JP2003323454A - Method, device and computer program for mapping content having meta-information - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reflect precisely meta-information imparted beforehand to each content to a mapping result, when mapping the content in a low-dimensional space. <P>SOLUTION: When mapping the content to which the meta-information such as a classification system is imparted beforehand in a low-dimensional space such as a two-dimensional plane, concept vectors are allocated to each content in a high-dimensional mapping process, and then position information in the low-dimensional space is allocated to each content based on the concept vectors in a low-dimensional mapping process. In this case, the distance between the concept vectors is corrected so that, the nearer classified category information imparted beforehand to each content is, the nearer the position information in the low-dimensional space becomes, to thereby enable to be clustered so that a plurality of contents reflect the meta-information. <P>COPYRIGHT: (C)2004,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for mapping a large amount of content having meta-information to a low-dimensional space such as a two-dimensional plane based on the content similarity between the content and a computer. And a recording medium having the computer program recorded therein.

[0002]

2. Description of the Related Art The spread of the Internet and the advancement of database technology have made it possible to use large-scale contents (information such as images, videos, music, text information) on personal computers.

[0003] However, as available contents become large-scaled and diversified, it is becoming difficult to efficiently find the required contents.

It has been awaited to establish a technique for discovering necessary contents while browsing contents by a simple operation using an intuitively understandable user interface. In particular, a content provider who wants to improve the user interface for the user to select the content so as to promote the user's content consumption, needs an effective solution to this problem.

Conventionally, a technique for visually classifying and arranging a large amount of contents on a two-dimensional plane has been proposed (for example, see Non-Patent Document 1).

In this document, when the content is a text document, the content of the text document is quantified to extract a feature vector, and a multidimensional scale construction method is applied to the feature vector to determine the two-dimensional arrangement of the content and its distribution. Has realized a browsing interface using. This makes it possible to search for content while visually grasping the similarity of a large amount of content on a two-dimensional plane.

[0007]

[Patent Document 1] Japanese Patent Laid-Open No. 2001-36875

[Patent Document 2] Japanese Patent Laid-Open No. 2001-229167

[Patent Document 3] Japanese Unexamined Patent Publication No. 2002-175321

[Patent Document 4] Japanese Unexamined Patent Publication No. 2002-157269

[Non-Patent Document 1] James A. Wise, et al, “Visualizin
g the non-visual: Spatial analysis and interaction
with information from text documents ”Proc. of IE
EE Information Visualization '95, p.51-58 (1995)

[Non-Patent Document 2] Inoue and 2 others, "AssociaView", a related discovery type information browser, Proceedings of the 2000 IEICE General Conference, IEICE, A-16-46, 2000 3
Month

[Non-Patent Document 3] Fujita et al., “Audience Rating Analysis Based on Visualization of Content Space”, Proceedings of the 2001 IEICE General Conference, IEICE, D-8-7, 2001 March
Month

[Non-Patent Document 4] Abe and 4 others, "AssociaView, a related discovery type information browser", Interaction 2001 Proceedings, Information Processing Society of Japan, March 2001, p.75-82.

[0008]

However, when the classification system is added to the contents and the classification category is added to each content as meta information, it is effective to use this meta information for searching the contents. Is. If this meta information can be reflected in the two-dimensional layout of the content, it will be easier to search for the desired content.

For example, in the case of searching for a web page, a user can efficiently narrow down a target web page by using such a content mapping system in a directory service provided by many portal sites and the like. .

Therefore, the present invention extends the above conventional method on the assumption that the classification category information is added to the contents, and visually classifies the contents (information such as image, video, music and text information). It is an object of the present invention to provide a method, an apparatus, a computer program, and a computer-readable recording medium having the computer program recorded therein, which also reflects the structure of classification category information.

[0011]

In order to achieve this object, the present invention provides a content mapping method for mapping a plurality of contents, to which meta information including classification category information is added, to a low dimensional space. A high-dimensional mapping step of assigning a concept vector to each of the plurality of contents, and a low-dimensional mapping step of assigning position information in a low-dimensional space to each of the plurality of contents based on the concept vector are provided. The process provides a content mapping method characterized by mapping contents such that position information in a low-dimensional space becomes closer as classification category information becomes closer.

According to one aspect of the present invention, when a content to which meta information such as a classification system is added is mapped in a low-dimensional space such as a two-dimensional plane, a concept vector is assigned to each content in the high-dimensional mapping process. ,
Next, in the low-dimensional mapping process, each content is mapped to the low-dimensional space based on the concept vector. At this time, based on the classification category information given to each content in advance, the position information (arrangement coordinates) in the concept vector or the low-dimensional space is corrected to map the plurality of contents so as to reflect the classification category information. It becomes possible to do.

Particularly, when a classification system having a hierarchical structure as meta information is given to each content and each content is classified into one of the classification categories at the lowest level of the classification system, the plurality of contents In a content mapping method that maps a to a two-dimensional space such as a two-dimensional plane, the contents can be clustered for each classification category, and the structure of the classification system can be reflected in the scatter diagram.

In another aspect of the present invention, the low-dimensional mapping process further includes a distance calculation process for obtaining a distance between the concept vectors assigned to each content in the high-dimensional mapping process and the classification category information. A distance correction process of correcting the distance calculated in the distance calculation process to obtain a shorter effective distance, and a lower dimension of each of the plurality of contents based on the effective distance calculated in the distance correction process. It is characterized in that a position information allocation process for allocating position information in space is provided.

When the two contents for which the distance is to be calculated are classified in the same classification category in the first layer, the distance between the two contents is corrected so as to be smaller. When the two contents are classified into different classification categories in the first layer, the distance between the two contents is not corrected, or conversely, the distance between the two contents is corrected to be large. .

Further, when the two contents are classified into the same classification category in the first hierarchy and the two contents are classified into the same classification category in the second hierarchy, the distance between the two contents is further increased. Correct so that it becomes smaller. Even if the contents are classified into the same classification category in the first hierarchy, but are classified into different classification categories in the second hierarchy, the distance between the two contents is not corrected, or conversely, between the two contents. Correct so that the distance is larger. However, the correction such that the distance becomes larger is not performed as much as the correction in the case where the first level (the level immediately above) is classified into different classification categories.

As long as the classification categories of the two contents match, this operation is repeated to the lowest layer of the classification system to correct the distance between the two contents. The correction amount of the distance is determined according to how much the classification categories match from the first layer to the bottom layer of the classification system. As a result, the stronger the degree of matching of the classification categories is, the smaller the distance between the contents is corrected, and conversely, the weaker the degree of matching of the classification categories is, the larger the distance between the contents is corrected.

[0018] In another aspect of the present invention, the position information assigning step applies a multidimensional scaling method to the effective distance to assign position information to each of the plurality of contents. . In the multidimensional scale construction method, contents with a short distance in a multidimensional space (concept vector) are mapped with a small distance even in a low dimensional space such as a two-dimensional plane. It will be clustered.

