JP2001318947A - Information integration system, its method and recording medium having information integration program recorded thereon - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make it possible to acquire variable information also from an information source of low reliability by providing the degree of confidence of information itself simultaneously with the possibility of the information. SOLUTION: Information development parts 11 to 13 extend the information of information sources by referring to respective tables 14 to 16 describing the degree of the relation between the pieces of information and an information integration part 18 integrates the extended information on the basis of a similarity function of all the information sources, finds out the possibility value of information from the similarity obtained by using similarity function and finds out the reliability of information as a confidence degree value from the scatter of the similarity. An expression generation part 19 selects an expression from a possibility expression table on the basis of the possibility value of information, selects the expression from a confidence degree expression table on the basis of the confidence degree value and generates a natural sentence by the combination of the selected results. An order determination part 20 selects one of the information by applying a selection function on the basis of the possibility value and confidence degree value of each information and determines the order of the arrangement of all the information.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a system for integrating a plurality of pieces of information by assigning a possibility value and a certainty factor, and a method for integrating the same, and a recording medium on which the program is recorded. The present invention relates to an information integration system, an integration method, and a recording medium that can acquire valuable information from a low information source and obtain a natural language expression that is easy for a user to understand based on certainty and possibility.

[0002]

2. Description of the Related Art In recent years, with the spread of a network infrastructure represented by the Internet, it has become possible for a user to easily acquire a large amount of various data. However, on the other hand, due to the ease of information transmission, such as the Internet, unreliable information sources, including missing or incorrect information, have been flooded. In order to extract highly reliable information from such unreliable information sources, information is generally obtained from a plurality of information sources, and various information presentation systems have been proposed. . Methods used in conventional information presentation systems can be roughly classified into the following four types. (1) A method in which information from a plurality of information sources is presented to the user as it is to leave the decision to the user. (2) A majority decision is made between the information sources, and the value adopted by the most information sources is determined. How to present it to the user,
(3) a method of presenting the average value of the information source to the user, (4)
A method of referring to the value of the most reliable information source.

[0003]

However, the method (1) has a problem in terms of convenience because the user has to judge by himself or herself a plurality of pieces of information. In the method (2), for example, if the information source is 3
There is a problem that if each of the information is different, a majority decision cannot be made between the different information, and a value cannot be selected. Further, the method (3) is a method applicable only to the case of the numerical attribute. For example, with the information of the nominal attribute such as the weather, the average value cannot be simply obtained. Although it is possible to arrange the nominal attribute information on a one-dimensional axis and assign a value to the pseudo-numerical attribute to integrate them, the following problem still remains.
For example, taking the weather forecast as an example, the nominal information of fine, cloudy, and rain is one-dimensionally fine = 1.0, cloudy = 0.5, rain = 0.0.
When the information source A is fine and the information source B is rain, the average value is 0.5, and the user is presented with nominal information of cloudy. Also, when both the information sources A and B are cloudy, the average value is 0.5.
, And the user is presented with nominal information of cloudy. In any case, the user is given a forecast of cloudiness, but intuitively, despite the fact that the reliability of the information is different, the user has no way of knowing that. In particular, in the former case, the obtained result remains questionable. Furthermore, in the former case, although the possibility of rain is low but existing, the information is lost due to the integration. For example, "Does it rain? However, there is a problem that the user's interest cannot be properly answered.

In the method (4), it is necessary to evaluate the reliability of the information source in advance, and even if the information source is highly reliable, a network malfunction or a server of the information source is required. Lack of information due to maintenance and the like is essentially unavoidable. This is inferior to the method of referring to multiple information sources. In addition, since the methods (1) and (4) do not have an index for evaluating the reliability of information, there is also a problem that the user has to judge how much the information is reliable. . Further, in a method of integrating information, such as the above methods (2) and (3), it is possible to assign a possibility value to information using a fuzzy inference method or a probability inference method. I don't know how reliable the degree is. For example, in the case of a weather forecast, as a result of integrating multiple pieces of information,
If the prediction that “the probability of rain is 0.7” is obtained as integrated information, the reliability of the obtained result itself differs depending on whether the information sources used for integration are large or small. Nevertheless, the user has no way of knowing it.

