JPH09114848A - Information relating device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To search related information even when the intention and viewpoint of a user are ambiguous and to search also unexpected information for the user. SOLUTION: An information storage part 101 stores information and the degree of association of each viewpoint of information. A viewpoint setting part 103 designates the criticality or the weighting factor of two pairs of viewpoints. Each of information relating part 104 and 105 synthesizes a composite viewpoint by linear and non-linear interpolations and further, calculates the composite degree of association based on each of the composite viewpoint and the degree of association of information. Each of the composite degree of association by a linear interpolation operation and a non-linear interpolation operation shows the degree of a normal relation and unexpectedness. A user can search desired information from the composite degree of association.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is based on an arbitrarily set viewpoint, even if the viewpoint indicating the viewpoint or intention of the user is ambiguous, and is surprising to the information related to the viewpoint or the viewpoint. The present invention relates to an information associating device suitable for searching various kinds of information.

[0002]

2. Description of the Related Art In information retrieval using a database system, a knowledge base system, etc., the relation between the elements of the information stored in the system is mainly used. At that time, generally, the information is related by the relationship between words such as keywords and attributes that are attached to each element of the information in advance.

For example, Hiroko Yuasa and Keiji Shiroshima "Broad catch system of information" (Information Processing Society of Japan Report IM1
3-6, pp37-44, 1993), based on the frequency of occurrence of words in full-text search of documents, search by matching degree to registered registered interest field or hierarchical clustering using similarity between information. It defines the relevance of information.

As another example, Jun Toyoura et al., "Automatic Document Classification by Self-Organizing Neural Network".
(IPSJ Research Report NL88-6, pp41-4
8, 1992), a document is associated with a distributive word pattern, a classification is learned by a neural network model based on the co-occurrence of words without giving a classification criterion from the outside, and the learning result is obtained. Relevance of information is defined by classification.

However, these methods are bound by the relationship between words, and it is difficult to correlate information according to a user's viewpoint or an arbitrary viewpoint expressing his / her intention, or to make an unexpected correlation.

As an example of a method for associating information according to viewpoints, Hiromichi Fujisawa's "Concept Browser and Personal Information Environment"
(The Information Processing Society of Japan Research Group Report IM7-6, pp41-4
8, 1992). In this method, we propose a method of selecting information related to a search sentence presented by a user by using a semantic network defined by semantic relationships between word concepts. At that time, it is considered that classification is performed according to a viewpoint (viewpoint) by hierarchical systematization for each individual, but the relation between concepts is Is-A relation or Part-O relation.
Because it is defined by the presence / absence of links such as f relationships (Fig. 11
In the conventional example, the information A to E are linked by the relation R), and the strength of the relation cannot be freely changed according to the viewpoint. In addition, the viewpoint of systematization for the whole information is not specified, so the viewpoint of systematization is unclear to other users. Moreover, it had a drawback that enormous effort was required to define the relationships between the concepts until they were systematized with overall consistency.

On the other hand, as an example in which the strength of the relationship can be treated, Yuji Sawada et al. “Realization of an association mechanism in consideration of viewpoint” (Information Processing Society of Japan, Vol. 35, No. 5, pp 714-7)
24, 1994). This method considers the conceptual adjacency defined by the similarity between the attributes of each case and the similarity of the relationships between the attributes, and expresses the viewpoint with a certain attribute or relationship. By weighting the conceptual adjacency according to the strength of the viewpoint defined by those attributes and the weight value of the relationship, it becomes possible to correlate the cases according to the viewpoint. Also, the strength of the viewpoint can be freely set by the user. However, in order to define conceptual adjacency, it is a prerequisite that a hierarchical conceptual system between attributes is obtained, but the effort to obtain a conceptual system with overall coherence is practical. It becomes enormous for the amount of information of the scale.

