DE102022207482B4 - Computer-implemented method for determining a data quality index, computer program and control unit - Google Patents

Abstract

The invention relates to a computer-implemented method for determining a data quality index (1) of data, comprising: receiving structured data, which in particular includes entries structured in columns and rows, receiving and/or determining information about the data, evaluating the information about the data to obtain at least one component (2, 3, 4, 5, 6, 7) of the data quality index (1), and determining the data quality index (1) of the data based on the at least one component (2, 3, 4, 5, 6, 7) of the data quality index (1).

Description

The present invention relates to a computer-implemented method for determining a data quality index of data, a computer program and a control unit designed to carry out the method.

Data quality plays an important role in today's data-driven world. For example, it is important to check the quality of data or data sets that you have not generated yourself before using them as a basis for further processes. For example, it is of particular interest for a data processor to understand the data background of the data and not just analyze the data superficially. However, it can be very time-consuming to identify and find the appropriate data owner or data creator who can provide detailed information about the respective data set in order to understand the data background. This usually requires regular communication with other people. It is often the case that the processor and the creator of the data are deployed in different departments or even different regions, which creates a high communication effort. Furthermore, the data background must be checked again every time the data is used again, for example if a new person is interested in working with the respective data set. This is not efficient and slows down development processes.

It is therefore an object of the present invention to provide a method and a device which can make a data background from a provided data set easily accessible in order to fundamentally understand this data and to base further processing of the data on it.

For example, methods or devices that deal with data are out US 2021 / 0 263 900 A1 , US 2014 / 0 229 456 A1 , US 2013 / 0 185 309 A1 and US 2020 / 0 380 379 A1 known.

The problem is solved with a method with the features of claim 1, with a computer program with the features of claim 15 and with a control unit with the features of claim 16.

• Empfangen von strukturierten Daten, welche insbesondere in Spalten und Zeilen strukturierte Einträge umfassen,
• Empfangen und/oder Bestimmen von Informationen über die Daten,
• Auswerten der Informationen über die Daten, um zumindest eine Komponente des Datenqualitätsindex zu erlangen,
• Bestimmen des Datenqualitätsindex der Daten auf Basis der zumindest einen Komponente des Datenqualitätsindex und
• Aussortieren und/oder Vorsortieren und/oder Auswählen von Bauteilen, wobei dazu der Datenqualitätsindex genutzt wird.

According to one aspect of the present invention, there is provided a computer-implemented method for determining a data quality index of data, the method comprising:

• Receiving structured data, which in particular includes entries structured in columns and rows,
• Receiving and/or determining information about the data,
• Evaluating the information about the data in order to obtain at least one component of the data quality index,
• Determining the data quality index of the data based on the at least one component of the data quality index and
• Sorting out and/or pre-sorting and/or selecting components, using the data quality index for this purpose.

Compared to the known prior art, the above method can be used to measure or determine a data quality of the received or acquired data. Furthermore, with the above method it is not necessary to fulfill a large number of different assumptions and/or requirements in order to analyze data with regard to its data quality. According to one embodiment, the data quality index includes six different components. Each component can be calculated by one or more mathematical measures or formulas. The mathematical formulas enable objective measurement of data quality. Each measure can be standardized, meaning that a uniform result interval can be used for all calculations. With different number intervals, there would be a risk that individual components of the data quality index would be artificially weighted due to their larger number interval. Due to the standardization, each component is equally important and therefore unweighted. Optionally, one or more components can be weighted by specifying weighting factors. To calculate individual components, information about the data or the data set is necessary (for example information about the columns and/or rows of the data). This information about the data (e.g. metadata) can be stored in a data dictionary as an interim solution be cherted. The data dictionary can describe the data and thus provide information about the data. The information can also be retrieved from a data catalog. In the data catalog, both the information about the data and the data itself can be stored centrally or decentrally.

According to one embodiment, the data quality index includes six different components and nine mathematical measures (i.e., mathematical formulas). The mathematical measures or formulas can also be referred to as subcomponents of the components of the data quality index. However, the data quality index can be flexibly expanded with additional components and/or subcomponents. It is conceivable that the data quality index is differentiated according to the type of data in its composition and calculation, such as for data relating to time series. The data quality index can be particularly advantageously determined for data (records) in the form of a table schema.

High data quality is an essential prerequisite for all data-driven projects and results. Without sufficient data quality, there is a risk of misinterpretations and wrong decisions. Data quality can be interpreted very subjectively. Without defining data quality using mathematical formulas, a transparent and objective evaluation of the data is not possible. According to one aspect of the present invention, structured data is provided as input data (i.e. as structured data or structured data set) to the method. Structured data can, for example, be arranged in table form in columns and rows. For the purposes of the present invention, structured data is in particular not semantic data, image data, sound data or video data. The data can include entries (i.e. values) in columns and rows. The entries can include numeric and/or written characters. Each entry can be inserted into a cell that can be clearly assigned by a column and a row. The data can be obtained from an information management system. Alternatively or additionally, the data can also be made available directly to the process. It is also conceivable that the method is connected to a database or several databases (e.g. “data lakes”) in order to assign a data quality index to all data stored there.

