DE10291392B4 - Method, system and data carrier for generating correlations and / or interactions and / or knowledge from a plurality of searched data sets - Google Patents

Abstract

Method for generating correlations and / or interactions and / or knowledge from a plurality of searched data sets,
characterized by the following process steps:
manual generation of a knowledge block (KG) with an individual provision of term tables, which have terms and / or words for characterization of a given knowledge area, and
machine generation of an information block (IG) from data sets and electrical reading units with automated provision of thesaurus tables having terms and / or words of a searched data set,
automatic generation of vector references (pointers), by which references between similar and / or different blocks of knowledge and information blocks (IG, IG, KG, KG, IG, KG) are marked,
Weighting of these pointers in accordance with matches of the knowledge and information blocks (IG, KG) connected via pointers, and
Display of a pointer as soon as a pointer exceeds a specified weight threshold.

Description

The The invention relates to a method for generating correlations and / or interactions and / or knowledge from a variety of searched records and a system that works according to this method. The invention also concerns a data medium, is the component of such an executable system and on the the program to carry out of the method is stored.

There are already a variety of publications known that deal with the automatic capture of data from documents. Thus, a search was carried out for the present invention in advance at the Austrian Patent Office under the file number R22 / 2002. As a technological background to the present invention, the following documents were found as prior art: DE 198 43 450 A1 ; US 5,297,039 ; WO01 / 86499 A2; US 5,446,891 ; WO 00/72256 A2 and EP 0 987 639 A1 ,

technological background

The varied and mostly incompatible data collections The modern information society is now posing a serious one Problem regarding of information processing and knowledge generation huge amounts of data stored electronically, the extraction of cohere and knowledge, however, is still for the most part work undertaken by humans must become.

Indeed the amounts of data are now so extensive that too. People often overwhelmed are to generate knowledge from it. It is therefore imperative Procedures to work out heterogeneous collections of data automatically generate correlations and knowledge.

In particular, two aspects of such a method are already out of the EP 0 987 639 A1 or WO 00/72256 A2.

From the EP 0 987 639 A1 It is known to automatically generate correlations between documents linked by hypertext links. This is done using a linkage table which records whether a given second document has already been requested from a given first document. In addition, the table may provide information on how often, starting from a given first document, the relevant second document has been requested. By doing so, one or more users automatically determine the correlation between the documents during use of the system.

It However, it should be underlined that in this system always the requirement a corresponding document by a user is necessary. Independently of which in the form of correlation existing knowledge is not used, a corresponding unsolicited information to the user to do.

Of the WO 00/72256 A2 is opposite another approach to classifying information refer to. This document is based on a neural network for computer-aided knowledge management, that on a neural in the memory area of a computer Network based. This neural network forms a system artificial Intelligence by thinking about it An underlying knowledge base in the form of computationally extends readable textual content. To manage and analyze the Sets of knowledge consists of the neural network of interrelated and weighted elements, which by their special properties as over Interaction pairs defined subsets form a Hilbert space and each of which is assigned a coordinate sector. By a selected accordingly Convergence property of associative data structure in a variety statistically independent Inputs by a user sink in the neural network formed Expert system redundancy, and it comes to a classification of knowledge.

Also However, with this approach it is not intended that independently User-generated knowledge would be brought to the knowledge.

Of the The present invention is therefore based on the object, a method indicate that from heterogeneous collections of data equal to which Type (eg sequences, documents, publications, patent applications or the like) automates correlations and / or interactions and / or Generates knowledge and displays this to the user without request. additionally supposed to be a system and a disk to carry out be specified in this method.

solution

These Tasks are for the method by the features of claim 1, for the system by the features of claim 21 and for the disk by the features of claim 22 solved.

further developments The invention are specified in the subclaims.

The Here presented method can largely independently and independently Find, rate, and catalog relationships from current user queries.

The inventive method assumes that information and knowledge are composed of basic blocks is. The following are these blocks Information Grain (IG) and Knowledge Grain (KG). Between these grains there are relationship interactions, which are called vectorial relations or pointers. The inventive method allows that these pointers are created automatically. Subsequently, will the topology of the resulting network by relationships The Grains searched each other and evaluated these associations. These associations are prepared in an appropriate manner User available posed.

There with growing Information networks very quickly the values to be calculated so be numerous, that the current computer for it too slowly would, can considered in a development of the invention limited subnetworks.

There the present method generally has no other goal than constantly information to relate and to weight these relationships and to document, the method also has no classic endpoint. Once triggered it works as long as the physical calculator supplies energy becomes. Queries to the procedure always bring only a current Knowledge as output result. It's like organic Information processing, which does not have a defined endpoint knows.

The Procedure allows it, on its own Results are communicated without the procedure being explicit afterwards was asked. The link The network elements through pointers allow it at any time every result is logically justified be attributed to the basic information and can be documented.

One Another advantage of the method is any information structures integrate and link together. The integrable information structures include, for example, relational and / or object-oriented databases, Texts of any kind (publications, Reports, summaries, articles), tables, sound documents, etc.

By a question and answer dialog between the user and the system according to the invention Is it possible, complex questions that can not (yet) be answered to decompose, to specify and to explain in less complex sub-questions to give the answer to the procedure in the Location is.

