DE10250990A1 - Procedure for automatic classification of through-put data-files e.g. for banking, requires firmly assigning substitute expression when no substitute expression has been assigned - Google Patents

Abstract

A method for automatic classification of data-files/sets by digital computer involves classifying an assignment of key-expressions to replacement/substitute expressions and priority values, and verifying the respective data-file/set for each key-expression if the respective data-file/set fulfills the actual key-expression, the replacement/substitute expression is firmly assigned when no replacement/substitute-expression has been assigned to the respective data-file/set, and when the respective data-file/set has already been assigned a replacement/substitute expression, and the priority value of the actual key-expression is greater than the priority value of the key-expression belonging to the already assigned substitute/replacement expression. The assignment pairs arising in this manner from the data files/sets and replacement/substitute expression are checked for at least part mutual agreement of the replacement/substitute expression, where the data-files/sets of con curring assignment pairs are evaluated as belonging to the same class. Independent claims are included for (a) a device for carrying out the method and (b) an application of the method or a device.

Description

The invention relates to a method according to the features of the preamble of claim 1 and a device for this, according to the features of the preamble of claim 7, and also the automated quantitative evaluation of sales data records Majority of bank movements.

For A variety of different applications make it desirable to look at a database Sales data from bank movements to determine which amounts for different Purposes are used and / or from which sources Inflow revenue. This is for example to determine a typical expenditure ratio measured for consumer goods on the income or to determine the savings rate, i.e. the amount of savings measured by revenue, practically of interest. So it is classification of sales data necessary, i.e. the sales data need the Type of sales assigned to specific classes and can be divided into groups.

According to the state of the art first known, by naively searching for key terms in the sales data records to every keyword to get a result set. However, it is disadvantageous that the numerical results in the quantitative evaluation of the result sets by "double hits" are strongly falsified. For example, the sales date INCOME SALES KAUFCITY EUR 1200.00 hit twice if, for example, the search term "INCOME" is used on the one hand and on the other hand after the term "SALARY" to include such sales as Counting income in which the word "SALARY" is used instead of the word "INCOME". Such "double hits" will both Number of sales of the "INCOME" class overcharged as well as possibly key figures derived from this are falsified.

Alternatively, it is from the prior art Technique known, instead of the sales text (purpose text) the numeric text key of the respective turnover for classification. adversely is here that this numeric text key (posting key) is a only very general classification allowed because this text key is only Disclosure allowed about the type of booking, such as direct debit, transfer or standing order. That way you can typically 100% of sales data is classified, however, the classification is for a practical application too rough.

Furthermore, is from the prior art known, the bank details instead of the sales text for classification of the person to whom payment is made or from a payment is received. However, this solution is hardly practical feasible because this requires precise knowledge of which provider of Goods and / or services and, if applicable, which service in particular, can be assigned to a specific bank account.

As an alternative to these "naive" search techniques Many attempts have been made according to the prior art with the methods of data mining. Here statistical methods, e.g. also through trained or untrained neural networks, tries to accumulations determine in the amount of sales data and determine whether a date belongs to a certain class or not.

In the present application, namely the Classification of revenue records into specific revenue- and / or expenditure categories and subcategories lead these Methods also fail. This is also because such sales data very often a big Share in such textual information, which for the assignment to classes is meaningless, but for statistical purposes adverse statistical noise is generated. An example could be here the 20-digit alphanumeric insurance number in the bank transaction text of a debit from a life insurance policy.

It is therefore the task of the present Invention to provide a method and a suitable device which overcomes the disadvantages of the prior art and in particular a way to Classification of sales data in classes and subclasses enables which has a high information content and at the same time guaranteed that a very big one Share of sales data can be assigned to a class.

This object is achieved by a Method according to the features of claim 1 and a device according to the features of claim 7 and use according to Claim 13 solved.

