(19)
(11)EP 2 595 065 B1

(12)EUROPEAN PATENT SPECIFICATION

(45)Mention of the grant of the patent:
14.08.2019 Bulletin 2019/33

(21)Application number: 11189099.2

(22)Date of filing:  15.11.2011
(51)Int. Cl.: 
G06F 16/2457  (2019.01)
G06F 16/35  (2019.01)
G06F 16/93  (2019.01)
G06Q 30/00  (2012.01)
G06F 16/33  (2019.01)
G06F 16/36  (2019.01)
G06F 16/38  (2019.01)

(54)

Categorizing data sets

Klassifizierung von Datensets

Classement de jeux de données


(84)Designated Contracting States:
AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

(43)Date of publication of application:
22.05.2013 Bulletin 2013/21

(73)Proprietor: Kairos Future Group AB
101 36 Stockholm (SE)

(72)Inventors:
  • Larsson, Tomas
    101 36 Stockholm (SE)
  • Lindgren, Mats
    171 73 Solna (SE)

(74)Representative: Kransell & Wennborg KB 
P.O. Box 27834
115 93 Stockholm
115 93 Stockholm (SE)


(56)References cited: : 
US-A- 6 128 613
US-A1- 2003 130 993
  
      
    Note: Within nine months from the publication of the mention of the grant of the European patent, any person may give notice to the European Patent Office of opposition to the European patent granted. Notice of opposition shall be filed in a written reasoned statement. It shall not be deemed to have been filed until the opposition fee has been paid. (Art. 99(1) European Patent Convention).


    Description

    TECHNICAL FIELD OF THE INVENTION



    [0001] The present invention relates to the categorizing of data sets. The invention more particularly relates to a method, device and computer program product for categorizing data sets obtained from a number of different sources.

    DESCRIPTION OF RELATED ART



    [0002] Data sets such as electronic documents, electronic articles, blogs and data in online discussion forums may exist on a variety of different computers. Such data sets are moreover often publicly provided. With the introduction of Internet it is then possible to access such data sets from virtually any computer. Hence, it is possible to obtain these data sets from a wide range of sources. There is thus a vast amount of information that is available on the Internet.

    [0003] This abundance of information would be interesting to use in a number of different fields, such as in order to determine different social and consumer trends. However, the amount of information is so vast that it is hard to obtain any comprehensive and useful knowledge from it.

    [0004] There is therefore an interest in organizing and categorizing data sets such that advanced information analysis can be performed on the data sets.

    [0005] There exist a number of techniques for organizing, structuring and searching data sets.

    [0006] Various methods of classifying data sets have for instance been proposed. As an example US 2010/0205525 describes automatic classification of a text based on the frequency of occurrence of a qualitative characteristic like a character shingle in the text. US 2009/0094021 describes determining a number of themes from a number of document clusters. US 6094653 describes classification of words into word clusters.

    [0007] US 2003/0130993 and US 6128613 are two more documents that are examples on the categorizing of data sets. US 2003/0130993 describes the use of machine learning algorithms for classification. In US 2003/0130993 a number of topics are first determined and for each topic the documents are processed to determine whether they satisfy threshold criteria and should be assigned to the topic. In US 6128613 words are divided into classes based on their frequency of appearance in a training set.

    [0008] There also exists various ways to group documents. EP 2045739 does for instance describe the selection of words in a document as keywords and clustering of the documents according to keywords to yield clusters, where each cluster corresponds to a topic. US 6078913 describes clustering of selected documents to a hierarchical tree structure. US 7809718 describes the finding of meta data of a document and the importance of words in the documents are emphasized if they also exist in the metadata. US 6778995 describes extracting terms from a document and the building of a concept space over a document collection, identifying terms correlated between documents and populating clusters with documents having differences between document angles and cluster angles in a document space falling within a certain variance. US 7720848 describes a probabilistic clustering system.

    [0009] Furthermore, efforts have been made in the field of search optimization. US 7483892 does for instance describe compiling a term-by-document matrix from documents representative of particular subject matter that represents the frequency of occurrence of each term per document and the creation of a weighted term dictionary. US 5926812 describes grouping of word clusters and combining of similar word clusters for forming a single word cluster.

    [0010] In a master thesis at the Lund University Sweden, Department of Industrial Management and Logistics, Production Management, "Textual Data Mining for Business Intelligence", 2010 Andreas Ek describes how information is obtained using hierarchical clustering, linear regression and probability based ranking. However, there still exists a need for an improvement in this field.

    SUMMARY OF THE INVENTION



    [0011] The present invention is therefore directed towards providing an improved categorization of a collection of data sets.

    [0012] One object of the present invention is to provide a method for categorizing data sets obtained from a number of sources, which simplifies the location of useful information in the data sets, as defined in claim 1.

    [0013] Another object of the present invention is to provide a device for categorizing data sets obtained from a number of sources, which enables the location of useful information in the data sets, as defined in claim 12.

    [0014] Another object of the present invention is to provide a computer program product for categorizing data sets obtained from a number of sources, which simplifies the location of useful information in the data sets, as defined in claim 15.