The information about the data can be stored together with the data and thus made available to the method (i.e. received) at the same time as the actual data. In contrast to the actual structured data, which includes measured values or simulated data for a technical component, the information about the data includes information that describes the structured data. In other words, the information about the data can be a description and/or a characterization of the structured data. The information about the data can be evaluated to obtain or determine at least one component of the data quality index. In other words, the structured data cannot be changed or edited to determine the at least one component of the data quality index. This means that only the information about the data can be evaluated. In other words, the actual data or data sets cannot be changed or edited in any other way. The data quality index of the data can then be determined based on the at least one component of the data quality index. An objective assessment of the structured data can thus be provided. The data quality index can then be saved in conjunction with the structured data so that future users can obtain it directly without having to repeat the above process. In other words, with the above method, an information label (i.e., an index) can be obtained for the data, which indicates the data quality of the data. Further, structured data may be characterized as data that includes an entry or value that is associated with another value or entry.

Evaluating the information in the data can involve applying mathematical measures or formulas. In other words, the information about the data can be input values that are used as the basis for the evaluation. Thus, the evaluation can be an application of mathematical formulas that at least partially include the information about the data as variables. The at least one component of the data quality index can be the result of a mathematical formula. This can ensure that the information is evaluated objectively and that the same information about the data leads to the same data quality index. Consequently, the data can be assessed objectively. Determining the data quality index may be a mathematical determination in the presence of multiple components of the data quality index. In particular, the data quality index can be an average of all components of the data quality index. In the case where there is only one component of the data quality index, the data quality index may correspond to the component of the data quality index.

Preferably, the structured data includes measurement data or simulated data relating to a component. The component can be, for example, a technical component of a vehicle or a vehicle component. For example, the component can be tested on a test stand or during normal use to obtain measurement data. Furthermore, the component can also be in a planning state and a test of the component can be simulated. In the latter case, simulated data can be obtained. In other words, the structured data can relate to a technical component and characterize this component. Therefore, the structured data can also be referred to as technical structured data.

Preferably, the information about the data includes metadata and/or calculated data. Metadata can be information stored in addition to structured data. The metadata can be generated automatically when the data is saved. Additionally or alternatively, the metadata can also be stored manually by a user who, for example, creates and stores the data. The metadata can be stored in a data dictionary, for example. The data owner or the data creator can fill in information about the data set, such as variable explanations, syntax rules, and the like. The calculated data, on the other hand, can be data that is calculated automatically or manually based on the structured data itself. In other words, no external information is required to create the calculated data, but only the structured data itself. In particular, the structured data is not changed or processed in any other way. Furthermore, the calculated data can be determined based on an assignment of entries into specific rows and/or columns of the structured data. For example, a variance of the entries in a column and/or in a row can be determined and calculated data can be determined based on this. Furthermore, the amounts of the entries (for numerical entries) can be compared and outlier entries can be determined and the calculated data can be determined based on this.

Preferably, the metadata includes an identifier, a column name, a row name, a data type, a description, a syntax rule, an example entry, necessary entries, optional entries, a free text field and/or a key or ID value of the data. Thus, the metadata can be a description of the actual structured data. An identifier can, for example, indicate what and/or whether an entry is present in the respective cell. A column name can indicate which values or entries are entered in the respective column. For example, a column can be indicative of speeds. Another column can be indicative of a temperature, for example. In both cases the lines can be indicative of time. A line name can be indicative of which entries are arranged or expected in a line. For example, a line can be assigned to a specific component or a specific test run. For example, a data type can distinguish between numeric or character entries. A description can describe the contents of a cell. For example, the description can specify a unit of an entry. A syntax rule can describe how the respective entry should be entered in the respective cell. For example, a syntax rule can describe the number of decimal places. Furthermore, the syntax rule can, for example, specify the representation of a date or other entries. An example entry can, for example, be an average of the entries usually present in this cell. Alternatively, an example entry can also serve to make it easier for a user to enter values manually. Necessary entries can be entries that are necessary to determine the at least one component of the data quality index. However, optional entries may not be necessary to determine at least one component of the data quality index. Rather, the at least one component of the data quality index can be determined even without an optional entry. If the optional entry is nevertheless present, the at least one component of the data quality index can be determined with higher precision. Whether an entry is necessary or optional can be displayed to the data owner and/or the data creator. A free text field can contain a further description of the data by the data owner and/or the data creator. This means that further explanations or information about the data may be included. The key or ID value of the data can identify the data. For example, a specific key or ID value can be used for recurring data so that it can be easily identified.