One important aspect of the present method is the possibility Feedback on the accuracy and valence of the found and issued answers and associations to process. This means that associations between the network elements through the feedback the user strengthened or weakened can be.

detailed description

- Definitions

Seed Table is a table of the Knowledge Grain, the terms, words and related terms and words contains which characterize the relevant field of knowledge or a functional one Own or produce connection with the relevant field of knowledge. The seed tables of different knowledge areas can contain common elements but differ from each other in at least one element. In This table will be a priori terms, words and related terms and words about that entered domain of knowledge. Furthermore, the total number the terms stored in the seed table also in the Saved table.

Growth Table is a table of the Knowledge Grain, in the terms, words and related terms and automatically include words that characterize the area of knowledge or that have or are functionally related to the area of knowledge. The growth tables of different knowledge areas can contain common elements. For each term stored in the growth table, a number is stored that indicates how many comparisons of the growth table were made with thesaurus tables of various information grains for which the term was not found in the thesaurus table. Furthermore, the total number of terms stored in the Growth Table is also stored in the table.

thesaurus table is a table of information Grain containing all terms, words and related terms and words of a record. A record can include a document, a document, a file, a result log, an image file, a sound recording, a Be part information of a database etc.

Storage Pointer is a saved reference to the source filing the information concerned. This can be, for example, a file name, a document tray, be a picture tray etc.

information Pointer is a bidirectional, weighted reference between one Information Grain and a Knowledge Grain. The weight of the A reference direction may not be equal to the weighting the other direction of reference.

Knowledge Pointer is a bidirectional, weighted reference between two Knowledge Grains. The weighting of a reference direction can but not equal to the weighting of the other direction of reference be.

information Pointer Table is a table of information Grain, the information Contains pointers.

Knowledge Pointer Table is a table of the Knowledge Grain, the Knowledge Contains pointers.

information Pointer Strength (IPS) is the measure of correspondence between one Information Grain and a Knowledge Grain. Like the value of the match will be described later in the text.

Knowledge Pointer Strength (KPS) is the measure of agreement between two Knowledge Grains. How to find the value of the match will be described later in the text.

Threshold parameter are limits where an action is taken when they are reached. Generally it should be noted that these parameters are in a preferred embodiment of the method both globally, that is to say for all subelements, as well as local, that is only a subelement can be set regarding.

pointer Strength Threshold is a limit to the degree of agreement. Thus an information pointer or a Knowledge Pointer is created, the value of the match must match at least reach the Pointer Strength Threshold.

Growth Storage Threshold is a limit to the inclusion of new terms into the Growth Table. Reached the weighting of an information pointer At least the growth storage threshold of the relevant knowledge Grain, they are the most common occurring in the relevant thesaurus table of the Information Grain Terms added to the Growth Table of the Knowledge Grain.

Growth Deletion Threshold is a limit at which the respective one is reached Term is deleted from the Growth Table of the considered Knowledge Grain.

Information Grain (IG) is a data structure; which preferably comprises at least the following constituents:
Name, Information Type, Storage Pointer, Thesaurus Table, Information Pointer Table

In 1 is an example of an Information Grain (IG) shown. This is the preface or abstract of an article entitled "New roles for structure in biology and drug discovery" from the journal "Nature Structural Genomics", Supplement Vol. 7, 2000, pages 928-930. In the Thesaurus Table N = 14 terms are stored. The term "structure" occurs three times in the document, which is published by the "(3)" after the word "structure". The total number of words in the source info The information unit IG has an information pointer which refers to the knowledge grain "protein structure" and has the weighting 118.

Knowledge Grain (KG) is a data structure that preferably comprises at least the following components:
Knowledge Pointer Table, Knowledge Pointer Table, Information Pointer Strength Threshold, Knowledge Pointer Strength Threshold, Growth Storage Threshold, Growth Deletion Threshold.

In 2 is an example of a Knowledge Grain (KG) shown. The total number of terms stored in the Seed Table and the Growth Table is K = 9. The Knowledge Grain (KG) has an Information Pointer that refers to the Information Grain "New roles for structure in biology and drug discovery" and has a weight of 118. It also has a Knowledge Pointer that points to the Knowledge Grain called "X-ray Cristallography" and has a weight of 250. Threshold values are listed as examples only. They are not complete.

information and association network (I + A network) denotes the network, Consisting of Information Grains and Knowledge Grains through information pointer and knowledge pointer.

Assoziationskette denotes a linear connection created by pointers between two grains of information and Association network. The nodes of the association chain are formed by grains.

General course of the inventive method

a) Initialization

- Creation of the initials Knowledge Grains

For the particular knowledge areas in question, a number of initial Knowledge Grains are created by humans. An example of a Knowledge Grain is in 2 given. The finer and more sharply defined the Knowledge Grains are, the more accurate and detailed is the association determination and the knowledge assignment and generation. To create a Knowledge Grain, the Seed Table must be created.

In a preferred embodiment can The seed tables of Knowldedge Grains are automatically generated. Here are well-known methods for the production of subject-specific Word lists and thesauri are used.

Of the Seed Table is one of the central tables of the Knowledge Grain. In it all terms, words and term combinations that cover the relevant knowledge area characterize or used as terms in this area become. The more accurate and appropriate the choice of terms is, the more accurate and sharper can that process according to the invention later Calculate history associations.