The invention according to claim 1 has the following advantageous features, which are briefly explained below:
Because an assignment of key expressions to replacement expressions and priority values is selected, in which each key expression is uniquely assigned a replacement expression, each key expression is assigned exactly one replacement expression and each replacement expression exactly one key expression, so that for each key expression a replacement expression and each replacement expression is unambiguously assigned eindeu tig the corresponding key expression can be determined; This enables the priority comparison, as is used below in methods, to determine the priority value of an associated key expression based on the replacement expression. The fact that the assignment is selected such that exactly one priority value is also assigned to each key expression makes it possible to give certain key expressions priority over other key expressions in the method. Overall, by selecting such an assignment, the classification information necessary for the functioning of the method is made available in a manner appropriate to the method.

Because each key expression each record is then checked (i.e. for every key expression each the amount of records is run through, the order of the records or the order of the key expressions is insignificant is) whether the respective record contains the current key expression Fulfills, is checked, whether for this data record the assignment of that replacement word into question which comes in this review round current key expression equivalent.

Because in the case of such Correspondingly, so if the record is indeed the key expression Fulfills, further checked whether the record has not yet been assigned a replacement expression and then, if applicable, the corresponding replacement expression Record is assigned and this assignment is saved in these cases a more complex priority check, such as they are explained below is saved, which results in improvements in runtime.

Because in the case of being fulfilled the key expression is checked by the data set, alternatively whether a replacement word has already been assigned to the data record and the priority value of the current key expression is bigger than the priority value of the key expression belonging to the already assigned replacement expression is achieved, that only such a further one, later in the course of the procedure, will be used exam Corresponding key expression to which a higher priority value is assigned is the existing classification of the data set, which is determined by the assignment of the replacement word to the data record has been overwritten. In the latter case, it is the one already assigned to the data record Replacement term with the replacement term assigned to the current key term replaced.

This unfolds its beneficial Effect for that Classification result especially if the assignment between priority values and key terms so is that with such key expressions, which both can be fulfilled by a certain data set, the one that higher priority value which has the classification of this particular record with higher level of detail can effect.

In this way it is achieved that if a record in the course of the procedure first has a key expression Fulfills and thus a replacement expression is assigned to the data record and then the record meets another key expression, only then a new replacement expression is assigned to it if so has the consequence that this (re) allocation and to be derived from it (Re) classification improves the overall classification result.

By doing so in the course assignment pairs created from the procedure from the data record and replacement expression for at least partial agreement of the replacement expression can be checked against each other on the basis of of the records assigned replacement expressions carry out a classifying assessment without renewed, temporal perform relatively complex assignments of data records to have to. With this "subsequent" evaluation the level of detail of the classification can be redetermined and thus Classification results parameterized according to the level of detail of the classification be generated.

The method according to claim 1 can be developed in accordance with the dependent claims referring back to this claim 1. Individual training courses are explained below:
The method can advantageously be further developed in such a way that the replacement expressions each contain a sequence of terms that become more specific and that the classification evaluation checks for at least partial correspondence between different assignment pairs by comparing the replacement expressions of the respective assignment pairs along the respectively more specific ones contained therein Consequences of terms. In this way it is achieved that a data record which has been assigned to a first, rough or general, category and a further data record which has been assigned to a second, more refined or more specific category, the second category being contained in the first category, both can be recognized as belonging to the first category and can additionally be recognized that the second data record also belongs to a more specific category, ie can also be distinguished from the first data record. This allows a more precise statistical preparation when evaluating the assignment results, such as determining the overall classification result or calculating derived key figures.

For this final evaluation, it is also advantageous to further develop the method in such a way that those data records which do not meet any of the key expressions during the method are assigned a special replacement expression which represents a null term. In this way The data records that are not to be classified can be processed further using the same methods as the other assignment pairs consisting of the data record and the replacement expression.

The device according to claim 7 to run of the method according to the invention or one of its further developments shows firstly the advantageous Feature on, the first storage means with a unique assignment of records to replace expressions as well as second storage means with an assignment of key expressions replacement expressions and priority values are included, so that the respective assignments within the scope of Matching and replacement procedures are accessible at all times. Secondly is an advantageous feature that the in the second storage means contained assignment trained according to the aforementioned method is, i.e. that each key expression contained in it also exactly one the replacement expression contained therein is uniquely assigned and also assigns exactly one priority value to each key expression contained therein is. This creates a "dictionary" of key phrases and replacement expressions as well as priority values for the Process optimized in a digital computer.