In still another aspect of the present invention,
The low-dimensional mapping step further includes a distance calculation step of obtaining a distance between the concept vectors assigned to each content in the high-dimensional mapping step, and a plurality of contents based on the distance obtained in the distance calculation step. A position information allocating process for allocating position information in the low-dimensional space and a position information correcting process for correcting the position information allocated in the position information allocating process are closer to each other as the classification category information is closer to each other. Is characterized by.

In this case, after mapping the content to the low-dimensional space, the content can be clustered for each hierarchy of the classification system.

Other objects, features, and advantages of the present invention will be apparent from the following detailed description when read with the accompanying drawings.

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[First Embodiment] FIG. 1 shows that a content mapping method according to a first embodiment of the present invention is implemented, and as a result, a scatter diagram of content created and modified by a content provider is provided to a user. FIG. 3 is a diagram showing a computer network that is used.

The computer network shown in FIG.
It consists of system administrators, content providers, and users who are interconnected by networks such as the Internet.

The system administrator is provided with a server (computer) 10 for executing the content mapping method according to the present invention and various databases used by the server 10. This will be described in detail later.

The content provider is provided with a content delivery server, and delivers the content (information such as images, videos, music, text information) stored in the content delivery server to the user. Further, the content provider has meta information (for example, a classification category of each content) of the content distributed by itself, and provides this meta information to the meta information database managed by the system administrator via the network.

The content provider is also the server 1 of the system administrator via the network.
0 can be accessed to create and modify scatter plots using the content mapping method of the present invention.

The user can access the server 10 of the system administrator via the network, display the scatter diagram image created and modified by the content provider on the user terminal, and search for the desired content. Further, if the target content is found as a result of the search, the content can be distributed from the content distribution server of the content provider.

FIG. 2 is a diagram showing the configuration of a content mapping system 10 for implementing the content mapping method according to the first embodiment of the present invention. Content providers can access this content mapping system 10 to create and modify scatter plots of content. The content mapping method according to the present invention is
This process is mainly carried out. The user can search the content while viewing the scatter diagram obtained as a result of this execution on the user terminal. However, the content mapping method according to the present invention is not limited to this, and may be performed while the user searches for content. It can also be implemented by the system administrator as a service to content providers.

The content mapping system shown in FIG. 2 includes a computer 10, a content database (content DB) 20 connected to the computer 10 via a network 30, a meta information database (meta information DB) 21, and a concept vector database. It is composed of a (concept vector DB) 22 and a position information database (position information DB or placement coordinate DB) 23.

FIG. 3 is a block diagram showing an example of a typical hardware configuration of the computer 10 that executes the content mapping method according to this embodiment.

The computer 10 is composed of a RAM, a ROM, a memory such as a magnetic disk, a CPU, a display unit 11 including a display, an instruction input unit 12 including a mouse and a keyboard, a network interface unit and the like. Concept vector calculation processing unit 4 realized by a software program executed by the CPU
A one- and two-dimensional coordinate calculation processing unit 42, a scatter diagram image creation processing unit 43, and a scatter diagram image communication processing unit 44 are provided.

In FIG. 2, the contents DB 20 contains
The content to be processed and its brief description are stored. Storing the content in the content DB 20 is not essential and may be acquired from the content distribution server. Then, it is also possible to store the outline description of each content in the meta information DB 21 and not have the content DB 20.

The contents include, but are not limited to, images, videos, music, text information and the like.
In the following, a case where the content is a knowledge system-related document (text information) will be described as an example. The description is the same for contents other than text information.

Further, as shown in FIG. 4, the meta information DB 21 stores classification category information for classifying each content stored in the content DB 20. The information of this classification category is given to each content according to the content classification system given in advance.

In this embodiment, the classification category information has a depth 2 hierarchical structure.

FIG. 5 shows an example of a classification category system for classifying contents.

When this classification system is followed, the contents DB
Each content stored in 20 has L shown in FIG.
Any appropriate classification category of ij (i, j = 1,2,3) is assigned in advance, and information on the assigned classification category is stored in the meta information DB 21.

In the concept vector DB 22 shown in FIG. 2, the concept vector for each content stored in the content DB 20 is stored as shown in FIG. 6 by the processing described below.

Further, in the position information DB 23 (arrangement coordinate DB) shown in FIG. 2, the position information in the two-dimensional plane of each content is stored as shown in FIG. 7, for example, by the processing described below. It

As described above, FIG. 5 shows an example of a classification system for classifying contents. This is a classification system with knowledge system related literature. The documents stored in the content DB 20 are Lij (i, j = 1, 2, 3) in FIG.
Are assigned to any appropriate classification category in advance, and the information of the assigned classification category is stored in the meta information DB 21. In addition, meta information DB2
When the outline description of each content is stored in 1, the present invention can be implemented in a configuration without the content DB 20.

The concept vector calculation processing unit 41 is a processing means for calculating concept vector data that quantifies the concept from the outline description of each content in the content DB 20. The two-dimensional coordinate calculation processing unit 42 uses the concept vector calculated by the concept vector calculation processing unit 41 and the classification category information of the contents stored in the meta information DB 21 to calculate the distance between the contents for all combinations of two contents. Is a processing means for calculating the position information of each content on the two-dimensional plane. The scatter diagram image creation processing unit 43 is a processing unit that creates a scatter diagram image of the content based on the two-dimensional coordinate information of each content calculated by the two-dimensional coordinate calculation processing unit 42. Scatter diagram image communication processing unit 44
Is a processing means for transmitting the scatter diagram image created by the scatter diagram image creating processing unit 43 to each terminal through the network in order to present it to the content provider and the user.

FIG. 8 is a flowchart of the content mapping system shown in FIG. Hereinafter, the processing of each processing unit will be described in detail.

The concept vector calculation processing unit 41 reads out the outline explanatory text of each content stored in the content DB 20 to the memory (step S1), and calculates the concept vector from the outline explanatory text. The concept vector is represented as a multidimensional real-valued vector. The calculated concept vector data is stored in the concept vector DB 22 in association with each content (step S2).

Here, the concept vector is a vocabulary regarded as a vector of a multidimensional space having many semantic categories as attributes, and each vocabulary has a degree of association (weight) with many semantic categories as a component. It is expressed as a multidimensional vector. Semantic categories include, for example, creatures, animals, plants, machines, metrology, vehicles, entertainment.
For example, in the vocabulary “horse”, 0.5 creatures, 0.3 animals,
Components such as vehicle 0.3 and entertainment 0.1 are assigned. Also,
For the "car" vocabulary, components such as machine 0.5, vehicle 0.3, entertainment 0.2 are assigned. The concept vector is generally normalized, but it does not have to be.

When assigning a concept vector to the outline sentence of the content, for example, the concept vector of the outline sentence can be generated by averaging the concept vectors of the vocabulary contained in the outline sentence and normalizing to 1.

The method of calculating the concept vector from the outline explanation is described in detail in the following document, so that further explanation is omitted. Mutsumi Kumamoto et al., "Application of Concept-Based Information Retrieval-Characteristic Evaluation of Retrieval Using Concept-Base", IEICE Technical Report AI98-63 (1999).