In order to solve the above problem, as a method of expressing the upper and lower bound probabilities in two values, for example, a dissertation “Uncertain knowledge expression method and inference method considering reliability of inference” (Suzuki) Makoto, Toshiyasu Matsushima, Shigeichi Hirasawa: There is a method of applying IPSJ Transactions, Vol. 35, No. 5, pp. 691 to 705, 1994). When this method is applied to the above example, The likelihood of rain is in the range of 0.5 to 0.8. " In this case, since the value of the possibility is represented by a section, the reliability of the integration result can be known as the width of the section value, but what information the user seeks is what information is obtained. And that the information is reliable, is a sufficient solution in that the interpretation of the interval values into the information desired by the user is still left to the user himself. Not. Furthermore, if a plurality of numerical values of the same scale (in this case, two possible values) are added to the information, the user's intuitive understanding may be impaired.

Accordingly, an object of the present invention is to solve these conventional problems, not only to present the possibility of information, but also to show how reliable the integrated information is, and to provide the user with low reliability. It is an object of the present invention to provide an information integration system and an information integration method capable of acquiring valuable information even from an information source and obtaining a natural language expression that is easy for a user to understand based on a certainty factor and a possibility.

[0007]

In order to achieve the above object, the information integration system of the present invention compares information from a plurality of information sources to give a reasonable possibility, and at the same time, gives a certainty to the integrated information itself. A system for assigning degrees of information, expands information of information sources into related information with reference to a table describing the degree of relevance between information, and integrates the expanded information for all information sources based on a similarity function. And the likelihood value of the information is obtained from the similarity obtained by the similarity function, and the reliability value of the information is obtained from the dispersion of the similarity. Also, based on the likelihood value and confidence value of the information obtained by the above system,
A method of generating a natural sentence that is easy for a user to understand, wherein the method selects an expression from a possibility expression table based on the possibility value of the information, and selects an expression from the confidence expression table based on the confidence value. Natural sentences are generated in combination. Furthermore, a method for determining information to be presented to a user based on the likelihood value and certainty value of a plurality of pieces of information obtained in the above system, wherein a selection function is determined based on the likelihood value and certainty value of each piece of information. This is to select one piece of information by applying.

[0008] In the present invention, the above-described system can compare information from a plurality of information sources to give a reasonable possibility, and at the same time, give the certainty of the integrated information itself. This is an advantage that the user can intuitively understand how reliable the information is. Further, according to the above method, an expression is selected from the possibility expression table based on the possibility value of the information, an expression is selected from the confidence expression table based on the confidence value, and a natural sentence is generated by combining the expressions. Therefore, there is an advantage that a natural sentence that can be easily understood by the user can be generated based on the possibility value and the certainty value of the obtained information. Further, the method described above has an advantage that information to be presented to the user can be determined based on the possibility value and the certainty factor value of the plurality of pieces of information obtained.

[0009]

Embodiments of the present invention will be described below in detail with reference to the drawings. FIG. 1 is a block diagram of an information integration system according to an embodiment of the present invention. The information integration system of the present invention integrates information from a plurality of information sources, outputs a plurality of information with a possibility value and a certainty factor, and generates a natural sentence expression for the information. FIG.
An information expansion unit 11 to 13 for assigning a possibility value to information obtained from a certain information source (input) based on the degree of relevance, and a table describing information of the information source in relation to the information 14 to 16 and an information integration unit 18 that integrates information from a plurality of developed information sources and gives a certainty factor
And an expression generation unit 19 that generates a natural language expression based on the assigned possibility value and certainty value, and an order determination unit 20 that determines the order based on the possibility value and the certainty value. ing. Here, if there is no need to perform natural language expression, there is no need to perform the processing of the expression generation unit 19. If the ordering is not required, there is no need to perform the processing of the order determining unit 20.

FIG. 2 is a processing flowchart of the information expanding unit in FIG. FIG. 6 shows an example of the data configuration of the information-related table, and FIG. 7 shows an example of the data configuration of the development information table. As a specific example, the integration of weather forecast information will be described as an example. Each input in FIG. 1 receives information from an individual information source. In the weather forecast example,
The information source indicates a WEB site that provides weather information. The information indicates a weather forecast such as "sunny" or "cloudy". In the present invention, information input from each information source is received (step 101), and is sent to the information expanding units 11 to 13 corresponding to each information source. Information development unit 1
1 to 13 refer to the information related table 10 as shown in FIG. 6 (step 102) and expand the input information. As shown in FIG. 6, the table 10 is created by various methods such as a result derived or integrated from a difference between an observed value and each information source and a result derived from the difference between each information source or manually created. be able to. here,
It does not mention how to make it.