[0008] Although associating unexpected information with a viewpoint is useful for creating an idea or idea, a method for artificially creating an unexpected association has not been proposed so much. On the other hand, the elements of the information are spatially arranged according to the distance defined according to the strength of the association of the information, and the user looks at the entire visualized relational diagram to identify the blank portions with weak relations. In the hope of finding one, a method of encouraging inspiration for ideas is being considered.

For example, Jun Tamura's "Concept map generation system SPRINGS based on dynamics between symbols" (IPSJ Journal Vol. 33, No. 2, pp. 465-47)
0, 1992) proposes a method of expressing the local dissimilarity between concepts by a dynamical model of spring and generating an overall concept map by the interaction of those springs.
In this method, unexpected information relevance may be obtained with respect to the relationship between predetermined concepts, but in order to handle the information relevance or unexpectedness with respect to the user's arbitrary intention or viewpoint, You have to redefine the relationships between concepts each time,
Therefore, great effort must be paid.

On the other hand, Koichi Hori, "A hypothesis for discussing the effect of the idea generation support system" (IPSJ Journal, Vo
l. 35, no. 10, pp 1988-2008, 19
94) considers a method of generating a concept map by interaction between concepts by a dynamical system model different from the above, and further changes the association between concepts by interactively controlling the parameters of the model so that the thoughts get stuck. Is proposing an idea to escape from. However, this method cannot be said to be parameter control that clearly defines the fluctuation of the viewpoint, and does not go beyond the scope of trial and error control.

In addition to these, Kazushi Nishimoto et al. “Trial of divergent group thinking support by information provision using associative memory”
(Proceedings of the 2nd Idea Support Tool Symposium, 19
In 93), the association between the keyword vectors in the associative model called associatoron is associated with the input search sentence, and the idea is supported by a method other than the map presentation. In this report, the chained associative method that uses the difference between the search sentence and what the model recalls as a new keyword is more surprising than the full-text search, random search, and the conventional associative memory model association method. It has been experimentally shown to be suitable for extracting information. However, because it is a model that uses an associative dictionary according to a predetermined fixed classification (relationship between keywords), it is possible to associate associative information according to the user's arbitrary viewpoint, It is not possible to flexibly make such connections.

[0012]

SUMMARY OF THE INVENTION The present invention has been made in consideration of the above circumstances. Even when the user's intention is ambiguous, information is related according to the intended range, and the relation By selectively presenting information according to strength, we aim to realize flexible information retrieval and search. Another object of the present invention is to allow a user to freely set a viewpoint representing an intention or a viewpoint, and to realize these with less effort as compared with the construction of an overall consistent concept system. .

[0013]

In order to achieve the above object, the present invention relates to a plurality of information records and a relation defining the strength of the relation between each of the information records and each of the plurality of viewpoints. And a means for setting two sets of weighting factors with respect to the viewpoint, and two sets of weighting factors are interpolated while changing the interpolation variable to 1
Means for synthesizing a set of weighting factors, means for accumulating the degree of association of each viewpoint of the information record while weighting with the synthesized weighting factor to calculate a composite degree of relevance, and an information record according to the degree of compositeness Is provided in the information associating device.

In this case, if the weighting factors are combined by linearly interpolating, a wide range of related information can be associated. Further, if the weighting factors are combined by performing non-linear interpolation, it is possible to associate unexpected information. In order to perform the non-linear interpolation, for example, in a coordinate system different from the coordinate system of the weighting factors mapped to the coordinate system of the weighting factors, the first coordinate and the second coordinate corresponding to the two sets of weighting factors. Is linearly interpolated to synthesize a third coordinate, a fourth coordinate of the coordinate system of the weighting coefficient corresponding to the third coordinate of the different coordinate system is obtained, and the fourth coordinate is used to synthesize the third coordinate. It is sufficient to find the weighting factor.

If the information record is displayed according to the result of comparison between the reference value and the composite relevance, and the reference value is changed, the related information can be narrowed down. Also, if the information record of a predetermined rank is displayed according to the comparison result of the reference rank and the rank of the composite relevance,
The associated information can be refined as well. The reference value or the reference rank can be changed interactively and heuristically, and the degree of narrowing down can be made a desired value.