In the two-dimensional coordinate calculation processing unit 42, the concept vector of each content stored in the concept vector DB 22 is read into the memory, and the classification category information of the content stored in the meta information DB 21 is read into the memory. (Step S3), distances between contents are calculated for all combinations of two contents included in the contents to be displayed (step S3).
4). Here, the combination of all two contents does not necessarily have to be all the contents stored in the content DB 20, and, for example, when the display target is narrowed down in advance by a search condition or the like, its display is displayed. It means all combinations of two contents that can be extracted from the target contents.

C_iAnd c_jBe any two contents
Then, c_iAnd c_jEffective distance between “disth *_ijIs calculated by the following formula
Issued: disth *_ij = disth_ij+ u_ij Where “disth_ij"Is content c_iAnd c_jThe concept of
Represents the Euclidean distance between tolls. Also u_ijIs the storyboard
C_iAnd c_jCategory determined by the category
It is the inter-distance (correction amount).

The inter-category distance u _ij is u _ij = 0 (if it matches up to the second hierarchy) u _ij = A (the first hierarchy) depending on to which hierarchical level of the classification system the classification categories of the contents c _i and c _j match. _U1 = B (when the first layer does not match, but the second layer does not match) u _ij = B (when the first layer does not match) where A and B are non-negative real values that satisfy 0 <A <B.

In FIG. 5, for example, when c _i is classified into L21 and c _j is also classified into L21, u _ij = 0. Also, c _i is L21 and c _j is
When classified in L22, u _ij = A. c _i is L21 and c _j is L
When classified into 31, u _ij = B.

Given the classification system of FIG. 5, in general, the classification categories for content c _i and c _j , respectively.
Given that Lmn and Lkl (m, n, k, l = 1,2,3) are assigned, it can be expressed as: u _ij = 0 (if m = k and n = l) u _ij = A (if m = k and n ≠ l) u _ij = B (if m ≠ k) The above is the calculation method of the inter-category distance when the depth of the hierarchy of the classification system is 2. Also in the case of N (N is a positive integer), the same definition as above is made based on the positional relationship in the classification system of the two classification categories.

Then, the position information of each content on the two-dimensional plane is calculated by the multidimensional scale construction method (step S5). The multidimensional scaling method is a compression algorithm from a high-dimensional vector space to a low-dimensional space, and can be solved as a minimization problem of the objective function shown below.

[0053]

[Equation 1] That is, give the minimum value of this objective function (x_i, y_i )
A set of (i = 1, 2, ..., N) is 2 for each content.
It is position information on the dimensional plane. However, the above objective function
Then, "dist2_ij"Is dist2_ij  = {(X_i −x_j )²+ (Y_i −y_j )²}^1/2 And n is the total number of contents to be displayed.
Represent The minimization problem of the above objective function is the so-called steepest descent.
Method, which is detailed in the following document.
Since it has been described, further explanation is omitted. J. W. Sammon. A nonlinear mapping for data structu
re analysis. IEEE Transactions on Computers, C-18
(5): 401-409 (1969).

As a result of the above calculation, (x _i , y _i ) (i = 1,
2, ..., N) are recorded in the position information DB 23. The above-described position information calculation processing is not simply given as the distance between contents by the distance between concept vectors, but is defined by considering dissimilarity between classification categories into which contents are classified, and thus multidimensional scale construction is performed. The feature is that the classification result information of the content can be included in the result of the law.

As a result, it is possible to obtain the effect that the contents included in the same classification category are arranged close to each other so as to form a group on the two-dimensional plane. In the example of FIG. 5, Li (i = 1, 2, 3) is arranged so as to form a group, respectively, and further inside the group of Li contents, Lij (i, j = 1, 2, 3) The contents included in are arranged so as to form a group.

The scatter diagram image creation processing unit 43 reads the two-dimensional coordinate information of each content stored in the position information DB 23 into the memory and creates a scatter diagram image of the content to be presented to the operator based on this. (Step S6). The scatter diagram image communication processing unit 44 transmits the scatter diagram image of the created content in response to a request from the operator terminal (step S7).

It is preferable that the scatter diagram image can be scaled up and down in an arbitrary range designated by the operator. Further, it is desirable that the group of contents on the scatter diagram image can be labeled with the category name of the classification category corresponding to the group. Furthermore, the information to be displayed on the label should be selectable according to the scale of the scatter plot image. If the scale is small, only the Li-level category name is displayed and the operator performs input operation to scale. When is set to some extent, it is desirable to be able to label Lij level information in addition to Li level information.

That is, the scatter diagram image creation processing unit 43
It has a processing function of continuously enlarging / reducing the scale of the scatter diagram image based on the operator's input operation, and a processing function of selecting various meta information to be displayed according to the scale of the scatter diagram image. As a method of selecting the meta information to be displayed according to the scale of the scatter plot image, it is possible to prepare a table in which the types of meta information to be displayed are stored in advance corresponding to the scale ratio, and to use that method. it can.

FIG. 9 shows an example of a scatter diagram image when the scale is small. The scatterplot image in FIG. 9 shows the classification of Li-level contents related to knowledge system-related documents. In the scatter diagram image of FIG. 9, by designating the rectangular area 50 and operating the enlargement / reduction operation knob 51 with a pointing device such as a mouse, a specific portion of the scatter diagram image can be enlarged and displayed. Instead of designating the rectangular area 50, a method of designating the center point of enlargement or reduction can be used.

FIG. 10 shows an example of a scatter diagram image when the rectangular area 50 of FIG. 9 is enlarged and displayed. In the scatter diagram image of FIG. 10, in the literature related to the knowledge system, particularly the category of pattern information processing (L3), the classification up to the Lij level is shown. Here, an example is shown in which the category name is displayed as a label, but various meta information other than the category name can be displayed, and the type of meta information to be displayed can be selected by the operator of the content provider from a menu or the like. You may

A scroll bar may be provided in the vertical and horizontal directions of the screen of the scatter diagram image shown in FIG. 9 or 10 so that the screen can be scrolled.

When the operation of further enlarging the scatter diagram image of FIG. 10 is performed and the scatter diagram image is in the most enlarged state, the scatter diagram image in which the individual contents are arranged on the two-dimensional plane is displayed. As the meta information of each content, a content name (reference name), an author name, and other information are displayed as a label based on preset information.

[Second Embodiment] This embodiment is a step further than the first embodiment, and even if the classification category information has a deep hierarchical structure, the classification category is used to calculate the distance between concept vectors. It is an object of the present invention to provide a new content mapping technique that enables information to be reflected easily.

Since the configuration of the content mapping system according to the second embodiment is almost the same as the configuration and operation of the content mapping system according to the first embodiment shown in FIG. 2, only the differences will be described in detail here. A description of the same points will be omitted.

The content mapping system according to the second embodiment is different from the content mapping system according to the first embodiment in the calculation process of the distance between contents executed in step S4 of the operation shown in the flowchart of FIG.

The classification category information according to this embodiment is
It is assumed to have a depth N hierarchical structure. N = in Figure 11
The case of 3 is shown. Each content is in the first layer
As a classification Li₁    : First level classification category Where i₁= 1, ...., M (M is a positive integer)
Tegoli Li₁Content classified as is in the second layer
As a classification Li₁i₂  : Li₁Second-level classification with parent category
Category Where i₂= I₂(i₁) = 1, ...., Mi₁ Is classified as one of (Mi₁Is a positive integer).

