A knowledge acquisition method based on relative steady-state analysis of data

Abstract

The present invention discloses a knowledge acquisition method based on relative steady-state analysis of data. To solve the problem of knowledge acquisition from a limited amount of multi-source data in the absence of domain expert knowledge, the present invention is based on the relative steady-state analysis of the data elements themselves. By defining a baseline data element and an accompanying data element, the trend degree and trend direction of the baseline data element and the accompanying data element, as well as the relative trend value of the accompanying data element, are calculated. Based on the relative trend value of the accompanying data element, the quantitative knowledge behind the data element, data unit, or entity node is acquired. The present invention provides a new method for knowledge acquisition based on a limited amount of multi-source data, which can improve the accuracy and other performance of digital decision-making driven by a limited amount of multi-source data and knowledge.

Description

Technical Field

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

[0002] In the field of digital decision-making, there are two common approaches to improving accuracy and other performance: expanding data volumes, and incorporating knowledge into the data. Increasing data volumes requires utilizing more computing resources to train big data, while incorporating knowledge into the data leverages available limited data and computing resources to achieve better decision-making accuracy and other performance. When data and computing resources are limited, incorporating knowledge into the data is undoubtedly the preferred approach. The key technology to address is knowledge acquisition.

[0003] Typically, the knowledge behind attribute data is primarily based on domain expert knowledge or acquired through machine learning. When domain expert knowledge is available, such as industry knowledge (e.g., healthcare, finance), or reference values ​​for indicators and attributes (e.g., medical testing, economic operations), it's relatively easy to acquire the knowledge behind attribute data based on this knowledge. This involves simply defining a quantitative knowledge function based on decision-making requirements and quantifying the domain expert knowledge. However, acquiring the knowledge behind attribute data through machine learning requires a large amount of training data, and the results are often less than ideal. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention uses a limited amount of available multi-source data without the knowledge of domain experts to obtain the quantitative knowledge behind the attribute data through relative steady-state analysis of the data elements themselves. This is a desirable and effective way. The relative steady-state analysis of data starts with defining the comparison value between one data element and another data element with different characteristics. Its purpose is to reflect the changes in the data elements in one vector relative to the data elements in another vector. This change may be stable or unstable. In order to better compare, the data elements arranged in sequence in the baseline vector are defined as baseline data elements, and the data elements in the companion vector constructed based on the data elements in the same data set are defined as companion data elements. Through the relative steady-state analysis between the baseline data elements and the companion data elements, the knowledge behind the attribute data is calculated.

[0005] The specific steps of the present invention are as follows:

[0006] (1) Definition of trend

[0007] Suppose there are n entity nodes e1,…,e i ,…,e n The m data sets generated are DS1,…,DSk ,…,DS m , where the dataset DS k ={s k,1 ,…,s k,i ,…,s k,n}, data unit L is the dimension of the data unit. According to the data set DS k Constructing a base vector from the data elements and adjoint vector and Based on the reference vector and adjoint vector The data elements in the table are used as the base data elements and accompanying data elements respectively; and The second data element in and First, define the trend degree as and

[0008] (2) Define benchmark data elements Trend direction and accompanying data elements Trend direction

[0009] (3) Knowledge acquisition based on relative trend value

[0010] Companion data elements The relative trend Depends on its trend and corresponding benchmark data elements Trend

[0011]

[0012] Companion data elements Relative trend direction Depends on its trend direction Corresponding benchmark data elements Trend direction

[0013]

[0014] Among them, "consistent", "half-consistent" and "inconsistent" mean "consistent", "half-consistent" and "inconsistent".

[0015] The relative trend direction Quantification, based on accompanying data elements The relative trend and the quantified relative trend direction By the relative trend function f cv Calculated The relative trend value of Relative trend function f cv It is the product of relative trend degree and relative trend direction.

[0016]

[0017] According to the data set DS k The base vector constructed from the data elements in Solve for the accompanying data elements The relative trend values ​​are Defining accompanying data elements The comprehensive relative trend value is:

[0018]

[0019] set up The maximum value in Z k,max , will be accompanied by the data element The knowledge is quantified as The normalized relative trend value of , that is:

[0020]

[0021] Get dataset DS k Quantitative knowledge of other accompanying data elements in, as well as other data sets DS1,…,DS k-1 ,DS k+1 ,…,DS m Quantitative knowledge accompanying data elements.

