of the villain t ^- specifically include:

If the value of y (or classification result) is different from t[y], select k (k>0) nearest neighbor samples from the historical records as villain t ^- . If the positive character t ^- cannot be found in the historical records, use the following method to generate a negative character t ^- :

的个数；x_j为用户t的可变属性集/>

的一个成员属性；x_j和x’_j分别为主人公t的第j个可变属性值及其微调后的取值。In the formula, p is the variable attribute of the protagonist t

The number of; x _j is the variable attribute set of user t/>

A member attribute of ; x _j and x' _j are respectively the jth variable attribute value of the protagonist t and its fine-tuned value.

In step S3, a story event is generated, and the story event includes a reliability test event, a fairness test event, and a solvability test event.

达到所期待的预测结果y^target。Among them, reliability refers to the reliability of the prediction results corresponding to multiple identical behaviors of the same protagonist t; fairness refers to whether there is discrimination or prejudice among characters t ⁼ similar to the protagonist t, that is, whether the bias rule (Bias_benchmark) is satisfied ; Solvability refers to whether for a specific user, by modifying its variable feature set

Achieve the expected prediction result y ^target .

(1) The generation method of the reliability test event is as follows: submit the feature information t[X] of user t n times (n≥2) to the predictive analysis system S, and check whether the corresponding label t[y]' is equal or floating Within the negligible reliability threshold range (ε ₂ ), namely:

(2) The method of generating the fairness test event is as follows:

The absolute value of the difference between the probability of the bias rule expression Bias occurring in the user group t ⁼ and the probability of the bias rule expression Bias appearing in all users is within the acceptable range _ε1 :

Bias＝{|P((y ^* ＝＝y ^target )|(X ^* ＝＝t[X ^＝ ]))-P(y＝＝y ^target )|<ε ₁ }

Among them, Bias represents whether there may be a bias in the detection of character features, t[X ⁼ ] represents the characteristics of the same type of character group t ⁼ in the same type of character as the protagonist wants to detect bias, _ε1 represents the acceptable range threshold of bias, P(( y ^* ＝＝y ^target )|(X ^* ＝＝t[X ^＝ ])) indicates the probability that the bias expression Bias appears in the same type of characters t ^＝ , and P(y＝＝y ^target ) indicates the probability that Bias appears in all characters The probability of bias expression Bias;

The meaning of the bias expression Bias is: calculate the difference between the probability of obtaining the predicted result expected by user t based on the features to be detected that may be biased and the probability of obtaining the predicted result expected by user t based on all features, if the difference is less than Threshold ε ₁ , it is considered that there is no bias in these features to be detected.

(3) The generation method of the solvability test event is as follows:

上进行最小变化来帮助其实现达到目的y^target，具体方法如下：For user t, through its variable features

Make minimal changes to help it achieve the goal y ^target , the specific method is as follows:

的个数；x_j为t的可变属性集/>

的一个成员属性；x_j和x’_j分别为主人公t的第j个可变属性值及其微调后的取值。In the formula, p is the variable attribute of the protagonist t

The number of ; x _j is the variable attribute set of t />

A member attribute of ; x _j and x' _j are respectively the jth variable attribute value of the protagonist t and its fine-tuned value.

In step S4, a storyline is generated. The storyline uses a pyramid model, divided into five stages:

The initial stage, rising stage, climax stage, descending stage and ending stage are shown in Figure 3.

(1) Initial stage: only one event is included, that is, the setting of story parameters, including the user's characteristic information and the prediction result of the predictive analysis system S.

(2) Ascending stage: set the events that occur on the premise that the user finds the y ^target , including three types of events:

Set the reliability test event as "inciting event" at the origin;

Draw the fairness test result at the fairness test event bit;

The remaining length of the ascending phase is divided into 4 parts equally, and the ^median event (Q ^b ₂ ), upper quartile event (Q ^b ₁ ) and The lower quartile event (Q ^b ₃ ) indicates that user t made various attempts and efforts before finding y ^target .