The calculated data preferably includes a determination of identical entries and/or outlier entries in the data. As already explained above, the calculated data can be determined based on the data alone without any additional external information. For example, the variance of the entries can be calculated or determined. The variance can later be used to determine the quality index of the data (more details on this below). When determining the calculated data The entries of a column and/or a row can be plotted and their distribution can be analyzed. This allows the variance of the entries to be determined and/or outlier entries to be made visible. For example, outlier entries can be entries that differ significantly from neighboring entries (that is, entries in adjacent cells). This means that the calculated data can only be determined or calculated on the basis of the structured data, without additional external information being available.

Preferably, the at least one component of the data quality index comprises a traceability value, which is determined based on a data creation value, which is determined in particular based on a data source of a data contact and/or a creation date, and an update value. The traceability value component can include all central information about the structured data, such as whether the data source and a contact person are known, in order to be able to clarify further questions if necessary. For example, the traceability value can be a value between 0 and 1 (ie traceability value ∈ [0;1]). The traceability value can be determined, for example, on a data creation value or data creation tracking value. The data creation value can be a value between 0 and 1 (i.e. data creation value ∈ [0 ;1]). The data creation value can have the following additional subcomponents: $$Quelle=\{\begin{array}{l}\mathrm{0,}Datenw\text{ö}rterbucH\to \mathrm{<figure-callout\; id="2"\; label="D\; D"\; filenames="DE102022207482B4\_0018.png"\; state="\{\{state\}\}">D</figure-callout>}D2\to Tabelle\to Daten\_Quellen\_Nameistlee\mathrm{<figure-callout\; id="1"\; label="r"\; filenames="DE102022207482B4\_0019.png"\; state="\{\{state\}\}">r</figure-callout>}\\ \mathrm{1,}DatenquellennameistnicHtleer\end{array}$$

Depending on the boundary conditions, the value for source can be 0 or 1. $$\begin{array}{l}KOntakt\\ =\{\begin{array}{l}\mathrm{0,}Datenw\text{ö}rterbucH\to \mathrm{<figure-callout\; id="2"\; label="D\; D"\; filenames="DE102022207482B4\_0018.png"\; state="\{\{state\}\}">D</figure-callout>}D2\to Tabelle\to EiGent\text{u}merderGescH\text{ä}ftsdaten\\ UNDEDv\_verantwOrtlicHer\\ UND\mathrm{e.g}us\text{ä}t\mathrm{e.g}licHerKOntaktistlee\mathrm{<figure-callout\; id="1"\; label="r"\; filenames="DE102022207482B4\_0019.png"\; state="\{\{state\}\}">r</figure-callout>}\\ \mathrm{1,}EiGent\text{u}merderGescH\text{ä}ftsdatenODERED{v}_{verantwOrtlicHer}ODER\mathrm{e.g}us\text{ä}t\mathrm{e.g}licHerKOntaktistnicH\end{array}\end{array}$$

Depending on the boundary conditions, the value for contact can be 0 or 1. $$DatenerstellunG=\{\begin{array}{l}\mathrm{0,}PWCQuerySQLQu\mathrm{<figure-callout\; id="1"\; label="e\; r\; y"\; filenames="DE102022207482B4\_0019.png"\; state="\{\{state\}\}">e</figure-callout>}ry1.sql\to ErstellunGistlee\mathrm{<figure-callout\; id="1"\; label="r"\; filenames="DE102022207482B4\_0019.png"\; state="\{\{state\}\}">r</figure-callout>}\\ \mathrm{1,}ErstellunGistnicHtleer\end{array}$$

Depending on the boundary conditions, the value for data creation can be 0 or 1. In other words, the data creation value can be used to determine whether or not there is information about the creation of the data.

To determine the data creation value, add the source, contact, and data creation values and divide the result by three.

Furthermore, an update value can take the value 0 or 1 as follows: $$AktualisierunGswert=\{\begin{array}{l}\mathrm{0,}UnterscHiedvOn\left(Heute,let\mathrm{e.g}te\text{Ä}nderunG\right)inTaGen>365\\ \mathrm{1,}UnterscHiedvOn\left(Heute,let\mathrm{e.g}te\text{Ä}nderunG\right)inTaGen\le 365\end{array}$$

The data creation value can then be added to the update value and the result can again be divided by two. The result then corresponds to the traceability value.