Similarly, a classification function in the (k-1) th layer is used.
Tegoli Li₁i₂i₃.... i_(k-1)Content classified as
As the classification in the kth layer, Li₁i₂i₃.... i_k  : Li₁i₂i₃.... i_(k-1)The parent category
-K category classification Where i_k= I_k(i₁, i₂, ...., i_(k-1)) = 1, ...., Mi₁i
₂i₃.... i_(k-1) It is classified as one of (Mi₁i₂i₃.... i_(k-1)Is positive
), And so on until k = N.

Then, as shown in FIG. 4, the information of the classification categories Li ₁ i ₂ i ₃ .... i _N in the Nth layer is stored in the meta information DB 21 as the above-mentioned classification category information. However, FIG. 4 shows a case where N = 2.

In the classification system shown in FIG. 11, all the branches of each classification category have the same number (3) of hierarchies for simplification of description. It goes without saying that the method can be applied even when the numbers are different. It is also applicable when the number of classification categories belonging to each branch is different. In this embodiment, the number of classification categories is set to 3 for both the first layer and the second layer in all the branches.

FIG. 11 shows the calculation process of the distance between contents executed in step S4 of the flowchart of FIG.
If we explain concretely based on the classification system shown in,
The content mapping system, as the distance matrix of the first hierarchy of the classification category, corresponds to that the first hierarchy of the classification category is composed of three classes. For example, as shown in FIG. A component having A ₁ (A ₁ <1) showing a value smaller than 1 and a non-diagonal component having B ₁ (B ₁ ≧ 1) showing a value larger than ₁ are used.

That is, when the first hierarchy of the classification categories of two contents is in the same category, A ₁ (A ₁ <1) showing a value smaller than 1 is assigned, and when not in the same category, it is larger than 1. B indicating the value
The distance matrix of the first hierarchy to which ₁ (B ₁ ≧ 1) is to be assigned is used.

Further, as the distance matrix of the second hierarchy of the classification category, corresponding to the fact that the second hierarchy of the classification category is composed of three classes, for example, as shown in FIG. A indicating a value smaller than 1 for the component
Those having ₂ (A ₂ <1) and B ₂ (B ₂ ≧ 1) showing a value larger than 1 for the non-diagonal component are used. However, B ₂ is set so that B ₂ <= B ₁ .

That is, when the second hierarchy of the classification categories of two contents is in the same category, A ₂ (A ₂ <1) showing a value smaller than 1 is assigned, and when not in the same category, it is larger than 1. B indicating the value
_This is because the distance matrix of the second hierarchy to which ₂ (B ₂ ≧ 1) is assigned is used.

Further, as the distance matrix of the third hierarchy of the classification category, corresponding to the fact that the third hierarchy of the classification category is composed of three classes, for example, as shown in FIG. A indicating a value smaller than 1 for the component
Those having ₃ (A ₃ <1) and B ₃ (B ₃ ≧ 1) showing a value larger than 1 for the non-diagonal component are used. However, B ₃ is set so that B ₃ <= B ₂ .

That is, when the third hierarchy of the classification categories of two contents is in the same category, A ₃ (A ₃ <1) showing a value smaller than 1 is assigned, and when not in the same category, it is larger than 1. B indicating the value
_This is because the distance matrix of the third layer to which ₃ (B ₃ ≧ 1) is to be assigned is used.

[0076] Content mapping system, with the correction coefficient w _{i j} specified by the distance matrix, among concepts vector calculated and concepts vector v _i of the content c _i, by the concept vector v _j of the content c _j Euclidean distance “disth _ij ” is corrected according to the calculation formula “w _{i j} × disth _ij ” (effective distance), so that the classification category information is reflected in the calculation of the distance between concept vectors. ing.

Correction coefficient w used at this time_ijIs, for example,
As shown in FIG. 13, two content classification categories
-Matches the levels of the first and second hierarchy,
Depends on whether or not the third hierarchy of classification categories matches
hand, w_ij= A₁× A₂× A₃(Match up to the third layer) w_ij= A₁× A₂× B₃(The third layer does not match) And the two content categories are on the first floor
Matches up to the level of the layer and at the level of the second layer
When not, w_ij= A₁× B₂ And the two content categories are on the first floor
When there is no match at the layer level, w_ij= B₁ I try to calculate.

When the correction coefficient w _ij calculated in this way is used, the more the classification categories of the two contents match at the deeper hierarchical level, the more the two calculation formulas “w _ij × disth _ij ” are calculated. The effective distance “disth * _ij ” of the content will be corrected to be smaller.

FIG. 14 shows a detailed flowchart of the calculation process of the effective distance between contents, which is executed in step S4 of the flowchart of FIG.

In executing the processing described in this flow chart, first, a distance matrix considering the distance between the classification categories is constructed in the memory as follows.

First, the classification category Li ₁ (i ₁ =
The distance matrix D [NULL] for 1, ..., M) is defined by D [NULL]: = S [NULL]. However, S [NULL] is defined as an M-th order non-negative symmetric matrix, has A (A <1) indicating a value smaller than 1 for diagonal elements, and 1 for non-diagonal elements. So that B also has a large value (B> = 1).

Next, all the classification categories L in the first layer
For i ₁ , the classification category Li of the second layer immediately below Li _1
The distance matrix D [Li ₁ ] for ₁ i ₂ (i ₂ = i ₂ (i ₁ ) = 1,…, Mi ₁ ) is transformed into D [Li ₁ ]: = D [NULL] (i ₁ , i ₁ ) XS [ Li ₁ ]. However, S [Li ₁ ] is defined as a Mi _1st order non-negative symmetric matrix, and the diagonal component is 1
It has Ai ₁ (Ai ₁ <1) showing a smaller value and Bi ₁ (Bi ₁ > = 1) showing a value larger than 1 for non-diagonal components. Note that Bi ₁ is set to satisfy Bi ₁ <= B. Further, D [NULL] (i ₁ , i ₁ ) is the (i ₁ , i ₁ ) element of D [NULL].

Similarly, for all classification categories Li ₁ ... i _k in the k-th layer (2 <= k <= N-1), the classification category Li in the (k + 1) -th layer immediately below Li ₁ ... i _{k 1} … i _k i _{(k + 1)} (i
_{(k + 1)} = i _{(k + 1)} (i ₁ , ..., i _k ) = 1, ... Mi ₁ ... i _k ), the distance matrix D [Li ₁ … i _k ] is _converted into D [Li ₁ .. .i _k ]: = D [Li ₁ ... i _(k-1) ] (i _k , i _k ) XS [L
i ₁ ... i _k ]. However, S [Li ₁ ... i _k ] is defined as a Mi ₁ ... i _k-th order non-negative symmetric matrix, and the diagonal component shows a value smaller than ₁ Ai ₁ ... i _k (Ai ₁ ... Bi ₁ with i _k <1) and greater than 1 for off-diagonal components
... i _k (Bi ₁ ... i _k > = 1). Note that Bi ₁ ... i _k is B
i ₁ … i _k <= Bi ₁ … i _(k-1) is satisfied. In addition, D [Li
_{1 ... i (k-1)} ] (i k, i k) is the (i _k, i _k) element of _{D [Li 1 ... i (k} -1)].

