Data processing method, device and program
The method efficiently determines causal relationships between numerous elements by using correlation coefficients and objective functions, addressing the challenge of high-dimensional causal structures and enhancing decision-making in various domains.
Patent Information
- Application Number
- JP2021563019
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2019-04-24
- Publication Date
- 2024-11-27
- Estimated Expiration
- 2039-04-24
AI Technical Summary
Existing methods struggle to quickly and accurately discover causal relationships between a large number of elements, leading to significant computational overhead and uncertainty in high-dimensional causal structures.
A data processing method and apparatus that determines a set of dependencies and causal relationships between multiple elements using a dependency determining unit and a causal relationship determining unit, employing correlation coefficients and objective functions to minimize computational overhead and enhance accuracy.
Enables rapid and precise discovery of high-dimensional causal structures, allowing for improved decision-making in fields like customer satisfaction, health management, and product sales by identifying influential factors and optimizing strategies.
Smart Images

Figure 0007590984000037 
Figure 0007590984000038 
Figure 0007590984000039
Abstract
Description
[Technical field]
[0001] FIELD OF THE DISCLOSURE Embodiments of the present disclosure relate to the field of machine learning, and more particularly, to data processing methods, apparatus and computer-readable storage media. [Background technology]
[0002] With the rapid development of information technology, the scale of data is rapidly increasing. Machine learning has been attracting more and more attention in this background and trend. Among them, causal discovery is widely applied in real life, for example, in fields such as supply chain, medical health, and retail. Here, the above causal discovery refers to discovering causal relationships between multiple elements from sample data related to multiple elements. For example, in the retail field, the results of causal discovery are used to assist in formulating various sales strategies, and in the medical health field, the results of causal discovery are used to assist in creating treatment plans for patients, etc.
[0003] However, with the development of technology, the number of various elements in a single system that may have causal relationships is increasing significantly. In addition, the interlocking between different systems is also often of interest. This leads to the number of elements for which causal relationships are to be discovered being hundreds to thousands or more. In this case, it is becoming increasingly important to quickly and accurately discover causal relationships that exist between many elements. Summary of the Invention [Problem to be solved by the invention]
[0004] SUMMARY OF THE DISCLOSURE Embodiments of the present disclosure provide a data processing method, an apparatus, and a computer-readable storage medium. [Means for solving the problem]
[0005] In a first aspect of the present disclosure, a data processing method is provided, the method including: obtaining a set of observation samples for a plurality of elements, each observation sample in the observation sample set including corresponding observations of the plurality of elements; determining a set of dependencies existing between the plurality of elements based on the observation sample set, a dependency in the dependency set indicating a pair of interrelated elements in the plurality of elements; and determining a causal sequence of the plurality of elements based on the dependency set, the causal sequence indicating that one element in the pair of interrelated elements is a cause of the other element.
[0006] In a second aspect of the present disclosure, a data processing apparatus is provided, the apparatus including at least one processing unit and at least one memory, the at least one memory being coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit, the instructions, when executed by the at least one processing unit, causing the apparatus to perform the following operations, the operations including: obtaining a set of observation samples for a plurality of elements, each observation sample in the set of observation samples including corresponding observations of the plurality of elements; determining a set of dependencies that exist between the plurality of elements based on the set of observation samples, a dependency in the dependency set indicating a pair of elements that are related to each other in the plurality of elements; and determining a causal sequence of the plurality of elements based on the dependency set, the causal sequence indicating that one element in the pair of elements that is related to each other is a cause of the other element.
[0007] In a third aspect of the present disclosure, a computer-readable storage medium is provided, storing machine-executable instructions that, when executed by a device, cause the device to perform a method according to the first aspect of the present disclosure.
[0008] The purpose of providing this Summary is to introduce a selection of concepts in a simplified form that are described in more detail in the specific embodiments below. The Summary is not intended to identify key or essential features of the disclosure, nor is it intended to limit the scope of the disclosure. Other features of the disclosure should be readily apparent from the following description. [Brief description of the drawings]
[0009] The objects, advantages and other features of the present invention will become more apparent from the following disclosure and claims. A non-limiting description of preferred embodiments is now given, by way of example only, with reference to the drawings, in which:
[0010] [Figure 1A] FIG. 1 is a block diagram illustrating an exemplary system for processing data, according to an embodiment of the present disclosure. [Figure 1B] FIG. 1 is a block diagram illustrating an exemplary system for processing data, according to an embodiment of the present disclosure.
[0011] [Diagram 2] FIG. 1 is a schematic diagram for determining causal relationships between multiple elements according to an embodiment of the present disclosure.
[0012] [Diagram 3] 1 is a flowchart illustrating an exemplary method according to an embodiment of the present disclosure.
[0013] [Figure 4] 1 is a flowchart illustrating an exemplary method according to an embodiment of the present disclosure.
[0014] [Diagram 5] 1 is a flowchart illustrating an exemplary method according to an embodiment of the present disclosure.
[0015] [Figure 6] 1 is a flowchart illustrating an exemplary method according to an embodiment of the present disclosure.
[0016] [Figure 7] FIG. 1 is a block diagram that illustrates a schematic of an exemplary device in which embodiments of the present disclosure may be implemented.
[0017] In the various drawings, the same or corresponding reference numerals indicate the same or corresponding parts. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0018] Hereinafter, the embodiments of the present disclosure will be described in more detail with reference to the drawings. Although the present disclosure shows some embodiments of the present disclosure with the drawings, the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein. On the contrary, it should be understood that the present embodiments are provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are merely illustrative and are not intended to limit the scope of protection of the present disclosure.
[0019] In describing embodiments of the present disclosure, the term "comprising" and similar terms should be interpreted as open ended, i.e., "including, but not limited to." The term "based on" should be interpreted as "based at least in part on." The term "one embodiment" or "the embodiment" should be interpreted as "at least one embodiment." The terms "first," "second," etc. may refer to different or identical objects. Other explicit and implicit definitions may be included below.
