JP3090233B2 - A method for identifying associations between complex information - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a database system utilizing information provided as video, information provided as print, and complex information provided through other information transmission media. The present invention relates to a method for identifying the association between complex information used for identifying the degree of association.

[0002]

2. Description of the Related Art In a conventional database system for handling multimedia information, the structure of data used in the database and the relationship between the data differ depending on the type and target of the media to be handled. Was. Therefore, the information that can be handled by each database depends on the originally planned application, such as map information and document information, and can be used only with the relationship between data defined in the database. Did not.

[0003] Furthermore, keywords are mainly used in the expression of relevance, but the task of giving appropriate keywords and attributes to a database in order to search for necessary data is as follows.
This was a heavy burden for the user.

[0004]

By the way, between individually created databases such as a map database and a document database, for example, it is desired to refer to map data from document data by using individual media or database. There is a demand to use multimedia data that exceeds the standard.

However, in the prior art, for example, in order to express a newly required relationship between two databases, a keyword capable of searching for a new relationship is added, and then a new reference relationship is created. It was necessary to create the database shown.

SUMMARY OF THE INVENTION The present invention provides a mechanism for identifying relevance to an arbitrarily created multimedia database, thereby creating an arbitrary multimedia database without creating a new reference database even when a new reference relationship is required. It is an object of the present invention to provide a method for identifying association between multiple pieces of information that can make various associations between pieces of information.

[0007]

According to a first aspect of the present invention, there is provided a method for identifying a relationship between complex information including image information, audio information, text information, and other media information. A feature expression sentence expressing each feature in natural language and the time of occurrence of the information are added, the feature expression sentence of each information is analyzed by natural language to extract each feature amount, and the generation of each information is generated. Using the time data and information on the time extracted by performing a natural language analysis on the characteristic expression sentence of each information, the information is converted into the absolute time at which the information actually occurred,
The degree of association between the pieces of information is calculated by comparing the feature amount of each piece of information and the absolute time.

According to a second aspect of the present invention, in the method for associating multiple pieces of information according to the first aspect, the characteristic expression sentence of each information is classified into a part of speech of an independent word and a modifier, and The degree of relevance of each part of speech is calculated based on the amount of words that match each word, and the occurrence time data of each piece of information is converted into absolute time using words representing the adverb and the date and time in the characteristic expression sentence of each piece of information. And the absolute time of each piece of information is assigned to each axis in an n + 1-dimensional space, and the degree of association between the pieces of information is determined by n + 1 It is characterized in that it is calculated as a pseudo distance in a dimensional space.

According to a third aspect of the present invention, in the method for identifying the relation between the complex information according to the second aspect, the amount of coincidence of the word for each part of speech and the absolute time of each information are determined on an n + 1-dimensional space. When assigning to each of the axes, a range of a distance that can be handled on a predetermined axis is designated to control the distance on the (n + 1) -dimensional.

According to a fourth aspect of the present invention, in the method for identifying the relation between complex information according to the second aspect, when calculating the amount of matching of the words of each part of speech, the relation of the characteristic expression sentence is calculated. Analyzing the information and separating it into a part that expresses the meaning of the information and a part that modifies it, weights the amount that matches in the part that expresses the significance with respect to the amount that matches in the modified part of each information. And

[0011]

According to the present invention, a feature expression sentence in which individual features are expressed in natural language is added to information corresponding to each media, and each feature expression sentence is analyzed by natural language to compare each feature amount extracted. Thereby, the degree of association between the pieces of information is calculated.
In addition, regarding the processing method of the natural language analysis, a plurality of relevances can be obtained between pieces of information by changing extraction of a regional attribute, extraction of a person, and the like.

[0012] Further, the occurrence time data of the information is added to each information, the characteristic expression sentence of each information is analyzed, and the absolute time when the information actually occurs is calculated from the occurrence time data, and each information is plotted on a time axis. Align.

Here, by comparing the feature amount and the absolute time of each information, it is possible to accurately obtain the degree of association between each information. That is, it is possible to identify the relevance of each information between a plurality of databases without being caught by a structure or a keyword.

[0014]

FIG. 1 is a block diagram showing the configuration of an embodiment of a multimedia database system for realizing a method for identifying a relationship between complex information according to the present invention. In the present embodiment, an article B and a news C similar to a certain article A are taken as examples of newspaper and television news in a database.
The process of searching for will be described.