Next, using the above distance matrix, any two
Content of c_iAnd c_jEffective distance between “disth *_ij”
Calculated by the following formula and recorded in memory: disth *_ij   : = w_ij X disth_ij Where “disth_ij"Is c_i, C_jConcept vector between you
Represents the crid distance, w_ijIs c_i Classification category is Li₁
... i_N, C_jClassification category is Lj₁… J_NWhen, (1) w_ij  = D [NULL] (i₁, j₁) if i₁  ≠
  j₁ (2) w_ij  = D [Li₁... i_(k-1)] (I_k, j_k) Otherw
ise Given as. However, k is two classification categories
-An integer indicating the matching hierarchical levels of Li1 ... iN and Lj1 ... jN
Is a number. That is, if Li1 ... iN and Lj1 ... jN are the same,
If k = N and Li1 ... iN and Lj1 ... jN are different
Satisfies i1 = j1, ..., i (k-1) = j (k-1), ik ≠ jk
It is an integer. That is, the equal sign holds up to the k-1th
And the k-th is different for the first time 2< k< An integer of N
is there.

Then, the effective distance "disth * _ij " in consideration of the distance between the classification categories is calculated for all combinations of contents.

Next, according to the flow chart of FIG.
The calculation process of the distance between contents to be executed in step S4 of the flowchart of FIG. 8 will be specifically described.

The content mapping system uses the content c _i for the multidimensional scaling method application processing described above.
When calculating the distance between the content and the content c _j ,
First, as shown in the flowchart of FIG.
1, and concepts vector v _i of the content c _i, content
Euclidean distance "di between the concepts vector v _j of c _j
sth _ij "is calculated.

Subsequently, in step S12, the variable k indicating the hierarchy level of the classification category is set to "1" indicating the first hierarchy.
Set.

Succeedingly, in a step S13, the contents c
The first-level classification category value of the classification category Lp of _i and the first-class classification category value of the classification category Lq of the content c _j are specified.

Subsequently, in step S14, the component value of the distance matrix of the first hierarchy pointed to by the specified classification category value is specified. That is, by referring to the distance matrix prepared in association with the first hierarchy of the classification category defined by the matrix shown in FIG. 12A, the component value indicated by the specified classification category value (see FIG. In the case of the example of (a), A ₁ or B ₁ ) is specified.

Then, in step S15, the specified component value is substituted into the variable w. Then, step S16
Then, it is determined whether or not the two classification category values identified in step S13 match, and when it is determined that they do not match, the process proceeds to step S17, and the value of the variable w and the distance “disth calculated in step S11 are calculated. The distance between the content c _i and the content c _j is calculated by multiplying by _ij ″, and the process ends.

On the other hand, in step S16, step S13
When it is determined that the two classification category values identified in step S18 match, the process proceeds to step S18, the value of the variable k is incremented by one, and in the subsequent step S19, the value of the variable k is determined from the depth N of the classification category. Also determines whether or not it has grown.

When it is determined by this determination process that the value of the variable k is not greater than the depth N of the classified category, the process proceeds to step S21 and the content c
The classification category value of the k-th hierarchy of the classification category Lp of _{i and} the classification category value of the k-th hierarchy of the classification category Lq of the content c _j are specified.

Then, in step S22, the component value of the distance matrix of the k-th layer pointed to by the specified classification category value is specified. That is, when k = 2, FIG.
By referring to the distance matrix prepared in association with the second layer of the classification category defined by the matrix as shown in
The component value indicated by the specified classification category value (see FIG. 12).
In the case of the example of (b), A ₂ or B ₂ ) is specified.

Subsequently, in step S23, the specified component value is multiplied by the value of the variable w, and the multiplication result is substituted into the variable w as the value of the new variable w. Succeedingly, in a step S24, it is determined whether or not the two classification category values specified in the step S21 are coincident with each other, and when it is determined that they are not coincident with each other, the process proceeds to a step S25, and the value of the variable w is compared with the step S11. The distance between the content c _i and the content c _j is calculated by multiplying the calculated distance “disth _ij ”, and the processing is ended.

On the other hand, in step S24, step S22
When it is determined that the two classification category values specified in step 3 match, the process returns to step S18 so as to proceed to the classification category at the next lower hierarchical level.

When it is determined in step S19 that the value of the variable k is larger than the depth N of the classification category by repeating the processing of steps S18 to S24, the process proceeds to step S20. The effective distance between the content c _i and the content c _j is calculated by multiplying the value of the variable w by the distance “disth _ij ” calculated in step S11, and the process ends.

In this way, the content mapping system
For example, as shown in FIG. 13, the stem has two components.
The classification category of Tenz is the level of the first and second hierarchy.
If it matches with, the third level of the classification category matches
Correction coefficient w depending on whether or not_ijTo w_ij= A₁× A₂× A₃(Match up to the third layer) w_ij= A₁× A₂× B₃(The third layer does not match) And the two content categories are on the first floor
Matches up to the level of the layer and at the level of the second layer
When not, w_ij= A₁× B₂ And the two content categories are on the first floor
When there is no match at the layer level, w_ij= B₁ To calculate the
Positive coefficient w_ijAnd the Euclidean distance “dis
th_ijBy multiplying with ", the content c_iAnd content
Tsu c_jProcessing to calculate the distance between
It

When the correction coefficient w _ij calculated in this way is used, the more the classification categories of the two contents are matched to the deeper hierarchical level, the more the two calculated according to the calculation formula “w _ij × disth _ij ”. It will be corrected so that the effective distance of the content becomes smaller.

Although the present invention has been described with reference to the illustrated exemplary embodiments, the present invention is not limited thereto. For example,
In the embodiment, the distance “disth _ij ” between the concept vectors is multiplied by the correction coefficient w _ij to obtain the distance “disth _ij ”.
However, the correction amount may be determined using such a correction coefficient, and the correction amount may be added or subtracted to perform the correction, as in the first embodiment. is there.

For example, when the correction coefficient is determined as described above, the correction coefficient to be added to the distance between concept vectors is determined by multiplying the determined correction coefficient by a predetermined correction amount, and the addition is performed. By performing the correction, the distance between the concept vectors is corrected (the smaller the classification category is, the smaller the distance is corrected), the correction coefficient A _i described above is made larger than 1, and the correction coefficient B _{i is set} to 1. The correction coefficient is determined in the above-described manner by making it smaller than the above, and the correction amount to be subtracted from the distance between concept vectors is determined by multiplying the determined correction coefficient by a predetermined correction amount, and the subtraction is performed. May be used to correct the distance between concept vectors (the closer the classification categories match, the smaller the distance will be corrected).

Further, in the embodiment, the distance matrices prepared in association with the same hierarchical level of the classification category are all the same. However, for example, the distance matrix directly below the classification category value L1 is used. The distance matrix of the second layer and the distance matrix of the second layer immediately below the classification category value L2 may be changed by changing the magnitude of the component value.