(3) Climax stage: only one event is included, that is, the event recommended by the predictive analysis system S or the event that the user finds for the first time in the what-if analysis that meets the predicted result y ^target expected by the user t.

(4) Declining stage: At least three events are included. On the premise that user t finds y ^target for the first time, a Why-not analysis is performed to analyze the success reasons of successful users. The length of the descending phase is equally divided into 4 parts, the ^median event (Q ^a ₂ ), the upper quartile event (Q a 1 ) and the lower quartile event (Q ^a ₁ ) and the lower quartile event ( The quartile event (Q ^a ₃ ) represents various causal analyzes performed by user t after discovering y ^target .

(5) Outcome stage: only one event is included, and the suggestion given by the predictive analysis system S to the user t can be determined by the business logic of the predictive analysis system S, such as corresponding to the corresponding commercial advertisement or algorithm recommendation information.

In step S5, a data story is generated. One is to generate a visual data story curve, and the other is to generate a readable data story report, both of which are complementary.

The generated data story curve is shown in Figure 4, and its specific drawing method is as follows:

·Abscissa: the development time of the storyline;

Vertical coordinate: the similarity with the y ^target discovered by the user for the first time;

Curve type: pyramid model, the spire is the y ^target discovered by the user for the first time;

· The shape of the dot: indicates whether it is the same as the user t and whether it is a positive person.

As can be seen from Figure 4, the protagonist first conducts a reliability test in the data story, and this event becomes an inciting event. Then the characters continue to act to find the y ^target , and the events at this stage are what-if analysis events, and the data story continues to advance. The development of the data story reaches its climax when the story characters first find the y ^target . Then the characters continue to take actions on the basis of obtaining the y ^target . The events that occur at this stage are why-not analysis events, and the y ^target at this stage will not change, and the development of the data story reaches a decline period. Finally, the story characters stop acting, the data story comes to an end, and the user finds a suggestion to make a decision or take an action.

The readable data story report consists of 3 columns and 6 rows, as shown in Table 1. The three columns are reliability, fairness, and solvability; the six rows are input samples, prediction results, analysis methods, analysis results, analysis conclusions, and suggestions from predictive analysis system S to user t.

Table 1 Structure of a readable data story report

Aiming at the data story description of predictive analysis results, the present invention proposes a model-independent post-explanatory data story description method provided by predictive analysis. The technical scheme of the present invention can be used as a new functional module in common data analysis software (such as SPSS, SAS, Excel). At present, data analysis software does not have the function of data storytelling. The data model in the data storytelling method for predictive analysis results proposed by the present invention, the training of agent analysis models, the definition and verification of formalized scripts, and the Python tool Packages can provide theoretical basis and tool support for adding new functional modules of data storytelling in such software.

Example 2

Embodiment 1 above provides a method for generating data stories oriented to predictive analysis results. Correspondingly, this embodiment provides a system for generating data stories oriented to predictive analysis results. The system provided in this embodiment can implement the predictive analysis result-oriented data story generation method in Embodiment 1, and the system can be implemented by software, hardware, or a combination of software and hardware. For example, the system may include integrated or separate functional modules or functional units to execute corresponding steps in the methods of Embodiment 1. Since the system of this embodiment is basically similar to the method embodiment, the description process of this embodiment is relatively simple. For relevant information, please refer to the part of the description of Embodiment 1. The embodiment of the system provided by this embodiment is only illustrative .

A data story generation system oriented to predictive analysis results provided in this embodiment includes:

a parameter generation module for treating the predictive analytics system as a black box and generating story parameters based on the input parameters of the user t;

A character generation module, configured to generate story characters based on story parameters;

An event generation module, configured to generate story events for story characters according to story parameters;

A plot generation module for generating storylines in multiple story stages based on story events;

The view generation module is used to generate a visual data story curve according to the storyline;

The report generation module is used to generate a data story report according to the storyline.

