WO2004095313A1

WO2004095313A1 - Data processing system for user-friendly data base searches

Info

Publication number: WO2004095313A1
Application number: PCT/EP2004/004136
Authority: WO
Inventors: Jürgen ANGELE
Original assignee: Ontoprise Gmbh
Priority date: 2003-04-19
Filing date: 2004-04-19
Publication date: 2004-11-04
Also published as: DE10318333A1

Abstract

The invention relates to a data processing system (1) comprising at least one unit (2) for storing data. The data is structured by an object model. An inference unit, which inter alia uses rules of the object model in order to ascertain concealed information, is used to evaluate data requests. A dialogue system (6), enabling natural-language sentence sequences to be inputted, is provided as an input-output system. The dialogue system (6) comprises a unit for translating the natural-vocal sentence sequences to be translated into a data request format which can be processed by the inference unit, enabling users to formulate requests in natural language which can be evaluated by the system in a highly comprehensive manner

Description

Data processing system for user-friendly database searches

description

Field of the Invention

The specific query or search for information in large amounts of data causes various difficulties. For example, entering search terms combined with a text search often results in search results that are not very relevant. In addition, relevant search results may be overlooked if they are expressed by terms other than those used for the search.

To solve these difficulties, DE 101 03 845 AI proposes a data processing system. This data processing system comprises a computer system with means for storing data. Furthermore, the data processing system has a query unit for determining output variables or search results with access to the stored data. The data are assigned to predefined classes which are part of at least one stored class structure which forms an object model. So-called ontologies are often used as the object model. The classes are assigned attributes that are inherited within a class structure. Attributes are characteristics of a class. The class "person" can have the attribute "hair color", for example. This attribute is assigned different values (called "attribute values") for different specific people (called "instances"), e.g. B. brown, blonde, black, etc.

The query unit contains an inference unit, by means of which rules can be evaluated, by means of which predetermined classes and / or attributes are linked. This makes it possible to find information that was not described by the search terms. The inference unit even makes it possible to generate new information by linking individual statements that was not explicitly contained in the data, but can only be inferred from the data.

The query unit can be used to enter query terms as input variables, which are not only those of the stored ones

Data are formed, but can also be formed by the classes or attributes of the class structure. In addition to a question: "What articles did Mr. Mustermann write?" - Mustermann would be a specific name from the data - are also questions of the type: "What are the names of all employees of company X?" possible. In such an example, a search term would search not for the specific name of an employee, but for the values of the "hatName" attribute of all instances of the employees who are related to instance X from the class of the company, namely in the relationship

"is employed". Such queries are often formulated in the language F-Logic (J. Angele, G. Lausen: "Ontologies in F-Logic" in S. Staab, R. Studer (Eds.): Handbook on Ontologies in Information Systems. International Handbooks on Information Systems, Springer Verlag, 2003, page 29). The following example, which depicts the relationships between well-known biblical persons, may serve as a rough, intuitive understanding of F-logic:

First some facts:

abraham: one. sarah: woman. isaac: man [fatherls-> abraham; motherIs-> sarah]. ishmael: man [fatherls-> abraham; motherIs-> hagar: woman]. jacob: man [fatherls-> isaac; motherls-> rebekah: woman]. esau: ma [fatherls-> isaac; motherls-> rebekah].

One recognizes certain classes: "man" and "woman", ie "man" and "woman". So Abraham is a man. The class "one" has z. B. the attributes "father" and "mother", so the attributes indicate the parents. For example, the man Isaac has the father Abraham and the mother Sarah.

F-logic is recognizably also suitable as a language to formulate a class structure. In addition to F logic, the ontology languages RDF, DAML + OIL and OWL are preferably used to define class structures.

Continue with some rules that show relationships;

FORALL X, YX [sonIs - >> Y] <- Y: man [fatherIs-> X]. FORALL X, YX [sonIs - >> Y] <- Y: man [motherls-> X]. FORALL X, YX [daughterls - >> Y] <- Y: woma [fatherIs-> X]. FORALL X, YX [daughterls - >> Y] <- Y: oman [motherls-> X].

