JPH11296523A - Method and device for filtering information - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable a user to easily grasp the relation between base words of a profile and efficiently correct the profile. SOLUTION: When the user inputs the profile (step S2), a filtering part performs filtering according to the profile (step S6). An interface management part displays the filtering result and the user evaluates it (step S7). The filtering part changes the profile according to the evaluation result (step S8). A visualizing interface management part visually display base words of the filter changed by the filtering part in two dimensions so that as the relation between the base words is closer and closer, the distance is shorter and shorter (step S11).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an information filtering method and apparatus for filtering information resources, and more particularly, to displaying a base word of a profile.

[0002]

2. Description of the Related Art There are the following documents that show conventional information filtering techniques. [1] Nicholas J. Belkin, W. Bruce Croft, Information
Filtering and Information retrieval: Two Sides of t
he Same Coin ?, Communication of the ACM, 35 (12), pp29
-38,1992

[2] Chris Beckley, Gerard Salton, Ja
mes Allan, The Effect of AddingRelevance Informati
on in a Relevance Feedback Environment, SIGIR'94 Pr
oceedings, 1994, pp292-300

[3] Scott Deerwester, Susan T. Dumais,
George W. Fornus, Thomas K. Landauer, Richard Harshma
n, Indexing by Latent Semantic Analysis, Journal of
the Society for Information Science, 41-6,1990, pp39
1-407

[0005] With the development of computer networks, information filtering technology has been developed that obtains only the information desired by the user for information that changes every day, such as e-mail and net news. In information filtering, a system is generally used in which filtering is performed by changing a user's profile using related feedback or a learning algorithm in order to follow a user's interest that changes little by day.

[0006] Here, the profile has almost the same meaning as a search sentence (or an inquiry sentence) in information search, but here, a search expressing a user's interest in search for a certain length of time. Statement. Such filtering is performed based on the similarity between the content of the text and a profile expressing information desired by the user.

[0007] The procedure of information filtering is generally performed by the following method. 1. The user enters an initial profile. 2. The information filtering device compares the user's profile with the delivered text, performs filtering, and displays the filtering result to the user. 3. The user inputs an evaluation for the filtering result. 4. Based on the information of the above 3, the information filtering device corrects the profile. Repeat steps 5.2 to 4.

Further, when the user intentionally wants to modify the profile, the following procedure is taken. 1. The user enters an initial profile or a modified profile. 2. The information filtering device compares the user's profile with the delivered text, performs filtering, and displays the filtering result to the user. 3. The information filtering device presents the profile to the user as it is. 4. The user modifies the profile. 5. The information filtering device corrects the profile based on the information of the above item 4. 6. Repeat steps 1-5.

[0009]

However, in the information filtering, when the user evaluates the filtering result and refines the profile by related feedback, the user may want to directly modify the profile. For example, a case where the filtering result does not match the user's interest.

However, in such a case, even if the system presents the base word of the profile to the user as it is, the user must know the relation between the base word and the relation between the base word and the filtering result (document). Can not. For this reason, there is a problem that even if the user attempts to directly modify the profile, the profile cannot be modified efficiently.

[0011] In view of the above, it has been desired to realize an information filtering method and apparatus which enable a user to easily grasp the relationship between base words of a profile and to efficiently modify a profile. .

[0012]

The present invention employs the following structure to solve the above-mentioned problems. <Structure of Claim 1> First, filtering is performed on an information source based on a base word of a profile input by a user, and the filtering result is displayed. Next, the user attaches the filtering result based on the display of the filtering result. When the base word of the profile is changed and displayed again according to the score, the base words are visually displayed such that the stronger the relation between the base words of the profile when performing filtering, the closer the distance between the base words becomes. The information filtering method is characterized in that

<Explanation of Claim 1> A profile is a search sentence expressing a user's interest in search for a certain length of time. In the case of English, for example, in the case of English, unnecessary words such as “the”, “a”, and “to” that do not make sense even in base words are excluded, and the stem of the inflected form is used as the basic form. A word. Displaying a filtering result means extracting information related to a user's profile from an information source and displaying the information.

