US 5,732,259 AGrant
Information Classifying System That Discovers the Hierarchical Structure of Keywords by Comparing Link Numbers
Issue Date:1998-03-24
•3 Claims
•40 Drawing Sheets
Abstract
A system classifies the information added with keywords by discovering the hierarchy relation with respect to all of keywords. There is generated a relation composed of a total link number and links to other keywords and information, each for keyword. The total link numbers of two relations with links are compared with each other thereby determining a high level relation having a larger total link number and a low level relation having a lower total link number, and a hierarchy structure can be discovered by making the link from the low level relation to the high level relation ineffective.
Metadata
Inventor
- Atsushi Konno
Application Information
Application Number:US 3934055
Filing Date:1995-02-23
Priority Date:1995-02-23
Art Unit:237
Classifications
IPC:
G06F 1730
Field of Search:
345395364600600;603;604;605;601;602283.1;968
Patent Drawings (40 sheets)
Description
Background of the Invention
Information systems such as document-retrieval systems handle information using keywords. Keywords help users to retrieve information.
However, existing information systems handling extensive information have a problem:
with many keywords available, which is the appropriate keyword to retrieve a given information?
The answer is to classify the information hierarchically. However, this leads to another problem, i.e.,
When is the hierarchical structure decided? If the hierarchical structure is decided before information is entered, it may differ from the actual structure of the data. If not, how is the hierarchical structure decided?
This invention resolves these problems.
Summary of the Invention
This invention classifies information by discovering the hierarchical structure of all keywords added to the information.
This invention creates relations from the keywords and discovers the hierarchical structure using the relations.
A relation contains an identifier, links and a total link number. Each identifier identifies a keyword; each link links to other relations or information.
The hierarchical structure of keywords is discovered as follows:
relations having a larger total link number are selected as the higher-ranking relation; and
links to higher-ranking relation are made ineffective.
Brief Description of the Drawings
FIG. 1 is a block diagram of an information-classifying system.
FIG. 2 is a block diagram of a keywords manager.
FIG. 3 is a combination of identifier and keyword, which is used by the keywords manager.
FIG. 4 is a block diagram of a relations manager.
FIG. 5 is an illustration of a relation.
FIG. 6 is a illustration of a link.
FIG. 7 is a block diagram of a particles manager.
FIG. 8 is an illustration of a particle.
FIG. 9 is an illustration of an alteration.
FIG. 10 is a hierarchical structure.
FIG. 11 is the numbers of combinations of keyword.
FIG. 12 is the numbers of times that keywords appear.
FIGS. 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26,27, 28, 29, 30 and 31 show the process used to discover the hierarchical structure of keywords.
FIG. 32 is a flow chart of an algorithm used to discovering the hierarchical structure of keywords.
FIG. 33 is a detailed flowchart of box 2005.
FIG. 34 is a detailed flowchart of box 2102.
FIG. 35 is a chart that shows the concept of hierarchical discovery.
FIG. 36 is a flowchart of an algorithm used to modify relations.
FIG. 37 is a detailed flowchart of box 3008.
FIG. 38 is a chart that shows modification of relations.
FIG. 39 is an illustration of information display and selection process.
FIG. 40 is a chart stack structure.
FIG. 41 is a chart of background "personality".
FIG. 42 is a chart of background "particlesBG".
FIG. 43 is a chart of background "relationsBG".
FIG. 44 is a chart of card "memorandum".
Detailed Description of the Invention
First, I will explain the idea of this invention.
In an information system that handles numerous information, keywords are added to the information.
If the keywords are combined correctly, keywords for higher-ranking concepts will be more frequently used.
For example, information on "money supply" is classified by the keywords "economics" and "social science". "Social science" is a higher-rank keyword than "economics". This is because concept of social science includes economics, politics and sociology. "Social science" is always used as keyword whenever the lower-ranked keywords "economics", "sociology" and "politics" are used. "Social science" is a more frequently used keyword than "economics", "politics" or "sociology"; more frequently used keywords are of higher rank.
The hierarchical classification process is shown in FIGS. 10 to 31.
FIG. 11 shows the numbers of combinations of keywords.
The letter A,B,C . . . P refer to keywords. Of these, the combination of A and E is used 3 times, and the combination of A,E and J is 4 times.
FIG. 12 shows the numbers of times that keywords appear. This number is equal to the total link number of relation.
FIG. 13 shows a sorted list of relations. The vertical character string shows relations; AEIJKMNO refers to the relation for keyword A , known as relation A which has links to relations E,I,J,K,M,N and O.