Preferably, the at least one component of the data quality index includes a usage value that is determined based on an elapsed time since the data was last used. The usage value component can indicate whether or not the data or data set was used at least once by a user in a defined period of time. This is due to the fact that more and more different data sets are being made available instead of checking which data sets are still being used regularly. Frequent use of the data can therefore indicate that the data set is current and is also used as a basis for other projects. The usage value can be a value between 0 and 1 (ie usage value ∈ [0;1]). The usage value can be determined as follows: $$verwendunGswert=\{\begin{array}{l}\mathrm{0,}\text{Difference from}\left(\text{TODAY},\text{Last Run}\right)\text{in days}>90\\ \mathrm{1,}\text{Difference from}\left(\text{TODAY},\text{Last Run}\right)\text{in days}\le 90\end{array}$$

• Der Wert für prc kann ein Wert zwischen 0 und 1 sein (d.h. prc ∈ [0;1]). $$prc=\left(1-\frac{\#\text{Spalten ohne Mussfeldinfo}}{\#\text{alle Spalten des Datensatzes}}\right)\ast \left(\frac{\#\text{Mussfeldspalten des Datensatzes}}{\#\text{alle Spalten des Datensatzes}}\right)$$
  
• Die Informationen, ob eine Spalte eine Mussfeldinfo aufweist kann in dem Datenkatalog gespeichert sein.
• Der Wert für pve kann ein Wert zwischen 0 und 1 sein (d.h. pve ∈ [0;1]). $$pve=1-\frac{\#SpaltendesDatensatzesmitdemselbenEintrag}{\#alleSpaltendesDatensatzes}$$
  

Preferably, the at least one component of the data quality index comprises an information content value which is determined based on a ratio of required entries and existing entries of the data as well as a ratio of the various entries. The ratio of required entries and existing entries can be denoted as prc. The ratio of the different entries can be referred to as pve. The ratio prc may include a ratio of the number of expected column data entries to the number of columns. The ratio pve can describe a ratio of the number of different entries within a column or row to the number of cells in a column or row of the data set. In other words, the quality of a data set is better when mandatory data is present. In contrast, optional data only has to be there optionally. The information content value can be a value between 0 and 1 (ie information content value ∈ [0 ;1]). The information content value can be determined based on the ratio prc and pve. In particular, the information content value can be the result of adding prc and pve divided by two. The ratios prc and pve can be determined as follows:

• The value for prc can be a value between 0 and 1 (i.e. prc ∈ [0;1]). $$prc=\left(1-\frac{\#\text{Columns without mandatory field info}}{\#\text{all columns of the data set}}\right)\ast \left(\frac{\#\text{Required field columns of the data record}}{\#\text{all columns of the data set}}\right)$$
  
• The information as to whether a column has mandatory field information can be stored in the data catalog.
• The value for pve can be a value between 0 and 1 (ie pve ∈ [0;1]). $$pve=1-\frac{\#SpaltendesDatensat\mathrm{e.g}esmitdemselbenEintraG}{\#alleSpaltendesDatensat\mathrm{e.g}es}$$
  

The ratio of the required columns is added to the ratio of the variance of the entries and the result of this is divided by two to obtain the information content value.

Preferably, the at least one component of the data quality index comprises a completeness value, which is determined based on a ratio of existing entries to a number of entries and required entries. The completeness value indicates how many missing values or entries there are in the columns of the data record or data. A distinction is made between which columns missing values can occur and in which columns data entries are to be expected.

The completeness value can be a value between 0 and 1 (completeness value ∈[0 ;1]). The completeness value can be denoted comp and can be determined as follows: $$cOmp=\frac{\left(cOmpf\text{u}rSpa\mathrm{<figure-callout\; id="1"\; label="l\; t\; e"\; filenames="DE102022207482B4\_0019.png"\; state="\{\{state\}\}">l</figure-callout>}t{e}_{}{}_1+\cdots +cOmpf\text{u}rSpalt{e}_{let\mathrm{e.g}teSpaltedesDatensat\mathrm{e.g}es}\right)}{\left(\#SpaltendesDatensat\mathrm{e.g}es-\#SpaltenmitMussfeld=falscH\right)}$$

The value “#Columns with mandatory field = false” indicates the number of columns in which a value is expected or in which a value should be but there is no entry.

Where $${\text{comp for column}}_i=\left(1-Spalt{e}_i\prime \text{s Mandatory field entry}\right)\ast \left(\frac{\#einGef\text{u}GteWertederSpalt{e}_i}{An\mathrm{e.g}aHlvOnZeilen}\right)$$

wobei
i = 1, ..., p.
p := letzte Spalte des Datensatzes
# := Anzahl von
Die obigen Variablen gelten für alle hier angeführten Formeln und Darstellungen.Where $$\begin{array}{l}Spalt{e}_i^{\prime }\text{s Mandatory field entry}\\ =\{\begin{array}{c}\mathrm{0,}\text{Mandatory field}=\text{true}\\ {\text{no comp for column}}_i\text{calculation}\left(\text{write}""\text{in the data quality model}\right),\text{Mandatory field}=\text{incorrect}\\ \text{1}\text{,Must field}=\text{no information}\end{array}\end{array}$$

where
i = 1, ..., p.
p := last column of the data set
# := Number of
The variables above apply to all formulas and representations listed here.