[0020] In embodiments of the present disclosure, the term "causal structure" generally refers to a structure that describes the causal relationships between elements in a system, and is also referred to herein as a "causal sequence." The term "element" is also referred to as a "variable." The term "observation sample" refers to a set of observations of directly observable elements, and the directly observable elements are also referred to as "observation variables."
[0021] As described above, in real life, it is desirable to quickly and accurately discover causal relationships that exist among many observed variables.
[0022] In the field of customer service, a large amount of customer consumption behavior data (e.g., customer level, amount of internet traffic used per month, percentage of free traffic, total amount of internet traffic used per month, etc.), satisfaction survey data, and carrier strategy data may be collected to determine which factors affect a customer's satisfaction with the telecommunications carrier. Each type of collected data is also called an observation of one kind of factor (or variable). By discovering the causal relationship between these factors, one or more factors that affect customer satisfaction can be determined. Furthermore, by changing the observation of the one or more factors or developing a corresponding strategy for the one or more factors, the customer's satisfaction with the telecommunications carrier can be improved.
[0023] In the field of health, a set of physiological indicators (i.e., observed values of a set of factors) of a patient may be collected in large quantities, such as heart rate, cardiac output, allergy index, total peripheral vascular resistance, release of catecholamines, blood pressure, etc. By discovering the causal relationships existing between these physiological indicators, the physiological indicators (i.e., factors) that affect the patient's blood pressure can be determined. Furthermore, by developing strategies to affect or respond to the physiological indicators, the patient's blood pressure can be kept stable.
[0024] In the field of product sales, to determine factors that affect the sales of a target product (e.g., umbrellas), external factor data (e.g., weather, season, temperature, date, store size, etc.), sales data of the product (e.g., sales of the product, price of the product, etc.), and sales data of one or more related products (e.g., ice cream), etc. may be collected. Each type of collected data is an observation value of one kind of factor. By discovering the causal relationship existing between these factors, one or more factors that affect the sales of the target product can be determined. Furthermore, by changing the observation value of the one or more factors or formulating a corresponding strategy for the one or more factors, the sales of the target product can be increased.
[0025] In the field of software development, information on various factors of software development may be collected to determine factors that affect the failure rate and / or development cycle of software, including, but not limited to, overall information of software development (e.g., development cycle, resources invested in development, etc.) and information on each stage of software development. Information on each stage of software development may include, for example, information on architecture stage (e.g., software architecture method, number of software architecture layers, etc.), information on coding stage (e.g., code length, number of functions, programming language, number of modules, etc.), information on testing stage (e.g., accuracy rate or failure rate of unit testing, accuracy rate or failure rate of black box testing, accuracy rate or failure rate of white box testing, etc.), and information on operation stage after software issuance (e.g., accuracy rate or failure rate of operation stage, etc.). Each type of collected data is regarded as an observation value of one kind of factor. By discovering the causal relationship existing between these factors, one or more factors that affect the development cycle and / or failure rate of software can be determined. Furthermore, by modifying the observed value of the one or more factors or developing a corresponding strategy for the one or more factors, the development cycle and / or failure rate of the software may be reduced.
[0026] Some conventional solutions mainly use constraint-based or score-based methods to search the entire variable space for systems with few elements (e.g., 100 elements or less) to find possible causal relationships. For example, constraint-based methods generally perform conditional independence tests on multiple elements to find their causal structures. However, when the number of elements that need to be searched is large, the results of the conditional independence tests become uncertain. Also, finding a causal structure always requires testing the dependency between any two elements using a set of conditions ranging from not including any element to including all other elements, which results in huge computational overhead. Score-based methods generally find a causal structure by optimizing a score that determines the degree of matching between the causal structure and sample data. However, because the search space grows at a rate greater than exponential, these solutions are not always applicable to finding a causal structure (also called a "high-dimensional causal structure") for many elements (e.g., hundreds to thousands or more).
[0027] According to an embodiment of the present disclosure, a solution for data processing is provided, which can quickly and accurately realize the discovery of high-dimensional causal structures, thereby solving the above problems and / or other potential problems. In the following, the embodiments of the present disclosure will be described in detail in accordance with the above exemplary scenario. It should be understood that this is merely for the purpose of explanation, and is not intended to limit the scope of the present invention in any manner.
[0028] FIG. 1A is an exemplary block diagram showing a system 100 for data processing according to an embodiment of the present disclosure. The system 100 can, for example, discover causal relationships between multiple elements. It should be understood that the system 100 shown in FIG. 1A is merely an example of an embodiment of the present disclosure and is not intended to limit the scope of the present disclosure. The embodiment of the present disclosure is equally applicable to other systems or architectures.
[0029] As shown in FIG. 1A, the system 100 may include a causal relationship determination device 120. The causal relationship determination device 120 may receive an observation sample set 110 for a plurality of elements and determine therefrom a causal relationship sequence 130 indicating a causal relationship between the plurality of elements. Preferably, in some embodiments, the system 100 may further include an observation sample collection device (not shown in FIG. 1A) for collecting the observation sample set 110 for the plurality of elements. The observation sample collection device may collect observations of the plurality of elements in real time, or periodically or irregularly, to obtain the observation sample set 110. In some embodiments, the observation sample collection device may include one or more collection units for collecting observations of different types of elements, respectively.