First, the E of relation A is focused, the rhombus in FIG. 14 as a focal point. This E is the link that links to relation E, which is the vertical character string EAIJKMNO.
Relation A has a higher rank than relation E because relation A has larger total link number than relation E. Relation E's A, which links to relation A, is an ineffective link because part of relation A is focused. The cross in FIGS. 14 to 31 signifies the ineffective condition. The I,J,K,M,N and O of relation A are also ineffective links, because they are included in relation E, which is of lower rank than relation A.
Next, in FIG. 15, relation E's I, which links to relation I, is focused. Relation I is 8th vertical character string from left.
Link A and E of relation I are ineffective, because parts of relation A and relation E are focused. Link M and link N of relation E are also ineffective, because they are included in relation I which is of lower rank than relation E.
In FIG. 16, relation l's M, which links to relation M, is focused. Link A, link E and link I of relation M are ineffective, because they are the links which link to focused relations.
The above process is repeated, as shown in FIGS. 17 to 30. Finally, in FIG. 31, the links which are not marked with a cross are effective. The hierarchical structure of keywords is determined from these links. FIG. 10 shows the hierarchical structure. In FIG. 10, L1 means "highest level".
An information-classifying system is shown in FIGS. 1 to 9 and FIGS. 32 to 39. FIG. 39 illustrates samples of information display and selection. The system accepts input of information which have keywords.
FIG. 1 shows a block diagram of the information-classifying system.
Input unit 101 accepts commands to add information and accepts selections of relations and information.
Analyzer 102 provides notification of the commands to total manager 103.
Total manager 1 03 manages the components of this information-classifying system.
Keywords manager 104 memorizes, encodes and decodes keywords.
Relations manager 105 generates relations and discovers the hierarchical structure.
Particles manager 106 memorizes, retrieves and generates particles. A particle is generated each information. One particle is equal to the combination of keywords and an information.
Display manager 107 controls display unit 108.
FIG. 2 shows keywords manager 104.
Keywords manager 104 accepts two types of request.
1. "Enter" request
Keyword-to-identifier converter 201 converts a keyword and returns an identifier.
If the keyword is not registered, keyword-to-identifier converter 201 registers the keyword.
2. "Retrieve" request
Identifier-to-keyword converter 202 converts an identifier and returns a keyword.
The combination of identifier 301 and keyword 302, shown in FIG. 3, is used by keyword-to-identifier converter 201 and identifier-to-keyword converter 202 in their conversion process.
FIG. 4 shows relations manager 105. A relation, which is shown in FIG. 5, comprises keyword identifier 501, total link number 502 and link array 503, 504 . . . A link, which is shown in FIG. 6, comprises identifier 601, link number 602 and flag 603.
Keyword identifier 501 identifies a keyword.
Total link number 502 is the sum of link numbers.
Identifier 601 identifies a particle or relation.
Link number 602 shows link thickness.
Flag 603 shows the relative position of a linked relation. Flag 603 is a four-valued flag:
1. the value "isHigher" means the linked relation is higher rank than this relation;
2. the value "isSame" means the linked relation is the same rank as this relation;
3. the value "isLower" means the linked relation is lower rank than this relation;
4. the value "isIncluded" means the link should be included in a lower relation.
Relations manager 105 accepts two types of request:
1. "Modify relation" request
When relation manager 105 accepts a modify relation request, relation modifier 401 modifies a link's identifier 601 and link number 602 using the alteration. An alteration is shown in FIG. 9.
Hierarchy discoverer 402 then discovers hierarchical structure.
2. "Retrieve relation" request
When relation manager 105 accepts a retrieve relation request, relation retriever 403 retrieves a relation.
FIG. 38 shows the relation modification process. In the figure, C . . . n means the relation's keyword identifier is n; L . . . means a link; L . . . Cm . . . n means a link links to the relation m and means the link number is n; L . . . Pn means a link links to the particle n.
When a new information is entered, a particle and alterations are created. In FIG. 38, the alteration is comprised of a particle identifier whose value is 2, first keyword identifier whose value is 1, second keyword identifier whose value is 2 and third keyword identifier w hose value is 4. Relation modifier 401 accepts the alteration, then modifies relations 1, 2 and 3, and creates relation 4.
Relation 1 is comprised as follows:
a keyword identifier whose-value is 1;
a first link to relation 2 with link number 1;
a second link to relation 3 with link number 1;
and a third link to particle 1.
After entering information, the link number of the first link of relation 1 increases, and links to relation 4 and particle 2 are added to relation 1.