Further, according to the present invention, a plurality of information records, storage means for storing the degree of association that defines the strength of the association between each of the information records and each of the plurality of viewpoints, and two sets of weights for the viewpoints. The means for setting the coefficient and the above two sets of weighting coefficients are interpolated while changing the interpolation variable to 1
Means for synthesizing a set of weighting factors, means for accumulating the degree of association of each viewpoint of the information record while weighting with the synthesized weighting factor to calculate a composite degree of relevance, and performing the interpolation linearly A means for displaying the coordinates having the composite degree of relevance and the composite degree of relevance obtained by performing the interpolation in a non-linear manner in a two-dimensional coordinate system, and selecting a coordinate in the two-dimensional coordinate system to display a corresponding information record The information associating device is provided with a means for selecting and a means for displaying the content of the selected information record. In this way, the normal relevance is expressed on one axis and the unexpectedness is expressed on the other axis, and the desired related information can be selected from both.

More specifically, the degree of association is regarded as a probability distribution for a set of information, and a family of probability distributions having a parameter indicating the degree of importance (weighting coefficient) of each viewpoint (a normal distribution is defined by an average or standard deviation). , We consider a set of probability distributions characterized by parameters, and perform a differential geometric operation on the function space of the probability distribution (Shun-ichi Ara
i, "Differential Geometry of Curved Exponential Fa
milies-Curvatures and Information Loss- "The Annal
s of statistics, Vol.10, No.2, pp357-385, 1982), based on the viewpoint range (two sets of weighting factors) set by the user, a virtual viewpoint (interpolated 1 By generating a set of weighting factors), it is possible to correlate information while shaking the viewpoint, and it is possible to correlate unexpected information.

[0018]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 shows the overall configuration of the above-described embodiment. In this figure, 101 is an information storage unit.
2 is a probability distribution function defining unit, 103 is a viewpoint setting unit, 104 is a first information relating unit, 105 is a second information relating unit, 106 is a related information presenting unit, and 107 is a presenting level setting unit. Each of the above elements may be configured as a stand-alone computer as a whole, or may be configured on a network. The information storage unit 101 stores the information record and also stores the degree of association indicating the degree of association with each viewpoint. The viewpoint setting device 103 defines a composite viewpoint by setting the importance (weighting coefficient) for each viewpoint. Two composite viewpoints are set. These two composite viewpoints indicate the range of viewpoints assumed by the user. The probability distribution function defining unit 102 extracts the degree of association of each viewpoint for each information record from the information storage unit 101 and defines a probability distribution function having the degree of association as a probability value. The first information associating unit 104 causes the current composite viewpoint to make a linear transition between the initially set two composite viewpoints. Substituting the importance into the probability distribution function, we calculate the composite relevance at that time. This degree of association is a normal degree of association. On the other hand, the second information associating unit 105
Is a non-linear transition between the current composite viewpoint and the initially set two composite viewpoints. The importance of each viewpoint representing the present composite viewpoint is calculated by a probability distribution function. Substitute in to calculate the composite relevance at that time. This degree of association represents unexpectedness. The related information presenting device 106 presents an information record whose overall degree of association exceeds a predetermined value as a search result, a search result, or a general composite viewpoint such as an ordinary degree of association or an unexpected degree of association. Present the information record from. The presentation level setting unit 107 sets the above-mentioned predetermined level.

Next, a schematic processing procedure of this embodiment will be described.