Example 3

This embodiment provides a processing device corresponding to the predictive analysis result-oriented data story generation method provided in Embodiment 1. The processing device may be a processing device for a client, such as a mobile phone, a notebook computer, a tablet computer, a desktop Machine computer etc., to carry out the method of embodiment 1.

The processing device includes a processor, a memory, a communication interface and a bus, and the processor, the memory and the communication interface are connected through the bus to complete mutual communication. A computer program that can run on the processor is stored in the memory, and the processor executes the predictive analysis result-oriented data story generation method provided in Embodiment 1 when running the computer program.

In some embodiments, the memory may be a high-speed random access memory (RAM: Random Access Memory), and may also include a non-volatile memory, such as at least one disk memory.

In some other embodiments, the processor may be a central processing unit (CPU), a digital signal processor (DSP) and other types of general-purpose processors, which are not limited herein.

Example 4

The method for generating a data story oriented to predictive analysis results in Embodiment 1 can be embodied as a computer program product, and the computer program product can include a computer-readable storage medium loaded with a method for executing the method described in Embodiment 1. Computer readable program instructions for a data story generation method for predictive analytics results.

A computer readable storage medium may be a tangible device that holds and stores instructions for use by an instruction execution device. A computer readable storage medium may be, for example, but is 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 of the above.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still be Modifications are made to the technical solutions described in the foregoing embodiments, or equivalent replacements are made to some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims (10)

1. A method for generating a data story oriented to predictive analysis results, the method comprising the steps of:

treating the predictive analysis system as a black box, and generating story parameters based on input parameters of a user t;

generating a story character based on the story parameters;

generating a story event for the story character according to the story parameters;

generating a storyline in a plurality of story stages according to the storyline;

and generating a visual data story curve and a data story report according to the story line.

2. The method for generating a data story for a predictive analysis result of claim 1,

the predictive analysis system is treated as a black box to generate story parameters, and the method specifically comprises the following steps:

acquiring input information t [ X ] of user t]Predicted outcome y expected by user t ^target Actual prediction result t [ y ] returned to user t by predictive analysis system]Wherein X is a feature set, and y is a target vector; acquiring feature subset X in feature set X, for which bias needs to be detected ^* The corresponding target vector is y ^* The method comprises the steps of carrying out a first treatment on the surface of the Acquiring feature subset X ^* As a subset of features without detection bias

The corresponding target vector is +.>

Obtaining an invariable feature subset X 'in a feature set X, wherein a corresponding target vector is y'; obtaining a complement of feature subset X "as a variable feature subset

The corresponding target vector is +.>

Will input information t [ X ]]Prediction result t [ y ]]Predicted outcome 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

As story parameters, for generating a story character.

3. The method for generating a data story for a predictive analysis result of claim 2,

the generating the story character according to the story parameters specifically comprises the following steps:

taking the user t as the host public t, and determining input information t [ X ] and actual prediction result t [ y ] of the host public t;

determining a character t of the same type as the host male t ^＝ Different types of characters t ^≠ The judgment basis is that the feature subset X needing to detect the bias ^* Whether the person who is the same as the host male t is the same,if the characters are the same, the character is called as a character t of the same type as a host ^＝ ", otherwise called" character t of a different type from the owner ^≠ ”；

Determining a frontal character t relative to a host man t ⁺ Reverse character t ^- Judging whether the predicted result is the same as the actual predicted result of the host male t, if so, the method is called as' front character t ⁺ ", otherwise called" reverse character t ^- ”。

4. The method for generating a data story for a predictive analysis result of claim 3,

generating a story event for a story character according to story parameters, including generating a reliability test event, a fairness test event and a resolvable test event;

the generation of the reliability test event refers to that whether the predictive labels t [ y ]' corresponding to the actual predictive results t [ y ] output for many times are the same or float within a range smaller than a reliability threshold value is judged by submitting input information t [ X ] of the host male t to the predictive analysis system for many times;

generating fairness test event refers to a character t of the same type as the owner ^＝ The absolute value of the difference between the probability that the expression of the occurrence bias detection rule is established and the probability that the expression of the occurrence bias detection rule is established in all users is in a range smaller than a bias threshold value;

the Bias rule expression Bias is:

Bias＝{|P((y ^* ＝＝y ^target )|(X ^* ＝＝t[X＝]))-P(y＝＝y ^target )|<ε ₁ }

wherein Bias represents whether the detected character feature may have a Bias, tx =]Representing the same type of character group t as the character of which the owner is willing to detect a prejudice ^＝ Is characterized by epsilon ₁ Represents the threshold of acceptable range for bias, P ((y) ^* ＝＝y ^target )|(X ^* ＝＝t[X＝]) Representing the same type of character group t ^＝ The probability that the Bias expression Bias holds, P (y= y) ^target ) Representing the probability of occurrence of Bias expression Bias in all characters;

generating a solvable test event refers to by a subset of features that are variable to the host's public t

Performing minimum change processing on a plurality of characteristic attribute values of (a) to realize the expected prediction result y of the user ^target 。

5. The method for generating a predictive analysis result-oriented data story according to claim 4, wherein,

the generating the storyline according to the story line and the story stages includes generating the storyline according to the start stage, the rise stage, the climax stage, the decline stage and the ending stage:

in the starting stage, story parameters are set, wherein the set story parameters comprise input information t [ X ] of a host male t and an actual prediction result t [ y ];

setting a reliability test event as an 'flaring event' at a starting point in a rising stage, and drawing a fairness test result at a fairness test event bit; finding y for the first time based on the host male t ^target On the premise of (1) and y by multiple what-if analysis events ^target After distance sequencing, set median event Q ^b ₂ Last quarter bit event Q ^b ₁ And a lower quartile event Q ^b ₃ ；

In the climax stage, setting climax events, namely, finding expected prediction results y conforming to the public t of the owner in the what-if analysis for the first time ^target The event or the event recommended by the predictive analysis system for the host male t;

during the descent phase, finding y for the first time based on the host male t ^target On the premise of (1) carrying out Why-not analysis, and carrying out Why-not analysis on the event according to the sum y ^target Setting median event Q after distance sequencing ^a ₂ Last quarter bit event Q ^a ₁ And a lower quartile event Q ^a ₃ ；

In the ending phase, suggested events are set, including suggestions made by the predictive analysis system for the owner.

6. The method for generating a data story for a predictive analysis result of claim 5,

the method for generating the visual data story line according to the story line specifically comprises the following steps:

taking the development time of the story line as the abscissa to find the expected prediction result y of the user t for the first time ^target The similarity value of (2) is taken as an ordinate, and different types of points are used for representing whether the host male t is the same type of person and whether the host male t is the front person, so that the expected prediction result y of the user t is found for the first time ^target Points (1) are the plotted curves of the tower tip according to the pyramid model.

7. The method for generating a data story for a predictive analysis result of claim 5,

the columns of the data story report include reliability, fairness and resolvable, and the rows of the data story report include input samples, predicted results, analysis methods, analysis results, analysis conclusions, and suggestions of the predictive analysis system to the user t.

8. A predictive analysis result-oriented data story generation system, comprising:

the parameter generation module is used for processing the predictive analysis system as a black box and generating story parameters based on input parameters of a user t;

a character generation module for generating a story character based on the story parameters;

the event generation module is used for generating a story event for the story character according to the story parameters;

the plot generation module is used for generating a plot according to the story event according to a plurality of story stages;

the view generation module is used for generating a visual data story line according to the story line;

and the report generation module is used for generating a data story report according to the story line.

9. A processing device comprising at least a processor and a memory, said memory having stored thereon a computer program, characterized in that the processor executes the steps of the predictive analysis result oriented data story generation method of any of claims 1 to 7 when running said computer program.

10. A computer storage medium having stored thereon computer readable instructions executable by a processor to implement the steps of the predictive analysis result oriented data story generation method of any of claims 1 to 7.

Images