Referring to FIG. 1, a multimedia database system according to the present embodiment has a television news database 1 corresponding to individual media such as a television and a newspaper.
0 and ₁ and newspaper articles database 10 _2, the traction link module 20, the structural link module 30, a search condition output module 40, the search result acquisition module 50
, A content link module 60, a time link module 70, and a control module 80.

The tow link module 20 associates individual data corresponding to each medium with a keyword added thereto. The structure link module 30 operates the structure of individual data corresponding to each medium. The search condition output module 40 accesses each of the databases 10 ₁ and 10 ₂ with the search conditions created by the towing link module 20 and the structure link module 30. The search result acquisition module 50 receives the information searched by the search condition output module 40, and receives the content link module 6
0 and time link module 70. The content link module 60 performs association using the characteristic expression sentence of each information. The time link module 70 performs association using the occurrence time of each piece of information. Control module 80
Controls each module while interfacing with the user, and creates a final search result from the information associated with the content link module 60 and the time link module 70.

Here, the configurations of the content link module 60 and the time link module 70 are shown in FIG. In the figure, a content link module 60 includes a search sentence analyzing unit 61.
, A result sentence analysis unit 62, a morphological analysis unit 63, and a comparison module 64. Search sentence analyzer 61
Analyzes the characteristic expression sentence of the article A to be searched, which is provided from the control module 80, based on the comparison element, and provides the result to the comparison module 64. The result sentence analysis unit 62
The search result of each database is fetched through the search result acquisition module 15, the characteristic expression sentences of the articles B and news C are similarly analyzed based on the comparison elements, and the result is given to the comparison module 64. The morphological analysis unit 63 assists the search sentence analysis unit 61 and the result sentence analysis unit 62 in analyzing the characteristic expression sentence. The comparison module 64 compares the analysis results of the search sentence analysis unit 61 and the result sentence analysis unit 62 with the threshold value given from the control module 80, and gives each feature amount of the article B and the news C.

The time link module 70 includes a search time determining section 71, a result sentence analyzing section 72, and an absolute time calculating section 73.
And a comparison module 74. The search time determination unit 71 determines a reference time of the search target from the characteristic expression sentence of the article A to be searched and the occurrence time provided from the control module 80, and provides the result to the comparison module 74. The result sentence analysis unit 72 takes in the search results of each database via the search result acquisition module 50, determines the time of each article from the characteristic expression sentences of the articles B and news C and the time of occurrence, and sends the result to the comparison module 74. give.
The absolute time calculation unit 73 selects “Yesterday”, “Today” from the characteristic expression sentences of the search time determination unit 71 and the result sentence analysis unit 72.
The time information such as is extracted, and the time is converted into an absolute time such as a date. The comparison module 74 converts the data of the searched article B and news C into the search time determination unit 7.
It is determined whether or not the time is within the time range determined in step 1, and the respective feature amounts of the article B and the news C are added.

Next, the configuration of the control module 80 is shown in FIG. In the figure, the control module 80 includes a search target sentence storage unit 81, a content link threshold storage unit 82, a time range storage unit 83, an analysis method storage unit 84, a comparison element determination unit 85, and an association degree calculation unit 86. And a search result storage unit 87. The search target sentence storage unit 81 stores the articles to be searched and gives them to the content link module 60. The content link threshold storage unit 82 stores the content link module 60
Are stored and given to the content link module 60. The time range storage unit 83 stores a time range used for calculating the feature amount in the time link module 70 and provides the time range to the time link module 70. The analysis method storage unit 84 stores an algorithm for calculating the degree of relevance and provides it to the comparison element determination unit 85 and the degree of relevancy calculation unit 86. The comparison element determination unit 85 determines comparison elements such as nouns and proper nouns necessary for calculating the degree of relevance and supplies the comparison elements to the content link module 60. The relevance calculating unit 86 takes in the feature amounts determined by the content link module 60 and the time link module 70, calculates the relevance using an algorithm for calculating the relevance, and provides the relevance to the search result storage unit 87. The search result storage unit 87 stores data that is finally determined to be relevant.