[0022] For dataset DS k Medium Data Unit Its quantitative knowledge consists of the accompanying data elements The quantitative knowledge is defined as follows:

[0023]

[0024] For generating data sets DS1,…,DS k ,…,DS m n entity nodes, entity node e i The data units are recorded in the matrix [s 1,i ,…,s k,i ,…,s m,i ] T In the entity node e iThe quantitative knowledge depends on the data units s contained 1,i ,…,s k,i ,…,s m,i Quantitative knowledge, namely:

[0025]

[0026] Preferably, define the reference data element Trend direction

[0027] 1) If Then the trend direction is "parallel" and

[0028] 2) If The trend direction is "upward" and

[0029] 3) If The trend direction is "downward" and

[0030] Preferably, define the accompanying data element Trend direction

[0031] 1) If and or and or and Then the trend direction is "parallel" and

[0032] 2) If and The trend direction is "upward" and

[0033] 3) If and The trend direction is "downward" and

[0034] 4) If and or and The trend direction is "semi-upward" and

[0035] 5) If and or and Then the trend direction is "semi-downward" and

[0036] Preferably, the relative trend direction of "consistent" is quantified as 1, and the relative trend directions of "semi-consistent" and "inconsistent" are quantified as 3 / 2 and 2. The purpose of quantifying the relative trend direction is to better reflect the trend value of the accompanying data element relative to the benchmark data element.

[0037] The present invention has the beneficial effects:

[0038] The present invention can better improve the performance of digital decision-making by integrating knowledge into data based on given limited data and computing resources. Knowledge acquisition is the key. In order to solve the problem of knowledge acquisition of limited multi-source data in the absence of domain expert knowledge, the present invention defines the trend degree, trend direction and relative trend value of the baseline data element and the accompanying data element through relative steady-state analysis of the data element itself, and obtains the quantitative knowledge behind the data element, data unit or entity node based on the relative trend value of the accompanying data element. The present invention provides a new method for knowledge acquisition based on limited multi-source data, which can improve the accuracy and other performance of digital decision-making driven by limited multi-source data and knowledge. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flow chart of a knowledge acquisition method based on relative steady-state analysis of data according to the present invention.

[0040] Figure 2 is a vector and Comparison diagram of the characteristic value and trend degree distribution of data elements, where (A) is the smooth and consistent state of the characteristic value and trend degree distribution of data elements with the same direction, (B) is the smooth and consistent state of the characteristic value and trend degree distribution of data elements with different directions, (C) is the smooth and inconsistent state of the characteristic value and trend degree distribution of data elements with the same direction, (D) is the smooth and inconsistent state of the characteristic value and trend degree distribution of data elements with different directions, (E) is the fluctuating and inconsistent state of the characteristic value and trend degree distribution of data elements with the same fluctuation, and (F) is the fluctuating and inconsistent state of the characteristic value and trend degree distribution of data elements with different fluctuations. DETAILED DESCRIPTION

[0041] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0042] like Figure 1As shown, the present invention provides a knowledge acquisition method based on relative steady-state data analysis. By defining trend degree, trend direction, and relative trend value, quantitative knowledge of data elements can be acquired. On this basis, quantitative knowledge of data units composed of data elements and quantitative knowledge of entity nodes composed of data units can be further defined. The specific steps are as follows:

[0043] (1) Definition of trend

[0044] Suppose there are n entity nodes e1,…,e i ,…,e n The m data sets generated are DS1,…,DS k ,…,DS m , where the dataset DS k ={s k,1 ,…,s k,i ,…,s k,n}, data unit L is the dimension of the data unit. k Central reference vector ( and ) and the accompanying vector The data elements in the table serve as the base data elements and accompanying data elements. and The second data element in and Begin to define the trend of the data elements. and Data elements and For example (i≥2), its trend degree is defined as and and The definition of trend degree of other data elements after the second data element in is similar. The data elements in are in ascending order, so The trend of the data elements in is non-negative.

[0045] The following vector analysis and The comparison status of the corresponding data elements in . The trend of the data elements in The distribution of the trend degree of the data elements in has three cases: smooth consistency, smooth inconsistency and fluctuating inconsistency, such as Figure 2As shown in the figure. In the smooth consistency state, the comparative distribution of the accompanying data element and the benchmark data element shows a consistent pace of change in the characteristic value (i.e., the value of the data element) and the trend degree. In the smooth inconsistency state, the comparative distribution of the accompanying data element and the benchmark data element shows a consistent proportion or regularity in the change of the characteristic value and the trend degree. In the fluctuating inconsistency state, the comparative distribution of the accompanying data element and the benchmark data element has no fixed proportion or regularity in the change of the characteristic value and the trend degree.