    [0015] The present invention has many advantages. It enables the location of useful information in data sets. More particularly it allows the obtaining of information about what data sets are relevant for various groups of symbols relating to the second collection of data sets, which may be a tool in for instance analysing trends among consumers or in society. In this way the invention enables analysis of various aspects of data sets through investigating the ranking of data sets in relation to various groups.

    [0016] It should be emphasized that the term "comprises/comprising" when used in this specification is taken to specify the presence of stated features, integers, steps or components, but does not preclude the presence or addition of one or more other features, integers, steps, components or groups thereof.

    BRIEF DESCRIPTION OF THE DRAWINGS



    [0017] The present invention will now be described in more detail in relation to the enclosed drawings, in which:

    fig. 1 schematically shows a device for categorizing data sets and being connected to a number of computers via a data communication network,

    fig. 2 schematically shows a collection of data sets, fig. 3 shows a block schematic of the device for categorizing data,

    fig. 4 schematically shows a flow chart of a method of characterizing data sets being performed by the device, and

    fig. 5 schematically shows a computer program product in the form of a CD Rom disc comprising computer program code for categorizing data sets according to the invention.


    DETAILED DESCRIPTION OF EMBODIMENTS



    [0018] In the following description, for purposes of explanation and not limitation, specific details are set forth such as particular architectures, interfaces, techniques, etc. in order to provide a thorough understanding of the present invention. However, it will be apparent to those skilled in the art that the present invention may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well known devices, circuits, and methods are omitted so as not to obscure the description of the present invention with unnecessary detail.

    [0019] Fig. 1 schematically shows a device 10 for categorizing data sets according to the invention being connected to a number of computers 14, 16, 18 and 20 via a computer communication network 12. The computers 14, 16, 18 and 20 are here sources of information on which data sets are provided. The computer communication network 12 may here typically be the Internet, which means that the data sets provided by the various sources may be freely available sources. The sources may thus be public and accessible by any computer connected to the Internet. It should however be realized that the invention is not limited to the Internet but may be used in relation to any computer communication network. The device 10 is with advantage provided as one or more computers or servers having such access to the Internet.

    [0020] Fig. 2 schematically shows a number of data sets DS1, DS2, DS3, .... DSn. The above-mentioned sources may each be provided with one or more data sets. In this example the data sets are data files or documents, comprising raw data D and meta data MD. The raw data D is made up of symbols such as words and can thus be for instance a text, and the meta data MD is data associated with the raw data D, such as a classification CL. A classification may be a categorization of the type of text or topic of the data set and may comprise indication of author, a short resume of the text etc. Such a metadata field may also comprise time information T indicating when the data set was created, last changed or made available or made public. The raw data field D of a data set may thus be made up of symbols. Here a first data set DS1 is shown as comprising two specimens of a first symbol S1 two specimens of a second symbol S2 and one specimen of a third symbol S3. A second data set DS2 is shown as comprising one specimen of the first symbol S1, one specimen of the second symbol S2, one specimen of a third symbol S3 and one specimen of an mth symbol. A third data set DS3 is shown as comprising one specimen of the first symbol, S1 one specimen of the third symbol S3 and one specimen of the mth symbol Sm. An nth data set DSn is finally shown as comprising one item of the mth symbol Sm.

    [0021] A symbol is here typically a number of characters, such as alphanumeric characters, formed into an entity that is separated from other symbols via a special symbol separating character, such as a "space" character. The number of symbols has here been deliberately limited in order to provide a clearer understanding of the invention. Normally each data set comprise several more symbols, often several thousands of symbols. The shown data sets are here a collection C of data sets comprising all the shown data sets DS1, DS2, DS3 and DSn. The data sets are furthermore divided into parts, where one such part, or subcollection, SC is indicated as comprising the first and the second data sets DS1 and DS2. It should also be realized that a subcollection C or part of a whole collection C comprises several more data sets.

    [0022] Fig. 3 schematically shows a block schematic of the structure of the device 10. The device comprises a data set collecting unit 22, which is connected to a data set database 23. There is also a symbol frequency determining unit 24. The symbol frequency determining unit 24 is also connected to the data set database 23 as well as to a symbol frequency database 25. There is furthermore a significance determining unit 26. The significance determining unit 26 is connected to the symbol frequency database 25 and to a significance database 27. There is also a grouping unit 28, which is connected to the dataset database 23, to the significance database 27 and to a group database 29. There is also a ranking unit 30 connected to the data set database 23 and the group database 29. Finally there is a location identifier providing unit 32 connected to the data set database 23.

    [0023] All the units are with advantage provided in the form of one or more processors with associated program memory comprising computer program code performing the functionality of these units. Here it should also be mentioned that as an alternative the data collecting unit 22 and the location identifier providing unit 32 may be omitted.

    [0024] Now will follow a description of the invention as provided by the device 10 with reference being made to the previously described fig. 1 - 3 as well as to fig. 4, which shows a flow chart of a method according to which the invention operates.

    [0025] It is today possible to access a vast number of different data sets, such as electronic documents, blogs, articles, chat forums etc. on the Internet. The information provided through these data sets is diverse and may cover a wide range of subjects.