Hereinafter, an operation procedure in a process of searching for an article B or a news C similar to the article A will be described with reference to the configuration of each unit described above and the flowchart shown in FIG. First, the user sets the article A to be searched, the algorithm of the method of analyzing the relevance calculation, the time range, and the threshold value in the control module 80, and the control module 80 starts operating.

The control module 80 includes a comparison element determining unit 8
In step 5, a comparison element required for calculating the degree of relevance is determined, and set in the content link module 60 together with the characteristic expression sentence of the article A and the threshold (step 1).

Next, the control module 80 sets the characteristic expression sentence, occurrence time, and time range of the article A in the time link module 70 (step 2). Next, the control module 80 sets a keyword for keyword search of each database in the towing link module 20 (step 3). Further, a condition for operating a structure such as obtaining only a photograph portion from a newspaper article, that is, a structure expression to be extracted is set in the structure link module 30 (step 4).

In the search condition output module 40, the
Link module 20 and structural link module 30
Using the set conditions, TV news database
10 ₁And newspaper article database 10_TwoSearch for
(Step 5). The search result is the search result acquisition module.
Module 50 and the content link module 60
Sent to the inter-link module 70. In addition, search results
The gain module 50 serves here as a buffer.
ing.

In the content link module 60, after performing a morphological analysis on the characteristic expression sentence of the article A by the search sentence analyzing unit 61, data necessary for the analysis is generated according to the comparison element.
It is stored in the comparison module 64. The result sentence analysis unit 6
In step 2, after performing a morphological analysis on the feature expression sentences of the retrieved articles B and news C, data to be compared is generated according to the comparison elements and stored in the comparison module 64. The comparison module 64 compares the analysis data of the article A with the analysis data of the article B and the news C, adds a feature amount of 0.00 or more and less than 1.00 to data having more matches than the threshold value, and adds data with less matches than the threshold value. To the control module 80 together with the data of each article (step 6).

In the time link module 70, the search time determination unit 71 uses the characteristic expression sentence of the article A and the occurrence time to, for example, convert the time information “yesterday” to “September 10, 1991”.
Converts to absolute time, such as days, and depending on the time range,
It is converted into a time section such as “August 1991 to October 1991” and stored in the comparison module 74. The result sentence analyzing unit 72 extracts, for example, data on the date of the incident and the date of the news report from the feature expression sentences of the retrieved articles B and news C, and stores the data in the comparison module 74. The comparison module 74 compares the time data of the article A with the time data of the article B and the news C, and sets the data existing within the time range to 0.
A feature value of not less than 00 and less than 1.00 is added, and a feature value of 1.00 is added to data outside the time range, and the data is sent to the control module 80 (step 7).

In the control module 80, the relevance calculating section 8
6, the feature amounts determined by the content link module 60 and the time link module 70 are fetched, and the degree of association is calculated using a given analysis method (step 8). As an example of the relevance determination, the feature amount obtained from the content link module 60 is α _i (i = 1 to n), the feature amount obtained from the time link module 70 is β, Regarding,

[0027]

(Equation 1) The relative distance is calculated by the following formula. Here, it is determined that there is relevance only for data within a predetermined distance designated during the analysis method.

The search result storage section 87 sorts and stores the data determined to be relevant in descending order of relative relevance from the relative distance (step 9). Here, an example will be described of a feature expression sentence, a data structure, a space in an algorithm for calculating the degree of association, a method of determining a feature amount in the temporal link module 70, and a method of determining a feature amount obtained in the content link module 60.

The characteristic expression sentence can be realized by using text data in print media such as newspapers and magazines, and by using scripts in video and audio media such as television and radio.

In the case of television news, the data structure is composed of video, audio, feature expression, broadcast date and time, and keywords. The keywords include, for example, place names such as Japan and genres such as economy.

The space in the relevance calculation algorithm includes, among the feature quantities generated by the content link module 60, a feature quantity by a noun which is present in a large number of feature expression sentences and roughly indicates information; It is generated from a feature quantity using a paranoun that indicates the state, a feature quantity using a proper noun that indicates the place where the incident occurred, the person involved, etc. And expressed as a four-dimensional space by using

For example, regarding the characteristic amount between the information X and
Step 8 of the operation procedure is performed on the information Y whose feature amount of the noun is 0.30, feature amount of the sa noun is 0.60, feature amount of the proper noun is 0.20, and the feature amount obtained by the time link module 70 is 0.00. According to equation (1), the pseudo distance is 0.
It becomes 70. In addition, for information Z in which the feature quantity of the noun is 0.30, the feature quantity of the sa-variant noun is 0.60, the feature quantity of the proper noun is 0.60, and the feature quantity obtained by the time link module 70 is 0.00, The distance is 0.90. Therefore, the information Y is determined to be information similar to the information X.