When an input is given, the information developing units 11 to 13 refer to the information-related table 10 shown in FIG. 6 to fetch a row having an input weather that matches the input (step 103). The matching line is output (step 104) and passed to the information integration unit 18 to end the processing. For example, when the input given from the information source A to the information developing units 11 to 13 is “fine”, the input weather of the related information table that matches the input is referred to by referring to the information related table 10 of FIG. By selecting the row, a development information table as shown in FIG. 7 is obtained. All the information developed as shown in FIG. 7 is sent to the information integration unit 18. Similar processing is performed for all information sources. However, information need not always be present in all sources.

FIG. 3 is a processing flowchart of the information integration unit in FIG. 8 is a data configuration example of a reliability coefficient table, FIGS. 9 and 10 are data configuration examples of each development information table from information sources B and D, and FIG. 11 is a data of a table in which development information of each information source is combined. FIG. 12 shows an example of a data configuration of a table supplementing the contents of FIG. 11, FIG. 13 shows an example of a table data configuration of a reliability coefficient of an information source, and FIG. 14 shows an example of a data configuration of an integrated table. The information integration unit 18 receives the expansion information tables from the information expansion units 11 to 13 of each information source, integrates the same information, and performs a process of assigning a possibility value and a certainty factor to each information.
Here, the likelihood value is a value indicating the possibility that the information is true, and the certainty factor is a value indicating how reliable the possibility value is. That is, the information integration unit 18 first reads the development information table (step 201), identifies the information source where information exists and the information source that does not exist, and counts the information sources that do not exist (step 202). Based on the number of information sources in which information exists (this is referred to as an integration target information source), the reliability coefficient of the integrated information is determined with reference to the reliability coefficient table 10c as shown in FIG. 8 (step 203). For example, if the information sources are A, B,
When there are four information sources C and D, and information cannot be obtained from the information source C, the integration target information source is 3, and the reliability coefficient is determined to be 0.9 from the reliability coefficient table 10c in FIG. .

Next, a union of the included information is generated with reference to the development information table received from the integration target information source (step 204). Then, a two-dimensional table including the union of the generated information and the information source to be integrated is created (step 205). For example, if the information source is A,
B, C, and D. When information cannot be obtained from the information source C, the development information table from the information source A is shown in FIG.
The expansion information table from the information source B is the expansion information table shown in FIG. 9 and the expansion information table from the information source D is the expansion information table shown in FIG. Next, a synthesis table shown in FIG. 11 is created from these tables. Then, the values of the information sources to be integrated are input into the cells of the synthesis table (step 206).
Next, with respect to the combined development information table of FIG.
A valueless cell is supplemented with 0.0 (step 207).
By this processing, the complement table shown in FIG. 12 is obtained. Next, integration of information is started based on the table of FIG. 12 complemented by the above processing (step 208), and the certainty factor and the possibility are determined. Steps 208 and 213 mean the start block and the end block of the same process, and indicate that the process of integrating information is continuously performed during this period. Next, a certainty factor is calculated for each possibility value (step 209).

The integration of information is performed as follows. That is, the information source set to be integrated is S (= ｛s ₁ ,...)
s _j ｝). At a certain information source s _j (∈S), the possibility that a certain weather event w occurs is given by Pos _w (s _j ). Possibility of weather event w
_w (s _j) is a confidence CF (Poss _w (s _j)) is calculated by the following equation (Equation 3).

(Equation 3) Here, the function sim () is a function indicating the degree of coincidence between two possibilities. Various functions can be applied to sim (). For example, the function of the following equation (Equation 4) can be used. This function indicates the similarity of the possibility obtained from each information source as the similarity. It is considered that the similarity is high if sim (ΔPoss w) is large, and the similarity is low if sim (ΔPoss w) is small.

(Equation 4) Α (s) is a weight indicating the reliability coefficient of the information source s, and takes a value of [0, 1]. That is, α (s) is 1
Is the highest reliability, and 0 is the lowest reliability.

Next, α (s) is determined with reference to the reliability coefficient table of the information source as shown in FIG. α (s)
Does not need to be static, and may vary according to the update frequency, the latest update date and time, the number of information sources, and the like. As the Pos value (possibility value) of the information w at the time of integration, a value that satisfies the following expression (Equation 5) is adopted.

