JP4082719B2 - Multi-dimensional database system and operation command translation method using natural language - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a multidimensional database using a natural language, and more particularly, to a multidimensional database operation technique, a text data and multidimensional database cooperation technique, and a qualitative analysis technique using a multidimensional database.
[0002]
[Prior art]
A multi-dimensional database is a database whose purpose is to analyze and analyze various types of data related to business from various viewpoints.
For example, when sales and marketing staff analyze product sales performance data, it is not just a monthly sales performance, but also analysis from various viewpoints such as by store, product attribute, customer attribute, price range, etc. Is required.
A multi-dimensional database is a very effective tool for collecting, compiling, and analyzing data from such various viewpoints.
Multi-level databases have operations such as dice for changing viewpoints, slicing for limiting the target of analysis, drilling down to divide the analysis target into lower levels, and drilling up to handle the higher levels collectively. They were realized using a spreadsheet for multidimensional display called multidimensional view.
[0003]
[Problems to be solved by the invention]
The multidimensional database has the following problems.
a. Problem of operability When data manipulation such as data analysis is performed on a multidimensional database, it is necessary to understand a unique concept and special data manipulation techniques, and it is difficult to sufficiently manipulate the multidimensional database. A multidimensional database has special operations such as dicing, slicing, drilling up, and drilling down, and these operations are performed using a spreadsheet called a multidimensional view. Therefore, when performing data operations such as data analysis, a special technique for realizing a data operation suitable for the purpose by combining these special operations on a multidimensional view is necessary.
For example, when analyzing “the color of a product whose sales increased by more than 10% compared to the previous month”, it was necessary to operate the multidimensional database in the following procedure on the multidimensional view.
Calculate the month-to-month ratio based on sales of last month and this month.
1. Select products that have exceeded 10% compared to last month.
2. Slice by selected product and color and sort by color.
3. For products, drill up to all products, classify by color, and aggregate.
[0004]
b. Cooperation with text data In a multidimensional database, there was no structure reflecting the semantic content of text data, so it was difficult to link with text data. The data to be analyzed includes text data described in a natural language in addition to quantitative data such as numerical data.
Text data such as analysis reports contain a lot of qualitative information that is difficult to analyze with numerical data.
However, the text data cannot be used effectively because the multi-dimensional database cannot handle the main language predicates and modifiers.
[0005]
c. Qualitative analysis It was difficult to conduct qualitative analysis with multidimensional databases. In data analysis, qualitative analysis is as important as quantitative analysis.
For example, if there is a fact that “the sales of a certain product has increased”, it is very important to analyze the factor in order to improve future performance.
However, since there is no direct means for qualitative analysis in a multidimensional database, it has been difficult to perform qualitative analysis in a multidimensional database.
Therefore, trial and error based on experience and intuition, and analysis by another means were necessary.
[0006]
Therefore, an object of the present invention is to solve the problem that the operation method of the multidimensional database is difficult, the problem of cooperation between the multidimensional database and text data, and the problem of qualitative analysis using the multidimensional database.
[0007]
[Means for Solving the Problems]
To achieve the above object, according to the present invention, in a multi-dimensional database system, information of general dimensions having a hierarchical structure having natural language noun elements as members and a hierarchical structure having natural language predicate elements as members. Predicate dimension information having the following is set in the storage device, and translation means for translating the operation instruction described in the natural language into the operation of the multidimensional database using the general dimension information and the predicate dimension information is provided. is doing.
Further, the information on the previous description part dimension includes information on the action dimension, information on the relation dimension, and information on the action dimension.
[0008]
DETAILED DESCRIPTION OF THE INVENTION
As an embodiment of the present invention, FIG. 1 shows a system configuration.
2 and 3 show the structure of a multidimensional database for reflecting the semantic structure of natural language.
FIG. 4 shows a dimension / member definition table for giving semantic contents to each dimension.
FIG. 5 shows a term definition table.
FIG. 6 shows an operation command generation process for issuing an operation command in a natural language.
FIG. 7 shows an operation translation / execution process in which the operation command is translated into a multidimensional database operation and executed.
FIG. 8 shows a result display process for displaying the operation result.
FIG. 9 shows an input example of the operation command.
FIG. 10 shows an example of natural language analysis used in text data and a multidimensional database.
FIG. 11 shows an example of an interface for realizing text cooperation.
FIG. 12 shows a table of additional information used for informationizing the contents of the cooperation.
FIG. 13 shows a member generation process for reflecting the modification relationship in the structure of the multidimensional database.
FIG. 14 shows an example of the result display.
FIG. 15 shows a description example of the operation command.
FIG. 16 shows a translation example in the case of “modification of predicate dimension”.
[0009]
Embodiments of the present invention will be described below with reference to the above drawings.
FIG. 1 shows a system configuration.
The system is roughly divided into a server 101 that interprets and executes operation instructions of a multidimensional database, and a client 102 that generates and transmits operation instructions and displays the results, and is connected by a network 103.
Server 101
An operation command analysis unit 104 for analyzing an operation command received from the client 102;
An operation procedure list generation unit 105 that translates the instruction into an operation procedure of the multidimensional database and generates an operation procedure list,
An operation control unit 106 that operates the multidimensional database according to the generated operation list and returns a result.
Term definition table 107, dimension / member definition table 108, linkage rule / additional information 109, used in the operation procedure list generation unit,
Multidimensional database 110,
It is composed of
[0010]
Client 102
An operation command generation unit 111 that generates an operation command for the multidimensional database 110,
Operation result display control unit 112 for controlling the display of the operation result,
Screen display unit 113 for displaying operation commands, display results, etc.