Relation 2 is comprised as follows:
a keyword identifier whose value is 2;
a first link to relation 1 with link number 1;
a second link to relation 3 with link number 1;
and a third link to particle 1.
After entering a information, the link number of the first link of relation 2 increases, and links to relation 4 and particle 2 are added to relation 2.
Since the alteration has no a keyword identifier whose value is 3, relation 3 is not influenced. And, because the alteration has a keyword identifier w hose value is 4, and there was not relation 4, relation 4 is created.
The relation modification algorithm is shown in FIG. 36.
Initial value of i is set to 1 in box 3001.
In decision box 3002, the existence of item i of keyword identifier array of the alteration is tested.
If the keyword identifier does not exist, the relation modification process is ended.
If the keyword identifier exists, relation retriever 403 is requested to retrieve a relation whose keyword identifier is same as item i of keyword identifier array of the alteration in box 3003.
If the requested relation does not exist, relation retriever 403 creates it; the created relation's keyword identifier is set to the same value as item i of the keyword identifier array of the alteration; the created relation's total link number is set to 0.
When a new information is entered, the requested relation does not link to the particle. In box 3004, therefore, a link to the particle is added to the relation; the link's link number is set to 1. Also, the total link number of the relation is incremented.
Then, initial value of j is set to 1 in box 3005.
In decision box 3006, the existence of item j of keyword identifier array of the alteration is tested.
If not exist, increment i in box 3011 and go to box 3002.
If exist, compare i with j in decision box 3007.
If same, go to box 3010.
If i is different from j, a link whose identifier is the same as item j of the keyword identifier array of the alteration, is added to the relation or link number of the link is incremented in box 3008.
In box 3009, the total link number is incremented.
In box 3010, increment j and go to decision box 3006.
FIG. 37 shows a detailed process of box 3008.
In box 3101, test whether the relation contains a link whose identifier is same as item j of the keyword identifier array of the alteration.
If the relation contains the link, increment the link number of the link in box 3102.
If not, add a new link, with link number 1, to the relation in box 3103.
The new link links to a relation whose keyword identifier is the same as item j of the keyword identifier array of the alteration.
The concept of hierarchical discovery is shown in FIG. 35.
In FIG. 35, there are relations A, B, C, D, E and F; relations have links described as arrows; total link number of relation A changes.
Arrows such as the arrow from relation D to E in FIG. 35 are marked with a cross; the cross means that the destination of the arrow is of higher rank and the link's flag is set to "isHigher". We call this an "isHigher" link.
An arrow marked with a triangle means a link whose link's flag is set to "isIncluded". We call this an "isIncluded" link.
An arrow marked with a circle means a link whose link's flag is set to "isSame". We call this an "isSame" link.
An arrow marked with nothing means a link whose link's flag is set to "isLower". We call this an "isLower" link.
In FIG. 35, relation A goes to a higher rank and the flags of some links change.
The link, marked with "a", was an "isIncluded" link, because:
relation A was a lower-rank relation than relation D;
relation A was a higher-rank relation than relation C;
relation D had links to relation A and relation C; and
relation A has a link to relation C.
After relation A's promotion the link marked with "a", becomes an "isLower" link which is marked with "a'", because:
relation A becomes the same rank as relation D; and
there is not a relation which is lower rank than relation D, higher rank than relation C, and linked from relation D and to relation C.
The links marked with "b" and "c" become "isSame" links, because relation A becomes the same rank as relation D.
The link marked with "d" becomes an "isLower" link, because there is no relation which is lower rank than relation E and higher rank than relation A.
The link marked with "e" becomes an "isLower" link.
The link marked with "f" becomes an "isHigher" link.
The link marked with "g" becomes an "isIncluded" link, because:
relation A is lower rank than relation E;
relation A is higher rank than relation B;
relation E has links to the relation A and relation B; and
relation A has a link to relation B.
If there was a link from relation B to relation C, the link marked with "h" would become an "isIncluded" link.
FIG. 32 shows the hierarchical discovery algorithm.
In box 2001, the initial value of i is set to 1.
In box 2002, test the existence of item i of the keyword identifier array of the alteration.
If item i of the keyword identifier array does not exist in the alteration, terminate the hierarchical discovery process.
If item i of the keyword identifier array exists in the alteration, retrieve a relation whose keyword identifier is same as item i of keyword identifier array of the alteration from relations memory 404, in box 2002.
In box 2003, the initial value of j is set to 1.
In box 2004, test the existence of item j of the retrieved relation's link array.
If item j does not exist in the relation, increment i in box 2007.