    [0026] Since such information is easily accessible it may also be of interest to analyse it, for instance to investigate the trends in various areas, like technology trends and political trends.

    [0027] However this cannot be done without in some way obtaining a categorizing of the data sets that enables such analysis.

    [0028] The object of the present invention is to handle such a situation, namely to provide a way to structure and categorize a data sets that enables such analysis to be made. One embodiment of the invention is directed towards structuring and categorizing a collection of data sets for enabling an analysis to be made on a part of the collection, for instance a part of the collection directed towards one area of interest to analyse.

    [0029] The operation of the invention for enabling this starts with the data collecting unit 22 collecting data sets from the various sources of information 14, 16, 18 and 20, step 34. The collected data sets here make up at least one collection C of data sets, which may then be stored in the data set database 23 for further analysis. In one variation of the invention a first and a second collection of data sets are collected.

    [0030] As was mentioned above, the data sets comprise symbols such as words based on alphabetic or alphanumeric characters. The symbols may also comprise other types of characters such as signs, like a pound sign, exclamation mark etc. Such symbols may in some variations of the invention be formed through a combination of hexadecimal data characters separated from other symbols through a separation character such as a "space character".

    [0031] In order to enable the categorisation of the data sets, the symbol frequency determining unit 24 accesses this data set database 23 and investigates all the data sets. In this first embodiment it investigates or determines the frequency of symbols in a whole collection C, here all data sets collected from the various sources, step 36. This means that the number of times that each symbol is present in the whole collection C is determined and registered. The statistics of this may be stored in the symbol frequency database 25. If the collection C in fig. 2 is used as an example, this means that for the first symbol S1 there is a determination of how many times this symbol appears in all the data sets DS1 - DSn.

    [0032] However, this is not all that is done. The symbol frequency determining unit 24 also determines the frequency of appearance of symbols in a part SC of the collection C of data sets, step 38. This part is therefore a subcollection of the whole collection. This means that the number of times symbols appear in the subcollection is determined. For the first symbol S1 this means that the number of times this symbol appears in the subcollection SC exemplified by the first and second datasets DS1 and DS2 is determined, which in the simplified example in fig. 2 is three times.

    [0033] The subcollection SC may be a subcollection dedicated to a certain information area, such as a certain area in which a lot of texts have been written of subjects such as technology, social sciences, sports, parenthood or health. The subcollection SC may also be a part of such a general field, like the politics of the US or Sweden etc. The subcollection may also be a subcollection according to type of data set, such as blog, chat or electronic document. As yet another possibility the subcollection SC may be based on time, for instance a specific point in time like a specific day or a time interval such as a week, a month or a year. A subcollection may be identified through a classification setting CL in the meta data MD of the data set, which meta data may here also comprise a time T setting out a time associated with the data set. It should also be realized that a subcollection SC may be based on a combination of such settings, such as a classification CL, like a classification such as politics, and time.

    [0034] The frequency of symbols in the subcollection may then also be stored in a database such as symbol frequency database 25.

    [0035] Once this has been done it is then possible for the significance determining unit 26 to determine the most significant symbols for the part based on both above-mentioned frequencies, i.e. based on the frequency of appearance in the whole collection and the frequency of appearance in the part, step 40. This may in one variation of the invention be done such the frequency of a symbol within the part of the collection C, a subcollection SC, is divided by the frequency of the same symbol in the whole collection. It is here possible that the most significant symbols are those for which the ratio between frequency in the part of the collection and the frequency in the whole collection is above a certain threshold, where the symbols having a ratio below this threshold are considered to be less significant. It is also possible to employ probabilities. The probability of a symbol in the subcollection is determined given the probability of the same symbol in the whole collection. In this way symbols that occur more often in the subcollection than what is probable given their occurrences in the whole collection, are deemed to be most significant.

    [0036] This can also be expressed mathematically in the following way:



    where

    nsc is the number of used symbols in the subcollection sc

    wjsc is the number of times symbol j has been used in the subcollection sc,

    p0j is the probability that a certain symbol taken from the whole collection is the symbol j,

    (1 - p0j) is the probability that the symbol in question is not symbol j, and

    p1,jsc(k) is the probability that symbol j appears k times in the subcollection sc under the presumption that the number of times symbol j occurs follows a binomial distribution with given parameters.



    [0037] The probability of the symbol in the subcollection sc given the probability of the same symbol in the whole collection may then be determined as p1,jsc(wjsc).

    [0038] The result may thereafter be stored in the significance database 27.

    [0039] It is as an alternative possible that the above mentioned activities are performed on the first and the second collection, where the second collection can be a separate collection. The second collection may then be related to the first collection. If for instance the first collection is related to data sets provided in a first time interval, like a certain year, it is then possible that the second collection is made up of data sets provided in a second time interval having a relationship to the first time interval, such as data collected in a following year. The second collection may also be seen as, just as in the first embodiment, a subcollection or subpart of the first collection.