Rules in F logic always have a control head on the left and a control body on the right. The first rule is translated z. E.g .: If Y is a man whose father was X, Y is one of the sons of X. The simple arrow "->" indicates that only one value is possible for a certain attribute, while the double arrow "- >> "indicates that several values are possible.

Finally, a query that asks about all women who have a son whose father is Abraham. In other words, with which women did Abraham have a son?

FORALL X, Y <- X: woman [sonIs - >> Y [fatherIs-> abraham]].

A query takes the form of a rule, but without the rule header on the left.

The answer is:

X = sarah X = hagar

In this way, the query options compared to conventional database systems are expanded considerably.

The disadvantage here, however, is that the user must be aware of the elements of the class structure that can be queried, so that the user can enter such elements via predetermined input formats in order to be able to carry out a search in a database of the data processing system. The elements of the class structure are classes, attributes, synonyms, relations, that is, relationships between elements or assignments, in short all that is used to build the object model or class structure.

task

The object of the invention is to simplify the carrying out of extensive searches in large databases as much as possible.

solution

This object is achieved by the inventions with the features of the ^■ independent claims. ^■ Advantageous developments of the inventions are characterized in the subclaims.

A data processing system is proposed with an information generation unit for storing or generating data forming a database. The information generation unit is in particular formed by a storage unit for storing data. In general, it can also be formed by information-generating systems such as sensors.

Both a stand-alone computer and a network of computers, for example an in-house, closed network, or computers which are connected to one another via the Internet can be considered as data processing system. Furthermore, the data processing system can be implemented by a client-server constellation, parts of the invention running on the server and others on a client. The data processing system has a computer unit for generating an object model, in particular an ontology, consisting of a class structure, the data being structurable using the object model. An object model with a class structure is thus assigned to the database.

The object model includes rules that contain statements about relationships. The rules form a declarative system. This means that the rules can be evaluated regardless of their order.

Furthermore, relations generally belong to the object model, i. H. Information about relationships between different elements of the class structure. For example, a relation can indicate that a city is "in" a country, or that person X is married to person Y.

The individual data in the database are arranged in the data structure by means of a link. If the data is in the form of a table, for example, the link is typically made by the table corresponding to a class. The rows or column headings of the table correspond to the attributes of the class. The values that the attributes assume for a specific data record can then be found in the fields of the table.

With the class structure, the data stock is structured in a predetermined manner, whereby the elements of the class structure can function as part of data queries in addition to the data of the data stock itself. Furthermore, the data processing system has an input / output unit for entering a data query and for outputting responses to the data query, and an inference unit for evaluating data queries. The inference unit uses the rules belonging to the object model to evaluate the data query in order to find information that is not explicitly available. Since the rules form a declarative system, the result of evaluating the rules is independent of the order in which the rules are specified.

A dialog system is provided as an input / output unit, via which natural-language sentence sequences can be entered as input variables from which the data queries are derived automatically.

The data processing system preferably has a unit for translating the natural language sentence sequences into a data query format, which can be processed by the inference unit for evaluating the data query. Ideally, this translation unit consists of a system for analyzing or parsing natural language texts. In the simplest case, the translation unit only contains a system for identifying certain key terms in the natural language inputs. Structured queries for the inference unit are generated in a predetermined manner from the key terms.

It is also possible for a set of permissible sentence types to be stored in the translation unit, with which the input variables are compared. If a record is entered that cannot be assigned to any of the saved record types, an error message is output via the dialog system so that the user can change the input size. Preferably the error message already gives an indication of the type of error, which gives the user an indication of how he has to change the input size.

A major advantage of the data processing system according to the invention is that natural language sentence sequences can be entered via the dialog system for performing searches in a database. An operator of the data processing system can therefore carry out searches in the data stock with freely formulated input variables without adhering to specified input formats and without knowledge of programming languages.

The class structure is preferably designed such that the individual elements of the classes have attributes and / or

Synonyms are assigned. This significantly extends the research options. Synonyms can also be assigned to the attributes. Furthermore, synonyms can also be assigned to the data in the database itself. The synonym structures can be designed as so-called meta attributes, which are assigned to predefined classes of the class structure.