(Equation 5) In this way, if the certainty is obtained by the calculation, the possibility value having the largest certainty is adopted as the possibility value of the integrated information and written into the integrated table (step 210). The certainty factor of the possibility value is adopted as the certainty factor of the integrated information (step 211), and the certainty value is multiplied by a confidence coefficient and written into the integrated table (step 212). Through these processes, an integrated table as shown in FIG. 14 is generated. The list of integrated information is stored in the expression generation unit 19.
(Step 214).

FIG. 4 is a processing flowchart of the expression generation unit in FIG. FIG. 15 is an example of a data configuration of the possibility value correspondence table, and FIG. 16 is an example of a data configuration of the certainty value correspondence table. The expression generation unit 19 generates a corresponding linguistic expression based on the likelihood value and the certainty factor value. However, if generation of an expression is not necessary, this processing may not be performed. First, the expression generation unit 19 reads the integration table passed from the information integration unit 18 (Step 30).
1), retrieval of each row of the read table is started (step 302). Steps 302 and 307 mean the start block and the end block of the same process, and indicate that the process of retrieving each row of the table is performed continuously during this period.
Next, the expression generation unit 19 first generates a linguistic expression based on the likelihood value of each piece of information. FIG. 15 is a table 10a showing the correspondence between the possibility values and the expressions. Where 0
In the case of <a <b <c <1, the value of the item of information in the integrated table (nominal attribute weather forecast, for example, “fine”) is inserted in [information]. By referring to the possibility value correspondence table 10a, an expression corresponding to the possibility value is extracted from the possibility value correspondence table 10a (step 303).

Next, the expression generator 19 generates a linguistic expression based on the certainty factor of each piece of information. FIG. 16 is a table 10b showing a correspondence relationship between a certainty factor value and an expression. By referring to the certainty factor value correspondence table 10b,
The expression corresponding to the certainty value is stored in the certainty value correspondence table 10b.
Is retrieved (step 304). Where 0 <d <e
<F <g <1, and [possibility] is the table 10 in FIG.
The expression determined by the likelihood value of b is inserted. For example, a = 0.2, b = 0.6, c = 0.8, d = 0.
Assuming that 6, f = 0.8, the above processing yields the integrated table with expressions shown in FIG. In this way, the expressions obtained in the above steps are synthesized (step 30).
5). By this processing, the generated expression is added to the integration table (step 306). The integrated table with expressions is output to be passed to the order determination unit 20 (step 30).
8).

FIG. 5 is a processing flowchart of the order determining unit in FIG. FIG. 18 is a diagram illustrating a data configuration example of the sorted expression-added integrated table. The order determining unit 20 reads the integration table passed from the expression generating unit 19 (Step 401), and reads the integrated table.
0 applies the order calculation function to the likelihood value and certainty value of each piece of information (step 402). That is, the order determining unit 20
Applies the following equation (Equation 6) based on the certainty factor value and the likelihood value given to each piece of information, sorts the obtained values in descending order, and determines the order (step 403).

(Equation 6) Here, the Pos value means the possibility value, and the CF value means the certainty value. Β is a parameter that determines which of the Pos value and the CF value is important, and is set in advance by the user or the system. When β is 1, the harmonic mean (n
1 / x1,1 / for the numbers x1, x2,.
x2,... 1 / xn). For example, when the integrated table in FIG. 17 is input and β = 1, this table is sorted and
Is obtained. It is sorted in descending order of likelihood value. The sorting result (table in FIG. 18) is output from the output unit 21 (step 404).

In this way, as shown in the table of FIG. 17, after a plurality of expressions are generated with the possibility value and the certainty factor value, the F value of the expression (Equation 4) is changed in order to order them. Used. The F value is calculated for each row, and the rows are rearranged in descending order of the F value. For example, "What's the weather tomorrow? ], It is inappropriate to present all the information as a dialogue system, so sort so that one of the most appropriate sentences is spoken. This is the result of sorting the table in FIG. 18, and from this table, the highest F-number, that is, the highest rank “certainly clear” is uttered. Here, one is selected, but a plurality such as the top three may be used.

Each step of the flowcharts of FIGS. 2 to 5 is converted into a program, and the converted program is stored on a CD-ROM.
By storing the recording medium in a recording medium such as a ROM or a hard disk, the recording medium can be mounted on an arbitrary personal computer and installed from the recording medium to the main memory, or
By downloading the program to the main memory of any other personal computer via a network and executing the program on those personal computers, the information integration method of the present invention can be realized at any place.

