Method and apparatus for retrieving data

• US 6,138,116 A
• Filed: 07/31/1997
• Issued: 10/24/2000
• Est. Priority Date: 08/01/1996
• Status: Expired due to Term

• Current Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Original Assignee: Canon Kabushiki Kaisha (Canon Inc.), T Kitagawa, Y Kiyoki
• Inventors: Kitagawa, Takashi, Kiyoki, Yasushi
• Primary Examiner(s): Amsbury, Wayne
• Assistant Examiner(s): PARDO, THUY N
• Application Number:
  US08/904,149
• Time in Patent Office: 1,181 Days
• Field of Search: 707/3, 707/5, 707/103, 707/526, 707/101, 707/532, 706/62, 704/9
• US Class Current: 707/752
• CPC Class Codes: G06F 16/3332 Query translation
• CPC Class Codes: Y10S 707/915 Image
• CPC Class Codes: Y10S 707/99933 Query processing, i.e. searching
• CPC Class Codes: Y10S 707/99935 Query augmenting and refining, e.g. inexact access

Abstract

A semantic data processing apparatus for retrieving one or a plurality of words, having a close meaning to that of an inputted word, in consideration of context or condition, comprises: a communication apparatus 101 for inputting a keyword and context word group; a projective transformation generating module 104 for transforming the inputted keyword and context word group into a keyword vector and a context word vector group indicative of respective meanings, and generating a projector which projects the keyword and a comparison-subjected vector group into a subspace corresponding to the context word group, in a semantic space which is generated in correspondence with a predetermined space generation word, on the basis of the obtained context word vector group; an input/output apparatus 102 for inputting a comparison-subjected word group; a comparison-subjected word group mapping module 107 for transforming the inputted comparison-subjected word group into a comparison-subjected vector group indicative of meanings of a comparison-subjected word; and a basic processing module 103 for projecting the keyword vector and comparison-subjected vector group into the subspace by utilizing the projector, calculating a distance between each vector of the comparison-subjected vector group and the keyword vector in the subspace, and rearranging the comparison-subjected word group in the order of the distance to be outputted.

Claims

1. A semantic data processing apparatus comprising:

first inputting means for inputting a keyword for a retrieval and a context word group designating a field of the retrieval;

first transforming means for transforming the keyword inputted by said first inputting means into a keyword vector representing a meaning of the keyword in a semantic space which has axes corresponding to a predetermined characteristic word group;

second transforming means for transforming the context word group into a context word vector group by transforming each context word of the context word group inputted by said first inputting means into a vector representing a meaning of the context word in the semantic space;

projector generating means for generating a projector for projecting a vector in the semantic space into a subspace of the semantic space, the subspace comprising axes corresponding to characteristic words related to the context word group, based on the context word vector group obtained by said second transforming means;

second inputting means for inputting a comparison-subjected word group;

third transforming means for transforming the comparison-subjected word group into a comparison-subjected vector group by transforming each comparison-subjected word of the comparison-subjected word group inputted by said second inputting means into a vector representing a meaning of the comparison-subjected word in the semantic space;

projecting means for projecting the keyword vector and the comparison-subjected vector group in the subspace by utilizing the projector generated by said projector generating means;

calculating means for calculating a correlation amount in the subspace between each vector of the comparison-subjected vector group and the keyword vector; and

retrieving means for retrieving at least one word from the comparison-subjected word group based on the correlation amount calculated by said calculating means.

2. The semantic data processing apparatus according to claim 1, wherein said selecting means rearranges words of the comparison-subjected word group in the order of the correlation amount with respect to each word and outputs the rearranged words.

3. The semantic data processing apparatus according to claim 1, wherein the correlation amount is a distance, and said selecting means selects, from the comparison-subjected word group, a comparison-subjected word having the minimum distance with respect to each word.

4. The semantic data processing apparatus according to claim 1, wherein said projector generating means calculates a semantic center of the context word vector group and generates the projector on the basis of the semantic center.

5. The semantic data processing apparatus according to claim 1, wherein said first, second and third transforming means transform each word into a vector corresponding to a combination of words which are utilized to describe a meaning of said word, with reference to a dictionary where meanings of words are described by the predetermined characteristic word group.

6. The semantic data processing apparatus according to claim 1, further comprising:

fourth transforming means for transforming the space generation word group into a space generation vector group by transforming each space generation word of the space generating word group into a space generation vector which has elements corresponding to a predetermined characteristic word group to represent a meaning of the space generation word; and

semantic space generating means for generating a semantic space on the basis of the space generation vector group obtained by said fourth transforming means.

7. The semantic data processing apparatus according to claim 6, wherein said fourth transforming means transforms each word into a vector corresponding to a combination of words which are utilized to describe a meaning of said word, with reference to a dictionary where meanings of words are described by the predetermined characteristic word group.