If, for example, a certain term is extracted from a sentence sequence forming an input variable in the dialog system to form a data query, this term is initially not only compared with the data or the elements of the classes, but also with the assigned synonyms. This means that even by entering fuzzy terms, evaluable data queries are generated by the translation unit and searches are carried out with a high success rate. To z. B. to be able to adapt expanded data sets or changed applications, the object model or the class structure can advantageously be changed, as can the rules belonging to the class structure. In addition, with unchanged data, the class structure and rules can be expanded to map the data in more detail and to expand the research options or to make them more precise. A changed object model can provide additional searchable terms that can be entered by the user as input variables. In this way, the usability of the data processing system is significantly expanded.

Furthermore, the natural language sentences forming the input variables can contain terms or relations which contain fuzzy relationships such as, for example, "inexpensive", "a little far from". Such terms or relations are processed using rules and thus converted into precisely defined relationships, on the basis of which the data in the database are researched.

In the examples mentioned, the term “inexpensive” can typically be converted into a rule “price less than X” and the relationship entered “a little far away” into a rule “distance less than Y”, where X and Y are predetermined quantities. An associated rule could e.g. B. read:

FORALL X, Y X: Auto [Price Estimation- Low Cost] <- X: Auto [Price-> Y] and Y <5,000.00.

In this case, too, reproducible research results can be obtained despite terms and relationships that are not clearly defined. It is particularly advantageous that the user have no knowledge of the function or structure of the rules to carry out the research.

If a user enters a sequence of sentences or questions as input variables for carrying out a search, the history of this sequence is also tracked and recorded in the data processing system. This enables the user to gradually approach the desired search result in an interactive dialog using the dialog system. For example, he can first search for a flight to the sea in a flight database. If, as a result, cities are too far away for him, he can use a vague term to formulate a restriction in a second question that the cities would be too far away for him.

An advantage of the data processing system according to the invention is that the implementation of the object model and of the dialog system is independent of the storage unit on which the database is stored. A wide variety of information sources, such as databases at different locations and in particular information from the Internet, can thus be used as a database for the data processing system according to the invention.

For example, the current temperature of a city can be an attribute of the class "Cities" with the further attribute "hatName". The specific values of these attributes can be detected by a sensor or they can be read in by an internet service. A possible rule would then be:

FORALL X, Y, Z

X [hatTemperature-> Z] <- X: Stadt [hatName-> Y] and internetservice- temp (Y, Z)]. A rule for the fuzzy term "when is a city warm" could be:

FORALL X, Y: city [hasCondition-> warm] <-

X: city [has temperature-> Y] and y> 28.

According to this, a city is defined as "warm" if its current temperature is more than 28 ° C.

The data processing system according to the invention can thus be used in a wide variety of applications. For example, the data stock can be formed from the data of a company, in particular its products, new developments and marketing data. In particular, the data stock can also be formed from data from several companies.

The invention was explained above using a suitable data processing system. It can be implemented in the same way in the form of a method which is carried out on a data processing system.

Furthermore, the scope of the invention includes a computer program that executes the method according to the invention in one of its configurations when it runs on a computer or computer network.

Furthermore, the scope of the invention includes a computer program with program code means in order to carry out the method according to the invention in one of its configurations when the program is executed on a computer or computer network. In particular, the program code means can be stored on a computer-readable data carrier. In addition, the scope of the invention includes a data carrier on which a data structure is stored which, after loading into a working and / or main memory of a computer or computer network, can carry out the method according to the invention in one of its configurations.

The scope of the invention also includes a computer program product with program code means stored on a machine-readable carrier in order to carry out the method according to the invention in one of its configurations when the program is executed on a computer or computer network.

A computer program product is understood to mean the program as a tradable product. In principle, it can be in any form, for example on paper or a computer-readable data carrier, and can in particular be distributed over a data transmission network.

Finally, the scope of the invention includes a modulated data signal which contains instructions that can be executed by a computer system or computer network for executing a method according to one of the preceding method claims.

The invention is explained in more detail below on the basis of exemplary embodiments which are illustrated schematically in the figures. However, the invention is not limited to the examples. The same reference numerals in the individual figures denote the same or functionally identical elements or elements which correspond to one another in terms of their functions. In detail shows: 1 shows a schematic representation of an exemplary embodiment of the data processing system according to the invention; and

2 shows an example of a class structure for the data processing system according to FIG. 1.