According to the first aspect of the present invention, first, information sources are filtered based on a profile input by a user.
Display the result. Next, the user looks at the displayed content and inputs whether or not the user is interested in the result. Thereby, a base word of a profile different from the profile first input by the user is extracted. When these base words are displayed, they are visually displayed, for example, in two dimensions so that the stronger the relation between the base words in the filtering is, the shorter the distance is.

[0015] With the above-described operation of the first aspect of the present invention, it is possible to efficiently remove unnecessary base words by a user, and therefore, only information resources closer to the user's interest are filtered. There is an effect that can be.

<Structure of claim 2> In the information filtering method according to claim 1, each base word is displayed, and the base words whose association between the base words is within a predetermined value range are clustered. This is an information filtering method for displaying, for each clustering, the importance indicating the degree of the weight of the clustering in the filtering result and the representative base word of the clustering.

<Explanation of claim 2> According to the invention of claim 2, in the invention of claim 1, a plurality of similar base words are displayed as one group, and a base word representative of the group is represented by: The importance of the group is displayed. Accordingly, in addition to the effect of the first aspect, even when the number of base words to be displayed increases, the relation between the base words is easy for the user to understand, and as a result, even when the user changes the profile, the efficiency is further improved. Can do well.

<Structure of claim 3> In the information filtering method according to claim 1 or 2, each base word of the profile is displayed, and the information of the result of filtering by the profile is distanced as the relation between the information is stronger. This is an information filtering method that visually displays such that is closer and the stronger the association between each information and each base word, the shorter the distance.

<Explanation of Claim 3> The invention of claim 3 is the invention of claim 1 or 2, wherein information indicating a filtering result is also visually displayed. Here, the information indicating the filtering result is, for example, a document related to the profile. Then, such a document is displayed together with the base words. This allows the user to easily grasp the relationship between the information such as the document and the like and the base word evaluated by the user, so that the profile can be efficiently changed even when the profile is changed.

<Structure of Claim 4> Based on the base word of the profile input by the user, filtering is performed on the information source, the result is output, and when evaluation information is input by the user, the evaluation information is output. A filtering unit that changes a profile, an interface management unit that receives an instruction from a user and receives a result from the filtering unit for the user, and a profile that performs filtering when the filtering unit changes the profile. An information filtering device, comprising: a visualization interface management unit that visually displays each base word such that the distance between the base words is closer as the relation between the base words is stronger.

<Explanation of Claim 4> The filtering unit comprises:
Based on the base word of the profile input by the user, filtering is performed on the information source, and the result is output to the interface management unit. The interface management unit displays the result to the user, and when the user inputs evaluation information for the result, passes the evaluation information to the filtering unit. Thereby, the filtering unit:
Change your profile. In addition, when the filtering unit changes the profile, the visualization interface management unit visually displays each base word such that the stronger the association between the base words of the profile in performing the filtering, the closer the distance between the base words becomes. I do.

According to the fourth aspect of the present invention having such an effect, it is possible to efficiently remove unnecessary base words by the user. Therefore, only information resources that are more interesting to the user are filtered. There is an effect that can be.

<Structure of claim 5> In the information filtering apparatus according to claim 4, the base words whose association between the base words is within a predetermined value range are clustered,
For each clustering, an importance indicating the degree of the weight of the clustering in the filtering result, a clustering unit that outputs a representative base word of clustering, a clustering importance output from the clustering unit and a representative base word are output. And a visualization interface management unit for visually displaying the information.

<Explanation of Claim 5> In the invention of claim 5, in the invention of claim 4, a clustering unit is provided, and the visualization interface management unit visually displays display data of the clustering unit. Things. Thereby, in addition to the effect of the invention of claim 4, even when the number of base words to be displayed increases, the relation between the base words is easy for the user to understand, and as a result, the user changes the profile. However, it can be performed more efficiently.