[0046] (2) Definition of trend direction

[0047] The trend value of the accompanying data element relative to the benchmark data element is not only related to their trend degree, but also to their trend direction. The trend direction of the benchmark data element has three cases: "upward", "downward" and "parallel", and the trend direction of the accompanying data element has five cases: "upward", "downward", "parallel", "half-upward" and "half-downward". For the convenience of calculation, the trend directions of "parallel", "upward", "downward", "half-upward" and "half-downward" are quantized to 0, 1, -1, 0.5 and -0.5. Medium benchmark data elements and adjoint vector Accompanying data elements For example, its trend direction and The definition of is as follows.

[0048] Benchmark Data Elements Trend direction

[0049] 1) If Then the trend direction is "parallel" and

[0050] 2) If The trend direction is "upward" and

[0051] 3) If The trend direction is "downward" and

[0052] Companion data elements Trend direction

[0053] 1) If and Then the trend direction is "parallel" and

[0054] 2) If and Then the trend direction is "parallel" and

[0055] 3) If and Then the trend direction is "parallel" and

[0056] 4) If and The trend direction is "upward" and

[0057] 5) If and The trend direction is "downward" and

[0058] 6) If and The trend direction is "semi-upward" and

[0059] 7) If and Then the trend direction is "semi-downward" and

[0060] 8) If and The trend direction is "semi-upward" and

[0061] 9) If and Then the trend direction is "semi-downward" and

[0062] (3) Knowledge acquisition based on relative trend value

[0063] The relative trend value of the accompanying data element is determined by its relative trend degree and relative trend direction. For example, its relative trend Depends on its trend and corresponding benchmark data elements Trend

[0064]

[0065] Companion data elements Relative trend direction Depends on its trend direction and corresponding benchmark data elements Trend direction

[0066]

[0067] Among them, "consistent", "half-consistent" and "inconsistent" mean "consistent", "half-consistent" and "inconsistent". Quantification, based on accompanying data elements The relative trend and the quantified relative trend direction The relative trend value is given by the relative trend function f cv Here, we define f cv It is the product of relative trend degree and relative trend direction to reflect the multiple effects of relative trend direction.

[0068]

[0069] If the "consistent" relative trend direction is quantified as 1, then the "semi-consistent" and "inconsistent" relative trend directions can be quantified as 3 / 2 and 2, that is, the quantified values ​​of the "semi-consistent" and "inconsistent" relative trend directions are 1.5 times and 2 times the quantified values ​​of the "consistent" relative trend direction. The purpose of quantifying the relative trend direction is to better reflect the trend value of the accompanying data element relative to the benchmark data element. Generally speaking, a consistent relative trend direction means that the trend value of the accompanying data element is smaller than that of the benchmark data element, while an inconsistent relative trend direction means that the trend value of the accompanying data element is larger than that of the benchmark data element. In general, the relative trend direction has a greater impact on the relative trend value than the relative trend degree.

[0070] According to the data set DS k A vector constructed from the data elements in As the reference vector, solve the accompanying data elements separately The relative trend value of Defining accompanying data elements The comprehensive relative trend value is:

[0071]

[0072] Assume that the dataset DS k The maximum relative trend value of all accompanying data elements in is Z k,max , then the data element The knowledge is quantified as The normalized relative trend value of , that is:

[0073]

[0074] Similarly, you can get the data set DS k Quantitative knowledge of other accompanying data elements in, as well as other data sets DS1,…,DS k-1 ,DSk+1 ,…,DS m Quantitative knowledge accompanying data elements.

[0075] For dataset DS k Medium Data Unit Its quantitative knowledge consists of the accompanying data elements The quantitative knowledge is defined as follows:

[0076]

[0077] For generating multi-source data DS1,…,DS k ,…,DS m n entity nodes, entity node e i The data units are recorded in the matrix [s 1,i ,…,s k,i ,…,s m,i ] T In the entity node e i The quantitative knowledge depends on the data units s contained 1,i ,…,s k,i ,…,s m,i Quantitative knowledge, namely:

[0078]

[0079] At this point, based on the relative trend values ​​of the data elements, we can obtain the m source data DS1,…,DS k ,…,DS m The quantitative knowledge behind each data element, data unit or entity node can be used in subsequent data processing processes such as fusion and learning to improve the accuracy and performance of digital decision-making based on fusion and learning results.