[0021]

As described above, according to the present invention,
By comparing information from multiple information sources and giving reasonable possibilities, and by giving the certainty of the integrated information itself, it is possible to present the possibility of information and at the same time how reliable the integrated information is Can be presented. Therefore, valuable information can be obtained even from an information source that is unreliable to the user. In addition, it is possible to acquire a natural language expression that is easy for the user to understand based on the certainty factor and the possibility.

[Brief description of the drawings]

FIG. 1 is an overall configuration diagram of an information integration system according to an embodiment of the present invention.

FIG. 2 is a processing flowchart of an information expanding unit in FIG. 1;

FIG. 3 is a processing flowchart of an information integration unit in FIG. 1;

FIG. 4 is a processing flowchart of an expression generation unit in FIG. 1;

FIG. 5 is a processing flowchart of an order determining unit in FIG. 1;

FIG. 6 is a diagram showing a data configuration example of an information relation table used in the present invention.

FIG. 7 is a diagram showing a data configuration example of a development information table used in the present invention.

FIG. 8 is a diagram showing a data configuration example of a reliability coefficient table used in the present invention.

FIG. 9 is a diagram showing a data configuration example of a development information table from an information source B used in the present invention.

FIG. 10 is a diagram showing a data configuration example of a development information table from an information source D used in the present invention.

FIG. 11 is a diagram showing a data configuration example of a table in which development information of each information source used in the present invention is synthesized.

FIG. 12 is a diagram showing an example of a data configuration of a table supplementing FIG. 11;

FIG. 13 is a diagram showing a table of information source reliability coefficients used in the present invention.

FIG. 14 is a diagram of an integration table used in the present invention.

FIG. 15 is a diagram of a possibility value correspondence table used in the present invention.

FIG. 16 is a diagram of a certainty factor value correspondence table used in the present invention.

FIG. 17 is a diagram of an integrated table with expressions used in the present invention.

FIG. 18 is a diagram of an integrated table with expressions used in the present invention.

[Explanation of symbols]

11 to 13: information development unit, 14 to 16: information relation table, 18: information integration unit, 19: expression generation unit, 20: order determination unit, 21: output unit.

Claims (7)

[Claims] 1. An information integration system for comparing information from a plurality of information sources to give a reasonable possibility and giving a certainty factor of information itself obtained by integrating the plurality of information, comprising: An information expansion unit that expands the information into related information with reference to a table describing the degree of relevance between information, and integrates the expanded information for all information sources based on a similarity function and obtains the similarity obtained by the similarity function. An information integration unit that obtains the likelihood value of the information from the degree of similarity and obtains the reliability of the information as a certainty value from the state of dispersion of the similarity, and selects an expression from the possibility expression table based on the possibility value of the information, An expression generator that selects an expression from the certainty expression table based on the certainty value and combines the selected expressions to generate a natural sentence, and an array order based on the likelihood value and certainty value assigned to each piece of information Determine Information integration system; and a sequence determination section for outputting. 2. A method of comparing information from a plurality of information sources to give a reasonable possibility and at the same time giving certainty of the information itself by integrating the information, comprising: Developing the related information with reference to a table describing the degree of relevance of the information, integrating the expanded information based on the similarity function for all information sources, and obtaining information from the similarity obtained by the similarity function. An information integration method, comprising the steps of: determining the likelihood of the information; and determining the reliability of the information as a confidence value from the state of dispersion of the similarity. 3. The information integration method according to claim 2, wherein, in addition to the steps, selecting an expression from a possibility expression table based on the possibility value of the information; An information integration method comprising: selecting an expression from a certainty expression table based on the expression; and generating a natural sentence that is easy for the user to understand by combining the selected expressions. 4. The information integration method according to claim 2, wherein, in addition to said steps, a step of applying a selection function based on a possibility value and a certainty factor value of said information; And a step of selecting information to determine information to be presented to the user by selecting one piece of information. 5. The information integration method according to claim 2, wherein the similarity function used in the step of integrating based on the similarity function is a function indicating the degree of coincidence between two possibilities, for example, w is a phenomenon, If ΔPoss is the difference between the possibilities, An information integration method characterized by using the following function. 6. The information integration method according to claim 4, wherein the function used in the step of sorting based on the value of the function is based on the confidence value Poss and the possibility value CF assigned to each piece of information. An information integration method comprising applying an expression and sorting the obtained values in descending order to determine a rank. (Equation 2) (β is a parameter that determines which of the Poss value and the CF value is important) 7. A program readable recording medium, wherein each step of the information integration method according to claim 2 is converted into a program, and the converted program is stored in the recording medium.