If item j exists in the relation, determine the value of the flag of item j, in box 2005.
In box 2006, increment j, and go to box 2004.
FIG. 33 shows the determination process for the link's flag in box 2005.
In this flow chart, suppose that relation A is the relation retrieved in box 2002, and that item j of relation A's link array is a link which links to relation B.
In box 2101, set LF to the same flag value as item j of relation A's link array.
In box 2102, set the flag of item j of relation A's link array to appropriate value.
In box 2103, test the value of LF.
If the value of LF is "isHigher", set the flags of relation B's "isLower" links and "isIncluded" links in box 2104.
If the value of LF is "isSame" and relation A is of higher rank than relation B, set the flag of relation B's link to relation A to "isHigher" in box 2106.
The link's flag shows relativity of two relations. This information-classifying system decides the relativity from a comparison of the two relation's total link number.
If 0.9 TLX (relation X's total link number)>TLY (relation Y's total link number) , relation X is of higher rank than relation Y.
If 0.9 TLY>TLX, relation Y is of higher rank than relation X.
Else, relation X is the same rank as relation Y.
FIG. 34 shows details of the process in box 2102.
In decision box 2201, test the condition that relation B is of higher rank than relation A.
If the condition is true, set the flag of relation A's link to relation B to "isHigher" in box 2202.
If the condition is not true, step into box 2203.
In decision box 2203, test the condition that relation B is of lower rank than relation A.
If the condition is not true, set the flag of relation A's link to relation B to "isSame" in box 2204.
If the condition is true, step into box 2205.
In decision box 2205, test the condition:
relation B is of lower rank than relation A;
relation C exists and is of higher rank than relation B;
relation C is of lower rank than relation A;
relation A has a link to relation C;
relation C has a link to relation B.
If the condition is true, set the flag of relation A's link to relation B to "isIncluded" in box 2206.
If the condition is not true, set the flag of relation A's link to relation B to "isLower" in box 2207.
FIG. 7 shows the particles manager.
The particles manager manages particles.
As described in FIG. 8, a particle comprises identifier 801, keyword array 802, 803 . . . , keyword identifier array 804, 805 . . . , and body 806.
Identifier 801 is a identification number which distinguishes one particle from another.
Keyword identifier 804, 805, . . . are identification numbers which identify keywords.
The main part of a particle is body 806, which comprises text data and picture data.
The particle manager accepts two types of requests.
1. "Enter" request
Particle creator 701 decides an identification number, sets the identifier 801 of particle to the number, sets the particle identifier 901 of alteration to the number, requests keywords manager 104 to enter the keywords, accepts return values from keywords manager 104, sets particle's keyword identifiers 804, 805 . . . to those values, sets alteration's keyword identifiers 902, 903, 904 . . . to those values, registers the particle to particles memory 703 and returns the alteration.
2. "Retrieve" request
Particle retriever 702 retrieves a particle with an identifier and returns it.
Analyzer 102 analyzes input and sends its results to total manager 103.
Total manager 103 accepts the results from analyzer 102 and requests other components to act.
When total manager 103 accepts a particle input indication from analyzer 102, total manager 103 requests particles manager 106 to enter the particle.
Particles manager 106 returns an alteration, then total manager 103 requests relations manager 105 to modify relation.
When total manager 103 accepts a select indication from analyzer 102 with an identifier, total manager 103 requests relations manager 105 to retrieve the relation or requests particles manager 106 to retrieve the particle. Total manager 103 prepares display data and requests display manager 107 to display it.
FIGS. 40 to 69 show another information-classifying system. The system runs on an Apple Computer.sub.-- .TM. PowerBook.sub.-- .TM. 520 with HyperCard.sub.-- .TM.J-2.2. The system may also runs on other Macintosh.sub.-- .TM. with HyperCard.sub.-- .TM. 2.2. FIG. 40 shows the structure of its information-classifying stack.
The stack consists of background "personality", background "particlesBG", background "classificationsBG" and an unnamed background.
FIG. 41 shows the background "personality", which comprises a field "classification" 4101, a field "particleName" 4102, a field 4103, a field "reference" 4104, a button "back track" 4105 and background picture.
FIG. 42 shows background "particlesBG", which comprises a field "particleName" 4201 and a field "classification" 4202.