[Third Embodiment] In the first and second embodiments described above, after correcting the distance between concept vectors corresponding to contents, each concept vector is converted into a two-dimensional plane by the multidimensional scaling method. Mapped to. On the other hand, before correcting the distance between the concept vectors, each concept vector is mapped on the two-dimensional plane by the multidimensional scaling method,
Then, the position information on the two-dimensional plane may be modified to reflect the classification category. The details will be described below.

FIG. 15 is a diagram showing a content mapping system according to the third embodiment of the present invention. Although it is almost the same as the content mapping system according to the first embodiment shown in FIG. 2, it is different in that a two-dimensional coordinate correction processing unit 45 is added.

The operator of the content provider is
When the scatter diagram data sent from the scatter diagram image communication processing unit 44 is displayed on the operator terminal and the contents are not clustered so as to reflect the classification category in each hierarchy of the classification system, the two-dimensional coordinate correction processing unit 45 can be activated to correct the location information of the content.

FIG. 16 is a flowchart showing the processing executed by the content mapping system according to the third embodiment shown in FIG. Each operation will be described below according to this flowchart.

The concept vector calculation processing unit 41 reads out the outline explanatory text of each content stored in the content DB 20 into the memory (step S1), and calculates the concept vector from the outline explanatory text. This concept vector is represented as a multidimensional real-valued vector. As described above, the calculated concept vector is stored in the concept vector DB 22 in association with each content as shown in FIG. 6 (step S2).

In the two-dimensional coordinate calculation processing unit 42, the concept vector of each content stored in the concept vector DB 22 is read into the memory, and the classification category information of each content stored in the meta information DB 21 is read into the memory ( In step S3), the distance between the contents is calculated for all combinations of the two contents included in the contents to be displayed (step S4).

It should be noted that all combinations of two contents do not necessarily have to be all contents stored in the contents DB 20, and, for example, when the display target is narrowed down in advance by search conditions or the like, It means all combinations of two contents that can be extracted from the contents to be displayed.

In the case of the present embodiment, the distance "disth _ij " between the contents c _i and c _j is set as "disth * _ij " without any correction, and the process proceeds to the next step S5.

Then, in the two-dimensional coordinate calculation processing section 42,
The position information of each content on the two-dimensional plane is calculated by the multidimensional scale construction method (step S5).

(X _i , y _i ) (i = 1,
2, ..., N) are recorded in the position information DB 23 as shown in FIG.

The position information of the content thus obtained does not reflect the meta information such as the classification system, but is in the state of being mapped only by the distance between the concept vectors corresponding to the content.

In the scatter diagram image creation processing section 43, the position information in the two-dimensional plane of each content stored in the position information DB 23 is read out into the memory, and the scatter diagram image of the contents to be presented to the operator based on this is read out. Is created (step S6). The scatter diagram image communication processing unit 44 transmits the scatter diagram image of the created content to the display unit of the operator terminal (step S7).

The operator browses the scatter diagram image displayed on the display unit of the operator terminal and confirms whether or not the classification / category is separated and clustered (step S8). If clustered, the process ends (No branch of step S8). The state of being clustered is specifically explained on the premise of the classification system shown in FIG. 5. In the classification categories of the first hierarchy and the second hierarchy of the classification system, for example, as shown in FIGS. , Content belonging to the same classification category is nearby,
The contents belonging to different classification categories are far apart.

On the other hand, in the scatter diagram image displayed on the display unit of the operator terminal, when the classification category of a certain hierarchy is not clustered (for example, FIGS. 17 and 22).
State), the process proceeds to step S9 (Yes in step S8).
branch).

In FIG. 17, an example in which the boundaries of the first-level classification category artificial intelligence (L1), computational linguistics (L2), and pattern information processing (L3) are vague and are not separated or clustered into classification category units Indicates. In addition,
In FIG. 17, reference numeral 1 denotes content whose classification category is L1, 2 denotes content L2, and 3 denotes content L3.

In the content mapping system of this embodiment, the operator can operate the operator terminal to cluster the scatter plots that are not clustered.

The operator operates the operator terminal, for example, by viewing the non-clustered scatter diagram screen shown in FIG. 17, and specifies the classification category to be clustered. FIG. 18 shows the case where the operator designates the classification category L3. According to the operator's designation, the content mapping system according to the present embodiment encloses the range in which the content belonging to the designated classification category L3 is distributed with a figure such as an ellipse in the scatter diagram image displayed on the operator terminal. At the same time, the center of gravity of the distribution on the two-dimensional plane is calculated and displayed by ◆. Further, the distance between each content and the center of gravity ◆ is reduced at a constant rate so that the contents belonging to the designated classification category L3 gather around the center of gravity ◆.

The operator can input this reduction ratio from the operator terminal.

FIG. 19 shows a scatter diagram image after being corrected by the operator. Note that the dotted ellipse indicates the existing range of the target content before the correction, but it is not actually displayed on the operator terminal.

From the scatter diagram image of FIG. 19, it can be seen that the contents of the classification categories L1 and L2 are included in the ellipse representing the cluster of the classification category L3, and the clustering is not sufficiently performed. In this case, the operator can operate the operator terminal to further correct the scatter plot image.

Next, in step S9, since the L3 classification category is separated from the other classification categories, the entire contents belonging to the designated classification category can be translated in a certain direction by a certain distance. The direction and distance of translation can be entered by the operator from the operator terminal. This is shown in FIG. The center of gravity of the designated classification category L3 moves as indicated by the arrow. Note that the dotted ellipse indicates the existing range of the target content before correction.

By repeating the processing from steps S6 to S9, the classification categories L1, L2, L3 of the first hierarchy can be clustered as shown in FIG.

The operator further selects the classification category L3 of the first hierarchy clustered as described above (see FIG. 22), and the contents belonging to the designated classification category L3 are classified in the second hierarchy in the same manner as above. L
31, L32, L33 can be clustered (see FIG. 23). Similarly, for other classification categories, it is possible to cluster each content.

In FIGS. 22 and 23, 31 indicates contents whose classification category is L31, 32 indicates contents of L32, and 33 indicates contents of L33.

The above process may be applied from any hierarchical classification category. Further, in the above example of L3, the repulsive force process of translational movement is performed after the attractive force process to the center of gravity, but this order may be arbitrary. Further, only one of the processes may be performed.

As described above, in the first or second embodiment, the contents are clustered by reflecting the classification category information given to the contents before mapping the contents in the low-dimensional space such as a two-dimensional plane. ing.
On the other hand, in the third embodiment, after the content is mapped in the low-dimensional space, the classification category information is reflected to cluster the content. As another method in which these two methods are fused, the contents may be intermediately clustered by reflecting the meta information and mapped in the low dimensional space, and further mapped after being mapped in the low dimensional space.

Other meta information that can be classified information such as classified category information is processed in the same manner as the classified category information and reflected in the distance between concept vectors or the position information (arrangement coordinates) in the low dimensional space. You can also

The preferred embodiments of the present invention have been described above. The present invention is not limited to these embodiments, and various variations can be made without departing from the scope of the present invention.