[0030] The observed sample set 110 may include observed samples of multiple elements for one or more target elements. For example, the observed sample set 110 may be a set of X={x1, x2, ..., x D}∈R N×D where N denotes the total number of observation samples in the observation sample set 110, D denotes the total number of elements, and each observation sample in the observation sample set 110 includes a set of observations having D elements. In this specification, a set having D elements is denoted as V={v1, v2, ..., v D} and v i (where 1≦i≦D) denotes the i-th element of the D elements. Vector x i ∈R N (where 1≦i≦D) denotes the N observations of the i-th element. For example, x i,n (where 1≦i≦D and 1≦n≦N) indicates the nth observation of the i-th element.
[0031] In the case of the above-mentioned scenario regarding customer satisfaction of a communication carrier, for example, the target element is "customer satisfaction," and the element set V may include one or more elements related to customer attributes (e.g., customer level, customer number, etc.), elements related to customer behavior (e.g., monthly used internet traffic, ratio of free traffic, total monthly used internet traffic, etc.), elements related to customer feedback (e.g., number of complaints, customer satisfaction), and strategy elements for customers (e.g., number of packet over presentations, timing, etc.). i ), for example, vector x i consists of N customer levels and x i,n may indicate the level of the nth customer among N customers. i ), for example, vector x i consists of the monthly internet traffic of N customers, and x i,n may indicate the amount of net communication traffic used by the nth customer per month among N customers. The causal relationship sequence 130 indicates the causal relationship between elements such as customer level, the amount of net communication traffic used per month, the ratio of free communication traffic, the total amount of net communication traffic used per month, and customer satisfaction, and can indicate, for example, which element is the cause of the target element "customer satisfaction."
[0032] Taking the above-mentioned scene related to the patient's blood pressure as an example, for example, the target element is "blood pressure", and the element set V may include heart rate, cardiac output, allergy index, total peripheral vascular resistance, catecholamine release, blood pressure, etc. i ), for example, vector x i consists of heart rates for N patients, and x i,n may denote the heart rate of the nth patient among N patients. i ), for example, vector x i consists of the cardiac outputs of N patients, and x i,nmay indicate the cardiac output of the nth patient among N patients. The causal sequence 130 indicates the causal relationships between elements such as heart rate, cardiac output, allergy index, total peripheral resistance, catecholamine release, blood pressure, etc., and can indicate, for example, which elements are the cause of the target element "blood pressure."
[0033] Taking the above-mentioned product sales scene as an example, for example, the target element is "sales of the target product", and the element set V may include one or more of external elements (e.g., weather, season, temperature, date, size of the store, etc.), elements related to the sales behavior of the target product (e.g., umbrella) (e.g., sales of the target product, price of the target product, etc.), elements related to the sales behavior of one or more related products (e.g., ice cream) (e.g., sales of the related products, price of the related products, etc.), and sales strategy elements for the target product (e.g., number of promotions, frequency, etc.). i ), for example, vector x i consists of the temperature on N days, and x i,n may indicate the temperature on the nth day. i ), for example, vector x i consists of the sales of umbrellas on N days, and x i,n may indicate the sales of umbrellas on the nth day. The causal sequence 130 indicates the causal relationships between elements such as weather, season, temperature, date, store size, sales of a target product, price of the target product, sales of related products, and price of related products, and can indicate, for example, which element is the cause of the target element "sales of the target product."
[0034] Taking the above software development scene as an example, for example, the target element is "software development cycle" or "failure rate of software in operation phase", and the element set V may include one or more of the overall elements of software development (e.g., development cycle, resources invested in development, etc.) and elements of each phase of software development. The elements of each phase of software development may include, for example, elements of the architecture phase (e.g., software architecture method, number of software architecture layers, etc.), elements of the coding phase (e.g., code length, number of functions, programming language, number of modules, etc.), elements of the testing phase (e.g., accuracy rate or failure rate of unit testing, accuracy rate or failure rate of black box testing, accuracy rate or failure rate of white box testing, etc.), and elements of the operation phase after software issuance (e.g., accuracy rate of operation phase, failure rate of operation phase, etc.). The element "development cycle" (i.e., v i ), for example, vector x i consists of N software product development cycles and x i,n may represent the development cycle of the nth software product. i ), for example, vector x i consists of the code length of N software products, and x i,n indicates the code length of the nth software product. The causal relationship sequence 130 indicates the causal relationships between elements such as software development cycle, resources invested in development, architectural method, number of architectural layers, code length, number of functions, programming language, number of modules, accuracy rate or failure rate of unit testing, accuracy rate or failure rate of black box testing, accuracy rate or failure rate of white box testing, accuracy rate in operation phase, and failure rate in operation phase. For example, it can indicate which element is the cause of the target element "development cycle" and which element is the cause of the target element "failure rate in operation phase".
[0035] As shown in FIG. 1A, the causal relationship determination device 120 may include, for example, a dependency relationship determination unit 121 and a causal relationship determination unit 122. In some embodiments, the dependency relationship determination unit 121 can determine a dependency relationship set that exists between a plurality of elements based on the observation sample set 110. Each dependency relationship in the dependency relationship set indicates a corresponding pair of elements that are related to each other in the plurality of elements. In some embodiments, the causal relationship determination unit 122 can determine an element relationship sequence 130 based on the dependency relationship set determined by the dependency relationship determination unit 121. The element relationship sequence 130 can indicate a causal relationship between the element pair in which a dependency relationship exists (i.e., one element is the cause of the other element).
[0036] It should be understood that these devices and / or units in devices included in system 100 are merely exemplary and are not intended to limit the scope of the present disclosure. It should be understood that system 100 may further include additional devices and / or units not shown. For example, in some embodiments, system 100 may further include a causality representation device (not shown) for representing aspects of causality sequence 130.