8. The semantic data processing apparatus according to claim 6, wherein said semantic space generating means generates a correlation matrix on the basis of the space generation vector group, and creates a semantic space transformation matrix by subjecting the correlation matrix to singular-value decomposition,

and said first, second and third transforming means perform transforming into a vector in the semantic space by utilizing the semantic space transformation matrix.

9. The semantic data processing apparatus according to claim 1, wherein said projector generating means comprises composite vector calculating means for calculating a composite vector of the context word vector group, and generates the projector on the basis of the composite vector.

10. The semantic data processing apparatus according to claim 9, wherein said composite vector calculating means calculates a mean vector, where the context word vector group is normalized, as the composite vector.

11. The semantic data processing apparatus according to claim 9, wherein the context word group is accompanied by a first operator operating on each context word and a second operator operating on a pair of context words, wherein the first operator operates at the time of transforming operation by said second transforming means, and the second operator operates at the time of calculating the composite vector by said composite vector calculating means.

12. The semantic data processing apparatus according to claim 1, wherein said projector generating means independently generates a projector for projecting the keyword vector and a projector for projecting the comparison-subjected vector group.

13. The semantic data processing apparatus according to claim 1, further comprising:

a database where the comparison-subjected group is applicable as a keyword; and

retrieving means for retrieving a word from said database by utilizing, as a keyword, the comparison-subjected word selected by said selecting means.

14. The semantic data processing apparatus according to claim 13, wherein said database stores an image in correspondence with one or more keywords which represent contents or impression of the image.

15. The semantic data processing apparatus according to claim 14, further comprising correcting means for correcting, in accordance with a user, a keyword group inputted by a user in correspondence with a desired image by utilizing said first inputting means.

16. The semantic data processing apparatus according to claim 14, further comprising correction amount calculating means for calculating a correction amount to be corrected by said correcting means, on the basis of an image provided from said database and the keyword group selected by the user in correspondence with the image.

17. A semantic data processing method, comprising:

a first inputting step of inputting a keyword for a retrieval and a context word group designating a field of the retrieval;

a first transforming step of transforming the keyword inputted in said first inputting step into a keyword vector representing a meaning of the keyword in a semantic space, which has axes corresponding to a predetermined characteristic word group;

a second transforming step of transforming the context word group into a context word vector group by transforming each context word of the context word group inputted in said first inputting step into a vector representing a meaning of the context word in the semantic space;

a projector generating step of generating a projector for projecting a vector in the semantic space into a subspace of the semantic space, the subspace comprising axes corresponding to characteristic words related to the context word group, based on the context word vector group obtained in said second transforming step;

a second inputting step of inputting a comparison-subjected word group;

a third transforming step of transforming the comparison-subjected word group into the comparison-subjected vector group by transforming each comparison-subjected word of the comparison-subjected word group inputted in said second inputting step into a vector representing a meaning of the comparison-subjected word in the semantic space;

a projecting step of projecting the keyword vector and the comparison-subjected vector group in the subspace utilizing the projector generated in said projector generating step;

a calculating step of calculating a correlation mount in the subspace between each vector of the comparison-subjected vector group and the keyword vector; and

a retrieving step of retrieving at least one word from the comparison-subjected word group based on the correlation amount calculated in said calculating step.

18. The semantic data processing method according to claim 17, wherein in said selecting step, words of the comparison-subjected word group are rearranged in the order of the correlation amount with respect to each word and outputted.

19. The semantic data processing method according to claim 17, wherein the correlation amount is a distance, and in said selecting step, a comparison-subjected word having the minimum distance with respect to each word is selected from the comparison-subjected word group.

20. The semantic data processing method according to claim 17, wherein in said projector generating step, a semantic center of the context word vector group is calculated and the projector is generated on the basis of the semantic center.

21. The semantic data processing method according to claim 17, wherein in said first, second and third transforming steps, each word is transformed into a vector corresponding to a combination of words which are utilized to describe a meaning of said word, with reference to a dictionary where meanings of words are described by the predetermined characteristic word group.

22. The semantic data processing method according to claim 17, further comprising:

a fourth transforming step of transforming the space generation word group into a space generation vector group by transforming each space generation word of the space generating word group into a space generation vector which has elements corresponding to a predetermined characteristic word group to represent a meaning of the space generation word; and

a semantic space generating step of generating a semantic space on the basis of the space generation vector group obtained in said fourth transforming step.

23. The semantic data processing method according to claim 22, wherein in said fourth transforming step, each word is transformed into a vector corresponding to a combination of words which are utilized to describe a meaning of said word, with reference to a dictionary where meanings of words are described by the predetermined characteristic word group.

24. The semantic data processing method according to claim 22, wherein in said semantic space generating step, a correlation matrix is generated on the basis of the space generation vector group, and a semantic space transformation matrix is created by subjecting the correlation matrix to singular-value decomposition,

and in said first, second and third transforming steps, transforming into a vector in the semantic space is performed by utilizing the semantic space transformation matrix.