<Structure of Claim 6> In the information filtering device according to claim 4 or 5, the information of the filtering result by the profile of the filtering unit is
A document visualization information generation unit that outputs data of a visual positional relationship such that the stronger the relationship between the pieces of information is, the closer the distance is, and the closer the relationship between each piece of information and each base word is, the closer the distance is. And a visualization interface management unit for visually displaying an output result of the document visualization information generation unit.

<Explanation of Claim 6> According to the invention of claim 6, in the invention of claim 4 or 5, a document visualization information generation section is provided, and the visualization interface management section is configured to display the document visualization information generation section. The data is displayed visually. This allows the user to easily grasp the relationship between the information such as the document and the like and the base word evaluated by the user, so that the profile can be efficiently changed even when the profile is changed.

[0027]

Embodiments of the present invention will be described below in detail with reference to the drawings. Embodiment 1 <Configuration> FIG. 1 is a flowchart showing an information filtering method according to Embodiment 1 of the present invention. Prior to the description, an information filtering apparatus for implementing the information filtering method will be described.

FIG. 2 is a block diagram showing a first embodiment of the information filtering apparatus of the present invention. The illustrated device includes a filtering unit 1, a profile storage unit 2, an information providing unit 3, an interface management unit 4, and a visualization interface management unit 5. These configurations are implemented on a general computer system including a CPU (central processing unit), a main storage device, an external storage device, and an input / output device.

The filtering unit 1 performs filtering on the information source 10 provided via the information providing unit 3 based on the base words of the profile input by the user 20 (or the profile stored in the profile storage unit 2). The result is output to the interface management unit 4, and when evaluation information from the user 20 is input from the interface management unit 4, the profile is changed based on the evaluation information and stored in the profile storage unit 2. It is a functional unit.

Here, a vector space model and related feedback (Relevance Feedback) will be described as information filtering techniques used in this example.

The vector space model is a retrieval method that takes a word in a text as a vector (for example, this is described in reference [3]). The text is divided into words, and the words are assigned importance levels, which are used as feature vectors. Here, in general, on a distributed network, information resources include image data, moving image data, compressed files, and the like, but in this specific example, information resources are limited to text. Also, a natural language sentence is used for the question sentence, and the question sentence is converted into a feature vector, similarly to a text (or document) word. The search result ranks the similarity between the text and the question sentence. Here, the word is hereinafter referred to as a base word. In addition, the base word is Japanese Patent Application No. Hei 9
As described in the specification of Japanese Patent No. 157909, it constitutes a vector axis in a vector space model. The importance of the base word is a value representing the weight of the base word with respect to a document or the like as a result of the filtering.

The above-mentioned base words are, for example, the following words when described in English information resources. That is, after the unnecessary words are deleted, stemming is performed. Stemming is an operation of making the stem of the inflected form of English into a basic form, and the word resulting from the stemming is the base word of the word. For example, te
Words that have stemmed, such as is → tie, hopping → hop, are base words. Words that do not make sense even in base words are called unnecessary words, for example, “the”, “a”, “to”
And so on.

A vector of a text is represented by the following equation. FIG. 3 is an explanatory diagram of the arithmetic expression of the vector space model. In FIG. 3, Expression (1) is an expression indicating a vector DW of a certain text. In equation (1), dw ₁ , dw ₂ ,.
dw _t is the base language of the importance of the text.

Similarly, the vector Q of the question sentence is expressed by equation (2). Here, q ₁ , q ₂ ,..., Q _t are the degrees of importance of the base words of the question sentence.

Further, the base word importance (weight vector word T _k of the document D _i) is given by equation (3). here,
The meaning of each symbol is as follows. · W _ik: document D _i base language T _k of a vector of weighting · tf _ik of: document D _i base language T _k of the number of occurrences · N of: The total number · n _k of the collected documents: in the collection documents And the number of documents containing the base word T _k

The similarity Sim is given by the following equation (4). Here, there is a threshold such as θ in equation (5). The search results are displayed in order of the degree of similarity among those satisfying Expression (3).