[0037] In some embodiments, the causal relationship representation device can represent aspects of the causal relationship sequence 130 in different forms, such as visually or aurally. For example, the causal relationship representation device can represent the causal relationship sequence 130 in the form of a graph, a map, text, etc. In some embodiments, the causal relationship representation device can represent all aspects of the causal relationship sequence 130, i.e., the causal relationships between all elements. Alternatively, in some embodiments, the causal relationship representation device can represent only some aspects of the causal relationship sequence 130, for example, only the causal relationships related to one or more target elements. In some embodiments, when the cause of the target element includes multiple elements, the causal relationship representation device can further represent the corresponding importance levels of the multiple elements, for example, in the form of different colors and / or numbers indicating different importance levels. The embodiments of the present disclosure are not limited in this respect.
[0038] FIG 1B is an exemplary block diagram illustrating a system 105 for data processing according to an embodiment of the present disclosure. The system 105 can, for example, apply and optimize the causality sequence 130 as shown in FIG 1A. It should be understood that the system 105 illustrated in FIG 1B is merely an example of an embodiment of the present disclosure and is not intended to limit the scope of the present disclosure. The embodiment of the present disclosure is equally applicable to other systems or architectures.
[0039] 1B, the system 105 may include an observation sample influencing device 140. The observation sample influencing device 140 may determine at least one element that causes the target element from the multiple elements based on the causal sequence 130. The observation sample influencing device 140 may affect the observation value of the target element by modifying the observation value of the at least one element to obtain a modified observation sample set 150. At least one observation sample in the modified observation sample set 150 includes a modified observation value of the at least one element.
[0040] In the above-mentioned scenario regarding customer satisfaction of a communication carrier, the target element is, for example, "customer satisfaction", and the causal sequence 130 can indicate which element (e.g., offer before using up packets, discount packets, etc.) is the cause of the target element "customer satisfaction". The observation sample influence device 140 can, for example, affect and change the observed values of these elements and / or develop corresponding strategies for these elements (e.g., offer more offers for the customer before using up packets, offer more discount packets for the customer), thereby improving the customer's satisfaction with the communication carrier.
[0041] In the example of the scene related to the patient's blood pressure, the target element may be, for example, "blood pressure", and the causal sequence 130 may indicate, for example, which physiological indicators are the cause of the target element "blood pressure". The observation sample influencing device 140 may, for example, influence, modify, and / or develop a corresponding strategy for these physiological indicators to keep the patient's blood pressure stable.
[0042] In the above-mentioned example of the product sales scene, the target element is, for example, "umbrella sales", and the causal sequence 130 can indicate, for example, which elements (e.g., weather, the number of umbrellas sold, etc.) are the causes of the target element "umbrella sales". The observation sample influence device 140 can, for example, influence, change, and / or develop corresponding strategies for these elements (e.g., increase the number of umbrellas offered for sale when it rains) to increase the sales of the target product, umbrellas.
[0043] In the above example of software development, the target element may be, for example, "development cycle", and the causal sequence 130 may indicate which element (e.g., the number of architecture layers, programming language, etc.) is the cause of the target element "development cycle". The observation sample influence device 140 may, for example, affect, change, and / or develop corresponding strategies for these elements (e.g., reduce the complexity of the software architecture, use a more friendly programming language, etc.) to shorten the software development cycle. Also, for example, the target element may be, for example, "operational phase software failure rate", and the causal sequence 130 may indicate which element (e.g., code length, number of modules, etc.) is the cause of the target element "operational phase software failure rate". The observation sample influence device 140 may, for example, affect, change, and / or develop corresponding strategies for these elements (e.g., shorten the code length, reduce the number of modules, etc.) to reduce the operation phase software failure rate.
[0044] As shown in FIG. 1B, the system 105 may include a causal relationship optimization device 160. The causal relationship optimization device 160 can optimize the causal relationship sequence 130 based on the modified observation sample set 150, thereby improving the accuracy of the causal relationship sequence 130. In some embodiments, the causal relationship optimization device 160 can rediscover causal relationships (e.g., similar to the execution process of the causal relationship determination device 120) between multiple elements based on the modified observation sample set 150, thereby obtaining an optimized causal relationship sequence. According to this form, the embodiment of the present disclosure can further improve the accuracy and robustness of causal discovery.
[0045] Although the causal relationship determination device 120 shown in FIG. 1A, the observation sample influence device 140 shown in FIG. 1B, and the causal relationship optimization device 160 shown in FIG. 1B are shown to be separate from each other, it should be understood that this is for illustrative purposes only and is not intended to limit the scope of the present disclosure. In some embodiments, the causal relationship determination device 120 shown in FIG. 1A, the observation sample influence device 140 shown in FIG. 1B, and the causal relationship optimization device 160 shown in FIG. 1B can be realized in the same physical device or multiple different physical devices. In some embodiments, the causal relationship determination device 120 shown in FIG. 1A and the causal relationship optimization device 160 shown in FIG. 1B can be realized as the same device. The embodiments of the present disclosure are not limited in that respect.
[0046] FIG. 2 is a schematic diagram for determining causal relationships among multiple elements according to an embodiment of the present disclosure. For simplicity and convenience of explanation, in FIG. 2, it is assumed that the number (i.e., D) of elements (i.e., observation variables) in the observation sample set 110 is 5. As shown in FIG. 2, the observation sample set 110 includes multiple observation samples for elements v1, v2, v3, v4, and v5. The dependency determination unit 121 can determine a dependency set that exists between the elements v1, v2, v3, v4, and v5 based on the observation sample set 110, which is shown as a skeleton diagram 210. For example, the dependency set 210 indicates that the elements v1 and v2 are related to each other, the elements v2 and v4 are related to each other, the elements v3 and v4 are related to each other, and the elements v3 and v5 are related to each other. The causality determination unit 122 can determine a causality sequence of elements v1, v2, v3, v4, and v5 based on the dependency set 210, e.g., shown as a directed acyclic graph 130. For example, the causality sequence 130 indicates that element v2 is a cause of element v1 (e.g., shown by edge v2→v1), element v2 is a cause of element v4 (e.g., shown by edge v2→v4), element v3 is a cause of element v4 (e.g., shown by edge v3→v4), and element v3 is a cause of element v5 (e.g., shown by edge v3→v5).