Farther are the name of the Grain, the Knowledge Type and the various Threshold values established. These initial values are entered based on experience. The inventive method is able to independently set these values in the course of information processing change. Also, new knowledge grains can independently by the process in the course of information processing new to be created.

Creation of an initial Global Thesaurus

to Simplification and standardization of information storage An initial global thesaurus is created. It can too for reasons the practical implementation for each Knowledge area separate thesauri are created. These initials Thesauri will be during the duration of the procedure constantly extended and checked.

Creation of an initial Word Exclusion List

To avoid filler words (eg in, at, where, who, etc.) in the Globalthesaurus and in the Thesau at least one word exclusion list is created, which is also continually expanded and revised during the runtime of the process.

b) information recording

Information Grain Creation - First phase

For each information unit to be processed (document, result file, data set, etc.), the method according to the invention generates information Grain with the elements described above. An example information Grain shows 1 ,

Centerpiece of information Grain is the thesaurus table, in which every term, every word is entered, which occurs in the information unit. In addition to the term becomes the number, how frequent this term occurs in the information unit, with stored. Furthermore, the thesaurus table notes how many different ones Total terms are stored in it and how many words or words Conceptual elements has the considered information unit in total. Before entering a term in the thesaurus table is the Term with the entries compared in the Global Thesaurus for a uniform writing and presentation of the terms can be guaranteed. comes a term not yet in the global thesaurus, it is in it added.

Of the Storage Pointer refers to the storage location of the information unit. This can be both a physical filing (z. Image) as well as a virtual file (eg file name, database entry) be.

Of the Information Type is automatically made from a list of possible Types assigned according to specified criteria.

Information Grain Creation - Second phase

After this for all currently to be processed information units the first phase The Information Grain creation was done in the second Phase the Information Pointer Table of the Information Grains and the Calculated Knowledge Grains.

For this purpose, the terms of the thesaurus table of an Information Grain are successively compared with the terms in the Seed Tables and the Growth Tables of all Knowledge Grains (see also 3 "Calculation of the Information Pointer Strength"). Terms can also be compound words (eg macro-molecular, high resolution, bridge pier). The Information Pointer Strength (IPS) is calculated.

In a preferred embodiment, the following formula is used for this purpose:

B

= (Begriff 1, Begriff 2, Begriff 3, ..., Begriff N); B ist die Gesamtheit der zwischen dem jeweils betrachteten Information Grain (gespeichert in Thesaurustabelle) und dem jeweils betrachteten Knowledge Grain (gespeichert in Seed Table und Growth Table) übereinstimmenden Begriffe

N

= Anzahl der Begriffe in der Thesaurustabelle des Information Grain

W

= Gesamtanzahl der Wörter/Begriffe in der Informationseinheit, und

K

= Gesamtanzahl der Begriffe in Seed Table und Growth Table des Knowledge Grain sind.

In which:

B

= (Term 1, term 2, term 3, ..., term N); B is the totality of the terms that match between the respective considered information Grain (stored in thesaurus table) and the considered Knowledge Grain (stored in Seed Table and Growth Table)

N

= Number of terms in the thesaurus table of the Information Grain

W

= Total number of words / terms in the information unit, and

K

= Total number of terms in Seed Table and Growth Table of the Knowledge Grain.

The factor 10 ^x only serves to transform the numbers into a more easily handled range. The modulo operation uses only the integer part of the calculated value as the final result. X is experimentally determined in the course of the trial of the method. The preferred initial value of x is 5.

In 3 an example for computing the Information Pointer Strength (IPS) is shown. The serve in 1 and 2 , shown Grains as a basis for calculation. The left side shows a section of an Information Grain (IG), the right side shows a section of a Knowledge Grain (KG). For the calculation of the Information Pointer Strength (IPS), the number of matching terms between Thesaurus Table on the one hand and Seed Table and Growth Table on the other hand are determined. In the thesaurus table, the terms additionally store how frequently these occur in the source information unit. In the example, the term "structures" occurs three times in the source information. In addition, the word "structure" still occurs in the compound term "structure determination", so that the total number of occurrences of the matching term "structure (s)" lies between the considered grains at four. If no number is stored, the term occurs only once. The calculation is done according to the above formula, with the values of the variables N, W and K taken from the grains.

After computing the Information Pointer Strength (IPS), this value is compared with the Information Pointer Strength Threshold of the Knowledge Grain in question. If the IPS is at least as high as the specified threshold, then the affiliation of the information unit to this knowledge area, represented by the knowledge grain, is established. An information pointer is saved (see also 3 "Calculation of the information pointer").

The However, this embodiment of Pointers is not preferred necessarily bidirectional and weighted. This means he points from the considered information grain to the considered knowledge grain and vice versa. A weighting is preferred with the pointer stored for both directions may be the same but also different. In In the initial phase, the weighting becomes identical for both directions be. The weight value is the Pointer Strength information used.

In 4a an example of this is shown. There are five Information Grains IG 1 to IG 5 and four Knowledge Grains KG A to KG D in the network.

From this, according to the above and in 3 "Calculation of the Information Pointer Strength (IPS)" computational rule described the information pointer of the Knowledge Grain "KG A" calculated and stored, as in 4b illustrated.