A method in which a user evaluates a retrieved document and feeds a vector of the document that the user is relevant to the question sentence and refines the retrieval formula is referred to as a related feedback.
For example, it is shown in reference [2]).

Although various methods have been proposed for related feedback, generally, equation (6) in the figure is used. Here, rel_docs is defined as:
This is a feature vector of a document of interest to the user. Also,
nonrel_docs is the retrieved document,
This is a feature vector of a document that the user is not interested in. Generally, values 8, 16, and 4 are used for α, β, and γ, respectively.

Returning to FIG. 2, the profile storage 2
This is a functional unit that stores a profile used when the filtering unit 1 performs filtering. The profile storage unit 2 is provided in a storage device such as a magnetic disk device.

The information providing unit 3 has a function of sending an identifier (Identifier (id)) of an information resource in the information source 10 to the filtering unit 1. Here, the information resource is the information source 1
0 indicates one unit of information included in the identifier, and the identifier is information for identifying each information resource. Also, in this specific example,
This identifier is called a document identifier.

The interface management unit 4 has a function of accepting an input from the user 20, transmitting information input by the user 20 to the filtering unit 1, and displaying a filtering result of the filtering unit 1 to the user 20.

When the filtering unit 1 changes the profile and saves the profile in the profile storage unit 2, the visualization interface management unit 5 determines between the base words of the profile when performing filtering based on the stored profile. It has a function of visually displaying each base word so that the stronger the association, the closer the distance between each base word is. Further, the visualization interface management unit 5 has a function of receiving a base word change instruction from the user 20 after visually displaying the same, and transmitting the changed profile to the profile storage unit 2.

That is, the visualization interface management unit 5
When a user's profile is visually displayed in two dimensions, an LSI (Latent Sem) is used to display a base word of the profile.
Using antic indexing), semantically similar base words are displayed collectively at a close place in two dimensions. Here, an LSI is used as one method of visualization, and this LSI will be described below.

LSI is a singular value analysis method.
decomposition) to perform orthogonalization of the feature vector (this is described in, for example, conventional literature [3]).

FIG. 4 is a diagram for explaining the operation of the singular value analysis.
That is, the singular value analysis of an m × n matrix A (m ≦ n, rank (A) = r) is defined as in equation (7). In the figure, rank
(A) shows the rank of the matrix. Here, in ^{1 ≦ i ≦ r, U T} U = V t V = I n, Σ = diag (σ 1, ..., σ n),
σ> 0, and σ _j = 0 when j ≧ r + 1.

Here, U is called a left singular value vector matrix and is a column orthogonal matrix. V is called a right singular value vector matrix and is a column orthogonal matrix. Also, in order to reduce the amount of calculation, Expression (8) in the figure in which the number of dimensions of the matrix is reduced is often used. For example, the matrix is reduced from m to k.
In the figure, A to which ~ is given on the upper side is a matrix A in which the rank of the matrix is lowered.

The search sentence is also compared with text as a k-dimensional space. The search sentence can be defined as in equation (9) in the figure. Here, q to which ~ is added on the upper side indicates a search sentence. Note that the search is performed based on the similarity between the feature vector of the search sentence and the feature vector of the document. In the present invention, only expression (8) is used, but expression (9) is shown for the purpose of describing the LSI.

Returning to FIG. 2, the information source 10 includes a plurality of information resources as described above, and includes data of documents and images. The filtering device is connected to the information source 10 through a computer network.

The filtering unit 1, the information providing unit 3, the interface management unit 4, and the visualization interface management unit 5 are programs for realizing the respective functions on a computer, and a CPU for executing these programs. And a memory or the like.