The information storage unit 101 is a prerequisite for the search process.
Stores a finite number of pieces of information Χ and each element of information x ∈ Χ
Information on each of the n independent viewpoint names.
Represents the degree of continuity (the degree of association with the information x for the i-th viewpoint)
Probability P_i(X) (i = 1, 2, ..., N) is defined.
It is assumed that In addition, it is a general view other than the n viewpoints.
P as the degree of association of information about points₀(X) is defined,
i = 0, 1, 2 ,. . . , N for each viewpoint
_i(X)> 0 and Σ x ∈ ΧP_iSatisfying (x) = 1
And Where P₀(X) belongs to any of the n viewpoints
Since it is the degree of relevance of the viewpoint that does not exist as a reference, P
₀Define (x) as the uniform distribution 1 / | Χ |, or
n each P_iSpecified by a probability distribution function other than (x)
Shall be.

For example, the former condition P_i(X)> 0 means that
Comparing the strength of association within the set of information Χ for each viewpoint i
, The degree of relevance of information x that is very weakly related
Instead of setting it to zero, P is set for a small value δ> 0._iIf we define (x)
Well, the latter condition Σ x ∈ ΧP_i(X) = 1 is a strong relationship
To a suitable value (eg, stepwise positive values 1, 2 ,,
3 ,. . . Etc.) and then their respective values (each viewpoint i
Sum of all information elements)
Just divide and standardize.

FIG. 3 shows a specific example of association between information records and viewpoints. In this figure, the neighborhoods of Tokyo (Shinjuku,
Shibuya, Akasaka), the viewpoint of the temporal closeness to the occurrence of the incident in Tokyo (the degree of association is defined as the difference between the incident occurrence time at each location and the incident occurrence time in Tokyo),
Two points of view called geographical proximity (the degree of association is defined by the distance from Tokyo to each point) are associated with each piece of information. In this example, using the perspective of temporal proximity and geographical proximity,
You can search for information about the incident.

FIG. 4 generally shows viewpoints, information records, and degrees of association. In this figure, two independent viewpoints 1 and 2 (in this case, viewpoint names whose numbers are 1 and 2) are provided. On the other hand, each element of the information from A to E is related to the relation r
_i (A) to r _i (E) (i = 1, 2) are associated with each other. In the following, it is assumed that a parameter indicating the importance of each of N independent viewpoints is θ ⁱ (i = 1, 2, ..., N), and the value can be freely set by the user next to the search. . A combination of N viewpoints i having the importance degree θ ⁱ is called a composite viewpoint.

As an initial setting for carrying out an artificial operation, a parameter θ ⁱ (i = 1, i = 1, i = 1, 1) indicating the degree of importance of each of n independent viewpoints by using the viewpoint setting unit 103 in FIG.
2,. . . , N). By this, the user defines two composite viewpoints, that is, parameter coordinate points P and Q according to the intention. Where θ ⁱ (> 0)
The larger the value of is, the higher the importance of the viewpoint is. If θ ⁱ <0, the viewpoint is the opposite. For example, in the example of FIG. 3, if the viewpoint i represents the distance from Tokyo to each point, θ
ⁱ <0 represents the importance of the viewpoint of distance.

Parameters (vectors) θ (P) and θ representing the degree of importance of (n independent) viewpoints corresponding to the two points P and Q
The value of (Q) may be input from the menu as shown in FIG. 5, or the value of each parameter may be set by a scroll bar or file input.

As another example, there is a method of designating coordinate points as shown in FIG. Usually, it is considered that attention is paid to a small number of viewpoints, such as two. Therefore, from the n independent viewpoints, the viewpoint names relating to the two viewpoint parameters θ ⁱ and θ ^j are selected from a menu or the like, and displayed on the displayed coordinate plane. , Specify the coordinate values of the points P and Q from the start point and the end point of a rectangle surrounded by a mouse or the like.

Next, a differential geometric operation on the function space of the probability distribution will be considered as an artificial operation.

For example, as shown in FIG. 7, the probability distribution function p (x; θ) is changed while changing the value of the parameter θ on the straight line P-Q and the curve P-R-Q with respect to the function space of the probability distribution.
Is derived. At that time, the probability distribution function defining unit 102 in FIG.
Then, the probability distribution function characterized by the parameter θ is sequentially obtained. On straight line P-Q and curve P-R-Q
Each of the above operations corresponds to the first information associating unit 10 in FIG.
4 and the second information associating unit 105.