Third is advantageous feature that data processing means are included, what functions have to check whether a data record meets a certain key concept, whereby in it also the means for sequential or suitably parallelized Selection of data for comparison are included, as well as funds to determine priority values, these means both the priority value starting from the replacement term and alternatively starting from keyword can determine and compare the priority values determined in this way can. Another, particularly important function, which is provided by the data processing means according to the invention functions are implemented to evaluate the conformity of replacement terms between different assignment pairs from data record and replacement term. This can be done after Mapping between records and replacement terms the automatic evaluation to a quantified Classification result.

After all, it is for the functioning this device requires that the storage means and the Data processing means are interconnected for informational purposes. In the context of the entire invention, "informational connection" means such a connection, when information is transmitted from a sender to a receiver, for example a data stream over a data channel, this data channel being an electrical, optical or other connection.

Advantageous further development of this Device are possible and according to the dependent claims referring to claim 7 are explained in part below.

To build the system according to the device greatly simplify and access at runtime of the process to the data at the time of the procedure with less technical effort design, it is advantageous to develop the device in such a way that that the first storage means or the second storage means besides the pure assignment information also the basis of the assignment Data itself included. That means that in the first storage means in addition to the assignment also contain the data records and the replacement expressions are in the second storage means in addition to the assignment the key phrases and / or the replacement expressions and / or the priority values are included. In this way, the realization of a Such a system is considerably simplified because of the memory structures can now be built up in tabular form and thus existing ones technical solutions can be used more easily to implement such a system, for example relational databases or spreadsheets. For the functioning It is irrelevant to the system whether the table structure is based lying memory management these values are actually stored in tabular form holds or the values are realized by pointer referencing.

A particularly advantageous form of further training The system provides that the key expressions are expressions as a Data query language available. As a result, the implementation in particularly simple in the context of a relational Database system and it can be considerably more accurate and the circumstances under which a such, realized by an expression of a data query language, key expression fulfilled by a data set will and when not. This allows a much better assignment between such a key expression fulfilling records to one with this key expression associated replacement expression than, for example, the use a simple search term, such a search term after the State of the art can then be regarded as a fulfilled key expression can, if the search term is contained in the data record.

Furthermore, an advantageous further development of the device provides that the replacement expressions which are contained in the second storage means are constructed in such a way that they each consist of several terms, which in their sequence are arranged in a conceptually more specific manner. In this way, as already described above, it is achieved that a data record which has been assigned to a first, rough or general, category and a further data record which has been assigned to a second, more refined or more specific category, the second category in the first category ent hold, both can be recognized as belonging to the first category and it can also be recognized that the second data record also belongs to a more specific category.

Because of the special properties of sales data from bank movements, whereby sales data in the sense of this entire data about the amount, month and year of sales, the bank statement text (or "Purpose Text"), the text key that Power key, as well as the account number of the recipient, the bank sort code of the Recipient and / or can be the customer number used to identify the customer this method and device are particularly suitable for classification and analysis of sales data from bank movements, albeit they for the Classification of other, preferably textual, data sets used can be. In this case, however, the application of such a multi-stage search and replacement procedures operated with particular effort for the Application to such sales data is therefore extremely surprising, because in this area of application the specialist is far superior than usual prestigious data mining practices did not produce satisfactory results able to perform.

The invention is explained below of an embodiment explained. Show it:

1 a tabular, schematic overview of the various intermediate states of the memory tables in the course of the execution of the method,

1a a table memory extract of an actual implementation form with replacement expressions, key expressions and priority values,

1b a table memory extract of an actual implementation form with a data record comprising eleven fields and a replacement expression assigned to this data record as the assignment result,

2 a screen extract with an input mask for the maintenance of the key expressions,

3 another screen extract for the selection of the key print to be maintained,

4 a screen extract in the practical application of the system, as well

5 another screen excerpt in the practical application of the system, in particular the determination of derived key figures.