$$understandf\ddot{\text{u}}rSpalt{e}_i=\frac{\#verf\ddot{\text{u}}gbareMetadateninDatenkatalogderSpalt{e}_i}{\#notwendigeMetadateninDatenkatalog}$$

Preferably, the at least one component of the data quality index comprises an understandability value, which is determined based on a ratio of existing metadata to required metadata. The understandability score can represent a ratio of how much of the required metadata information is provided per column. For this purpose, a data catalog can, for example, be provided in a central location to describe the content of data. The metadata can be the metadata defined above. The comprehensibility value can take a value between 0 and 1 (comprehensibility value ∈[0;1]). The comprehensibility value can be referred to as understand and can be determined as follows: $$\begin{array}{l}understand\\ =\frac{understandf\ddot{\text{u}}rSpa\mathrm{<figure-callout\; id="1"\; label="l\; t\; e"\; filenames="DE102022207482B4\_0019.png"\; state="\{\{state\}\}">l</figure-callout>}t{e}_{}{}_1+\cdots +understandf\ddot{\text{u}}rSpalt{e}_{let\mathrm{e.g}teSpaltedesDatensat\mathrm{e.g}es}}{\#alleSpaltendesDatensat\mathrm{e.g}es}\end{array}$$

$$understandf\ddot{\text{u}}rSpalt{e}_i=\frac{\#verf\ddot{\text{u}}GbareMetadateninDatenkatalOGderSpalt{e}_i}{\#nOtwendiGeMetadateninDatenkatalOG}$$

Note: Required metadata in the data catalog includes: ID; column name; tablename; data type; description_explanation; syntax rule; example_value; mandatory field; free text field; Key or ID_value.

• • Verhältnis der syntaktisch korrekten Werte der Syntaxspalten pac ∈ [0;1] $$pac=\frac{\left(pacvonSpalt{e}_i+\cdots +pacvonSpalt{e}_p\right)}{\left(\#SpaltendesDatensatzes-\#SpaltenmitSyntaxregel=falsch\right)}$$
  
  $$\text{pac f}\ddot{\text{u}}\text{r }Spalt{e}_i=\left(1-Spalt{e}_i^{\prime }\text{s Sytaxregeleintrag}\right)\ast \left(\frac{\#syntaktischkorrekteEintr\ddot{\text{a}}gederSpalt{e}_i}{AnzahlvonZeilen}\right)$$
  
  wobei $$\begin{array}{l}Spalt{e}_i^{\prime }\text{s Sytaxregeleintrag}\\ =\{\begin{array}{c}\mathrm{0,}\text{Syntaxregel}=\text{wahr}\\ \text{kein pac f}\ddot{\text{u}}\text{r }Spalt{e}_i\text{ Berechung}\left(\text{schreibe}""\text{in dem Datenqualit}\ddot{\text{a}}\text{tsmodel}\right),\text{Syntaxregel}=\text{<figure-callout id="1" label="falsch" filenames="DE102022207482B4\_0019.png" state="{{state}}">falsch</figure-callout>}\\ \mathrm{1,}\text{Syntaxregel}=\text{keine Information}\end{array}\end{array}$$
  
• • Verhältnis der Spalten mit potenziellen Ausreißern (Ausreißerwert) outlier ∈[0;1] $$\begin{array}{l}outlier\\ =(1\\ -\left(\frac{\#outliersvonnumerischerSpla\mathrm{<figure-callout\; id="1"\; label="t\; e"\; filenames="DE102022207482B4\_0019.png"\; state="\{\{state\}\}">t</figure-callout>}{e}_{}{}_1+\cdots +\#outliersvonnumerischerSplat{e}_{letztenumerischeSplate}}{AnzahlvonnumerischenSplaten}\right)\end{array}$$
  
  $$outliervonSplat{e}_i=\frac{\#outliersvonnumerischerSplat{e}_i}{AnzahlvonReihen}$$
  

Preferably, the at least one component of the data quality index comprises an accuracy value which is determined based on a ratio of syntactically correct entries to the number of entries and/or a ratio of possible outlier entries to the number of entries. The precision value can define possible outliers and a ratio of syntactically correct columns. In particular, the precision value is only applicable for numerical entries. The precision value can be a value between 0 and 1 (accuracy value ∈[0;1]). The precision value can also be referred to as acc and can be determined based on a ratio of the syntactically correct entries of the syntax columns pac and a ratio of the columns with potential outliers. Here pac and outlier can be ∈[0;1] and can be determined as follows:

• • Ratio of the syntactically correct values of the syntax columns pac ∈ [0;1] $$pac=\frac{\left(pacvOnSpalt{e}_i+\cdots +pacvOnSpalt{e}_p\right)}{\left(\#SpaltendesDatensat\mathrm{e.g}es-\#SpaltenmitSyntaxreGel=falscH\right)}$$
  
  $$\text{pac f}\ddot{\text{u}}\text{r}Spalt{e}_i=\left(1-Spalt{e}_i^{\prime }\text{s syntax rule entry}\right)\ast \left(\frac{\#syntaktiscHkOrrekteEintr\ddot{\text{a}}GederSpalt{e}_i}{An\mathrm{e.g}aHlvOnZeilen}\right)$$
  
  where $$\begin{array}{l}Spalt{e}_i^{\prime }\text{s syntax rule entry}\\ =\{\begin{array}{c}\mathrm{0,}\text{Syntax rule}=\text{true}\\ \text{no pac f}\ddot{\text{u}}\text{r}Spalt{e}_i\text{Calculation}\left(\text{write}""\text{in the data quality}\ddot{\text{a}}\text{tsmodel}\right),\text{Syntax rule}=\text{incorrect}\\ \mathrm{1,}\text{Syntax rule}=\text{no information}\end{array}\end{array}$$
  
• • Ratio of columns with potential outliers (outlier value) outlier ∈[0;1] $$\begin{array}{l}Outlier\\ =(1\\ -\left(\frac{\#OutliersvOnnumeriscHerSpla\mathrm{<figure-callout\; id="1"\; label="t\; e"\; filenames="DE102022207482B4\_0019.png"\; state="\{\{state\}\}">t</figure-callout>}{e}_{}{}_1+\cdots +\#OutliersvOnnumeriscHerSplat{e}_{let\mathrm{e.g}tenumeriscHeSplate}}{An\mathrm{e.g}aHlvOnnumeriscHenSplaten}\right)\end{array}$$
  
  $$OutliervOnSplat{e}_i=\frac{\#OutliersvOnnumeriscHerSplat{e}_i}{An\mathrm{e.g}aHlvOnReiHen}$$
  

Furthermore, an outlier threshold can be defined, above which a value is defined as an outlier. For example, at least three methods for determining outliers can be provided and an outlier can be determined if three out of three methods recognize a data point as an outlier. In particular, it is only possible to calculate outliers for numeric columns.

For example, syntactically correct values can only be positive values. Furthermore, the syntax can specify that only values between 2 and 5 are to be expected in a cell, and everything that does not fall within this number range is defined with a value 0. The outliers can be determined using machine learning methods, for example. Here, an algorithm can be trained to output output data based on input data that indicates whether there are outliers in the input data. To do this, the algorithm can be trained in advance with a training data set. This can increase the reliability of determining outliers.

Thus, each component of the data quality index can be calculated by a defined formula. Each component result can be standardized and have the same range of values as all other component results. This range of values is preferably between 0 and 1. 0 can mean that this component has a bad result and 1 can mean that this component has a very good result. For example, the usage value component may have a value of 0.0, which is a poor result. The component of the accuracy value can have a value of 0.8, which is close to 1, which is a good value. The data quality index can be the average of the component values and can be determined from the sum of the individual component values divided by the number of components.

Preferably, the data quality index is determined at a central data storage location. This means that a variety of data can be analyzed by a local body and provided with a data quality index. Alternatively, the data quality index can also be determined decentrally, for example at the location of data storage and/or data generation. This allows the amount of data to be transferred to be reduced since the data does not have to be transferred for the purpose of determining the data quality index.

The values obtained for several or all components of the data quality index are preferably in the same value range, preferably in a range between 0 and 1. The values of the components of the data quality index are therefore comparable and not artificially weighted.

Preferably, the obtained values for several or all components of the data quality index are added and the result is divided by the number of components of the data quality index to obtain the data quality index. In other words, the data quality index can be the average of all components of the data quality index. This means that all components of the data quality index can be weighted equally and have the same influence on the data quality index.

When determining the data quality index, the individual components of the data quality index are preferably multiplied by a factor, in particular between 0 and 10, in order to weight all or individual components of the data quality index. Each component of the data quality index can be weighted with a weighting factor λ ≥ 0. The value of the weighting factor λ can be any positive number. A value close to 0 means a lower weighting of the corresponding component and therefore a smaller impact. A value of 1 means that the component is not weighted because multiplying something by 1 gives a neutral result in terms of weighting. A value > 1 means that the corresponding component has a greater influence in determining the data quality index. Theoretically, the weighting factor cannot be limited. However, the weighting factor is preferably in a numerical interval between 0 and 10. This allows clarity and comparability to be maintained. This means that factors that are particularly important for a data processor (ie components tens of the data quality index) are weighted more heavily than others. This means that the data quality index can be individually adapted to the needs and/or requirements of the data user.

According to a further aspect of the present invention, a computer program is provided with program code which, when executed on a computing unit, is designed to carry out the above method. The computer program may be in any code, particularly code suitable for analyzing data sets. According to a further aspect, the invention is directed to a computer-readable medium comprising a computer program as defined above. The computer-readable medium can be any digital data storage device, such as a USB flash drive, hard drive, CD-ROM, SD card, or SSD card. Of course, the computer program does not have to be stored on such a computer-readable medium in order to be made available to a customer, but can also be obtained via the Internet.