According to the present invention, when the content is mapped to a low-dimensional space such as a two-dimensional plane, the classification category information given in advance to each content is reflected so that the content is clustered. It becomes possible to do. This makes it possible to visualize a large amount of content as a scatter diagram that reflects the structure of the classification system in a low-dimensional space. Content is clustered in each layer of the classification system. Therefore, the user can finally reach the content he / she wants by proceeding from the upper layer to the lower layer while selecting the cluster in which he / she is interested in each layer of the classification system.

[Brief description of drawings]

FIG. 1 is a diagram showing a computer network according to a first embodiment of the present invention.

FIG. 2 is a diagram showing a content mapping system according to the first exemplary embodiment of the present invention.

FIG. 3 is a block diagram showing a hardware configuration of a content mapping system according to the first embodiment of the present invention.

FIG. 4 is a diagram illustrating a meta information database (DB) according to the first embodiment of the present invention.

FIG. 5 is a diagram illustrating a classification system that is an example of meta information according to the first embodiment of the present invention.

FIG. 6 is a diagram illustrating a concept vector database (DB) according to the first embodiment of the present invention.

FIG. 7 is an explanatory diagram of a position information database (DB) according to the first embodiment of the present invention.

FIG. 8 is a flowchart showing an operation executed by the content mapping system according to the first embodiment of the present invention.

FIG. 9 is a diagram showing an example of a first layer scatter diagram image according to the first embodiment of the present invention.

10 is a diagram showing an example of a second-level scatter diagram image displayed when pattern information processing (L3) is selected in the scatter diagram image of FIG.

FIG. 11 is a diagram for explaining a classification system which is an example of meta information according to the second embodiment of the present invention.

FIG. 12 is a diagram illustrating a distance matrix used in the content mapping system according to the second exemplary embodiment of the present invention.

FIG. 13 is a diagram for explaining a correction coefficient used in the content mapping system according to the second embodiment of the present invention.

FIG. 14 is a flowchart showing the operation of the content mapping system.

FIG. 15 is a diagram showing a content mapping system according to a third exemplary embodiment of the present invention.

FIG. 16 is a flowchart showing an operation executed by the content mapping system according to the third embodiment of the present invention.

FIG. 17 is a diagram showing an example of a first layer scatter diagram image according to the third embodiment of the present invention.

FIG. 18 is a diagram showing an example of a scatter diagram image of the first layer according to the third embodiment of the present invention.

FIG. 19 is a diagram showing an example of a scatter diagram image of the first layer according to the third embodiment of the present invention.

FIG. 20 is a diagram showing an example of a scatter diagram image of the first layer according to the third embodiment of the present invention.

FIG. 21 is a diagram showing an example of a scatter diagram image of the first layer according to the third embodiment of the present invention.

FIG. 22 is a diagram showing an example of a second layer scatter diagram image according to the third embodiment of the present invention.

FIG. 23 is a diagram showing an example of a second layer scatter diagram image according to the third embodiment of the present invention.

[Explanation of symbols]

10 computers 11 Display 12 Instruction input section 20 Content DB 21 Meta Information DB 22 Concept Vector DB 23 Arrangement coordinate DB 30 network 41 Concept vector calculation processing unit 42 Two-dimensional coordinate calculation processing unit 43 Scatter plot image 44 Scatter diagram image communication processing unit

Continued front page    (72) Inventor Shinji Abe             2-3-1, Otemachi, Chiyoda-ku, Tokyo             Inside Telegraph and Telephone Corporation (72) Inventor Yasuhito Hayashi             2-3-1, Otemachi, Chiyoda-ku, Tokyo             Inside Telegraph and Telephone Corporation F-term (reference) 5B050 AA08 BA06 BA10 CA07 CA08                       EA19 FA02 FA13 FA19 GA08                 5B069 AA01 DD01                 5B075 ND02 NK02 NK43 NR05 NR12                       PP22 PQ02 PQ13 PQ16 PQ46                       PQ69 PQ74 PR06 QM08

Claims (31)