1 shows an exemplary embodiment of the data processing system 1 according to the invention. In the present case, the data processing system 1 has two storage units 2, 2 'as information generation units, on which a database is stored. Furthermore, sensors providing data or other data supplying units can be provided as information generation units. The number of storage units 2 on which the database is stored can be selected as desired. The storage units 2 can be formed by database systems, file systems and the like.

In the present case, the first storage unit 2 consists of a database system of a travel agency, on which data about travel, in particular air travel, are stored as a database. The second storage unit 2 'is formed by databases accessible via the Internet, on which further data from the database are stored, e.g. B. Current temperature data of different cities.

A server 3 is assigned to the storage units 2, 2 ', via which the data stored in the storage units 2, 2' are accessed.

A computer unit 4 is assigned to this server 3. The computer unit 4 has a processor system on which a first software module 5 is implemented, by means of which an object model forming an ontology can be generated. Furthermore, the computer unit 4 has a further software module 5 ', which forms an inference unit for evaluating data queries.

Finally, the computer unit 4 has a dialog system 6, which has the function of an input / output unit. For this purpose, the dialog system 6 preferably has a terminal, not shown separately, via which data can be input and output. Furthermore, the dialog system 6 has its own processor system. Alternatively, the dialog system 6 can use the processor system of the computing unit 4. In general, the dialog system 6 can thus be part of the computer unit 4 or form a separate unit.

Part of the dialog system 6 is a translation unit (not shown) which can translate natural language sentence sequences into a data query format, preferably F-logic queries, which can be processed by the inference unit.

The data of the database is structured using the object model installed on the computer unit 4. It is essential here that the data is structured solely by specifying the object model, so that no interventions in the structures of the data stock on the storage units 2, 2 'are necessary. The database on the storage units 2 can be, for example, a database on an SQL server. This remains unchanged. The object model merely creates a class structure with attributes, synonyms and rules and refers to the data in database 2. The data stored in the first storage unit 2 are, for example, travel data. A data record for a flight is defined by the start and destination airport, the travel time and the price. It may also include information about a hotel that may have been booked with the flight.

Correspondingly, there is a class for air travel in the object model with the attributes: departure airport, destination airport, travel time, price and hotel.

To structure the data, a geographic model is also generated, which is shown as an example in FIG. 2.

2 has a hierarchical class structure for forming the geographic model. The hierarchical structure consists of classes and subclasses. The "States" class is subordinate to the "Continents" class. The "States" class is subordinate to the "Regions" class. The "Cities" class is subordinate to the "Regions" class.

The data of the data stored in the first storage unit 2 are assigned to this class structure as so-called instances. If the data in the first memory unit 2 is in the form of a table, the link is made in that a table corresponds to a class, and the rows or column headings of a table correspond to the attributes of the class.

For example, Las Palmas is assigned as an instance of the "City" class. According to the class structure this instance is also associated with the "Canary Islands" region as an instance of the parent class'regions'^'. Furthermore, the regulation for the class "state", in which it is defined that Las Palmas is located in Spain as an instance of the class "state". Finally, the assignment is made that Spain is located in Europe as an instance of the "continent" class.

The geographical model according to FIG. 2 contains, as further components, "seas" and "mountains" as landscape forms, which each form separate classes.

The classes of the class structure are assigned attributes that are passed on within a class structure. The simplest attribute in the example considered is the attribute "hatName" or "Name". The instance assigned to this attribute then specifies the name of the state, the region, etc. In general, attributes can be used to assign certain properties to certain classes.

Synonyms are defined as a special case of such attributes. Here, synonyms for the name of attributes can be assigned. For example, instead of the attribute "lies" for one

State that lies within a specified continent, the designation "TeilVon" can also be used for the attribute.

Similarly, synonyms for the name of the class itself can be defined. For example, the "State" class can be assigned the meta attribute "Country" as a synonym, or the "Seas" class can be assigned the meta attribute "Oceans" as a synonym. These attributes are referred to as so-called meta attributes.