As in 4c The further information pointers are calculated graphically and stored. The weighting of the respective pointers is illustrated by the numerical values at the pointers.

A possible embodiment of the method may not be the terms of the thesaurus tables to compare all tables of all Knowledge Grains, but only with a limited Circle of Knowledge Grains. This can be achieved, for example become that only the most prominent terms of the thesaurus table with the names of the knowledge grains or with the terms of the seed tables The knowledge grains are checked for similarity. If you reach with this shortened review one Minimum match, this is the complete comparison of the concepts described above performed between these grains.

When Result of the calculation of the information pointer described above After that, the information grains have an information pointer Table, whose weighted pointer to the associated Knowledge Grains refer. In turn, the Knowledge Grains will also be included an information pointer table whose weighted pointers to the Associated Information Grains Direct. This is a weighted one Relationship between the information unit and a knowledge area produced. Of course is it possible that from one information unit relationships to several different ones Be prepared knowledge areas. Just as natural Relationships from one knowledge area to several information units possible.

c) Generation of basic knowledge relationships

Inclusion of new Terms in the Growth Table of a Knowledge Grain

exceeds considering the information correlation between information Grain and a Knowledge Grain of Information Pointer Strength (IPS) the Growth Storage Threshold of the Knowledge Grain, then the most common in the thesaurus table of the Information Grain occurring terms in the Growth Table of the Knowledge Grain. The calculation of A is described below in the text. Come several Terms equally common before and can not all are included in the Growth Table, so vote Random generator the items to be included among the most common out.

The number (A) of the terms to be included depends in a preferred embodiment on the following formula: A = ln (N 2 / W)

N

= Anzahl der Begriffe in der Thesaurustabelle des Information Grain und

W

= Gesamtanzahl der Wörter/Begriffe in der Informationseinheit ist.

In which:

N

= Number of terms in the thesaurus table of the Information Grain and

W

= Total number of words / terms in the information unit.

Deletion of terms from the Growth Table of a Knowledge Grain

P

= Gesamtanzahl der Information Grains und

Q

= Gesamtanzahl der Knowledge Grains ist.

If a term of the growth table is not found in at least one of the thesaurus tables within a certain number of comparisons with thesaurus tables of different information grains, it is deleted from the growth table of the knowledge grain. The number (V) of the comparisons is determined in the Growth Deletion Threshold and, in a preferred embodiment, is based on the following formula: V = P / Q in which

P

= Total number of information grains and

Q

= Total Number of Knowledge Grains.

Calculation of the Knowledge Pointer, method 1

The Knowledge Pointers are an essential part of knowledge correlations and represent the relationships between Knowledge Grains. They are bidirectional, weighted references, which in a preferred embodiment are calculated as follows (see also FIG 5a and 5b "Calculation of Knowledge Pointers").

outgoing from a Knowledge Grain, the procedure looks at this Output Knowledge Grain belonging Information Pointer (IP) in succession. Starting with the first Information Pointer is in the associated information Grain the Presence of further information Pointer checked. If at least one more Pointer, yes to another Knowledge Grain, the target Knowledge If Grain points, the Knowledge Pointer becomes Strength for this Calculated association chain. The calculation is preferred after following calculation rule.

The Sum of the weights of the involved information pointer divided by the number of nodes passed minus one gives the Knowledge Pointer Strength (Each Information Grain and each Knowledge Grain count as one node each, starting grain and target grain count).

Corresponding This calculation rule will be all possible connections via information Pointer calculated between the two considered Knowledge Grains.

If the highest calculated Knowledge Pointer Strength is at least as high as each of the Knowledge Pointer Strength Threshold values involved Knowledge Grains, so it is an association between the involved Knowledge Grains detected. A knowledge pointer is stored. The design of this knowledge pointer is preferably bidirectional and weighted. This means that it points to the considered source knowledge Grain to the considered target knowledge grain and vice versa. Together the pointer stores a weighting for both Directions can be the same but also different. In the initial phase is the weighting for be identical in both directions. The weighting value will be the highest calculated Knowledge Pointer Strength used.

The Knowledge Pointer Strength calculation described above is performed for each Knowledge Grain. As a result of these calculations, the Knowledge Grains then have a Knowledge Pointer Table whose weighted pointers point to the connected Knowledge Grains. This creates weighted relationships between knowledge areas. Of course, it is possible for a Knowledge Grain to establish relationships with several different areas of knowledge. It is equally possible that knowledge domains are not associated via Knowledge Poin own each other.

The following example illustrates the calculation of a Knowledge Pointer:

outgoing from the Knowledge Grain (KG) A leads the Information Pointer (IP) to the Information Grain (IG) 1. From there introduces Another pointer to the Knowledge Grain (KG) B. Therefore, now a Knowledge Pointer (KP), KG A <-> KG B, in both cases Knowledge Grains saved the weight (200 + 100) / (3 - 1) = 150 gets.

According to the above All possible chains of association are described calculated between two Knowledge Grains. The Knowledge Pointer with the highest Weighting is stored in the involved knowledge grains, if the weight of the pointer is at least the Knowledge Pointer Strength Thresholds of participating Knowledge Grains reached. Subsequently, will the same procedure with all other Knowledge Grains.