In addition, it is also possible to change the pseudo distance by changing the feature amount of each axis. For example, the axial strength μ _i (i = 1) indicating the degree of influence of each axis in equation (1)
~n), added mu _B,

[0034]

(Equation 2) And Here, μ _i and μ _B are real numbers of 0.00 or more, and the standard is 1.0. In this case, for example, the axial strength of the proper noun is
Assuming that it is 0.010, the feature amount of the noun is 0.30, the feature amount of the sa-variable noun is 0.60, the feature amount of the proper noun is 0.20, and the time link module 7
The pseudo distance of the information Y whose feature amount obtained at 0 is 0.00 is 0.67. In addition, the pseudo-distance of the information Z in which the characteristic amount of the noun is 0.30, the characteristic amount of the sa-variant noun is 0.60, the characteristic amount of the proper noun is 0.60, and the characteristic amount obtained by the time link module 70 is 0.007. Becomes Therefore, it is determined that the similarity of information Y and information Z to information X is the same. In this way, by changing the axis strength, for example, if the influence of the axis of the proper noun is weakened, the similarity in terms of the similarity of the cases represented by the sa noun is emphasized, and conversely the axis of the proper noun is emphasized. If the influence of is strengthened, the similarity of a person or a place is emphasized, and diversity can be given to the identification of relevance.

The method of determining the feature quantity in the time link module 70 is 0.00 when the time link module 70 handles only information on the same day as the article A, 1.00 when it handles infinitely old information, and the time width is hyperbolic. This can be achieved by expressing it functionally. For example, if the time difference X from the article A is represented by the number of days and the feature amount Y is represented by Y = 1.00-1 / (X + 1) (3), the feature extracted by the time link module 70 for the article on the same day as the article A The amount is 0.00, and the information one year ago is 0.997.

Another method of determining the characteristic amount Y is that if the time link module 70 determines the characteristic amount according to the type of the medium, the information is generated for different media such as television and newspaper after the information is generated. It can absorb the time lag before being reported. For example, time link module 7
When the article A input to 0 is TV news in the afternoon, the feature amount obtained by the time link module 70 for an article published in the morning on the next day in a newspaper is set to 0.00, and the hyperbolic curve is obtained for other information. Express with a function. Also, if the article A input to the time link module 70 is the morning TV news, an evening newspaper article, if the article A is a newspaper morning newspaper, the previous day's television news, and if the article A is an evening newspaper, lunch time For the TV news up to, the feature amount obtained by the time link module 70 is set to 0.00. As a result, it is possible to absorb the time difference between the newspaper and the television news due to the media, and to easily search for the article most similar to the article.

The method of determining the feature amount obtained by the content link module 60 is such that the comparison target data obtained from the feature expression sentence of the article A to be searched and the word obtained from the search sentence from the database completely match. Assuming that the number of matched words is a and the number of words of the comparison target data is b when the case is set to 0.00 and the case of no match is set to 1.00, the characteristic amount W is W = 1.00−a / b (4) ). For example, in article A, the noun is 10
When there are multiple matches, the noun that matches in the searched article is 8
If there are, the feature amount is O.20. The threshold of the content link module 60 is set from outside.

In another method of determining the characteristic amount W, the characteristic expression sentence is subjected to the analysis in the calculation of the characteristic amount of the content link module 60, and is separated into a phrase expressing the subject of the sentence and a phrase modifying the sentence. That is, the article A input to the content link module 60 and the characteristic expression sentence of each of the articles obtained from the database are subjected to the parsing and analyzed, and the words obtained from the words representing the subject (general) and the words to be modified are obtained. Classified words and. For example, in the data to be compared, ten nouns are obtained from the phrase representing the subject, there are seven matching nouns in the phrase representing the subject in the retrieved article, and three out of six nouns are used in the modifying phrases. If they match,
Assuming that the effect of the clause to be modified is 0.30 and the effect of the subject clause is 0.70, the feature amount W is 0.70 × 0.70 + 0.30 × 0.50 = 0.36. As a result, it is possible to eliminate adverse effects on the calculation of relevance due to the side information.