    [0040] Thereafter the grouping unit 28 groups the most significant symbols into groups G according to their appearance in the same data set, step 42. This means that groups or clusters of symbols are being formed. These groups may be formed using a number of different types of techniques. It is for instance possible to use principal component analysis, clustering, such as Ward clustering, or multidimensional scaling. The groups may also be formed through a combination of two or more of these techniques. The groups may then be stored in a group database 29. As an example one such group is formed through the first symbol S1 and the second symbol S2.

    [0041] When this has been done the ranking unit 30 ranks the data sets in relation to the symbol groups according to a ranking scheme, step 44. The ranking scheme may be based on the frequency of the symbols of the group in the data sets. In the ranking scheme used in the first embodiment a data set comprising more of the symbols of a group has a higher rank than a data set comprising fewer symbols of the same group. This means that as an example the first data set DS1 is ranked above the second data set DS2, since it comprises two specimens of the first symbol S1 and two specimens of the second symbol S2 and in totality four such specimens, while the second data set DS2 comprises one specimen of the first symbol S1 and one specimen of the second symbol S2. The absolute frequency of the symbols is thus higher in the first data set DS1 than in the second data set DS2. In a variation of this ranking scheme the frequencies are relative and related to the size of the data set. This variation of the ranking scheme thus employs the absolute frequency per symbol. These are just a few examples of ranking schemes that may be used. It should be known that it is possible with other types of ranking schemes.

    [0042] The ranking may here be stored for the data sets in the database 23.

    [0043] In this way it is possible for interested users to see what data sets in a part SC of a collection C of data sets are the most relevant for a certain subject area, such as a certain classification.

    [0044] It is here also possible that the location identifier providing unit 32 provides location identifiers for the highest ranked data sets, step 46. This may be done for just one such group or for a number of groups of being associated with a subcollection or a part of the whole collection of data sets. A location identifier may for instance be Uniform resource locator (URL) or some other pointer to the computer on which the data set is provided. This data may be stored for the data sets in the data set database 23. A user desiring to access the highest ranked data set may therefore be provided with the location identifier of the data set.

    [0045] In this way it is possible to obtain information about what data sets are relevant for various groups of symbols relating to a certain part of a collection of data sets, which may be a tool in for instance analysing trends in society. It is thus possible to analyse various aspects of a certain category through investigating the ranking of data sets in relation to various groups, which is a significant improvement in the location of relevant information concerning one or more such aspects.
    The device 10 may, as was mentioned above, be implemented using software in a computer. This software may furthermore be provided in the form of a computer program product, for instance as a data carrier carrying computer program code for implementing the units of the categorizing device 10 when being loaded into a computer and run by that computer. One such data carrier 48 with computer program code 50, in the form of a CD ROM disc is generally outlined in fig. 5. A CD ROM disc is only one example of a data carrier. It is feasible with other data carriers, like memory sticks as well as hard disks.

    [0046] While the invention has been described in connection with what is presently considered to be most practical and preferred embodiments, it is to be understood that the invention is not to be limited to the disclosed embodiments, but on the contrary, is intended to cover various modifications and equivalent arrangements. Therefore the present invention is only to be limited by the following claims.


    Claims

    1. A method for categorizing data sets (DS1, DS2, DS3, ... DSn) obtained from a number of sources (14, 16, 18, 20), said data sets being composed of symbols (S1, S2, S3, ... Sm), the method comprising:

    determining (36) the frequency of appearance of symbols in a first collection (C) of data sets,

    determining (38) the frequency of appearance of symbols in a second collection of data sets,

    determining (40) the most significant symbols (S1, S2) for the second collection based on the frequency of appearance in the first collection and on the frequency of appearance in the second collection,

    said determining of the most significant symbols for the second collection being based on the probability p1,jsc(wjsc) of a symbol in the second collection given the probability of the same symbol in the first collection employing wjscBin(nsc,p0j) and

    where nsc is the number of used symbols in the second collection, wjsc is the number of times a symbol j has been used in the second collection, p0j is the probability that a certain symbol taken from the first collection is the symbol j, (1 - p0j) is the probability that the symbol in question is not symbol j, and p1,jsc(k) is the probability that symbol j appears k times in the second collection under the presumption that the number of times symbol j occurs follows a binomial distribution with given parameter,

    grouping (42) the most significant symbols into groups (G) according to their appearance in the same data set, and

    ranking (44) the data sets in relation to the symbol groups using a ranking scheme.


     
    2. The method according to claim 1, further comprising collecting (34) the data sets from said sources.
     
    3. The method according to any previous claim, further comprising providing location identifiers (46) for the highest ranked data sets for at least one group.
     
    4. The method according to any previous claim, wherein the grouping of symbols is performed using principal component analysis.
     
    5. The method according to any previous claim, wherein the grouping of symbols is performed using clustering, such as Ward clustering.
     
    6. The method according to any previous claim, wherein the grouping of symbols is performed using multidimensional scaling.
     
    7. The method according to any previous claim, wherein the second collection is a part (SC) of the first collection.
     
    8. The method according to claim 7, wherein the division of the collection of data sets into parts is based on time, such as a specific point in time or a time interval.
     
    9. The method according to claim 7, wherein the division of the collection of data sets into parts is based on a classification of the data sets.
     
    10. The method according to claim 9, wherein the data sets are documents and the classifications comprise subject, author and/or type of document.
     