[0047] 3 is a flowchart of a method 300 for determining a causal relationship between multiple elements according to an embodiment of the present disclosure. For example, the method 300 may be performed by the causal relationship determination device 120 shown in FIG. 1A. It should be understood that the method 300 may further include additional operations not shown and / or omit some operations shown. The scope of the present disclosure is not limited in its aspects.
[0048] In block 310, the causal relationship determiner 120 obtains an observation sample set for a plurality of elements (eg, the observation sample set 110 shown in Figures 1A and 2). An observation sample in the observation sample set includes corresponding observations of a plurality of elements.
[0049] In block 320, the causal relationship determination device 120 (e.g., the dependency determination unit 121) determines a dependency relationship set that exists between the multiple elements based on the observation sample set (e.g., the skeleton diagram 210 shown in FIG. 2). A dependency relationship in the dependency relationship set indicates a pair of elements that are related to each other in the multiple elements.
[0050] In some embodiments, to determine the dependency set, the causal relationship determination device 120 can estimate a correlation coefficient between any two elements in the plurality of elements based on corresponding observation values of these two elements. For example, the correlation coefficient may be any one of the Spearman correlation coefficient or the Kendall correlation coefficient. Based on the estimated correlation coefficient between each pair of elements, the causal relationship determination device 120 can establish a correlation coefficient matrix S. For example, assuming that the total number of elements is D, S is a D×D matrix. If the element in the jth row and kth column of the matrix S is S, then the correlation coefficient matrix S is S. jk Assuming that, S jk is determined as follows:
number
number
number
[0051] In some embodiments, the causal relationship determination device 120 can establish an objective function (also referred to herein as a "first objective function") for determining a dependency set (i.e., skeleton diagram 210) based on the estimated correlation coefficient matrix S. The causal relationship determination device 120 can determine the dependency set by minimizing the first objective function. In some embodiments, for example, the causal relationship determination device 120 can learn a precision matrix Ω indicating corresponding dependencies between elements based on a graphical Lasso algorithm. For example, Ω is determined as follows:
number
number
number
number
number
[0052] Alternatively, in some embodiments, the causal relationship determination device 120 can determine the dependency relationship set M by performing a conditional independence test on multiple elements. The conditional independence test can determine whether two elements are independent of each other in a given condition set. In a conventional conditional independence test, when determining whether two elements are independent, it is necessary to set the condition set to any combination of all other elements. When the number of elements (i.e., D) is large, the computational overhead is very large. When the condition set includes many other elements, the two elements are likely to be determined to be independent of each other. In order to quickly and accurately realize the discovery of high-dimensional causal structures, in some embodiments, when determining the dependency relationship set by performing a conditional independence test on multiple elements, the causal relationship determination device 120 can limit the number of other elements in the condition set to 1. According to this embodiment, not only can the computational overhead of the conditional independence test be reduced, but the number of element pairs determined to be independent of each other can be reduced, which is useful for discovering subsequent causal relationship sequences.
[0053] In this manner, by determining a dependency set, an embodiment of the present disclosure can quickly discover causal relationships between many elements by reducing the size of the variable space to be searched.
[0054] In block 330, the causality determination device 120 (e.g., the causality determination unit 122) determines a causality sequence of the multiple elements based on the dependency set. The causality sequence may indicate that one element in a pair of related elements is a cause of the other element.
[0055] In some embodiments, to determine the causal relationship sequence, the causal relationship determination device 120 may determine, for a plurality of interrelated element pairs indicated by the dependency set, the influence of one element in each interrelated element pair on the other element. In some embodiments, to determine the influence of one element in each element pair on the other element, the causal relationship determination device 120 may establish a second objective function based on a predetermined distribution (e.g., a Gaussian distribution or other distribution) and determine the influence of one element in each element pair on the other element by minimizing the second objective function. The second objective function may be established, for example, based on the following two factors: first, to ensure that the discovered causal structure has a good degree of data fitting for the observed data sample; and second, to ensure that the discovered causal structure remains sparse.
[0056] In some embodiments, assuming a total number of elements is D, and denoting the determined influence for each of a plurality of element pairs by a matrix B, B is determined as follows:
number
number
number
number
number
number
number
number
[0057] In some embodiments, the causal relationship determination device 120 can determine the causal relationship sequence 130 based on the corresponding influences determined for a plurality of element pairs and the observation sample set. For example, FIG. 4 is a flowchart of a method 400 for determining a causal relationship sequence according to an embodiment of the present disclosure. The method 400 may be performed by the causal relationship determination device 120 shown in FIG. 1A. It should be understood that the method 400 may further include additional operations not shown and / or omit some operations shown. The scope of the present disclosure is not limited in its aspects.
[0058] In block 410, the causality determination device 120 may obtain a historical causality sequence and a score for the historical causality. Here, the historical causality sequence is denoted as Q S and the score of historical causality is given by f(Q S ) is assumed to be denoted as
[0059] In some embodiments, in an initial situation, the causality determiner 120 can initialize the historical causality sequence to an empty sequence, i.e., Q S The causality determination device 120 can determine the score of the initial causality corresponding to the empty sequence as the score of the historical causality, i.e.
number
number
number
number
number
number
number
number
number
number
number
number
number
number
[0060] In block 420, the causality determiner 120 determines the historical causality sequence Q S and the plurality of element pairs indicated by the dependency set (e.g., skeleton diagram 210), determine one or more candidate elements for possible addition to the causal sequence. In some embodiments, the one or more candidate elements are included in a candidate node set V\Q S The candidate node set V\Q S is a node set V with S Indicates all nodes other than those contained in .