Furthermore, synonyms can also be defined at the level of instances, i.e. data. For example, the instance "Canary Islands", which belongs to the class "Regions", can be assigned the synonym "Canaries". As further elements of the geographic model, FIG. 2 contains “climate profiles” and “longitude and latitude”, which are preferably assigned as attributes to the “cities” class.

In the present case, the data assigned to these attributes are stored in the second storage unit 2 '. These data are preferably available on the Internet.

The class structure defined in this way is not fixed, but can be expanded and supplemented if necessary or even replaced by a new class structure. Since the class structure is defined by a software module, the class structure for structuring the stored data can be changed in a simple manner by suitable programming.

The functioning of the data processing system 1 is explained below using examples.

As a first example, the following natural language sentence sequence is entered by a user as input variable: "I would like to fly to the sea for a week on March 29th".

In the simplest case, the translation unit extracts the following key terms from this: "03/29", "a week" and "to the sea". Since the example in question is a flight database, the word "fly" does not have to be recognized. In the case of a more general travel database, the sentence entered would also be searched for such a keyword. The three keywords found can easily be identified by their format or by comparison with a few examples:

Start of the trip: 29.03.

Travel time: one week

This results in the end of the trip: 05.04.

Destination: is by the sea.

The translation unit uses this to generate the following F-logic request:

FORALL X <- X: Region [liesAm-> Sea]. FORALL X, Y <- X: Region and X [hatstadt-> Y]. FORALL Y, Z <-

Z: Flight [destination-> Y; Abflugdatum-> 3.29. ; Rückflugdatum-> 4.5. ].

In the class structure according to FIG. 2, knowledge of regional relationships is defined, for example, by linking the classes “states”, “regions” and “cities” on the one hand and the class “seas” on the other, so that a selection of regions is displayed as the output variable, that lie by the sea. A choice of states would be unsuitable because in a large country by the sea a city can be very far from the sea.

The list of regions is further processed in such a way that all cities in the flight database which are located in these regions are output. These are - since it is a flight database - only cities with airports.

This means that all travel destinations are known. Then all flights to these destinations are issued with the departure date 29.03. and the return flight date 05.04. As a regular service for flights "to the sea", the current temperature values in the individual cities are also queried via the Internet, since the input size implies that the user would like to book a beach holiday. This implies that the temperatures in the destinations should be sufficient. With this query, information from sensors, namely temperature values, is queried. This further limits the possible trips.

If the user then enters the further input variable: "But I want to go further away", the translation unit extracts the keywords "further away" from it. An F logic request is formed from this, namely:

FORALL X, Y, D1, D2

<- X: City and Y: current destination and distance (current location, Y, Dl) and distance (current location, X, D2) and D2> D1

The result is all cities that have a greater distance to the current location than the current flight destinations. This data query quantifies the fuzzy term "further away".

To calculate the distance, the attributes "Longitude and Latitude" for cities in the previously selected regions can be used to determine the distance to the travel destinations.

The user is then shown a selection of travel destinations in the previously selected regions as the output variable, the distance of which lies above a predetermined limit value. Finally, if the user enters an additional input variable: "I want to go to the Canary Islands", the synonym structures defined in the geographical model recognize that the region of the Canary Islands is desired as the travel destination.

The user can then make the desired booking on the basis of the destinations on the Canary Islands which are then output, and depending on the design of the dialog system 6, the user can alternatively be offered a booking of an air travel with or without a hotel booking.

As a second example, a user formulates the input variable: "I want to climb and swim on vacation".

The key words this time are "climbing" and "bathing".

The system has a database in which keywords and, if applicable, their synonyms can also be found as wildcards (e.g. "kletter *"), as well as the respective translation into an F-logic query or the associated assignment an element of the class structure from which a suitable F-logic query can be generated.

"Climbing" is converted into the F logic query:

FORALL X <- X: Region [liesAm-> Mountains].

"Baden" is implemented in the F-logic query:

FORALL X <- X: Region [liesAm-> Sea].

We are looking for a region that has a sea and mountains: FORALL X <- X: Region [liesAm-> Sea].

FORALL X <- X: Region and [liesAm-> Mountains].

The user is shown a suitable holiday region as output variable, for example Corsica.