1 shows a schematic representation of various intermediate states of the memory tables during the course of the method. The table memory I can be designed as a table in a spreadsheet or in a relational database and contains in the first column in each row a key expression formulated as an expression of a data query language, with which the data records are checked for compliance with the key expression. The second column of the same table or table memory contains a replacement expression for each row. This replacement expression can itself be implemented as a character string or as a data symbol and can be composed of one or more words or partial symbols. In the present case, the implementation is selected by means of a character string, which can contain one or more words, and the terms represented by the words are arranged from left to right in a sequence which starts from the general and becomes more special. Taking the replacement expression in column 2, row 2 of this table as an example, "INCOME" is the most general category, followed by the more specific "SALARY", which further specifies general income and characterizes the special form of income as "SALARY", as well as further specified by Word "KAUFCITY", whereby this salary income is marked as one that was paid by the employer Kaufcity. In this way, a more and more specific categorization is made possible along the sequence of terms, while at the same time the information about belonging to the more general category is retained. Each row in column 3 of the table contains a priority value, in the present case implemented as an integer numerical value. Alternatively, this can also be realized by letters or other symbols, which are subject to a strict order.

This arrangement in the table memory realizes the line-by-line assignment between a key expression, a replacement expression and a priority value per line. For this Example can also be seen from each key expression exactly one priority value is assigned in this way, and Moreover every key expression exactly one replacement expression, the latter assignment being unambiguous is, i.e. it is equally clear every replacement expression exactly one key expression assigned. In this way you can print the key at any time on the associated Replacement expression and vice versa.

In the present embodiment is this table memory, like all the others in this exemplary embodiment, realized by a Microsoft Excel spreadsheet. The key phrases that in this embodiment through a pseudo data query language for one Easier to understand are presented in a concrete Embodiment realized by SQL statements.

1 II shows a further table memory with still uncategorized raw data records in column 1, with a raw data record in each row. Column 2 of the table memory in still unoccupied, because no categorizing replacement expressions have yet been assigned to the raw data records net. In column 1, the raw data records of sales data consist of a text of the intended use as a text string and the textual amount of sales. Such a raw sales record can be clearly imagined, for example, as an entry in an account statement.

In 1 III are different intermediate states of the in 1 II described table memory shown during the execution of the method. The presentation of this sequence of intermediate states (one state per table) is based on the assumption that for each key expression the 1 I shown table over the entire raw data set inventory of sales data (as in the table storage in 1 II shown) iterates. Alternatively, however, another access and comparison method is also conceivable, depending on the underlying database system, which makes such a table memory available to the outside, for example parallelized methods.

So it is first based on the first key expression in 1 I , Row 1, column 1 all in the table memory 1 II If the raw sales data records contained are checked for compliance with the key expression, the system determines that the data records from row 1 and row 3 of the table memory (as in 1 IIIa ) shown) fulfill the key expression. Correspondingly, in rows 1 and 3, the data records in column 1 are assigned the corresponding replacement expression in that it is stored in the table in 1 I , Line 1, column 2 containing the replacement expression corresponding to the current key expression in line 1, column 1 and into the table memory, as in 1 IIIa ) is shown, in rows 1 and 3, column 2.

This insertion of the replacement expression to those for it certain places in the table memory are carried out after checking whether at these points there is already a replacement expression in the table memory is included. This is not the case here.