FIG. 8 is an example in which the user gives two viewpoints 1 and 2 regarding the possibility of future technology and business potential, and conceptually represents an artificial operation in the parameter space. Due to the fluctuation of the viewpoint, a virtual viewpoint parameter value between the viewpoint 1 and the viewpoint 2 is generated along the broken line, while a surprising axial parameter value in the axial direction is also generated along the dotted line. The degree of association of is sequentially obtained.

As a method of selectively presenting information from the degree of association obtained as a result of the artificial operation, for example, the degree of association of information obtained at each time is set by the presentation level setting unit 107 in FIG. Information ({x | P (x; θ)> = P _th }) for a stronger one than the set value P _th is presented to the related information presentation unit 106. At this time, when the number of presented information is large or the relevance is weak, the value P _{th of the} presentation level is reset to a large value so as to limit the range to the information having strong relevance. On the contrary, the number of presented information is small,
Alternatively, when the relevance is too strong and only limited information is presented, the presentation level P _th is reset to a smaller value so as to expand the range of the relevant information. In this way, the search can be performed while adjusting the range of related information. On the other hand, the information may be presented in a rank according to the value of the degree of association.
In this case, the presentation level value P _th is the ranking number to be presented. In any case, by gradually decreasing (or increasing) the value P _{th of the} presentation level, the range of related information can be changed stepwise (stepwise clustering).

An example of another presentation method is shown in FIG. In the example of FIG. 9, the related information is presented as a two-dimensional distribution in the coordinates with the degree of association obtained by two artificial operations relating to the shaking of the viewpoint and the unexpectedness as coordinate axes. The small circles in the figure represent each information, and when the coordinate point is designated, the corresponding information is selected. If the horizontal axis represents the degree of relevance obtained by the operation of shaking the viewpoint, and the vertical axis represents the degree of relevance related to the unexpectedness, the following can be said as a guide when selecting information from the distribution graph. That is, the area on the upper right of the graph represents information that is likely to be associated with what is intended and that may be unexpected. In the upper left area of the graph, information that is less relevant to the intended one but may be expected to be unexpected (large hole-like), and in the lower right area of the graph is related to the intended one Expresses little (solid) information. The area near the origin represents information that does not correspond to any of the above (which is not very useful to the user). In the example of FIG. 9, the distribution graph for each time 0.1 among the degrees of association obtained at each time by the operation during the time t = 0 to 1 is displayed in a list. It doesn't have to be. That is, it is possible to present a distribution graph for an arbitrarily designated time, or present it while changing the distribution over time. Further, as shown in FIG. 10, the specified partial area may be enlarged and displayed on the distribution graph, and detailed information may be selected as necessary.

As described above, the information corresponding to the change in the parameter value θ of the virtual viewpoint generated by the artificial operation for associating the information corresponding to the fluctuation of the viewpoint and the unexpected information association. Information with strong relevance is selectively presented as the degree of relevance is sequentially obtained. The related information is presented while the viewpoint parameter moves from the point P to the point Q, or until the presentation level setting unit 107 instructs the forced termination.

Next, the procedure of the information associating operation according to the fluctuation of the viewpoint and the associating operation of surprising information complementary to this operation will be described in detail with reference to the flowchart of FIG.

[0035]

[Table 1]

[0036]

[Table 2]

[0037]

[Table 3]

[0038]

[Table 4]

[0039]

[Table 5]

[0040]

[Table 6]

[0041]

[Table 7]

[0042]

As described above, according to the present invention, information is associated in accordance with a composite viewpoint arbitrarily set by the user. Further, even if the user's intention is ambiguous, information can be associated through the shaking of the viewpoint based on two composite viewpoints that represent a rough range of the intention by an artificial operation. You can also associate unexpected information.