[0061] If one or more candidate elements do not exist (i.e., in the set V\Q S is empty), then in block 470 the causality determiner 120 S may be output as a determined causality sequence 130.
[0062] If there are one or more candidate elements, then in block 440, the causal relationship determiner 120 may determine a score for one or more candidate causal relationships corresponding to the one or more candidate elements. For example, for a set of candidate nodes V\Q S Each candidate node in
number
number
number
number
[0063] In block 450, the causal relationship determiner 120 may select a candidate element to add to the causal relationship sequence from the one or more candidate elements based on the determined score of the one or more candidate causal relationships.
[0064] In some embodiments, the causal relationship determination device 120 can determine a minimum candidate causal relationship score from the scores of one or more candidate causal relationships and select the candidate element associated with the minimum candidate causal relationship score to add to the causal relationship sequence 130.
[0065] Additionally or alternatively, in some embodiments, in order to more quickly determine the causal relationship sequence, the causal relationship determiner 120 may obtain constraints associated with the causal relationship sequence awaiting determination.
[0066] In some embodiments, the causal relationship determination device 120 can obtain expert information indicating the constraint condition, and determine the constraint condition based on the obtained expert information. Taking the example shown in Fig. 2 as an example, the expert information can indicate, for example, that node v3 is before node v4, that is, the element corresponding to node v3 may be the cause of the element corresponding to node v4, but the element corresponding to node v4 may not be the cause of the element corresponding to node v3.
[0067] Additionally or alternatively, in some embodiments, the causality determiner 120 can determine constraints based on the historical causality sequence and a number of interrelated element pairs indicated in the skeleton diagram 210. For example, in the example shown in FIG. Sassumes that an element corresponding to node v3 is a cause of an element corresponding to node v4 (i.e., there is an edge v3→v4 in the causality sequence 130). Also, based on the skeleton diagram M, it can be determined that nodes v1 and v2 are related to each other, nodes v2 and v4 are related to each other, elements v3 and v4 are related to each other, and elements v3 and v5 are related to each other. Thus, nodes v1, v2, and v4 constitute a strongly connected node set, and nodes v3 and v5 constitute a strongly connected node set. In this case, for example, the causality determination device 120 can determine that node set {v3, v5} is before node set {v1, v2, v4}. That is, a node in node set {v3, v5} may be a cause of a node in node set {v1, v2, v4}, but any node in node set {v1, v2, v4} may not be a cause of a node in node set {v3, v5}.
[0068] In some embodiments, in response to obtaining a constraint associated with the causal relationship sequence to be determined, the causal relationship determiner 120 may select a candidate element to add to the causal relationship sequence from one or more candidate elements such that the addition of the selected candidate element meets the obtained constraint. For example, if the addition of a candidate element associated with a minimum candidate causal relationship score violates the constraint, the causal relationship determiner 120 may select the other candidate element (e.g., the candidate element associated with the next smallest candidate causal relationship score) to add to the causal relationship sequence 130.
[0069] According to this aspect, by using constraint conditions, the number of candidate elements is limited in the process of determining the causal relationship sequence, so that the causal relationship sequence can be determined more quickly.
[0070] In response to the candidate element being selected, at block 460, the causality determiner 120 determines the historical causality sequence Q S and the historical causality score f(Q SFor example, the causality determination device 120 may update the historical causality sequence Q S Instead of, we use the candidate causal sequence Q corresponding to the selected candidate element. S ' to obtain the score of historical causality f(Q S ), we use the candidate causal sequence Q S ', the score f(Q S ') can be used.
[0071] In some embodiments, the causal relationship determiner 120 may repeatedly execute blocks 410-460 in the method 400 (ie, up to block 470) until all possible candidate elements have been searched.
[0072] 5 is a flow chart of a method 500 for influencing an observation value of a target element, according to an embodiment of the present disclosure. For example, the method 500 may be performed by the observation sample influencing device 140 shown in FIG. 1B. In some embodiments, the method 500 may be performed after the method 300. It should be understood that the method 500 may further include additional operations not shown and / or omit some operations shown. The scope of the present disclosure is not limited in its aspects.
[0073] In block 510, the observation sample influencing unit 140 determines at least one element from the plurality of elements that is causal to the target element based on the causal sequence. Then, in block 520, the observation sample influencing unit 140 influences the observed value of the target element by modifying the observed value of the at least one element. In some embodiments, for example, the observation sample influencing unit 140 can influence the observed value of the target element by influencing, modifying, and / or developing a corresponding strategy for the at least one element.
[0074] In the above-mentioned scenario regarding customer satisfaction of a communication carrier, the target element is, for example, "customer satisfaction". The observation sample influence device 140 can determine which element (e.g., offer before using up packets, discount packets, etc.) is the cause of the target element "customer satisfaction" based on the causal sequence 130. The observation sample influence device 140 can further influence, change, and / or develop corresponding strategies for these elements (e.g., provide more offers for customers before using up packets, provide more discount packets for customers) to improve the customer's satisfaction with the communication carrier.
[0075] In the above example of a scene related to the patient's blood pressure, the target element is, for example, "blood pressure". The observation sample influencing device 140 can determine which physiological indicators are the cause of the target element "blood pressure" based on the causal sequence 130. The observation sample influencing device 140 can further influence, modify, and / or develop corresponding strategies for these physiological indicators to keep the patient's blood pressure stable.
[0076] Taking the above-mentioned product sales scene as an example, the target element is, for example, "umbrella sales". The observation sample influencing device 140 can determine which element (e.g., weather, the number of umbrellas sold, etc.) is the cause of the target element "umbrella sales" based on the causal sequence 130. The observation sample influencing device 140 can further influence, change, and / or develop corresponding strategies for these elements (e.g., increase the number of umbrellas offered for sale when it rains) to increase the sales of the target product, umbrellas.