    11. The method according to any previous claim, wherein the ranking scheme is based on the frequency of the symbols of the symbol group in the data sets.
     
    12. A device (10) for categorizing data sets (DS1, DS2, DS3, ... DSn) obtained from a number of sources (14, 16, 18, 20), said data sets being composed of a number of symbols (S1, S2, S3, ... Sm), the device comprising:

    a symbol frequency determining unit (24) configured to determine the frequency of appearance of symbols in a first collection (C) of data sets and to determine the frequency of appearance of symbols in a second collection of data sets,

    a significance determining unit (26) configured to determine the most significant symbols (S1, S2) for the second collection based on the frequency of appearance in the first collection and the frequency of appearance in the second collection, the determining of the most significant symbols for the second collection being based on the probability p1,jsc(wjsc) of a symbol in the second collection given the probability of the same symbol in the first collection employing wjscBin(nsc,p0j) and

    where nsc is the number of used symbols in the second collection, wjsc is the number of times a symbol j has been used in the second collection, p0j is the probability that a certain symbol taken from the first collection is the symbol j, (1 - p0j) is the probability that the symbol in question is not symbol j, and p1,jsc(k) is the probability that symbol j appears k times in the second collection under the presumption that the number of times symbol j occurs follows a binomial distribution with given parameter,

    a grouping unit (28) configured to group the most significant symbols into groups (G) according to their appearance in the same data set, and

    a ranking unit (30) configured to rank the data sets in relation to the symbol groups using a ranking scheme.


     
    13. The device according to claim 12, further comprising a data set collecting unit (22) configured to collect the data sets from said sources.
     
    14. The device according to claim 12 or 13, further comprising a location identifier providing unit (30) configured to provide location identifiers for the highest ranked data sets of at least one group.
     
    15. A computer program product for categorizing a collection (C) of data sets (DS1, DS2, DS3, ... DSn) obtained from a number of sources (14, 16, 18, 20), said data sets being composed of a number of symbols (S1, S2, S3, ... Sm), the computer program product comprising a computer readable storage medium (48) comprising computer program code (50) causing a computer to:

    determine the frequency of appearance of symbols in a first collection (C) of data sets,

    determine the frequency of appearance of symbols in a second collection of data sets,

    determine the most significant symbols (S1, S2) for the second collection based on the frequency of appearance in the first collection and the frequency of appearance in the second collection, said determining of the most significant symbols for the second collection being based on the probability p1,jsc(wjsc) of a symbol in the second collection given the probability of the same symbol in the first collection employing wjscBin(nsc,p0j) and

    where nsc is the number of used symbols in the second collection, wjsc is the number of times a symbol j has been used in the second collection, p0j is the probability that a certain symbol taken from the first collection is the symbol j, (1 - p0j) is the probability that the symbol in question is not symbol j, and p1,jsc(k) is the probability that symbol j appears k times in the second collection under the presumption that the number of times symbol j occurs follows a binomial distribution with given parameter,

    group the most significant symbols into groups (G) according to their appearance in the same data set, and

    rank the data sets in relation to the symbol groups using a ranking scheme.


     


    Ansprüche

    1. Verfahren zum Kategorisieren von aus einer Anzahl von Quellen (14, 16, 18, 20) bezogenen Datensätzen (DS1, DS2, DS3, ... DSn), wobei sich die Datensätze aus Symbolen (S1, S2, S3, ... Sm) zusammensetzen, wobei das Verfahren Folgendes umfasst:

    Bestimmen (36) der Häufigkeit des Auftretens von Symbolen in einer ersten Sammlung (C) von Datensätzen,

    Bestimmen (38) der Häufigkeit des Auftretens von Symbolen in einer zweiten Sammlung von Datensätzen,

    Bestimmen (40) der signifikantesten Symbole (S1, S2) für die zweite Sammlung basierend auf der Häufigkeit des Auftretens in der ersten Sammlung und der Häufigkeit des Auftretens in der zweiten Sammlung,

    wobei das Bestimmen der signifikantesten Symbole für die zweite Sammlung auf der Wahrscheinlichkeit p1,jsc(wjsc) eines Symbols in der zweiten Sammlung bei gegebener Wahrscheinlichkeit des gleichen Symbols in der ersten Sammlung unter Verwendung von wjscBin(nsc,p0j) und

    basiert,

    wobei nsc die Anzahl verwendeter Symbole in der zweiten Sammlung ist, wjsc angibt, wie viele Male ein Symbol j in der zweiten Sammlung verwendet wurde, p0j die Wahrscheinlichkeit ist, dass ein bestimmtes aus der ersten Sammlung genommenes Symbol das Symbol j ist, (1 - p0j) die Wahrscheinlichkeit ist, dass das fragliche Symbol nicht das Symbol j ist, und p1,jsc(k) die Wahrscheinlichkeit ist, dass das Symbol j k Male in der zweiten Sammlung auftritt, wenn angenommen wird, dass die Anzahl an Malen, die das Symbol j auftritt, einer Binomialverteilung mit gegebenem Parameter folgt,

    Gruppieren (42) der signifikantesten Symbole in Gruppen (G) nach ihrem Auftreten im gleichen Datensatz und

    Erstellen (44) einer Rangordnung der Datensätze in Bezug auf die Symbolgruppen mittels eines Rangordnungsschemas.