<Operation> In the first embodiment, it is assumed that the information filtering apparatus is connected to the information source 10 for delivering a document. In this specific example, it is assumed that the user 20 wants to filter out articles of magazines related to “multimedia, intranet, and the Internet” in computer-related books. If the user 20 registers a profile in the information filtering device, after the registration, the information filtering device sets the profile of the magazine every time an article related to “multimedia, intranet, Internet” comes in as the information source 10. The article is delivered to the user 20.

Here, it is assumed that the user 20 considers that the filtered data indicates an article slightly different from the interest of the user 20, and the user 20 wants to change the profile by himself. The operation in this case will be described below with reference to FIG.

[Step S1] The information filtering device is activated by a user 20 activation command.

[Step S2] Here, the information filtering apparatus waits for the user 20 to input data. The user 20 enters the word "multimedia, intranet, internet" as a profile.

[Step S3] Interface management unit 4
Sends the data of the user 20 input in step S2 to the filtering unit 1.

[Step S4] The filtering unit 1
The profile of the user 20 is decomposed into base words, the importance is calculated, and stored in the profile storage unit 2.

[Step S5] Based on the profile stored in the profile storage unit 2 by the filtering unit 1 in step S4, the information providing unit 3 collects documents suitable for the profile from the information source 10, and extracts the collected documents. Send to filtering section 1.

[Step S6] Each time a new book or magazine enters the digital library, the filtering unit 1 filters the documents collected by the information providing unit 3.
Then, the filtering result is sent to the interface management unit 4, and the interface management unit 4 displays the filtering result to the user 20.

[Step S7] The user 20 evaluates the filtering result. For example, as an evaluation method, whether or not the user is interested in the document is input.

[Step S8] The filtering unit 1
Using the above-described related feedback equation (6), the profile of the user 20 is changed, and the profile storage unit 2 is changed.
To be stored.

FIG. 5 is an explanatory diagram of the profile in this case. As shown, the profile is a pair of the base word and the importance.

[Step S9] Whether the user 20 wants to change the profile is input.

[Step S10] The user 20 inputs whether he wants to end the filtering.

[Step S11] If the user 20 wants to change the profile in step S9, the visualization interface management section 5 sets the profile storage section 2
Is visually displayed to the user 20.
Here, the base words are displayed in two dimensions. That is, to display the base words in the profile, the dimension of the feature vector is set to 2
And is displayed using U ₁ in equation (8).

FIG. 6 is an explanatory diagram of a display result by the visualization interface management unit 5. As shown, each base word in the profile is displayed two-dimensionally. The horizontal axis in the figure indicates the value of a one-dimensional vector, and the vertical axis indicates the value of a two-dimensional vector.

[Step S12] Here, the user 20
Is for a base word shown on two dimensions as shown in FIG.
Eliminate base words that are different from your interests. here,
It is assumed that the user 20 has deleted the base words “music” and “server”. This is performed through the visualization interface management unit 5.

FIG. 7 is an explanatory diagram of the profile after deletion. As shown in the figure, “music” and “server” that the user 20 instructed to delete are deleted from the base words.

[Step S13] The visualization interface management section 5 sends the profile changed in step S12 to the profile storage section 2, and the profile storage section 2 rewrites the profile with a new profile of the user 20.

[Step S14] Filtering unit 1
Performs filtering on the newly entered information resource using the new profile stored in the profile storage unit 2 and sends the result to the interface management unit 4. The interface management unit 4 displays the filtering result to the user 20.

[Step S15] If the user 20 wants to end the filtering in step S10, the filtering device ends the filtering.

<Effects> As described above, according to the first embodiment, when the user changes the profile, the base words in the profile are visually displayed in two dimensions. Can efficiently delete unnecessary base words, so that there is an effect that only information resources closer to the user's interest can be filtered.

<< Specific Example 2 >> In the specific example 2, when there are many base words in the profile when visualization is performed and cannot be displayed in two dimensions, clustering of the nearby base words is performed and the clustering of the cluster is performed. A representative value is determined and displayed.