It is different in the following run, the result of which is shown in the table storage status III b). Here the key expression from the in 1 I shown table memory, line 2, column 1 used for checking with all raw data sets, the data set in line 1 of the in 1 IIIb ) table storage state shown fulfills this replacement expression. However, this data record was previously created, as in 1 III a ) shown in the appropriate place, assigned the replacement expression "INCOME". So after determining that the current key expression ( 1 I , Row 2, column 1) through the data set ( 1 III b ), Line 1, column 1) is fulfilled, based on the previous replacement expression assigned to the data record ( 1 IIIa ), Row 1, column 2) in table memory I determines the priority value belonging to this replacement expression (or to the corresponding key expression) ( 1 I , Row 1, column 3), which in this case is 1. This is now with the priority value of the current key expression ( 1 I , Row 2, column 1), namely the priority value 3 (same row, column 3) compared. In this case, it is determined in this case that the priority value of the current key expression, namely the value 3, is higher than the priority value of the previously assigned replacement expression, namely the value 1. In accordance with the method, the replacement expression corresponding to the current key expression ( 1 I , Line 2, column 2) in the place of the previous replacement expression and thus the data record ( 1 III b ), Line 1) assigned a new replacement expression. This key printout is not met by any of the other data records listed here as examples.

In the next example III c) was then the key printout from table memory I, line 3, column 1 used for the check, this key expression also being used in an analogous manner is met by the data set in III c), line 1. Also in analog A comparison is also made of the priority values of the existing one Replacement expression with the priority value of the key expression currently used for the check carried out. In this case, however, the existing replacement expression is assigned (III b), row 1, column 2) retained (III c), row 1, column 2) because the priority value the current key expression is less than the priority value the existing assigned replacement expression.

In a manner analogous to III a), a further key expression (I, line 3, column 1) was used for checking in III d) and it was found that the data record in line 2 of the table memory as shown in III d) was fulfilled and thus the Replacement value is assigned in a procedural manner. The peculiarity of this assignment compared to that explained in III a) is that the replacement expression as part of the device-related assignment contained in table memory I in row 4 not only assigns the key expression to a replacement expression on the conceptual generalization level, but also the replacement expression Has information about the classification of the overlying, generalized categorical level. The present key expression is thus fulfilled by the occurrence of the word "toys" in the data record, and then assigns to this data record the information that it is toy sales that at the same time fall into the upper class of consumer sales. The fact that the main category is contained in storage table I makes subsequent quantitative processing at various categorical general book levels much easier, as described in fol is still described.

By gradually pulling it up of the various key phrases to check with everyone raw data sets could, as the different storage states III a) to III d) show, the data record in the fourth line in memory status III d) is not be provided with a replacement expression and therefore not immediately be classified. To the process in terms of final processing simplify assignments and assess categories will be in a final Pass III e) all not yet with a replacement expression provided records with a zero replacement expression provided that in this case "no Assignment "is (III e), row 4, column 2).

After running this main part The method is originally in memory state II existing table memory now in the memory state as in III e) shown. Further processing now takes place from here the initial categorization by assigning the replacement expressions. On the one hand, a simple numerical entry per class is possible across from the search and counting methods known in the prior art already has the clear advantage that due to the arrangement several terms per replacement expression, which are conceptual in their sequence become more special, the quantitative evaluation without renewed Assignments and search for terms according to the desired subtlety configured parameterized can be. For example, memory table III e) can be evaluated according to the main categories at the top level, such as that such records as belonging to the same category which match in the first partial term of the replacement expression. In this way, two data records result from III e) Income category, a record of the Category "consumption", as well as a data set the category "none Assignment ". Off same memory table without having to search again and / or Assignment would have to take place also a finer categorization can be made by using analog Way such records as belonging to the same category which are evaluated in the first two terms (analogously for any number other or all terms) match. Practical such a categorical structure appears through several sub-terms a replacement expression when using approximately six partial terms or category hierarchy levels.

When processing this Data can particularly advantageously also those contained in the sales data records amounts can be used to calculate derived key figures. Such Key figures can about the income itself, as well as saving (absolute and divided by bank products) and the savings rate (i.e., the ratio of the Savings amount to the income amount) the own and third party share with Saving (i.e. the share that the customer's house bank has towards third parties Financial service providers on the customer's savings, as well as risk provisioning and / or freely available Income. Based on these key figures, savings behavior can be (savings) attractiveness and (savings) prospect of success be assessed. All these key figures are calculated based on the classified sales. Unclassified sales flow into the Calculation not one. The key figure "income of a customer" is calculated - as a simple one Example - as the sum of all sales of the customer classified as income of any kind where the sales records have been provided with a replacement expression, the first or only sub-term is "income". Be on this Salary, child benefit, income from rental, interest income etc. all counted towards income. The absolute savings amount, for example is the sum of the individual savings, which in an analogous manner were classified in all Areas: traditional savings, building savings, repayment of financing, Fund savings and life insurance. It can be divided after payments to the house bank and to external financial service providers. The freely available amount can also be determined as the derived key figure which is the income less expenses for rent, Consumption, risk provisioning, saving and other things.