[0080] Example

[0081] Given a multi-source medical dataset DS1,…,DS generated by several patient entities, including blood routine, urine routine, stool routine, and emergency biochemistry k ,…,DS m , the dataset DS k (such as the "blood routine data set") feature vector (such as "leukocyte feature vector", where and )and (such as "red blood cell feature vector", where ) as the base data element and accompanying data element, calculate and The trend degree and trend direction of all data elements in the The adjoint vector is the base vector Then, calculate the relative trend value of all data elements in the dataset DS in the same way. k When the other vectors are the base vectors, the adjoint vector The relative trend value of all data elements in , thus obtaining the accompanying vector The comprehensive relative trend value of all data elements in the dataset DS k The comprehensive relative trend value of all data elements in , calculate DS k Similarly, we can obtain the quantitative knowledge of data elements and data units in other datasets, thereby obtaining the quantitative knowledge of each entity node in a given multi-source medical dataset.

Claims

1. A knowledge acquisition method based on relative steady-state analysis of data, characterized by: The specific steps include: (1) Definition of trend Suppose there are n patient entities e1,…,e i ,…,e n The m data sets generated are DS1,…,DS k ,…,DS m , m data sets are specifically multi-source medical data sets consisting of blood routine, urine routine, stool routine, and emergency biochemistry, among which data set DS k ={s k,1 ,…,s k,i ,…,s k,n }, data unit L is the dimension of the data unit; according to the data set DS k Constructing a base vector from the data elements and adjoint vector and Based on the reference vector and adjoint vector The data elements in the table are used as the base data elements and accompanying data elements respectively. and The second data element in and First, define the trend degree as and (2) Define benchmark data elements Trend direction and accompanying data elements Trend direction (3) Knowledge acquisition based on relative trend value Companion data elements The relative trend of: Companion data elements Relative trend direction Among them, "consistent", "half-consistent" and "inconsistent" mean "consistent", "half-consistent" and "inconsistent"; The relative trend direction Quantification, based on accompanying data elements The relative trend and the quantified relative trend direction By the relative trend function f cv Calculated The relative trend value of Relative trend function f cv It is the product of relative trend degree and relative trend direction; According to the data set DS k The base vector constructed from the data elements in Solve for the accompanying data elements The relative trend values ​​are Defining accompanying data elements The comprehensive relative trend value is: set up The maximum value in Z k,max , will be accompanied by the data element The knowledge is quantified as The normalized relative trend value of , that is: Get dataset DS k Quantitative knowledge of other accompanying data elements in, as well as other data sets DS1,…,DS k-1 ,DS k+1 ,…,DS m Quantitative knowledge accompanying data elements; For dataset DS k Medium Data Unit Its quantitative knowledge K(s k,i ) consists of the accompanying data elements The quantitative knowledge is defined as follows: For generating data sets DS1,…,DS k ,…,DS m n patient entities, patient entity e i The data units are recorded in the matrix [s 1,i ,…,s k,i ,…,s m,i ] T In the patient entity i The quantitative knowledge depends on the data units s contained 1,i ,…,s k,i ,…,s m,i Quantitative knowledge, namely:

2. The method for acquiring knowledge based on relative steady-state data analysis according to claim 1, characterized in that: Defining baseline data elements Trend direction 1) If Then the trend direction is "parallel" and 2) If The trend direction is "upward" and 3) If The trend direction is "downward" and 3. The method for acquiring knowledge based on relative steady-state data analysis according to claim 2, characterized in that: Defining accompanying data elements Trend direction 1) If and or and or and Then the trend direction is "parallel" and 2) If and The trend direction is "upward" and 3) If and The trend direction is "downward" and 4) If and or and Then the trend direction is "semi-upward" and 5) If and or and Then the trend direction is "semi-downward" and 4. The method for acquiring knowledge based on data relative steady-state analysis according to any one of claims 1 to 3, characterized in that: The relative trend direction of "consistent" is quantified as 1, and the relative trend directions of "semi-consistent" and "inconsistent" are quantified as 3 / 2 and 2.

Images