According to 5a After completion of the calculation of the information pointer in accordance with the above-described calculation rule, the knowledge pointers between the knowledge grains "KG C" and "KG D" are calculated in a first step: KG C <-> IG 3 <-> KG D = 200 + 400 = 600; 600: 2 = 300. There are other possibilities to calculate a Knowledge Pointer between these two Knowledge Grains (eg KG C <-> IG 2 <-> KG A <-> IG 1 <-> KG D; 200 + 100) + 200 + 50 = 550; 550: 4 = 137.5), but the association chain KG C <-> IG 3 <-> KG D achieves the highest Knowledge Pointer Strength.

Finally, all information pointers and knowledge pointers of the information and association network are calculated. In 5b is shown this ready calculated network.

From a certain size of the network It may not be necessary to calculate all Knowledge Pointers anymore allow.

P

= Gesamtanzahl der Information Grains

Q

= Gesamtanzahl der Knowledge Grains

y

wird experimentell im Zuge der Erprobung des Verfahrens bestimmt.

In a preferred embodiment, the method is then limited to tracking only the pointers having the highest J weights, the J being set as a global variable or in tabular form, or independent of the method, depending on the number of existing ones Node is calculated in a preferred embodiment according to the following formula: V = Q / P × 10 y in which:

P

= Total number of information grains

Q

= Total number of Knowledge Grains

y

is determined experimentally in the course of testing the process.

One potential preferred initial value of y is 5.

C

= Konstante; zum Beispiel 3000

R

= Anzahl der vorhandenen Knoten

In a further preferred embodiment, the search depth (S) is limited to a maximum number of nodes. The number of nodes to be searched is determined either in a global variable or calculated independently by the method, depending on the number of existing nodes, for example with the following formula: S = C × e -Log R in which:

C

= Constant; for example 3000

R

= Number of existing nodes

A further preferred embodiment uses a hybrid form of both aforementioned embodiments.

Calculation of the Knowledge Pointer, method 2

A further preferred embodiment The calculation of the Knowledge Pointer is characterized as follows.

It will be made in accordance with the "Information Grain Creation - Second Phase "described Process the Seed and Growth Tables of a Knowledge Grain one after the other compared with the Seed and Growth Tables of all other Knowledge Grains. This will be for every knowledge grain performed. The method calculates the respective Knowledge Pointer Strength (KPS).

To The calculation of the Knowledge Pointer Strength (KPS) becomes this Value with the Knowledge Pointer Strength Threshold of the respective Knowledge Grains compared. Is the KPS at least as high as the threshold, so it is a correlation of these fields of knowledge, represents through the two Knowledge Grains, noted. It becomes a knowledge Pointer saved. The design of this knowledge pointer is preferably bidirectional and weighted. This means he is pointing looked at from the first considered knowledge grain to the second Knowledge Grain and vice versa. Together with the pointer becomes a weighting stored for both directions may be the same but also different. In In the initial phase, the weighting becomes identical for both directions be. The weight value used is the Knowledge Pointer Strength.

d) association finding, Presentation of results and interaction with users

As a general rule applies to all in this section "Association finding, Presentation of results and interaction with users ", that a "holographic Hierarchy "the considered knowledge Grains an essential feature of the method according to the invention is. The term "holographic Hierarchy "should following property of the information generated by the method according to the invention and association network (I + A network).

Of the Entry point into the I + A network happens in a preferred Shape over a knowledge grain. This grain is for this request by definition the hierarchically highest. The Knowledge Pointers refer to this grain now on related knowledge areas, sub-areas or higher-level ones Knowledge structures. All of these related Knowledge Grains are however, subordinated and hierarchical in consideration of this request, on the Knowledge Pointer, connected to the entry point and allow thereby an iterative question-answer dialogue with the user leading to a delimitation of the origin question and to more and more precise Answers leads.

provides Now you have a new request to the I + A network, so now the new entry point is the hierarchically highest knowledge grain. This means that the hierarchy of the network with the asked Question changed. Since this is not a static hierarchy and the entry point can be at least any knowledge grain and it can be always the same as the network, this hierarchy is called "holographic".

It is possible, the entry point into the network, e.g. the first one asked Question on any Knowledge Grain to choose. In this point differs The present method is fundamentally different from the so-called neural Networks.

Representation of the direct Knowledge Relations (KG - KG)

In a preferred embodiment, all knowledge relationships represented by the Knowledge Pointers between Knowledge Grains that exceed a Minimal Association Score (MAS) can be presented to the user as a result. The association score is basically calculated by dividing the sum of the weights of the pointers passed in the considered association chain by the number of pointers passed through. The limit value MAS can be selected both by the user directly and by the program as a function of various other parameters, such as, for example, the computing power of the hardware used, the number of available knowledge grains and information grains. The presentation of the results can be done in all technically feasible ways, for example by output in writing, printed or on a screen, graphically, spoken, as a file etc.

In a further preferred embodiment the weighting of each of the considered pointers must be one Exceed minimum pointer score (MPS), so that the association is included in the results presentation can be. This means that all pointers are an information chain exceed this MPS have to. The value for The MPS can be set as a global variable or depending on be considered by the Knowledge Grain or by the procedure independently in dependence be calculated by other process parameters.