According to a further aspect of the present invention, a control unit designed to carry out the above method is provided. The control unit can be a computer-like device that has a processor and memory elements. For example, such a control unit can be provided in a test stand in order to directly analyze recorded measurement data and provide it with a data quality index. Furthermore, it is conceivable that a control unit is provided in a simulation machine in order to add a data quality index to simulated results. Alternatively or additionally, the control unit can be provided in a computer. Such a computer can be used, for example, in the development of components, especially in the vehicle industry, to sort out components that were developed on the basis of data with a specific data quality index. The data quality index is therefore used in development to sort out and/or pre-sort and/or select components. It is also conceivable that the control unit is used to provide evidence of the data basis on which a specific component was developed. This means that liability issues or the like can also be clarified retrospectively. The above method can therefore be used to evaluate technical components with regard to their data basis and, in particular, to select them based on this evaluation.

Furthermore, the present invention is directed to using the above method to select a prototype of a technical component from a plurality of prototypes. In other words, a variety of prototypes can be produced based on different data and selected based on the data quality index at a later point in time when it is clear which data quality index is required by the prototype. This also offers the opportunity during a development process to select a respective technical component based on the data quality.

The description refers to columns and/or rows of the structured data; the terms columns and rows can be interchanged as desired. The structured data can, for example, be data entered in a spreadsheet. An entry can be entered in a cell, whereby the cell can be clearly determined by a column value and a row value. If reference is made to an entire column or row, this refers to all cells and any entries entered therein that are in the row of columns or rows.

Individual features of embodiments can be combined with other features or other embodiments to form new embodiments. The advantages and refinements mentioned in connection with the individual features also apply analogously to the new embodiments. Advantages and refinements that are described in connection with the device also apply analogously to the method and vice versa.

• 1 zeigt ein schematisches Diagramm, welches einzelne Komponenten des Datenqualitätsindex gemäß einer Ausführungsform der vorliegenden Erfindung zeigt.
• 2 ist ein schematisches Diagramm, welches Komponenten des Datenqualitätsindex zusammen mit Subkomponenten der Komponenten des Datenqualitätsindex gemäß einer Ausführungsform der vorliegenden Erfindung zeigt.
• 3 ist ein Ablaufdiagramm, welches den Ablauf eines Verfahrens gemäß einer Ausführungsform der vorliegenden Erfindung schematisch darstellt.

Embodiments of the present invention are described below by way of example using the attached figures.

• 1 shows a schematic diagram showing individual components of the data quality index according to an embodiment of the present invention.
• 2 is a schematic diagram showing components of the data quality index along with subcomponents of the components of the data quality index according to an embodiment of the present invention.
• 3 is a flowchart that schematically illustrates the flow of a method according to an embodiment of the present invention.

1 is a schematic representation of the data quality index and six different components of the data quality index. The Data Quality Index 1 consists of six different components: Traceability Score 2, Usage Score 3, Information Content Score 4, Completeness Score 5, Comprehensibility Score 6 and Accuracy Score 7. The components are determined by mathematical formulas and/or measurements. Some measurements can only be calculated if the required information is present in the data catalog. The required information is referred to as metadata and/or information about the record. All required metadata is collected in the data catalog and/or a data dictionary. The traceability value 2 component includes all central information about the data set, such as whether the data source and a contact person are known in order to be able to clarify further questions if necessary. The usage value 3 indicates whether the data record was used at least once by a user in a defined period of time. This is due to the fact that more and more different data sets are being made available instead of checking which data sets are still being used regularly. The completeness value of 5 indicates how many missing entries there are in the columns (or rows) of the data set. A distinction is made between which columns missing values can occur and which column data entries are to be expected. The precision value 7 detects possible outliers and the ratio of syntactically correct columns. This component is only applicable to numeric columns. The information content value 4 contains a ratio of required or expected data entries to the total columns and a ratio of columns with the same entry to the total columns. The understandability value 6 is a ratio of the required metadata information per column of the data set.

Each component can be calculated by a defined formula. Each component result is standardized and has the same range of values. It lies between 0 and 1. For example, traceability 2 can provide a value of 0.4, the usage value can provide a value of 0.0, the completeness value 5 can provide a value of 0.6, the accuracy value can provide a value of 0.8, and the understandability value can provide a value of 0.8 Value of 0.8 and the information content value 4 has a value of 0.8. From this, a data quality index of 0.6 can be determined.