     
    2. Verfahren nach Anspruch 1, ferner umfassend ein Sammeln (34) der Datensätze aus den Quellen.
     
    3. Verfahren nach einem der vorhergehenden Ansprüche, ferner umfassend ein Bereitstellen von Positionskennungen (46) für die am höchsten eingestuften Datensätze für mindestens eine Gruppe.
     
    4. Verfahren nach einem der vorhergehenden Ansprüche, wobei das Gruppieren von Symbolen mittels Hauptkomponentenanalyse erfolgt.
     
    5. Verfahren nach einem der vorhergehenden Ansprüche, wobei das Gruppieren von Symbolen mittels Clusterbildung wie beispielsweise Ward-Clusterbildung erfolgt.
     
    6. Verfahren nach einem der vorhergehenden Ansprüche, wobei das Gruppieren von Symbolen mittels multidimensionaler Skalierung erfolgt.
     
    7. Verfahren nach einem der vorhergehenden Ansprüche, wobei es sich bei der zweiten Sammlung um einen Teil (SC) der ersten Sammlung handelt.
     
    8. Verfahren nach Anspruch 7, wobei die Unterteilung der Sammlung von Datensätzen in Teile basierend auf der Zeit erfolgt, beispielsweise einem bestimmten Zeitpunkt oder einem Zeitintervall.
     
    9. Verfahren nach Anspruch 7, wobei die Unterteilung der Sammlung von Datensätzen in Teile basierend auf einer Klassifizierung der Datensätze erfolgt.
     
    10. Verfahren nach Anspruch 9, wobei es sich bei den Datensätzen um Dokumente handelt und die Klassifikationen Thema, Autor und/oder Dokumenttyp umfassen.
     
    11. Verfahren nach einem der vorhergehenden Ansprüche, wobei das Rangordnungsschema auf der Häufigkeit der Symbole der Symbolgruppe in den Datensätzen basiert.
     
    12. Einrichtung (10) zum Kategorisieren von aus einer Anzahl von Quellen (14, 16, 18, 20) bezogenen Datensätzen (DS1, DS2, DS3, ... DSn), wobei sich die Datensätze aus einer Anzahl von Symbolen (S1, S2, S3, ... Sm) zusammensetzen, wobei die Einrichtung Folgendes umfasst:

    eine Symbolhäufigkeitsbestimmungseinheit (24), die dafür konfiguriert ist, die Häufigkeit des Auftretens von Symbolen in einer ersten Sammlung (C) von Datensätzen zu bestimmen und die Häufigkeit des Auftretens von Symbolen in einer zweiten Sammlung von Datensätzen zu bestimmen,

    eine Signifikanzbestimmungseinheit (26), die dafür konfiguriert ist, die signifikantesten Symbole (S1, S2) für die zweite Sammlung basierend auf der Häufigkeit des Auftretens in der ersten Sammlung und der Häufigkeit des Auftretens in der zweiten Sammlung zu bestimmen, wobei das Bestimmen der signifikantesten Symbole für die zweite Sammlung auf der Wahrscheinlichkeit p1,jsc(wjsc) eines Symbols in der zweiten Sammlung bei gegebener Wahrscheinlichkeit des gleichen Symbols in der ersten Sammlung unter Verwendung von wjscBin(nsc,p0j) und

    basiert,

    wobei nsc die Anzahl verwendeter Symbole in der zweiten Sammlung ist, wjsc angibt, wie viele Male ein Symbol j in der zweiten Sammlung verwendet wurde, p0j die Wahrscheinlichkeit ist, dass ein bestimmtes aus der ersten Sammlung genommenes Symbol das Symbol j ist, (1 - p0j) die Wahrscheinlichkeit ist, dass das fragliche Symbol nicht das Symbol j ist, und p1,jsc(k) die Wahrscheinlichkeit ist, dass das Symbol j k Male in der zweiten Sammlung auftritt, wenn angenommen wird, dass die Anzahl an Malen, die das Symbol j auftritt, einer Binomialverteilung mit gegebenem Parameter folgt,

    eine Gruppierungseinheit (28), die dafür konfiguriert ist, die signifikantesten Symbole nach ihrem Auftreten im gleichen Datensatz in Gruppen (G) zu gruppieren, und

    eine Rangordnungserstellungseinheit (30), die dafür konfiguriert ist, eine Rangordnung der Datensätze in Bezug auf die Symbolgruppen mittels eines Rangordnungsschemas zu erstellen.


     
    13. Einrichtung nach Anspruch 12, ferner umfassend eine Datensatz-Sammeleinheit (22), die dafür konfiguriert ist, die Datensätze aus den Quellen zu sammeln.
     
    14. Einrichtung nach Anspruch 12 oder 13, ferner umfassend eine Positionskennungsbereitstellungseinheit (30), die dafür konfiguriert ist, für die am höchsten eingestuften Datensätze mindestens einer Gruppe Positionskennungen bereitzustellen.
     