[Claims] 1. A content mapping method for mapping a plurality of contents to which meta information including classification category information is added to a low-dimensional space, the high-dimensional mapping step of assigning a concept vector to each of the plurality of contents, Based on the concept vector, a low-dimensional mapping process for assigning position information in a low-dimensional space to each of the plurality of contents is provided, and the low-dimensional mapping process includes position information in a low-dimensional space as classification category information is closer. A content mapping method, characterized in that the contents are mapped so that they are close to each other. 2. The content mapping method according to claim 1, wherein the low-dimensional mapping step further includes a distance calculation step of obtaining a distance between the concept vectors assigned to each content in the high-dimensional mapping step, The closer the classification category information is, the shorter the distance calculated in the distance calculation process is to be corrected to obtain the effective distance, and the plurality of contents based on the effective distance calculated in the distance correction process. A content mapping method, characterized in that a position information allocation process for allocating position information in a low-dimensional space is provided for each of the. 3. The content mapping method according to claim 2, wherein the classification category information is a classification category of each content in a classification system having a hierarchical structure, and the distance correction process is performed up to a hierarchy with a low classification category. A content mapping method, characterized in that the effective distance is corrected as the number of matches increases. 4. The content mapping method according to claim 3, wherein in the distance correction process, a correction amount corresponding to a hierarchical level of the classification system in which classification categories assigned to two contents for which a distance is determined match each other. A content mapping method, characterized in that the distance between the two contents is corrected by adding. 5. The content mapping method according to claim 3, wherein the distance correction process multiplies a correction coefficient corresponding to a hierarchical level of the classification system in which classification categories assigned to two contents for which a distance is determined match. By
A content mapping method, characterized in that the distance between the two contents is corrected. 6. The content mapping method according to claim 5, wherein the correction coefficient has a size corresponding to each hierarchical level of the classification system and has a size corresponding to the number of classification categories belonging to the hierarchical level. A content mapping method characterized by being calculated from a distance matrix having as a matrix element. 7. The content mapping method according to claim 1, wherein the low-dimensional mapping step further includes a distance calculation step of obtaining a distance between the concept vectors assigned to each content in the high-dimensional mapping step, A position information allocation process of allocating position information in a low-dimensional space to each of the plurality of contents based on the distance obtained in the distance calculation process, and a position allocated in the position information allocation process as the classification category information is closer. A content mapping method comprising a position information correction process for correcting information so that the information becomes closer to each other. 8. The content mapping method according to claim 7, wherein the classification category information is a classification category of each content in a classification system having a hierarchical structure, and the position information correction process is a hierarchy with a low classification category. The content mapping method is characterized in that the position information is corrected so that the position information becomes closer to the above. 9. The content mapping method according to claim 8, wherein in the position information correction process, the center of gravity of a content belonging to a specific layer in a low-dimensional space is determined according to an input from the user, and the specific information is determined. A content mapping method, wherein contents belonging to a hierarchy are drawn at a fixed rate so as to approach the center of gravity. 10. The content mapping method according to claim 1, wherein the low-dimensional mapping step further includes a distance calculation step of obtaining a distance between the concept vectors assigned to each content in the high-dimensional mapping step, The closer the classification category information is, the shorter the distance obtained in the distance calculation step is corrected to obtain an effective distance, and the distance correction step of obtaining the effective distance, and the plurality of contents based on the effective distance obtained in the distance correction step. A position information allocating process for allocating position information in the low-dimensional space and a position information correcting process for correcting the position information allocated in the position information allocating process are closer to each other as the classification category information is closer to each other. Content mapping method characterized by. 11. The content mapping method according to claim 2, 7, or 10, wherein the position information allocation process applies a multidimensional scale construction method to each of the plurality of contents in a low dimensional space. A content mapping method characterized by allocating position information. 12. The content mapping method according to claim 2, 7, or 10, further comprising a position information transmitting step of transmitting the position information, wherein in the position information transmitting step, together with the position information of each content. A content mapping method comprising transmitting the classification category of each content. 13. The content mapping method according to claim 12, wherein in the position information transmitting process, a range of contents for transmitting the position information and the classification category is continuously changed in response to an input from a user. A content mapping method characterized by the above. 14. A content mapping device for mapping a plurality of contents to which meta information including classification category information is added to a low-dimensional space, and a high-dimensional mapping means for assigning a concept vector to each of the plurality of contents, Based on the concept vector, a low-dimensional mapping unit that allocates position information in a low-dimensional space to each of the plurality of contents is provided, and the low-dimensional mapping unit is a position information in the low-dimensional space as classification category information is closer. A content mapping device characterized in that contents are mapped in such a manner that they are close to each other. 15. The content mapping device according to claim 14, wherein the low-dimensional mapping means further calculates a distance between the concept vectors assigned to each content by the high-dimensional mapping means. The closer the classification category information is, the shorter the distance obtained by the distance calculating means is corrected to obtain the effective distance, and the distance correction means for obtaining the effective distance, based on the effective distance obtained by the distance correcting means. A content mapping device, wherein position information assigning means for assigning position information in a low-dimensional space is provided to each of them. 16. The content mapping device according to claim 15, wherein the classification category information is a classification category of each content in a classification system having a hierarchical structure, and the distance correction unit is for a hierarchy with a low classification category. A content mapping device, which corrects so that the effective distance becomes shorter as the numbers match. 17. The content mapping device according to claim 16, wherein the distance correction means sets a correction amount corresponding to a hierarchical level of the classification system in which classification categories assigned to two contents for which distances are to be matched. A content mapping device, wherein the distance between the two contents is corrected by adding them. 18. The content mapping device according to claim 16, wherein the distance correction means multiplies a correction coefficient corresponding to a hierarchical level of the classification system in which classification categories assigned to two contents whose distances are to be matched. By
A content mapping device, which corrects a distance between the two contents. 19. The content mapping device according to claim 18, wherein the correction coefficient has a size corresponding to each hierarchical level of the classification system and has a size according to the number of classification categories belonging to the hierarchical level. A content mapping device characterized by being calculated from a distance matrix having as a matrix element. 20. The content mapping apparatus according to claim 14, wherein the low-dimensional mapping means further calculates a distance between the concept vectors assigned to each content by the high-dimensional mapping means. A position information allocating unit that allocates position information in a low-dimensional space to each of the plurality of contents based on the distance obtained by the distance calculating unit, and a position allocated by the position information allocating unit as the classification category information is closer. A content mapping device comprising position information correction means for correcting information so that the information becomes closer to each other. 21. The content mapping device according to claim 20, wherein the classification category information is a classification category of each content in a classification system having a hierarchical structure, and the position information correction means is a hierarchy with a low classification category. The content mapping device is characterized in that the position information is corrected so that the position information becomes closer to the above. 22. The content mapping device according to claim 21, wherein the position information correction unit further obtains a center of gravity of a content belonging to a specific layer in a low-dimensional space according to an input from a user, and determines the specific point. A content mapping device, wherein contents belonging to a layer are attracted at a fixed rate so as to approach the center of gravity. 23. The content mapping device according to claim 14, wherein the low-dimensional mapping means further calculates a distance calculation means for obtaining a distance between the concept vectors assigned to each content by the high-dimensional mapping means, The closer the classification category information is, the shorter the distance obtained by the distance calculating means is corrected to obtain the effective distance, and the distance correction means for obtaining the effective distance, based on the effective distance obtained by the distance correcting means. Positional information allocating means for allocating positional information in a low-dimensional space to each, and positional information correcting means for correcting so that the positional information allocated by the positional information allocating means becomes closer as the classification category information is closer to each other. Content mapping device characterized by. 24. The content mapping device according to claim 15, 20 or 23, wherein the position information assigning means applies a multidimensional scale construction method to each of the plurality of contents in a low dimensional space. A content mapping device characterized by allocating position information. 25. The content mapping device according to claim 15, 20 or 23, further comprising position information transmitting means for transmitting the position information, wherein the position information transmitting means includes position information of each content. A content mapping device, which transmits a classification category of each content. 26. The content mapping device according to claim 25, wherein, in the position information transmitting means, a range of contents for transmitting the position information and the classification category is continuously changed in response to an input from a user. A content mapping device characterized by the above. 27. A computer program for causing a computer to execute a content mapping method for mapping a plurality of contents, to which meta information including classification category information is added, to a low-dimensional space, wherein each of the plurality of contents has a concept vector. And a low-dimensional mapping process that allocates position information in a low-dimensional space to each of the plurality of contents based on the concept vector, and the low-dimensional mapping process has similar classification category information. A computer program characterized by mapping contents so that position information in a low dimensional space becomes closer to each other. 28. The computer program according to claim 27, wherein the low-dimensional mapping step further includes a distance calculation step of obtaining a distance between the concept vectors assigned to each content in the high-dimensional mapping step, The closer the classification category information is, the shorter the distance obtained in the distance calculation step is corrected to obtain an effective distance, and the distance correction step, and the plurality of contents based on the effective distance obtained in the distance correction step. A computer program, characterized in that a position information allocation process for allocating position information in a low-dimensional space is provided in. 29. The computer program according to claim 27, wherein the low-dimensional mapping step further includes a distance calculation step of obtaining a distance between the concept vectors assigned to each content in the high-dimensional mapping step, A position information allocation process of allocating position information in a low-dimensional space to each of the plurality of contents based on the distance obtained in the distance calculation process, and position information allocated in the position information allocation process as the classification category information is closer. A computer program, characterized in that a position information correction process for correcting so as to be closer is provided. 30. The computer program according to claim 27, wherein the low-dimensional mapping step further includes a distance calculation step of obtaining a distance between the concept vectors assigned to each content in the high-dimensional mapping step, The closer the classification category information is, the shorter the distance obtained in the distance calculation step is corrected to obtain an effective distance, and the distance correction step, and the plurality of contents based on the effective distance obtained in the distance correction step. A position information allocating process of allocating position information in a low dimensional space and a position information correcting process of correcting the position information allocated in the position information allocating process to be closer to each other as the classification category information is closer to each other are provided. A characteristic computer program. 31. A computer-readable recording medium having the computer program according to claim 27 recorded therein.