Calculation and presentation indirect information relations (KG - IG - KG - IG; KG - KG - IG)

indirect Information relations are associations between a knowledge Grain and not directly over an information pointer associated with information Grain. The Connection between the observed knowledge grain and the information Grain is running over it at least one intermediate node.

to Calculation of the information chain of these indirect relationships is calculated the method is the weighting of all relations between a knowledge Grain and all information Grains. This is the sum of the pointer weights between the starting point and the end point of the information chain divided by the number of used pointers. This will be for all knowledge Grains performed. Subsequently can the associations that exceed a weighting value M, the user be presented as a result. The assumption of the weighting value M and the type of result representation is the same as under "Presentation of direct knowledge relationships (KG - KG) ".

at The search depth S can be determined as described under point "Calculation of the Knowledge Pointer, Method 1 "be limited.

In a further preferred embodiment Every single one of the considered pointers has to have a minimal pointer Exceed score (MPS), so that the association is included in the results presentation can be. This means that all pointers are an information chain exceed this MPS have to. The value for The MPS can be set as a global variable or depending on be considered by the Knowledge Grain or by the procedure independently in dependence be calculated by other process parameters.

processing direct user requests

direct User requests to the system according to the invention are possible. The User request, preferably in natural language form, is preferred edited as follows.

Under Exploitation of standard database techniques will be the terms of Request with a) the names of the Knowledge Grains, b) with the Terms of the Seed Tables and Growth Tables of the Knowledge Grains and c) the source information contents of the information grains compared. This query process may preferably hierarchically according to the above Sequence, i. if there are knowledge grains whose Designation match with terms of the request, so form these Knowledge Grains and the over the pointers stored in it are linked to the Knowledge Grains Association chains the limited, secondary Search space for the user request.

results there is no sufficient agreement with terms of the Knowledge Grains, the search space of the Request on the terms of Seed and Growth Tables of all Knowledge Grains expanded. It will be the Knowledge Grains and the associated Information Grains, which are the N highest matches between request and terms of seed and growth tables have displayed as result. The number N can be set as a global variable, enter by the user or by the AAA method itself depending calculated from other process parameters.

results If there is no satisfactory answer, then the search space will be extended to all source information units. The chains of association, which are connected to the source information units that the N highest matches with the request are displayed as the result. Also here can set the number N as a global variable, to be entered by the user or from the system according to the invention even in dependence calculated from other process parameters.

These Hierarchical search may be in another preferred embodiment with the consideration the pointer weights be combined. The search is preferred to such network elements continued the above highly weighted pointers connected to the output network element are. In each search step, the N are the highest weighted Pointer, where the number N is set as a global variable, Enter by the user or the system according to the invention itself depending can be calculated from other process parameters.

Question-answer dialogue

A typical, preferred interaction between user and system according to the invention happens over a question-answer dialog. The user asks a question that System generates a preferably natural language query, which according to the under "processing direct User Requests " hierarchical order to the information and association network (I + A network). Here are the intermediate results issued as partial answers, with a question for further Delimitation of the original question of the user is connected. Of the User can have one or more of the terms listed in the partial answer use as a delimitation or introduce new, additional query terms. With the limited question will continue through the hierarchical search path. After one or more constraints, the process becomes one limited number of knowledge grains have arrived. Now follows the limitation at the information pointer level. The inventive method search with the limited question in the thesaurus tables Information Pointer related information Grains and provides the Results again as a partial answer to the user. This in turn can further narrow the scope of the question.

This Question-answer dialog leads iteratively to a refinement and specification of the original question. Generally speaking this dialogue allows for (at the moment) unanswerable questions Interaction with the method according to the invention in answerable Sub-questions are decomposed and the answers to these sub-questions then be given.

e) plasticity of the information and association network

Autonomous generation of Knowledge Grains

In a preferred embodiment is the inventive method and system able to generate Knowledge Grains automatically.

Receives an information Grain no information pointer to the existing Knowledge Grains, Thus, the inventive method generates or system automatically a new Knowledge Grain consisting of the essential components of the information Grain and calculated the information pointer between these two network elements. As essential components become the N-most frequent terms of the Thesaurus table of the Information Grain into the new seed table of the Knowledge Grain, where the number N is set as a global variable, entered by the user or depending on the system itself can be calculated from other process parameters.

In a further preferred embodiment, the inventive method or system can split existing knowledge grains into new knowledge grains. This is done when a certain number N of Knowledge Pointers and / or Information Pointers are exceeded on a Knowledge Grain. The number N can be defined as a global variable, entered by the user or calculated by the method or system according to the invention itself as a function of other method parameters. Then, for example, for each term that appears in the Seed Table or the Growth Table, it can be determined to what extent and what information Grains a match with a notion of their thesaurus tables. A cluster analysis of the term information Grain Relations is performed and the original Knowledge Grain is split into new Knowledge Grains according to the clusters. The original Knowledge Grain receives new Knowledge Pointers that reference the Grains it has spawned, and the Seed Table Terms are the names of the new Knowledge Grains. Be as a name used the terms with the highest relational values of cluster analysis. Finally, the information pointers of the new grains are recalculated.