[0077] In the above example of software development, the target element is, for example, "development cycle". The observation sample influence device 140 can determine which element (e.g., the number of architecture layers, programming language, etc.) is the cause of the target element "development cycle" based on the causal sequence 130. The observation sample influence device 140 can further influence, change, and / or develop corresponding strategies for these elements (e.g., reducing the complexity of the software architecture, using a more friendly programming language, etc.) to shorten the software development cycle. In addition, for example, the target element may be "failure rate of software in operation phase". The observation sample influence device 140 can determine which element (e.g., code length, number of modules, etc.) is the cause of the target element "failure rate of software in operation phase" based on the causal sequence 130. The observation sample influencing device 140 can further influence, modify, and / or develop strategies to address these factors (e.g., shortening code length, reducing the number of modules, etc.) to reduce the failure rate of the production-phase software.
[0078] 6 is a flowchart of a method 600 for optimizing causal relationships according to an embodiment of the present disclosure. For example, the method 600 may be performed by the causal relationship optimizer 160 shown in FIG. 1B. In some embodiments, the method 600 may be performed after the method 500. It should be understood that the method 600 may further include additional operations not shown and / or omit some operations shown. The scope of the present disclosure is not limited in its aspects.
[0079] In block 610, the causal relationship optimization device 160 obtains a modified observation sample set for the multiple elements. In some embodiments, at least one observation sample in the modified observation sample set may include a modified observation value of at least one element (e.g., at least one element is a cause of the target element). Then, in block 620, the causal relationship optimization device 160 can optimize the causal relationship sequence based on the modified observation sample set. In some embodiments, for example, the causal relationship optimization device 160 can newly discover the causal relationships (e.g., similar to the execution process of the causal relationship determination device 120) between the multiple elements based on the modified observation sample set 150, and thus obtains an optimized causal relationship sequence. According to this form, the embodiment of the present disclosure can further improve the accuracy and robustness of causal discovery.
[0080] FIG. 7 is a schematic block diagram of an exemplary device 700 suitable for implementing embodiments of the present disclosure. For example, the causal relationship determination device 120 shown in FIG. 1A, the observation sample influence device 140 and / or the causal relationship optimization device 160 shown in FIG. 1B may be implemented by the device 700. As shown in the figure, the device 700 includes a central processing unit (CPU) 701 that can perform various suitable operations and processes based on computer program instructions stored in a read-only memory (ROM) 702 or loaded from a storage unit 708 into a random access memory (RAM) 703. The RAM 703 may further store various programs and data required for the operation of the device 700. The CPU 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.
[0081] Several components in device 700 are connected to I / O interface 705, including input unit 706 such as a keyboard, mouse, etc., output unit 707 such as various types of displays, speakers, etc., storage unit 708 such as a magnetic disk, optical disk, etc., and communication unit 709 such as a network card, modem, wireless communication transceiver, etc. The communication unit 709 allows device 700 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.
[0082] The processing unit 701 is configured to perform each of the above-mentioned steps and processes, for example, methods 300, 400, 500 and / or 600. For example, in some embodiments, the methods 300, 400, 500 and / or 600 may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 708. In some embodiments, some or all of the computer program may be loaded and / or installed in the device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the CPU 701, one or more steps of the above-mentioned methods 300, 400, 500 and / or 600 may be performed.
[0083] The present disclosure may be a system, a method, and / or a computer program product, which may include a computer-readable storage medium having computer-readable program instructions stored thereon for carrying out various aspects of the present disclosure.
[0084] A computer-readable storage medium may be a physical device capable of holding and storing instructions for use by an instruction execution device. A computer-readable storage medium may 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 suitable combination of the foregoing. More specific examples (non-exhaustive list) of computer-readable storage media include mechanical coding devices such as portable computer diskettes, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disk read-only memory (CD-ROM), digital versatile disk (DVD), memory stick, floppy disk, punch cards or protruding structures in recessed grooves having instructions stored therein, and any suitable combination of the foregoing. A computer-readable storage medium as used herein should not be construed as a transitory signal itself, such as radio waves or other electromagnetic waves propagating freely, electromagnetic waves propagating through a wave guide or other propagation medium (e.g., light pulses through fiber optic cables), or electronic signals transmitted through wires.
[0085] The computer readable program instructions described herein can be downloaded from computer readable storage media into various computing / processing devices or downloaded to an external computer or external storage device over a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network may include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium in each computing / processing device.
[0086] The computer program instructions for carrying out the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state setting data, or source or target code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, and traditional procedural programming languages such as "C" or similar programming languages. The computer readable program instructions may be executed entirely or partially on the user's computer, may be executed as one independent software package, may be executed partially on the user's computer and partially on a remote computer, or may be executed entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may execute computer readable program instructions by utilizing state information of the computer readable program instructions to customize the electronic circuitry to perform aspects of the present disclosure.
[0087] Aspects of the present disclosure will now be described with reference to flowcharts and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the present disclosure. It will be understood that each block of the flowcharts and / or block diagrams and combinations of blocks in the flowcharts and / or block diagrams can all be implemented by computer readable program instructions.
[0088] These computer readable program instructions may be provided to a processing unit of a general purpose computer, special purpose computer, or other programmable data processing apparatus to create a machine that, when these instructions are executed by a processing unit of the computer or other programmable data processing apparatus, creates an apparatus that implements the functions / operations specified in one or more blocks in the flowcharts and / or block diagrams. These computer readable program instructions may be stored on a computer readable storage medium, and these instructions cause a computer, programmable data processing apparatus, and / or other device to operate in a determined manner, such that the computer readable medium on which the instructions are stored includes an article of manufacture that includes instructions for each aspect of implementing the functions / operations specified in one or more blocks in the flowcharts and / or block diagrams.