At this time, the related information corresponding to the virtual viewpoint generated by the artificial operation is presented according to the strength of the relationship (between the viewpoint and the information), or the related information is two-dimensionally displayed. By selectively presenting from various distributions, it is possible to interactively execute information search and search according to the user's intention (intended range).

Furthermore, since it is only necessary to combine a relatively simple rule of associating only the information of the part necessary for the user according to the individual viewpoint (intention or viewpoint),
The effort is less than when the relation between information is defined by the concept system with overall consistency.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 is a flowchart illustrating the operation of the above-described embodiment.

FIG. 3 is a graph for explaining the relationship between information and viewpoints according to a specific example.

FIG. 4 is a graph illustrating information, viewpoints, and degrees of association.

FIG. 5 is a diagram showing a method of setting composite viewpoints.

FIG. 6 is a diagram illustrating another example of a composite viewpoint setting method.

FIG. 7: Associating information using the shaking of the viewpoint,
FIG. 6 is a diagram for generally explaining association of surprising information.

[Fig. 8] Fig. 8 is a diagram for explaining the association using the fluctuation of the viewpoint and the association of the surprising information according to a specific example.

FIG. 9 is a diagram showing a mode in which two types of degree of association obtained by associating information using shaking of viewpoints and associating information with unexpectedness are selected as coordinate axes and corresponding information is two-dimensionally distributed and displayed. is there.

FIG. 10 is a diagram showing an option of enlarging and displaying a part of the display of FIG. 9;

FIG. 11 is a diagram illustrating a conventional example.

[Explanation of symbols]

 101 information storage unit 102 probability distribution function defining unit 103 viewpoint setting unit 104 first information relating unit 105 second information relating unit 106 related information presenting unit 107 presentation level setting unit

Claims (6)

[Claims] 1. A storage unit that stores a plurality of information records, a degree of association that defines the degree of association between each of the information records and each of a plurality of viewpoints, and two sets of weighting factors are set for the viewpoints. Means for interpolating the two sets of weighting factors while changing the interpolation variables to combine one set of weighting factors, and weighting the degree of association of each viewpoint of the information record with the combined weighting factors. An information associating device comprising: means for accumulatively calculating a composite degree of relevance; and means for displaying information records according to the degree of composite relevance. 2. The information associating device according to claim 1, wherein the weighting factors are combined by linearly interpolating. 3. The information associating device according to claim 1, wherein the weighting factors are synthesized by performing non-linear interpolation. 4. The first coordinate and the second coordinate corresponding to the two sets of weighting factors are linearly interpolated in a coordinate system different from the coordinate system of the weighting factor mapped to the coordinate system of the weighting factor. And synthesizes the third coordinate to obtain a fourth coordinate of the coordinate system of the weighting coefficient corresponding to the third coordinate of the different coordinate system, and obtains the synthesized weight coefficient from the fourth coordinate. The information associating device according to claim 3. 5. The means for displaying an information record according to the composite degree of relevance compares a reference value or a standard rank with the reference value and the degree of composite relevance, and depending on the result of this comparison. An information record by displaying the information record or comparing the reference rank with the rank of the composite degree of relevance, and displaying the information record according to the comparison result.
The information associating device described in 2, 3, or 4. 6. A storage unit that stores a plurality of information records, a degree of association that defines the strength of association between each of the information records and each of the plurality of viewpoints, and two sets of weighting factors for the viewpoints. Means for interpolating the two sets of weighting factors while changing the interpolation variables to combine one set of weighting factors, and weighting the degree of association of each viewpoint of the information record with the combined weighting factors. A means for cumulatively calculating a composite degree of relevance, coordinates having the complex degree of relevance when the interpolation is linearly performed, and a degree of complex relevance obtained by performing the non-linear interpolation are elements in a two-dimensional coordinate system. An information associating device comprising: a means for displaying, a means for selecting coordinates of the two-dimensional coordinate system to select a corresponding information record, and a means for displaying the content of the selected information record.