<Structure> FIG. 8 is a diagram showing the structure of the second embodiment. The illustrated device includes a filtering unit 1, a profile storage unit 2, an information providing unit 3, an interface management unit 4, a visualization interface management unit 5, and a clustering unit 6. Here, the filtering unit 1 to the interface management unit 4 have the same configuration as that of the first embodiment, and a description thereof will be omitted.

The clustering unit 6 collectively displays the base words visually displayed in a certain two-dimensional range, and at this time, displays the representative words of the clustering and also displays the importance of the clustering. For outputting display data for the display.

Further, the visualization interface management unit 5
The display data output from the clustering unit 6 is visually displayed, and the profile storage unit 2 is displayed based on the base word change instruction input from the user 20 by this display.
Has the function of changing the profile of

<Operation> Also in the specific example 2, it is assumed that the information filtering device is connected to the information source 10 for delivering a document. In this specific example, it is assumed that the user 20 wants to filter out articles of magazines related to “multimedia, intranet, and the Internet” in computer-related books. If the user 20 has registered a profile in this information filtering device, after the registration, the information filtering device reads articles relating to “multimedia, intranet, and the Internet” from the information source 10.
Every time it comes in, the article of the magazine is delivered to the user 20.

Here, it is assumed that the user 20 considers that the filtered data indicates an article slightly different from the interest of the user 20, and the user 20 wants to change the profile by himself. The operation in this case will be described assuming that the user wants to change the profile in step S9 of FIG.

FIG. 9 is a flowchart showing the operation of the second embodiment. [Step S21] If the user 20 wants to change the profile, the visualization interface management unit 5 clusters the profiles in the profile storage unit 2 and visually displays them to the user 20.

FIG. 10 is an explanatory diagram when visualized and displayed. Here, the base words are displayed two-dimensionally, and the importance is displayed for each clustering. Here, the base words in the profile are displayed by using U in equation (7), with the dimension of the feature vector being 2. "Music", "server", "internet", "intranet",
“Multi” or the like indicates a representative value of each clustering. Then, the importance of each clustering is also displayed next to it.

Here, the representative importance of clustering indicates the sum of the importance of the base words. For example, if the representative value is intranet importance (intranet, 0.
543), (mounted, 0.037), (constructed, 0.033), and (software, 0.029), the sum of the importance of 0.642 is displayed.

[Step S22] The user 20 enters the state shown in FIG.
With respect to the base words displayed on the two-dimensional plane indicated by, the base words that are different from the user's interest are deleted. Here, it is assumed that the user 20 has deleted the base words “music” and “server”. This is performed through the visualization interface management unit 5.

[Step S23] The visualization interface management section 5 sends the profile changed in step S2 to the profile storage section 2. Thereby, the profile storage unit 2 rewrites the new profile of the user 20.

[Step S24] Filtering unit 1
Performs filtering on a newly entered information resource using a new profile, and sends the result to the interface management unit 4. Interface management unit 4
Displays the filtering result to the user 20.

<Effects> As described above, according to the specific example 2, when the user changes the profile, the base words in the profile are visually displayed two-dimensionally, and
Since similar base words are clustered and displayed, the following effects are obtained in addition to the effects of the first embodiment. That is, if the number of base words in the profile increases due to the related feedback, the number of base words increases even if the profile is simply displayed in two dimensions, and the problem that the relationship between the base words is difficult to understand can be eliminated. This makes the display easier to see for the user, and has an effect that when the user changes the profile, the display can be performed efficiently.

<< Specific Example 3 >> In the specific example 3, at the time of visualization, both the base word of the profile and the filtered document are displayed.

<Structure> FIG. 11 is a view showing the structure of the third embodiment. The illustrated device includes a filtering unit 1, a profile storage unit 2, an information providing unit 3, an interface management unit 4, a visualization interface management unit 5, and a document visualization information generation unit 7. Here, since the filtering unit 1 to the interface management unit 4 have the same configuration as those of the first and second specific examples, description thereof will be omitted.