[0089] The computer readable program instructions can be loaded into a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be executed by the computer, other programmable data processing apparatus, or other device to create a computer-implemented process, such that the instructions executed on the computer, other programmable data processing apparatus, or other device implement the functions / acts specified in one or more blocks in the flowcharts and / or block diagrams.
[0090] The flowcharts and block diagrams in the drawings illustrate possible implementation architectures, functions, and operations of the systems, methods, and computer program products according to the embodiments of the present disclosure. In this regard, each block in the flowcharts or block diagrams may represent a module, program segment, or part of instructions, which includes one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions shown in the blocks may occur in a different order from the order shown in the drawings. For example, two blocks shown in succession may actually be executed essentially in parallel or in the reverse order, as determined by the related functionality. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented by a dedicated hardware-based system that executes the specified functions or operations, or may be implemented by a combination of dedicated hardware and computer instructions.
[0091] Although each embodiment of the present disclosure has been described above, the above description is not exhaustive, but merely illustrative, and is not limited to each of the disclosed embodiments. Many modifications and changes will be apparent to those skilled in the art without departing from the scope and spirit of each of the described embodiments. The selection of terms used in this specification is intended to best explain the principles, practical applications, or improvements to the technology in the market of each embodiment, or to enable those skilled in the art to understand each of the embodiments disclosed in this specification.
Claims
1. obtaining a set of observation samples for a plurality of elements, an observation sample in the set of observation samples including corresponding observations of the plurality of elements; determining a set of dependencies that exist between the plurality of elements based on the observation sample set, where a dependency in the dependency set indicates a pair of elements that are related to each other in the plurality of elements; determining a causality sequence of the plurality of elements based on the dependency set, the causality sequence indicating that one element in the pair of elements that are related to each other is a cause of the other element; Determining the dependency set includes: For any two elements in the plurality of elements, estimating a correlation coefficient between the two elements based on corresponding observations of the two elements in the observation sample set; Establishing a first objective function for determining the dependency set based on the estimated result, the first objective function being expressed by the following equation, where Ω is a precision matrix indicating the dependency relationships, λ is a predefined coefficient, S is a correlation coefficient matrix established based on the correlation coefficients, and tr() is a trace of the matrix; [0010] determining the dependency set by minimizing the first objective function while satisfying Ω≧0. A data processing method executed by a data processing device.
2. the plurality of elements includes a target element, and the method further comprises: determining at least one element from the plurality of elements that is a cause of the target element based on the causal sequence; The method of claim 1 , further comprising: modifying an observed value of the at least one element to affect an observed value of the target element.
3. obtaining a modified set of observation samples for the plurality of elements, wherein at least one observation sample in the modified set of observation samples includes a modified observation value of the at least one element; The method of claim 2 , further comprising optimizing the causal sequence based on the modified sample set of observations.
4. Determining the causality sequence includes: determining, for each of a plurality of pairs of related elements represented by the dependency set, an influence of one element on the other element in the pair; determining the causality sequence based on the sample set of observations and corresponding influences determined for a plurality of pairs of elements.
5. Determining the influence of one element on the other in each element pair is establishing a second objective function for determining the corresponding influences for the plurality of element pairs based on a predetermined distribution, the second objective function being expressed as follows, and a matrix B indicating the determined influences for each of the plurality of element pairs, X={x 1 , x 2 , . . . , x D}∈R N×D denotes the observed sample set, N denotes the total number of observed samples in X, D denotes the total number of elements, and vector x i ∈R N (where 1≦i≦D) denotes the N observations of the i-th element, and x i,n (where 1≦i≦D and 1≦n≦N) denotes the n-th observation of the i-th element, and the vector β i ∈R D-1 denotes the influence of each element on the i-th element, and |B| 0 denotes the total number of non-zero elements in matrix B; and [0025] 5. The method of claim 4, further comprising: determining the influence of one element on the other element in each element pair by minimizing the second objective function.
6. Determining the causality sequence includes: obtaining a historical causality sequence and a historical causality score corresponding to the historical causality sequence; determining one or more candidate elements for potential addition to the causality sequence based on the historical causality sequence and the plurality of element pairs; responsive to the presence of the one or more candidate elements, determining scores of one or more candidate causal relations corresponding to the one or more candidate elements based on the scores of the historical causal relations, the corresponding effects determined for the plurality of element pairs, and the sample set of observations; selecting a candidate element from the one or more candidate elements to add to the causal relations sequence based on a score of the one or more candidate causal relations; and updating the historical causality sequence and the scores of the historical causality based on the selected candidate element at least once.
7. The method of claim 6 , further comprising: in response to the one or more candidate elements not being present, determining the historical causality sequence as the causality sequence.
8. Obtaining the historical causality sequence and the scores of the historical causality includes: initializing the historical causality sequence to an empty sequence; and determining an initial causality score corresponding to the empty sequence as the historical causality score.
9. Selecting the candidate element from the one or more candidate elements includes: determining a minimum candidate causal relation score from the scores of the one or more candidate causal relations; and selecting from the one or more candidate elements the candidate element associated with the smallest candidate causal relationship score.
10. Selecting the candidate element from the one or more candidate elements includes: obtaining constraints associated with the causality sequence awaiting determination; and selecting the candidate element to add to the causality sequence from the one or more candidate elements such that the addition of the candidate element meets the constraints.
11. Obtaining the constraints includes: obtaining information indicative of the constraint; and determining the constraint based on the information.
12. Obtaining the constraints includes: The method of claim 10 comprising determining the constraint based on the historical causality sequence and the plurality of element pairs.
13. A computer program product for causing a data processing apparatus to carry out the method according to any one of claims 1 to 12.
14. at least one processing unit; and at least one memory coupled to said at least one processing unit for storing a computer program as claimed in claim 13 for execution by said at least one processing unit.