FIG. 43 shows background "classificationsBG". This background comprises a field "makeupedName" 4301, a field "classificationName" 4302, a field "totalLinkNumber" 4303, a button "back track" 4304, a button "new card" 4305, a field "links" 4306, a field "button1" 4307, a field "button2" 4308, a field "button3" 4309, a field "button4" 43 10, a field "button5" 43 11, a field "button6" 4312, a field "button7" 4313, a field "button8" 4314, a field "otherButton" 4315 and a field "others" 4316 . The field "others" 4316 overlays the field "otherButton" 4315. The LockText properties of all fields on this background are true. Field 4306 and 4316 are scroll fields.
FIG. 44 shows the card "memorandum" of the unnamed background. Card "memorandum" comprises field "particles" 4401, field "classifications" 4402 and field "alteration" 4403.
The script of stack is as follows:
The script contains:
a function "textPos" which calculates the position of a character string in a container;
a handler "closeStack" which stores the value of global variable "classificationsM" and "particlesM"; and
a handler "openStack" which restores the value of both global variables:
"classificationsM", which memorizes all keywords; and
"particlesM", which memorizes all particle's name.
The script of card "memorandum" is as follows:
When this card opened, the values of global variable "classificationsM", "particlesM" and "alterationM" are displayed. The global variable "alterationM" memorizes a alteration.
The script of background "personality" is as follows:
The handler "closeCard" passes a combination of a particle name and keywords to background "particlesBG". The handier "normalize" normalizes the card.
The script of button "back track" of background "personality" is as follows:
The script of background "particlesBG" is as follows:
The handler "openBackground" accepts the combination of a particle name and keywords from background "personality", creates the alteration and passes the alteration to background "classificationsBG".
The script of background "classificationsBG" is as follows:
In the following, relation A is a current relation and relation A has a link to relation B. The script comprises:
1. a handler "openBackground" which modifies link numbers and discovers the hierarchical structure of keywords;
2. a handler "alterateLinkNumber" which modifies link numbers;
3. a handler "discoverHierarchy" which discovers the hierarchical structure of keywords;
4. a handler "setLowerLinkOfHigher" which sets flags of relation B's "isLower" and "isIncluded" links;
5. a handler "setLinkFlag" which sets flags of a relation's link;
6. a function "isIncludedLink" which examines whether a link to relation B is an "isIncluded" link; p1 7. a handler "makeupCard" which makes-up cards of this background.
The name of first card of background "classificationsBG" is ".about.root.about.".
The script of field "links" of background "classificationsBG" is as follows:
The script comprises handlers which increment or decrement link numbers.
The script of field "others" of background "classificationsBG" is as follows:
The script of field "otherButton" of background "classificationsBG" is as follows:
The script displays field "others".
The script of field from "button1" to "button8" is as follows:
The script of button "backtrack" of background "classificationsBG" is as follows:
The script of field "new card" of background "classificationsBG" is as follows:
The invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The present embodiment is therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.
Also it is to be understood that the phraseology or terminology employed herein is for the purpose of description and not of limitation.
Claims
What is claimed is:
1. An apparatus for classifying information comprising: (a) means for generating a relation having the following elements each a keyword, (i) a keyword identifier, (ii) a total link number, (iii) links to information or other relations, (b) means for discovering a hierarchy structure of a total information and a total keyword by comparing the total link numbers of two relations with links with each other thereby making a high level relation a larger total link number, (c) means for memorizing said relations, and (d) means for memorizing said information.
2. An apparatus according to claim 1 wherein said means (b) for discovering a hierarchy structure discovers said total information and said total keyword by: (i) comparing the total link numbers of two relations with links with each other, thereby making a high level relation having a larger total link number, and (ii) making ineffective the link to said high level relation, said link being included in all of relations.
3. An apparatus according to claim 1, wherein said means (b) for discovering a hierarchy structure discovers said total information and said total keyword by: (i) comparing the total link numbers of two relations with links with each other, thereby making a high level relation having a larger total link number; (ii) making ineffective the link to said high level relation, said link being included in all of relations; (iii) making ineffective the p-th link L.sub.ip included in a relation R.sub.i under the following conditions; said link L.sub.ip being linked to a low level relation R.sub.j lower than said relation R.sub.i, the q-th link L.sub.iq included in said relation R.sub.i being linked to the relation R.sub.k lower than said relation R.sub.i and higher than said relation R.sub.j, and said relation R.sub.k having the link to be linked to said relation R.sub.j ; (iv) making ineffective the p-th link L.sub.ip included in said relation R.sub.i under following conditions; said link L.sub.ip being linked to an information I.sub.m, the q-th link L.sub.iq included in said relation R.sub.i being linked to the relation R.sub.k lower than said relation R.sub.i, said relation R.sub.k having the link to said information I.sub.m.