25. The semantic data processing method according to claim 17, wherein said projector generating step includes a composite vector calculating step of calculating a composite vector of the context word vector group, wherein the projector is generated on the basis of the composite vector.

26. The semantic data processing method according to claim 25, wherein in said composite vector calculating step, a mean vector, where the context word vector group is normalized, is calculated as the composite vector.

27. The semantic data processing method according to claim 25, wherein the context word group is accompanied by a first operator operating on each context word and a second operator operating on a pair of context words, wherein the first operator operates at the time of transforming operation in said second transforming step, and the second operator operates at the time of calculating the composite vector in said composite vector calculating step.

28. The semantic data processing method according to claim 17, wherein in said projector generating step, a projector for projecting the keyword vector and a projector for projecting the comparison-subjected vector group are independently generated.

29. The semantic data processing method according to claim 28, wherein in said second inputting step, a keyword, which is applicable as the comparison-subjected word in a predetermined database, is inputted,

said method further comprising a retrieving step of retrieving data from said database by utilizing, as a keyword, the comparison-subjected word selected in said selecting step.

30. The semantic data processing method according to claim 29, wherein said database stores an image in correspondence with one or more keywords which represent contents or impression of the image.

31. The semantic data processing method according to claim 30, further comprising a correcting step of correcting, in accordance with a user, a keyword group inputted by a user in correspondence with a desired image in said first inputting step.

32. The semantic data processing method according to claim 30, further comprising a correction amount calculating step of calculating a correction amount to be corrected in said correcting step, on the basis of an image provided from said database and the keyword group selected by the user in correspondence with the image.

33. A data retrieving method, comprising:

a projecting step of projecting a keyword vector representing a meaning of an inputted keyword for a retrieval in a semantic space and a comparison-subjected vector group representing a meaning of a comparison-subjected word group in the semantic space into a subspace of the semantic space by utilizing a predetermined projector; and

a calculating step of calculating a correlation amount of each vector of the projected keyword vector and the projected comparison-subjected vector group; and

a retrieving step of retrieving at least one word from the comparison-subjected word group based on a calculation result,

wherein the predetermined projector is a projector for projecting a vector in the semantic space, which comprises axes corresponding to a predetermined characteristic word group, into the subspace, which comprises axes corresponding to characteristic words related to an inputted context word group designating a field of the retrieval, and is generated based on a context word vector group corresponding to the context word group.

34. A data retrieving apparatus comprising transforming means for transforming an inputted keyword to an internal data keyword to perform data retrieval operation, the transforming means comprising:

projecting means for projecting a keyword vector representing a meaning of the inputted keyword and a comparison-subjected vector group representing a meaning of a comparison-subjected word group in the semantic space into a subspace of the semantic space by utilizing a predetermined projector;

calculating means for calculating a correlation amount of each vector of the projected keyword vector and the projected comparison-subjected vector group; and

selecting means for selecting a word as the data keyword, from the comparison-subjected word group based on a calculation result,

wherein the predetermined projector is a projector for projecting a vector in the semantic space, which comprises axes corresponding to a predetermined characteristic word group, into the subspace, which comprises axes corresponding to characteristic words related to an inputted context word group designating a field of the retrieval, and is generated based on a context word vector group corresponding to the context word group.

35. A product comprising a computer readable medium storing a program for retrieving data, the program comprising:

program code for projecting a keyword vector representing a meaning of an inputted keyword for a retrieval in a semantic space and a comparison-subjected vector group representing a meaning of a comparison-subjected word group in the semantic space into a subspace of the semantic space utilizing a predetermined projector, the predetermined projector being a projector for projecting a vector in the semantic space, which comprises axes corresponding to a predetermined characteristic word group, into the subspace, which comprises axes corresponding to characteristic words related to an inputted context word group designating a field of the retrieval;

program code for calculating a correlation amount of each vector of the projected keyword vector and the projected comparison-subjected vector group; and

program code for retrieving at least one word from the comparison-subjected word group based on a calculation result.

36. A product comprising a computer readable medium storing a program for processing semantic data, the program comprising:

program code for transforming an inputted keyword for a retrieval into a keyword vector representing a meaning of the keyword in a semantic space, which has axes corresponding to a predetermined characteristic word group, transforming an inputted context word group designating a field of the retrieval into a context word vector group representing a meaning of the context word group in the semantic space, and transforming an inputted comparison subjected word group into a comparison-subjected vector group representing a meaning of the comparison-subjected word group in the semantic space;

program code for generating a projector for projecting a vector in the semantic space into a subspace of the semantic space, the subspace comprising axes corresponding to characteristic words related to the context word group, based on a context word vector;

third program means for projecting the keyword vector and the comparison-subjected vector group into the subspace by utilizing the projector;

fourth program means for calculating a correlation amount in the subspace between each vector of the comparison-subjected vector group and the keyword vector; and

fifth program means for retrieving at least one word from the comparison-subjected word group based on a calculated correlation amount.