    15. Computerprogrammprodukt zum Kategorisieren einer aus einer Anzahl von Quellen (14, 16, 18, 20) bezogenen Sammlung (C) von Datensätzen (DS1, DS2, DS3, ... DSn), wobei sich die Datensätze aus einer Anzahl von Symbolen (S1, S2, S3, ... Sm) zusammensetzen, wobei das Computerprogrammprodukt ein computerlesbares Speichermedium (48) umfasst, das Computerprogrammcode (50) umfasst, der einen Computer zu Folgendem veranlasst:

    die Häufigkeit des Auftretens von Symbolen in einer ersten Sammlung (C) von Datensätzen zu bestimmen,

    die Häufigkeit des Auftretens von Symbolen in einer zweiten Sammlung von Datensätzen zu bestimmen,

    die signifikantesten Symbole (S1, S2) für die zweite Sammlung basierend auf der Häufigkeit des Auftretens in der ersten Sammlung und der Häufigkeit des Auftretens in der zweiten Sammlung zu bestimmen, wobei das Bestimmen der signifikantesten Symbole für die zweite Sammlung auf der Wahrscheinlichkeit p1,jsc(wjsc) eines Symbols in der zweiten Sammlung bei gegebener Wahrscheinlichkeit des gleichen Symbols in der ersten Sammlung unter Verwendung von wjscBin(nsc,p0j) und

    basiert,

    wobei nsc die Anzahl verwendeter Symbole in der zweiten Sammlung ist, wjsc angibt, wie viele Male ein Symbol j in der zweiten Sammlung verwendet wurde, p0j die Wahrscheinlichkeit ist, dass ein bestimmtes aus der ersten Sammlung genommenes Symbol das Symbol j ist, (1 - p0j) die Wahrscheinlichkeit ist, dass das fragliche Symbol nicht das Symbol j ist, und p1,jsc(k) die Wahrscheinlichkeit ist, dass das Symbol j k Male in der zweiten Sammlung auftritt, wenn angenommen wird, dass die Anzahl an Malen, die das Symbol j auftritt, einer Binomialverteilung mit gegebenem Parameter folgt,

    die signifikantesten Symbole nach ihrem Auftreten im gleichen Datensatz in Gruppen (G) zu gruppieren, und

    eine Rangordnung der Datensätze in Bezug auf die Symbolgruppen mittels eines Rangordnungsschemas zu erstellen.


     


    Revendications

    1. Procédé pour catégoriser des ensembles de données (DS1, DS2, DS3, ... DSn) obtenus à partir d'un nombre de sources (14, 16, 18, 20), lesdits ensembles de données étant composés de symboles (S1, S2, S3, ... Sm), le procédé comprenant de:

    déterminer (36) la fréquence d'apparition de symboles dans une première collection (C) d'ensembles de données,

    déterminer (38) la fréquence d'apparition de symboles dans une seconde collection d'ensembles de données,

    déterminer (40) les symboles les plus pertinents (S1, S2) pour la seconde collection sur la base de la fréquence d'apparition dans la première collection et de la fréquence d'apparition dans la seconde collection, ladite détermination des symboles les plus pertinents pour la seconde collection étant basée sur la probabilité p1,jsc(wjsc) d'un symbole dans la seconde collection étant donné la probabilité du même symbole dans la première collection employant wjscBin(nsc,p0j) et

    où nsc est le nombre de symboles utilisés dans la seconde collection, wjsc est le nombre de fois qu'un symbole j a été utilisé dans la seconde collection, p0j est la probabilité qu'un certain symbole pris dans la première collection est le symbole j , (1 - p0j) et la probabilité que le symbole en question n'est pas le symbole j, et p1,jsc(k) est la probabilité que le symbole j apparaît k fois dans la seconde collection en postulant que le nombre de fois que le symbole j apparaît suit une distribution binomiale avec un paramètre donné,

    grouper (42) les symboles les plus pertinents en groupes (G) selon leur apparition dans le même ensemble de données, et

    classer hiérarchiquement (44) les ensembles de données par rapport au groupe de symboles en utilisant un mécanisme de classement hiérarchique.


     
    2. Procédé selon la revendication 1, comprenant en outre de collecter (34) les ensembles de données à partir desdites sources.
     
    3. Procédé selon une quelconque des revendications précédentes, comprenant en outre de fournir des identifiants de localisation (46) pour les ensembles de données au classement hiérarchique le plus haut pour au moins un groupe.
     
    4. Procédé selon une quelconque des revendications précédentes, dans lequel le groupage des symboles est effectué en utilisant une analyse de composant principal.
     
    5. Procédé selon une quelconque des revendications précédentes, dans lequel le groupage des symboles est effectué en utilisant un groupement, tel qu'un groupement Ward.
     
    6. Procédé selon une quelconque des revendications précédentes, dans lequel le groupage des symboles est effectué en utilisant une mise à l'échelle multidimensionnelle.
     
    7. Procédé selon une quelconque des revendications précédentes, dans lequel la seconde collection (SC) est une partie de la première collection.
     