[0039]

As described above, in the method for identifying the relationship between complex information of the present invention, the relationship existing between information of different media such as newspaper articles and television news can be easily identified. be able to. That is, a high-performance information retrieval service can be provided for multimedia databases constructed by different creators.

Further, in the method of the present invention, it is only necessary to add a characteristic expression sentence and occurrence time data to individual information, and it is not necessary to consider the relevance of keywords and the like across a plurality of databases. The burden on the user can be greatly reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an embodiment of a multimedia database system for realizing a method for identifying association between complex information according to the present invention.

FIG. 2 is a block diagram showing a configuration example of a content link module 60 and a time link module 70.

FIG. 3 is a block diagram illustrating a configuration example of a control module 80.

FIG. 4 is a flowchart illustrating an operation procedure of the embodiment of the present invention.

[Explanation of symbols]

10 ₁ TV news database 10 ₂ Newspaper article database 20 Tow link module 30 Structural link module 40 Search condition output module 50 Search result acquisition module 60 Content link module 61 Search sentence analysis unit 62 Result sentence analysis unit 63 Morphological analysis unit 64 Comparison module 70 Time link module 71 Search time determination unit 72 Result sentence analysis unit 73 Absolute time operation unit 74 Comparison module 80 Control module 81 Search target sentence storage unit 82 Content link threshold storage unit 83 Time range storage unit 84 Analysis method storage unit 85 Comparison element determination Unit 86 relevance calculation unit 87 search result storage unit

Continuation of the front page (56) References JP-A-2-287876 (JP, A) JP-A-2-287768 (JP, A) JP-A-3-176690 (JP, A) JP-A-4-54564 (JP) JP-A-4-84271 (JP, A) JP-A-1-243116 (JP, A) JP-A-61-248160 (JP, A) JP-A-1-278171 (JP, A) 2-98779 (JP, A) JP-A-2-232771 (JP, A) JP-A-64-26252 (JP, A) JP-A-4-24871 (JP, A) (58) Fields investigated (Int. Cl. ^7, DB name) G06F 17/30 G06F 17/28 JICST file (JOIS)

Claims (4)

(57) [Claims] 1. A method for identifying the association between complex information including image information, audio information, text information, and other media information, wherein each of the information includes a feature expression sentence expressing individual features in a natural language. , Adding the occurrence time data of the information, extracting the feature amounts by analyzing the characteristic expression sentence of each information in natural language, and extracting the occurrence time data of each information and the characteristic expression sentence of each information in natural language. Using the information on the time extracted by analysis, the information is converted into an absolute time at which the information actually occurs, and the degree of association between the pieces of information is calculated by comparing the feature amount of the information and the absolute time. A method for identifying the relationship between featured complex information. 2. The method according to claim 1, wherein the characteristic expression sentence of each information is classified into a part of speech of a self-sufficient word and a modifier, and an amount of matching of the word of each part of speech is obtained. , The degree of relevance of each part of speech is calculated using the adverb and the word indicating the date and time in the feature expression sentence of each information.
The occurrence time data of each piece of information is converted into absolute time, and when the number of each part of speech is n, the amount of word matching for each part of speech and the absolute time of each piece of information are plotted on each axis in an n + 1-dimensional space. A method for identifying associations between complex information, comprising: allocating and calculating a degree of association between the pieces of information from the sizes thereof as a pseudo distance in an n + 1-dimensional space. 3. The method according to claim 2, wherein the amount of coincidence between words of each part of speech and the absolute time of each piece of information are n + 1.
A method for identifying relevance between complex information, comprising specifying a range of a distance that can be handled on a predetermined axis and controlling a distance on an (n + 1) -dimensional plane when assigning to each axis in a dimensional space. 4. A method according to claim 2, wherein, when calculating the amount of matching of the words of each part of speech, the dependency of the characteristic expression sentence is analyzed by calculating the amount of matching of the words. The complex information is separated into a part that expresses the meaning and a part that modifies it, and weights the amount that matches in the part that indicates the meaning with respect to the amount that matches in the modification part of each information. Relevance identification method.