The document visualization information generation unit 7 has a function of passing information for visualizing a filtered document in two dimensions to the visualization interface management unit 5.
That is, the document visualization information generation unit 7 includes the filtering unit 1
The information of the filtering result by the profile of is such that the stronger the association between the information, the shorter the distance, and
It has a function of outputting data of a visual positional relationship to the visualization interface management unit 5 so that the stronger the association between each information and each base word, the closer the distance.

Further, the visualization interface management unit 5
A function of visually displaying the display data output from the document visualization information generation unit 7 and changing the profile of the profile storage unit 2 based on the base word change instruction input from the user 20 by this display. Have.

<Operation> In the third embodiment as well, it is assumed that the information filtering device is connected to the information source 10 for delivering a document. In this specific example, it is assumed that the user 20 wants to filter out articles of magazines related to “multimedia, intranet, and the Internet” in computer-related books. If the user 20 has registered a profile in this information filtering device, after the registration, the information filtering device reads articles relating to “multimedia, intranet, and the Internet” from the information source 10.
Every time it comes in, the article of the magazine is delivered to the user 20.

Here, it is assumed that the user 20 considers that the filtered data indicates an article slightly different from the interest of the user 20 and that the user 20 wants to change the profile by himself. The operation in this case will be described assuming that the user wants to change the profile in step S9 of FIG.

FIG. 12 is a flowchart showing the operation of the third embodiment. [Step S31] If the user 20 wants to change the profile, the visualization interface management unit 5 visualizes the profile stored in the profile storage unit 2 and the information indicating the document generated by the document visualization information generation unit 7 so that the user 20 20 is displayed.

FIG. 13 is an explanatory diagram when visualized and displayed. Here, the base word in the profile is displayed using U in equation (7), with the dimension of the feature vector being 2. Here, when the filtered document is displayed, the dimension of the feature vector is set to 2, and V
Display using ^T.

In FIG. 13, D1, D2, D2, D4, D
Reference numeral 39 denotes a two-dimensional position of the filtered document. In the figure, it can be said that the closer the distance to the base word or another document is, the more similar the base word or document is. here,
(○) is a document that the user 20 has input in step S7 of FIG. 1 in the specific example 1, and (x) is
This is the document to which the user 20 has input in step S7 that he is not interested. That is, D1, D2, D4, and D39 in the figure are documents that the user 20 has input as being interested, and D3 is a document that the user 20 has input as not being interested.

[Step S 32] The user 20
With respect to the base words displayed on the two-dimensional plane indicated by, the base words that are different from the user's interest are deleted. Here, since the document D3 input that the user 20 is not interested is close to the base word “music” in FIG.
Assume that the base word "music" has been deleted. This is performed through the visualization interface management unit 5.

[Step S33] The visualization interface management section 5 sends the profile changed in step S2 to the profile storage section 2. Thereby, the profile storage unit 2 rewrites the new profile of the user 20.

[Step S34] Filtering unit 1
Performs filtering on the newly entered information resource using the new profile stored in the profile storage unit 2 and sends the result to the interface management unit 4. The interface management unit 4 displays the filtering result to the user 20.

<Effects> As described above, according to the third embodiment, when the user changes the profile, the base words in the profile are visually displayed two-dimensionally, and
Since the filtered document is displayed two-dimensionally, the following effects are obtained in addition to the effects of the first embodiment. That is, since the user can easily grasp the relationship between the document evaluated by the user and the base word, the user can efficiently change the profile when the user changes the profile.

<< Usage Mode >> The present invention is not limited to the above specific examples, but can be applied to the following modified examples.

[0098] Instead of the information providing unit 3, the present invention can be applied to a connection to a normal information retrieval system.

Instead of the information providing unit 3, the present invention can be applied to a case where the system is connected to a system for receiving articles such as texts and images from a plurality of information sources (for example, newspaper companies, publishers, and news agencies). .

In each of the specific examples, the vector space model is used as the information filtering technique. However, a probability model used in information retrieval may be used.

In each of the specific examples, the related feedback is used as a method for correcting the user profile.
A machine learning method may be used.