    8. Procédé selon la revendication 7, dans lequel la division de la collection d'ensembles de données en parties est basée sur le temps, tel qu'un point dans le temps spécifique ou un intervalle de temps.
     
    9. Procédé selon la revendication 7, dans lequel la division de la collection d'ensembles de données en parties est basée sur une classification des ensembles de données.
     
    10. Procédé selon la revendication 9, dans lequel les ensembles de données sont des documents et les classifications comprennent un sujet, un auteur et/ou un type de document.
     
    11. Procédé selon une quelconque des revendications précédentes, dans lequel le mécanisme de classement hiérarchique est basé sur la fréquence des symboles du groupe de symboles dans les ensembles de données.
     
    12. Dispositif (10) pour catégoriser les ensembles de données (DS1, DS2, DS3, ... DSn) obtenus à partir d'un nombre de sources (14, 16, 18, 20), lesdits ensemble de données étant composés d'un nombre de symboles (S1, S2, S3, ... Sm), le dispositif comprenant:

    une unité de détermination de fréquence de symbole (24) configurée pour déterminer la fréquence d'apparition des symboles dans une première collection (C) d'ensemble de données et déterminer la fréquence d'apparition des symboles dans une seconde collection d'ensembles de données,

    une unité de détermination de pertinence (26) configurée pour déterminer les symboles les plus pertinents (S1, S2) pour la seconde collection sur la base de la fréquence d'apparition dans la première collection et la fréquence d'apparition dans la seconde collection, la détermination des symboles les plus pertinents pour la seconde collection étant basée sur la probabilité p1,jsc(wjsc) d'un symbole dans la seconde collection étant donné la probabilité du même symbole dans la première collection employant wjscBin(nsc,p0j) et

    où nsc est le nombre de symboles utilisés dans la seconde collection, wjsc est le nombre de fois qu'un symbole j a été utilisé dans la seconde collection, p0j est la probabilité qu' un certain symbole pris dans la première collection est le symbole j, (1 - p0j) est la probabilité que le symbole en question n'est pas le symbole j, et p1,jsc(k) est la probabilité que le symbole j apparaît k fois dans la seconde collection en postulant que le nombre de fois que le symbole j apparaît suit une distribution binomiale avec un paramètre donné,

    une unité de groupage (28) configurée pour grouper les symboles les plus pertinents en groupes (G) selon leur apparition dans le même ensemble de données, et

    une unité de classement hiérarchique (30) configurée pour classer hiérarchiquement les ensembles de données par rapport aux groupes de symboles en utilisant un mécanisme de classement hiérarchique.


     
    13. Dispositif selon la revendication 12, comprenant en outre une unité de collecte d'ensembles de données (22) configurée pour collecter les ensembles de données à partir desdites sources.
     
    14. Dispositif selon la revendication 12 ou 13, comprenant en outre une unité de fourniture d'identifiants de localisation (30) configurée pour fournir des identifiants de localisation pour les ensembles de données au classement hiérarchique le plus haut d'au moins un groupe.
     
    15. Produit de programme informatique pour catégoriser une collection (C) d'ensembles de données (DS1, DS2, DS3, ... DSn) obtenus à partir d'un nombre de sources (14, 16, 18, 20), lesdits ensemble de données étant composés d'un nombre de symboles (S1, S2, S3, ... Sm), le produit de programme informatique comprenant un support de mémorisation lisible par ordinateur (48) comprenant un code de programme informatique (50) amenant un ordinateur à:

    déterminer la fréquence d'apparition de symboles dans une première collection (C) d'ensembles de données,

    déterminer la fréquence d'apparition de symboles dans une seconde collection d'ensembles de données,

    déterminer les symboles les plus pertinents (S1, S2) pour la seconde collection sur la base de la fréquence d'apparition dans la première collection et de la fréquence d'apparition dans la seconde collection,

    ladite détermination des symboles les plus pertinents pour la seconde collection étant basée sur la probabilité p1,jsc(wjsc) d'un symbole dans la seconde collection étant donné la probabilité du même symbole dans la première collection employant wjscBin(nsc,p0j) et

    où nsc est le nombre de symboles utilisés dans la seconde collection, wjsc est le nombre de fois qu'un symbole j a été utilisé dans la seconde collection, p0j est la probabilité qu'un certain symbole pris dans la première collection est le symbole j , (1 - p0j) et la probabilité que le symbole en question n'est pas le symbole j, et p1,jsc(k) est la probabilité que le symbole j apparaît k fois dans la seconde collection en postulant que le nombre de fois que le symbole j apparaît suit une distribution binomiale avec un paramètre donné,

    grouper les symboles les plus pertinents en groupes (G) selon leur apparition dans le même ensemble de données, et

    classer hiérarchiquement les ensembles de données par rapport au groupe de symboles en utilisant un mécanisme de classement hiérarchique.


     




    Drawing












    REFERENCES CITED IN THE DESCRIPTION



    This list of references cited by the applicant is for the reader's convenience only. It does not form part of the European patent document. Even though great care has been taken in compiling the references, errors or omissions cannot be excluded and the EPO disclaims all liability in this regard.

    Patent documents cited in the description




    Non-patent literature cited in the description