In addition to these general key figures can specific analyzes can also be carried out, such as can be determined which income group has a particularly high savings rate with have a high proportion of American financial service providers.

As a simple example, too the sum of all sales from those records are formed, which are assigned the replacement expression with the first partial terms "consumer media Internet" to have.

In the practical implementation of the system according to the invention The classification of the main categories proves to be particularly advantageous in the areas of income, savings, rent, consumption, risk provisioning and other. This enables a practical calculation of banking and sales-related key figures at customer and household level allows.

1a shows a table memory dump of a practical implementation with replacement expression, key expression and priority value. In this representation of an actually implemented system, the key expression is shown in particular as an SQL statement that is common in practice.

Using the example of the first line, all data records that fulfill the key expression are assigned the replacement expression "income-salary-AOK" with priority 1; in the present case, these are all data records that fall in the month of January 2002, whose sales amount is greater than 0, and continue to do so whose text key has the value 53 or 56 and whose intended text contains the character string "AOK - the health insurance company".

1b shows a table memory extract of an actually implemented system as an exemplary embodiment of the invention, in which data records are each assigned exactly one replacement expression, the data records each containing eleven data fields.

In the specific case shown, the data record consists of eleven data fields. The twelfth field is reserved for the replacement expression. Typically, the key expression refers to, such as in 1a shown, only on two of the three data fields, for example "KONTOAUSZUGSTEXT" in combination with "GEGENBANKLEITZAHL". The key expression very often refers to a single data field, such as "ACCOUNT EXTRACT TEXT". Combinations of elements of a key expression are - as is usual for SQL statements - logically connected with AND or OR operations. Key expressions can in turn consist of combinations of key expressions. In this embodiment, too, each key expression has an associated replacement expression, the number of key expressions being equal to the number of replacement expressions.

One versus the too 1 The procedure described above is an advantageous alternative in this embodiment, the table memory - as in 1a shown - to be sorted according to priority, which makes redundant priority determination and comparison processes as described in 1 IIIc ) have been omitted because the data records are always checked for the key expression in the order of increasing priority. The in 1 IIIc ) illustrated case that the key term is met by the data record, but the replacement has a lower priority than the previously assigned one and therefore the replacement expression is retained, does not occur. By keeping the table memory sorted as in 1a represented according to priority values, thus advantageously avoids unnecessary comparison processes.

2 shows a screen excerpt of a tool - in this example - implemented in the Microsoft Excel spreadsheet for entering and maintaining the table memory ( 1 I ) contained key expressions for a specific replacement expression and for assigning a priority value. The dialog window, which is clearly highlighted in the screen extract, shows a SQL expression (Structured Query Language) in full and realistic complexity. Various buttons with standard operators are available for the simplified formulation of such a key expression. The priority can be indicated in the selection field on the right and can be added to the table memory by pressing the "Apply" button.

3 shows a screenshot of the already in 2 tool shown in another aspect. Based on the data stored in the table memory ( 1 I ) available replacement expressions hierarchically selectable from the table memory, which can be deleted or edited by this tool, among other things. After pressing the "Edit" button, the already in 2 Shown and described there is carried out. Furthermore, a new triple of replacement expression, key expression and priority value can be generated from here by first generating the desired replacement expression by pressing the "Add" button and then assigning the corresponding key expression and priority value by "editing".