4.5.2 Autonomous deletion of Knowledge Grains

In a further preferred embodiment Is it possible, Also to delete Knowledge Grains. This essentially means the reversal of the previously described splitting. This may be necessary if the number N is that of the one considered Knowledge Grain Outgoing Information Pointer and / or Knowledge Pointer falls below a minimum value. The number N can be global Set variable, enter by the user or by the method according to the invention even in dependence can be calculated from other process parameters. The terms the Seed Table and Growth Table of the grain to be deleted are in the Growth Table of the Knowledge Grains, which is the highest weighted Knowledge Pointer to this Grain owns. The information pointer new ones are calculated.

processing from feed back from outside the method according to the invention

One An essential aspect of the method according to the invention is the possibility feedback from outside take. feedback strengthen or weaken connections between grains.

To are in a preferred embodiment all pointers of a considered association chain according to the feedback from outside the method according to the invention with a bonus or penalty. Bonus and penalty can be one To be a factor set by the user as a global variable or from the method according to the invention even in dependence can be calculated from other process parameters. It can too Absolute values are used as bonus and penalty, too as previously described are generated internally by the user or process can. Bonus and penalty can for information Pointer and Knowledge Pointer but also be different. You can also in dependence of a possible value (see chapter "Possible Process refinement ") of the grain. Bonus and penalty values can with every feedback changed become. Are pointers of an association chain multiply positive feedback strengthened, this is what is referred to as "routing".

In a preferred embodiment the bonus and penalty values are separated with the weight of a Pointers stored, so that better traceability of the Association weighting guaranteed is.

It It should be noted that a new information of a already by the inventive method used database also be a feed back from the outside can (update a database).

With the user feedback "matures" once in progress set inventive system. It will be in a later Ripe state possible be that the process produces concepts, hypotheses, and models by showing association chains with high weighting factors become. These hypotheses can with the help of the feedback from outside the system according to the invention checked and evaluated become. Furthermore you can this contradictions be shown in existing hypotheses and models.

f) process details

One direct access to and search in source information respectively Information Grains is possible. This is ensured by standard database procedures.

at the presentation of results Data derived from trusted sources, identified as facts being represented. For other results, for example Can be based on association chains Probability values, which estimate the correctness of the result, with be issued. These probabilities can change in a preferred embodiment calculated using the weights of the pointers. Adds to that for example, the highest in the overall network occurring pointer weighting and assumes that this weighting represents a 95% correctness. The correctness value of the considered pointer is now calculated in relation to it.

A further preferred embodiment is over the feedback from outside of the system given an empirical value for the correctness Build pointer weights.

The method may have a correlation between positive feedback (correct answer) and pointer calculate weighting and use it to determine the probabilities of correctness of future answers.

By The link all members of an association chain via pointers is intrinsic Documentation guaranteed. The reasoning for the Showing an association are traceable at any time for each user, over there the pointers traced the chain of association both forward and backward can be.

Possible process refinements Grain quality factor

In a preferred embodiment is for Each grain introduced a weighting factor, the importance and Indicates correctness of the information represented by the grain. This weighting factor flows in the calculation of the association chain weighting. A For example, preferred calculation of this weighting factor by adding all the weights of the information coming from that grain Pointer and subsequent Division done by the number of information pointers. At Knowledge Grains will use both this quotient for the information pointer also the quotient for calculates the knowledge pointer. The sum of the weighting factors of a knowledge grain either be evaluated separately or it will be the mean of the formed two quotients.

W

= Wertigkeitsfaktor des betrachteten Grains

N

= Anzahl der Pointer des betrachteten Grains

In a further preferred embodiment, the number of pointers is used to form the weighting factor of the grain. The following formula is preferably used: W = ln N in which:

W

= Weighting factor of the considered grain

N

= Number of pointers of the considered grain

Distance weighted

In a preferred embodiment a distance weighting of the associations is calculated. This should consider, that chains of association over The more Grains, the more unlikely they become are connected in this chain, the distance weighting as set global factor, enter by the user or the system according to the invention or process itself depending can be calculated from other process parameters.

5.3 Analogies

In a preferred embodiment can the procedure in a later Maturity automatically perform analogies by: it internally compares chains of associations. This can be for Example by tracing of Conceptual Affinities of Different Seed Tables and Growth Tables of Knowledge Grains happen or by comparison of all chains of information emanating from an information grain.

5.4 "Mutation" and "Creativity"

In a preferred embodiment can selected by random number generators weak or nonexistent Knowled ge pointers reinforced or be generated. this leads to to new chains of association. These chains will be through the feedback from outside the method according to the invention selected for meaningful and meaningless new associations. A Further selection can be done by matching with the source information respectively. Another mutation principle may be the change of the weighting factors of the Knowledge Grains. Of decisive Importance of all mutations is checking for correctness feedback from outside the erfidnungsgemäßen method or system.

Possible forms of realization

The inventive method can be stored as software on optical and / or magnetic disk is to be realized on computers.

The Procedures can be used both on local computers as well as on networked ones Computers are realized. The networked computers can also be accessed via intra- or Internet connected.

In In another preferred embodiment, the algorithms for generating the pointer and / or the algorithms for the calculation of the association chain weighting and / or the algorithms in an application specific chip (ASIC) can be realized. This ASIC will be on an expansion board introduced into the computer.

