Knowledge Acquisition Method Based on Data Relative Steady-State Analysis

Abstract

To provide a method for knowledge acquisition based on data relative steady-state analysis configured to improve the precision and other performances of digital decision jointly driven by limited multi-source data and knowledge.SOLUTION: A method to solve the problem of knowledge acquisition of limited multi-source data under the condition of no domain expert knowledge comprises: based on relative steady-state analysis of data elements, defining reference data elements and accompanying data elements, and calculating trend degrees and trend directions of the reference data elements and the accompanying data elements and a relative trend value of the accompanying data element; and obtaining quantitative knowledge behind the data element, a data unit or an entity node based on the relative trend value of the accompanying data element.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to knowledge acquisition techniques in the field of digital decision making, and in particular to a knowledge acquisition method based on data relative steady-state analysis. [Background technology]

[0002] In the field of digital decision-making, there are two common methods for improving the accuracy and other performance of digital decision-making: one is to expand the data scale, and the other is to integrate knowledge into data. Expanding the data scale requires training on big data using more computing resources, while integrating knowledge into data uses limited data and computing resources to achieve relatively good decision-making performance such as accuracy. In situations where data and computing resources are limited, integrating knowledge into data is undoubtedly a better choice, and the key technology to be solved is knowledge acquisition.

[0003] Typically, the knowledge behind attribute data is mainly based on the knowledge of domain experts or acquired through machine learning methods. When domain expert knowledge is available, such as industry knowledge (medical care, finance, etc.) or attribute reference values ​​of indicators (medical testing, economic management, etc.), it is relatively easy to acquire the knowledge behind attribute data based on the domain expert knowledge. That is, it is sufficient to define a quantification knowledge function according to the decision requirements and quantify the domain expert knowledge. Acquiring the knowledge behind attribute data using machine learning methods requires a large amount of training data and often has poor results. Summary of the Invention

[0004] In response to the shortcomings of the prior art, the present invention utilizes a limited amount of available multi-source data without the knowledge of domain experts, and is a desirable and effective approach to obtain quantitative knowledge behind attribute data through relative steady-state analysis of the data elements themselves. Data relative steady-state analysis begins with defining a contrast value between one data element and another data element with different characteristics, with the goal of reflecting the change status of data elements in one vector relative to data elements in another vector. This change can be stable or unstable. For better comparison, data elements arranged in order in the reference vector are defined as reference data elements, and data elements in an adjoint vector constructed based on data elements in the same data set are defined as adjoint data elements. Knowledge behind attribute data is calculated through relative steady-state analysis between the reference data elements and the adjoint data elements.

[0005] JPEG0007792156000001.jpg139170

[0006] JPEG0007792156000002.jpg28170

[0007] JPEG0007792156000003.jpg28170

[0008] JPEG0007792156000004.jpg17170

[0009] JPEG0007792156000005.jpg11170

[0010] JPEG0007792156000006.jpg22170

[0011] JPEG0007792156000007.jpg27170

[0012] JPEG0007792156000008.jpg22170

[0013] JPEG0007792156000009.jpg66170

[0014] Preferably, the "coincident" relative trend direction is quantified as 1, and the "semi-coincident" and "incident" relative trend directions are quantified as 3 / 2 and 2, respectively, with the purpose of quantifying the relative trend direction to better reflect the trend value of the accompanying data element relative to the reference data element.

[0015] The beneficial effects of the present invention are as follows: The present invention, based on given finite data and computing resources, uses a method of incorporating knowledge into data to better improve digital decision-making performance. Knowledge acquisition is key to solving the problem of knowledge acquisition from finite multi-source data without the knowledge of domain experts. The present invention uses relative steady-state analysis of the data elements themselves to define the trend degree, trend direction, and relative trend value of the reference data element and associated data element, and acquires quantified knowledge behind the data element, data unit, or entity node based on the relative trend value of the associated data element. The present invention provides a new method for knowledge acquisition based on finite multi-source data, thereby improving the accuracy and other performance of digital decision-making jointly driven by finite multi-source data and knowledge. [Brief explanation of the drawings]

[0016] JPEG0007792156000010.jpg53170 DETAILED DESCRIPTION OF THE INVENTION

[0017] The present invention will be further explained below in conjunction with the accompanying drawings and examples. As shown in Figure 1, the present invention is a knowledge acquisition method based on data relative steady-state analysis, which can obtain quantification knowledge of data elements through trend degree definition, trend direction definition, and relative trend value definition, and then further define quantification knowledge of data units consisting of data elements and quantification knowledge of entity nodes consisting of data units. The specific steps are as follows:

[0018] JPEG0007792156000011.jpg68170

[0019] JPEG0007792156000012.jpg65170

[0020] JPEG0007792156000013.jpg48170

[0021] JPEG0007792156000014.jpg22170

[0022] JPEG0007792156000015.jpg105170

[0023] JPEG0007792156000016.jpg28170

[0024] JPEG0007792156000017.jpg119170

[0025] JPEG0007792156000018.jpg29170

[0026] JPEG0007792156000019.jpg17170

[0027] JPEG0007792156000020.jpg10170

[0028] JPEG0007792156000021.jpg22170

[0029] JPEG0007792156000022.jpg22170

[0030] JPEG0007792156000023.jpg21170

[0031] JPEG0007792156000024.jpg177170

[0032] JPEG0007792156000025.jpg109170

[0033] The method of the present invention involves creating an application program, storing it in memory, and arranging it to be called and executed by a processor or computer.

Claims

1.

2.

3.

4. 4. The method for knowledge acquisition based on data relative steady-state analysis according to claim 1, wherein the "concordant" relative trend direction is quantified as 1, and the "semi-concordant" and "incongruent" relative trend directions are quantified as 3 / 2 and 2, respectively.

5.

6. The method for knowledge acquisition based on data relative steady-state analysis according to any one of claims 1 to 3, characterized in that the method is arranged to create an application program, store it in a memory, and call and execute it by a processor or computer.

Images