In each specific example, the information source 10 is one, but the present invention can be applied to a plurality of information sources.

In the second embodiment, the sum of the importance of the base words included in the clustering is used to calculate the importance of the clustering. The highest value of the importance of the base word in the clustering may be displayed.

[Brief description of the drawings]

FIG. 1 is a flowchart illustrating an information filtering method according to a first embodiment of the present invention.

FIG. 2 is a configuration diagram of an information filtering device according to a first embodiment of the present invention.

FIG. 3 is an explanatory diagram of an operation expression of a vector space model.

FIG. 4 is an explanatory diagram of calculation of singular value analysis.

FIG. 5 is an explanatory diagram of a profile displayed by the information filtering device according to the first embodiment of the present invention.

FIG. 6 is an explanatory diagram showing a display result in the information filtering device according to the first embodiment of the present invention.

FIG. 7 is an explanatory diagram of a profile after deleting base words in the information filtering device according to the first embodiment of the present invention.

FIG. 8 is a configuration diagram of an information filtering device according to a second embodiment of the present invention.

FIG. 9 is a flowchart illustrating an operation of the information filtering device according to the second embodiment of the present invention.

FIG. 10 is an explanatory diagram showing a display result in the information filtering device according to the second embodiment of the present invention.

FIG. 11 is an explanatory diagram of an information filtering device according to a third embodiment of the present invention.

FIG. 12 is a flowchart illustrating an operation of the information filtering device according to the third embodiment of the present invention.

FIG. 13 is an explanatory diagram showing a display result in the information filtering device according to the third embodiment of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Filtering part 2 Profile storage part 4 Interface management part 5 Visualization interface management part 6 Clustering part 7 Document visualization information generation part

Claims (6)

[Claims] 1. First, filtering is performed on an information source based on a base word of a profile input by a user, and a result of the filtering is displayed. Next, the user attaches a filter based on the display of the filtering result. When the base words of the profile are changed and displayed again according to the score, the respective base words are visually displayed such that the stronger the relation between the base words of the profile at the time of filtering, the closer the distance between the base words becomes. An information filtering method, comprising: 2. The information filtering method according to claim 1, wherein each of the base words is displayed, and the base words whose association between the base words is within a predetermined value range are clustered. And an information filtering method for displaying importance indicating the degree of weight of the clustering in the filtering result and a base word representative of the clustering. 3. The information filtering method according to claim 1, wherein each base word of the profile is displayed, and information on a result of filtering by the profile is set such that the closer the information is, the shorter the distance is. And an information filtering method for visually displaying such that the stronger the association between the respective information and the respective base words, the shorter the distance. 4. Filtering of an information source is performed based on a base word of a profile input by a user, a result is output, and when evaluation information is input by the user, a profile is generated based on the evaluation information. A filtering unit to be changed; an interface management unit that receives an instruction from a user and performs a result from the filtering unit to the user; and a case where the filtering unit changes a profile.
Information filtering comprising: a visualization interface management unit for visually displaying each base word such that the stronger the association between the base words of the profile when filtering is performed, the closer the distance between the base words becomes. apparatus. 5. The information filtering apparatus according to claim 4, wherein base words whose association between the base words is within a predetermined value range are clustered, and for each clustering, the weight of the clustering in the filtering result is obtained. A clustering unit that outputs a degree of importance indicating the degree of the clustering and a base word of the representative of the clustering, and a visualization interface management that visually displays the importance of the clustering and the base word of the representative outputted from the clustering unit And an information filtering device. 6. The information filtering apparatus according to claim 4, wherein the information of the filtering result by the profile of the filtering unit is set such that the stronger the relation between the pieces of information is, the shorter the distance is, and the information is filtered. A document visualization information generation unit that outputs data of a visual positional relationship such that the stronger the association between the base words is, the closer the distance is; a visual display that visually displays an output result of the document visualization information generation unit An information filtering device, comprising: an interface management unit.