2 is a schematic representation of the data quality index 1 together with the components of the data quality index and their subcomponents. The subcomponents are information about the data on the basis of which the data quality index 1 is determined. The information about the data differs in metadata and calculated data. Metadata is data that is added to the respective structured data by a user or automatically. For example, for the traceability value 2, a data creation value 21 and an update value 22 are the subcomponents determined based on the metadata. The data creation value 21 consists of a source, a contact and a creation date. This can be determined automatically. The update value 22 can also be determined automatically and indicates when the data was last updated. The usage value 3 has usage tracking 31 as a subcomponent. This can also be automatically assigned to the structured data. The completeness value 5 has a completeness ratio 51, which states how many cells of the data record are completely filled out. The accuracy value 7 has an accuracy ratio 71, which indicates how many syntactically correct cells are present in relation to the total number of cells in the data set. Furthermore, the accuracy value is based on a potential outlier detection or outlier value 72. The potential outlier detection 72 is a calculated value. The outlier detection 72 can also be viewed as a ratio of columns with potential outliers. For this it is not necessary to provide external data; it is sufficient to simply analyze the data set itself. The understandability value 6 has a ratio 61, which indicates how many cells have metadata. In other words, this can determine how much information about the data is available. The information content has a required column ratio 41, which indicates how many required columns are filled in relation to the total number of columns. Furthermore, the information content value 4 has a determination of identical entries 42. The determination of the same entries 42 is calculated data that can only be determined based on the structured data. This allows you to analyze how many identical entries there are in a column.

3 is a flowchart that schematically shows the process flow. First, the structured data is received in step S1. In step S2, information about the data is received and/or determined. The information about the data is then evaluated in step S3 in order to obtain at least one component of the data quality index. In step S4, the data quality index of the data is then determined based on the at least one component of the data quality index. The procedure can also be carried out several times in a row.

Reference symbols

1

Data quality index

2

Traceability value

3

Use value

4

Information content value

5

Completeness value

6

Intelligibility value

7

Accuracy value

21

Data creation value

22

Update value

31

Usage tracking

41

Column ratio

42

same entries

51

Completeness ratio

61

Ratio of metadata

71

Accuracy ratio

72

Outlier detection

Claims (16)

Computer-implemented method for determining a data quality index (1) of data, comprising: Receiving structured data, which in particular includes entries structured in columns and rows, Receiving and/or determining information about the data, Evaluating the information about the data in order to obtain at least one component (2, 3, 4, 5, 6, 7) of the data quality index (1), Determining the data quality index (1) of the data based on the at least one component (2, 3, 4, 5, 6, 7) of the data quality index (1) and Sorting out and/or pre-sorting and/or selecting components, using the data quality index (1) for this purpose. Procedure according to Claim 1 , wherein the structured data includes measurement data or simulated data relating to a component. Procedure according to Claim 1 or 2 , wherein the information about the data includes metadata and/or calculated data. Procedure according to Claim 3 , wherein the metadata includes an identifier, a column name, a row name, a data type, a description, a syntax rule, an example entry, necessary entries, optional entries, a free text field and / or a key or ID value of the data. Procedure according to Claim 3 or 4 , wherein the calculated data includes a determination of identical entries and/or outlier entries of the data. Method according to one of the preceding claims, wherein the at least one component (2, 3, 4, 5, 6, 7) of the data quality index (1) comprises a traceability value (2), which is based on a data creation value (21), which is in particular based on a data source, a data contact and/or a creation date is determined, and an update value (22) is determined. Method according to one of the preceding claims, wherein the at least one component of the data quality index (1) comprises a usage value (3) which is determined based on an elapsed time since the data was last used. Method according to one of the preceding claims, wherein the at least one component of the data quality index (1) comprises an information content value (4) which is determined based on a ratio of required entries (41) and existing entries of the data as well as a ratio of the different entries (42). . Method according to one of the preceding claims, wherein the at least one component of the data quality index (1) comprises a completeness value (5), which is determined based on a ratio (51) of existing entries to a number of entries and the required entries. Method according to one of the preceding claims, wherein the at least one component of the data quality index (1) comprises an understandability value (6) which is determined based on a ratio (61) of existing metadata to required metadata. Method according to one of the preceding claims, wherein the at least one component of the data quality index (1) comprises an accuracy value (7), which is based on a ratio of syntactically correct entries to the number of entries (71) and / or a ratio of possible outlier entries the number of entries (72) is determined. Method according to one of the preceding claims, wherein the values obtained for several or all components of the data quality index (1) lie in the same range of values, preferably in a range between 0 and 1. Method according to one of the preceding claims, wherein the obtained values for several or all components of the data quality index (1) are added and the result is divided by the number of components of the data quality index to obtain the data quality index. Procedure according to Claim 13 , whereby the individual components of the data quality index are multiplied by a factor, in particular between 0 and 10, when determining the data quality index, in order to weight all or individual components of the data quality index. Computer program with program code which, when executed on a computing unit, is designed to carry out the method according to one of the preceding claims. Control unit designed to implement the method according to one of Claims 1 until 14 to carry out.

Images