If the user enters an additional input variable: "Is there something similar?", The geographic model is used to search for comparable regions.

If, for example, the class structure according to FIG. 2 is expanded in such a way that the class “lakes” is provided in addition to the class “seas”, then a similarity relationship between the classes “seas” and “lakes” can be defined using rules.

Then, as the output variable for the second request, the region of Lake Garda, for example, can be displayed to the user as a travel region similar to Corsica.

As a result of the data queries derived from the input variables in this way, the user can also carry out searches with generally and vaguely formulated queries, for example in order to book a trip. The advantage here is that he does not have to specify a specific travel destination exactly as in known database systems, but can carry out the research independently of the data formats of the data stored in the travel agency.

As a result of a data query, a subset of the data stock is preferably output via the dialog system 6 as the output variable. In the case of a search to be carried out, the user can enter a sequence of several input variables into the dialog system 6, with a corresponding output variable preferably being output for each input variable. In the dialog system 6, the history of the user inputs is recorded in that the input variables of the sequence are linked with AND links. This ensures that information found during research is not lost.

reference numeral

Data processing system Storage unit for data 'Storage unit for data server computer unit Software module for generating the object model' Inf erence unit dialog system

Claims

claims

1. Data processing system a) with at least one information generation unit (2) for storing or generating data forming a database; b) with at least one computer unit (5) for generating an object model, consisting of a class structure and a declarative system of rules, the data being structurable by means of the object model; c) with an input / output unit (6) for entering a data query and for outputting responses to the data query; and d) with an inference unit (5 ') for evaluating the data query; characterized in that e) a dialog system (6) is provided as the input / output unit, via which natural-language sentence sequences can be entered as input variables from which the data queries are automatically derived.

2. Data processing system according to the preceding claim, characterized by a unit for translating the natural-language sentence sequences into a data query format, which can be processed by the inference unit.

3. Data processing system according to one of the preceding claims, characterized in that that in the dialog system (6) a set of permissible record types is stored, with which the input variables are compared.

, Data processing system according to one of the preceding claims, characterized in that the class structure has synonym structures assigned to the classes.

5. Data processing system according to the preceding claim, characterized in that the synonym structures are designed as meta-attributes, which predefined classes are assigned to the class structure.

6. Data processing system according to one of the two preceding claims, characterized in that synonyms are assigned to the data of the database.

7. Data processing system according to one of the preceding claims, characterized in that the class structure and / or the associated rules and / or associated synonym structures can be changed.

8. Data processing system according to one of the preceding claims, characterized. that elements of the class structure, attributes or synonym structures are assigned to form a data query from an input variable in terms contained therein.

9. Data processing system according to one of the preceding claims, characterized in that a fuzzy term in an input variable in the form of a natural-language sentence sequence is translated by a rule into a quantitative variable.

10. Data processing system according to one of the preceding claims, characterized in that a plurality of input variables entered one after the other are logically linked by means of AND operations.

11. Method for querying data from a database, the following steps being carried out in particular using a data processing system: a) the database is stored; b) an object model is generated, consisting of a class structure and rules, the data being structurable with the aid of the object model; c) a data query is entered; d) an inference unit (5 ¹ ) evaluates the data query; e) a response to the data query is output; characterized in that f) that a dialog system (6) is provided as an input / output unit, via which natural-language sentence sequences can be input as input variables; and g) that the natural language sentence sequences are translated into a data query format which can be processed by the inference unit.

12. Computer program, characterized in that it executes the method according to the preceding method claim when it runs on a computer or computer network.

13. Computer program with program code means to carry out a method according to one of the preceding method claims when the computer program is executed on a computer or the computer network.

14. Computer program with program code means according to the preceding claim, which are stored on a computer-readable data carrier.

15. Data carrier or computer system on which a data structure is stored, which executes the method according to one of the preceding method claims after loading into a working and / or main memory of a computer or computer network.

16. Computer program product with program code means stored on a machine-readable carrier in order to carry out all steps according to one of the preceding method claims when the program is executed on a computer or computer network.

17. Modulated data signal which contains instructions for executing a method according to one of the preceding method claims which can be executed by a computer system or computer network.