4 shows a screen excerpt that a user of the system sees during the actual data evaluation. In the present case, the sales of a specific customer are based on the classification that has already taken place (i.e. on the basis of a stored table as in 1 III e )) examines whether there are any sales which fall into the category "consumer media Internet America Online Inc." suffice. This is the present example of sales for the particular customer in July of the year 2001 the case. This revenue is displayed in the bottom left of the "Bank Statement Text" dialog box.

5 shows a screen shot of another application scenario. In this application, a compilation of derived key figures of the turnover of a specific customer is calculated and displayed. Again starting from the already fully assigned data records ( 1 III e )) The sales of the customer "Jamon Pan" are filtered out in July 2001 as a subset of the data records to be processed, such filtering again using SQL expressions and corresponding queries about the fully categorized data record table as in 1 III e ) can be implemented, as is also the case with the other connected search and filter functions, for example in 4 , Applies. From this subset of categorized data records, the sales per category falling into these categories are added and displayed according to the categories of savings, building society savings, financing, fund savings, life insurance, property insurance, each for own business and external business as well as rent and income brought, from which further values are added up, shares are calculated according to percent, and the freely available amount and the savings rate are determined and displayed.

Claims (14)

Automatic classification procedure of data records, the method being carried out in a digital computer, characterized in that a) an assignment of key expressions to replacement expressions and priority values is selected, in which each key expression is uniquely assigned a replacement expression and each key expression is also uniquely assigned a priority value, b) for each The key expression of each data record is checked to determine whether the respective data record fulfills the current key expression and, if this is the case, the substitution expression belonging to the current key expression is clearly and permanently assigned to the respective data record if: (1) the replacement data record has not yet been assigned to the respective data record or (2) a replacement expression has already been assigned to the respective data record and the priority value of the current key expression is greater than the priority value of the already assigned replacement expression end of key expression. c) the assignment pairs thus created from the data record and the replacement expression are checked for at least partial correspondence of the replacement expression with one another, the data records of such matching assignment pairs being rated as belonging to the same class. A method according to claim 1, characterized in that the assignment is selected so is that the replacement expressions each contain a sequence of more specific terms. A method according to claim 2, characterized in that the exam for at least partial agreement different assignment pairs by comparing the replacement expressions along the each of the term sequences contained therein takes place. Method according to one of claims 1 to 3, characterized in that that the records each an alphanumeric string corresponding to the sales data contain. Method according to one of claims 1 to 4, characterized in that that as belonging to the same class records numerically per class be recorded. Method according to one of claims 1 to 5, characterized in that that records, that do not meet any of the key terms Replacement expression that contains a null term is assigned. Device for carrying out a method according to one of the claims 1 to 6, characterized in that - First storage means with contain a clear assignment of data records to replacement expressions are, as well as second storage means with an assignment of key expressions replacement expressions and priority values, - in which by assigning exactly one to each key expression contained therein the replacement expression contained therein is uniquely assigned and also assigns exactly one priority value to each key expression contained therein is - data processing equipment are included with - means to check whether a record a key term enough, - means to determine priority values based on the replacement term or the key term and for comparison of these values, - means for comparison between strings in selectable areas - means to assess compliance of replacement terms between different assignment pairs from data record and replacement term, - the first and the second Storage means and the data processing means are interconnected for informational purposes are. Apparatus according to claim 7, characterized in that the first storage means in addition to the assignment also the data records and the replacement expressions contain. Device according to one of claims 7 or 8, characterized in that that the second storage means in addition to the assignment also the key expressions and / or the replacement terms and / or the priority values contain. Device according to one of claims 7 to 9, characterized in that that the key phrases as expressions in a data query language. Device according to one of claims 7 to 10, characterized in that that the replacement expressions each consist of several terms, which are conceptual in their sequence are more specifically arranged. Device according to one of claims 7 to 11, characterized in that that means are available to add numerical values from the data sets remove and further process arithmetically. Use of a method or device according to one of the preceding claims for automatic quantitative Evaluation of automatically generated and electronically saved sales data strings a variety of bank movements. Use according to claim 13, characterized in that the quantitative analysis is also the derivation of derived Indicators includes.