A Another preferred embodiment is the realization in autonomous operating machines (robots). This robot is then in the Location, independent to make associative conclusions.

application areas

One preferred application for the inventive method is the information processing in the field of molecular Biology, including Biochemistry, genetics, genomics, proteomics, etc. Here are the to disposal stationary data collections and databases, including the Publications and Research reports included as origin information in the process. Especially in this field, the increase of information is overexponient runs, it becomes inevitable be, the processing of this information and the generation of new knowledge to automate this information. The present Method is capable of doing so.

Of course you can the present method applies equally to all scientific fields of research be applied.

One Special case of scientific-technical fields of application exists in the possibility with the help of the presented method simulations complex networked To carry out the procedure.

One Another application of the presented method is in company-wide information processing. Large companies have a not to be overlooked Variety of individual information, data collections, databases, personal Knowledge etc. As source information in the presented system or procedure, it becomes possible from this information To generate knowledge, or new associations and relations show.

One Special case of enterprise-wide application is the use in Procedures for handling technical customer requests ("technical hotline"). The procedure allows in the initial phase, the automated construction of a response library and in a later one Phase at least partially fully automatic answering the Requests. This information about the products to be supported filed as information Grain and generate the initial answers Knowledge Grains. After a short time, there is an association network built that a targeted documentation of requests and after all allows an automated answer.

In 6 Two examples of requests to the system according to the invention (example 1 and example 2) are shown. In the present examples, English is provided as language. However, it could as well be communicated German or French. In this context, it should also be mentioned that the system or method according to the invention can also automatically translate the recorded document into a specific language when automatically creating thesaurus tables. So if a German-language document is detected, the individual terms - preferably freed from filler words - can be stored translated into English language.

Claims (25)

Method for generating correlations and / or interactions and / or knowledge from a plurality of searched data sets, characterized by the following method steps: manual generation of a knowledge block (KG) with an individual provision of term tables, which concepts and / or words for the characterization of a given Knowledge area, and machine-generating an information block (IG) from data sets and electrical reading units with automated provision of thesaurus tables having terms and / or words of a searched data set, self-generating vector references (pointers), by which references between like and / or different blocks of knowledge and information blocks (IG, IG, KG, KG, IG, KG), weighting of these pointers in accordance with matches of the knowledge and information blocks (IG, KG) connected via pointers, and display of a pointer as soon as possible a pointer exceeds a given weight threshold. Method according to claim 1, characterized in that that the weighting of the pointers is influenced manually. Method according to claim 1 or 2, characterized that for initialization of the method continues the name and type of the knowledge block (KG) and the thresholds are manually specified. Method according to one of claims 1 to 3, characterized that for initialization of the method an initial, changeable Thesaurus table is specified. Method according to one of claims 1 to 4, characterized that when creating the thesaurus tables, each word of a searched Record is evaluated. Method according to claim 5, characterized in that that in the search of the records predeterminable Füllwörter hides become. Method according to one of claims 1 to 6, characterized that in addition for each word of a thesaurus table a number is stored the frequency this occurring in the respective searched record word indicates. Method according to one of claims 1 to 7, characterized that words the thesaurus tables with other thesaurus tables and the knowledge blocks (KG) with regard to matching words be compared and as required of matches the pointers are evaluated. A method according to claim 8, characterized in that the calculation of the weighting (IPS) takes place according to the following formula

where: B = (term 1, term 2, term 3, ..., term N); B is the totality of the terms between the respective considered information block (stored in the thesaurus table) and the respectively considered knowledge block (stored in the seed table and growth table), N = number of terms in the thesaurus table of the information block, W = total number of words / terms in the information block, and K = total number of terms in the term table and growth table of the knowledge block. Method according to claim 8 or 9, characterized that all Thesaurus tables are compared. Method according to claim 8 or 9, characterized that only prescribed thesaurus tables were compared become. Method according to claim 8 or 9, characterized that all words and / or terms of thesaurus tables subjected to comparison become. Method according to claim 8 or 9, characterized that only given words and / or terms of thesaurus tables subjected to comparison become. Method according to one of claims 1 to 13, characterized in that a plurality of Wis sensor blocks (KG) is specified in the process. Method according to one of claims 1 to 14, characterized that a knowledge block (KG) contains a growth table into which automatically words and / or terms that contain the relevant knowledge area characterize. Method according to one of claims 1 to 15, characterized that when requesting the method any knowledge block (KG) selected as starting point will and over the pointer connections in a question-answer dialog iteratively responses are generated. Method according to one of claims 1 to 16, characterized that it is in accordance with internal factors at least two blocks of knowledge (KG) automatically to one Knowledge block (KG) can summarize. Method according to one of claims 1 to 17, characterized that it is in accordance with internal factors can change the weighting of pointers. Method according to one of claims 1 to 18, characterized that it is in accordance with external factors change the weighting of pointers can. Method according to one of claims 1 to 19, characterized that the records from data, sequences, documents and / or images are formed. System for generating correlations and / or Interactions and / or knowledge from a variety of searched Records characterized in that it according to the method according to a the claims 1 to 20 works. disk with one on the disk stored program, by which in at least one of claims 1 to 20 claimed method is executable. disk according to claim 22, characterized in that this is an optical or magnetic storage media is. disk according to claim 23, characterized in that the data carrier a Compact Disc, a flash memory or a microdrive is. disk according to one of the claims 22 to 24, characterized in that this data carrier component a system according to claim 21.