Natural language analysis unit 114 for structural analysis of operation instructions,
Coordination rules / additional information 115 used for associating elements of the multidimensional database 110 with operation instructions,
It is composed of
[0011]
2 and 3 show the structure of the multidimensional database.
An important feature of the present invention resides in the structure of the multidimensional database system that can reflect the semantic structure of natural language.
For this purpose, the concept of “dimension / member definition” is introduced.
The dimension / member definition defines how natural language phrases are reflected in the dimensions and members of a multidimensional database.
In the dimension / member definition, each term or phrase corresponds to which member of which dimension, and what kind of semantic content is indicated by a function such as a mathematical expression (exactly “mapping”, but the term “function”). Because it is easier to grasp the image when using, the term “function” is used here).
In the present invention, in consideration of cooperation with natural language, the following dimension, which is an extension of the normal dimension of the multidimensional database, is introduced.
[0012]
General dimensions (Figure 2)
The general dimension is a dimension for defining an ordinary member of the multidimensional database 110 and has a hierarchical structure (level).
Here, nounistic (common nouns, pronouns, proper nouns, numbers, etc.) elements of natural language, that is, objects corresponding to objects, numerical values, and the like are associated.
As an example, a product dimension 201 and a color dimension 202 used in the description of the operation described later are given.
In the product dimension 201, “all products” 203 is the first level, “home appliances” 204, “OA products” 205,... "TV" 206, "Video" 207, ... are supported.
In the color dimension, “all colors” 208 corresponds to the first level which is the highest level, and “red” 209 and “blue” 210 correspond to the second level.
The general dimension is almost the same as the dimension handled in a normal multi-dimensional database, but it is possible to define semantic content, and in the sense that members are automatically added to the lower level during text linkage. It is an extension of the database dimension.
[0013]
Predicate dimension (Figure 3)
The predicate dimension is a dimension newly introduced in the present invention in order to operate a multi-dimensional database in a natural language, and corresponds to a natural language predicate (verb, adjective verb, etc.).
The structure of the predicate dimension is shown in FIG.
The predicate dimension is divided into an action dimension 301, a relation dimension 302, and an action dimension 303.
The action dimension action dimension 301 is a dimension corresponding to the action of the object.
In the action dimension, the first level, which is the highest level, corresponds to actions such as “all actions” 304, “increase” 305, “decrease” 306,.
In addition, when it is modified with an adverb or the like, a member is added under “increase”, for example, “increase by 10% or more” 307.
The relation dimension relation dimension 302 is a dimension corresponding to the relation between objects.
In the relationship dimension, the first level, which is the highest level, corresponds to “all relationships” 308, the second level “faster” 309, “slower” 310,.
In addition, when it is modified with an adverb or the like, a member is added under “faster”, for example, “5 times faster” 311.
The action dimension action dimension 303 is a dimension corresponding to the action of an object on another object.
In the action dimension, the first level, which is the highest level, corresponds to “all actions” 312, and the second level corresponds to actions such as “good influence” 313, “adverse influence” 314,.
In addition, when it is modified by an adverb or the like, for example, a member is added under “having a good influence” like “increase sales” 315.
[0014]
In the present invention, the multidimensional database 110 has three types of predicate dimensions such as an action dimension 301, a relation dimension 302, and an action dimension 303, a general dimension (201, 202), etc., and the semantic content is defined as a dimension / member definition. It is defined in table 108.
The dimension / member definition table 108 is shown in FIG.
The dimension / member definition table 108 includes “word / phrase” 401, “corresponding dimension” 402, “member name” 403, “position” 404, and “definition contents” 405.
“Phrase” 401 corresponds to a natural language phrase.
“Corresponding dimension” 402 is a dimension to which the word belongs, “Member name” 403 is the name of the member to which the word corresponds, “Position” 404 is the position of the member (such as upper members), and “Definition contents” 405 is This is the definition of the member's role and semantic content.
For example, taking the phrase “increased” 406 as an example, this is included in the motion dimension 301 under the name “<increase>” 305.
Members not included in the action dimension 301 are automatically added under “all actions” 304.
The “definition contents” 405 is defined by a function such as a member like “value (t2) −value (t1)> 0.
[0015]
In the present invention, the predicate part of the natural language is made to correspond to the slicing operation on the member of the predicate dimension (FIG. 3), and the modification relationship is an operation of “creating a new member at the lower level of the member corresponding to the modified phrase” Text cooperation is realized by making it correspond to.
In the present invention, in order to realize text cooperation, the concept of “term definition” is also introduced.
The term definition is to define terms in a natural language as components, operations, and operations of the multidimensional database 110.
The contents of the term definition are registered in the term definition table 107.
The term definition table is shown in FIG.
The term definition table 107 is a table in which the definition contents of terms are defined by the components and operations of the multidimensional database 110.
The term definition table 107 includes “term” 501 and its “definition contents” 502.
For example, the term “comparison with last month” 503 is defined as “(current month value) / (last month value)” 504.
[0016]
In the present invention, general (general-purpose) dimension / member definition table and term definition table, dimension / member definition table and term definition table for each purpose such as for financial business and medical industry, and user-definable user definition By preparing a dimension / member definition table and a term definition table, the operability and flexibility of the system are improved.
[0017]
Hereinafter, a method for realizing an operation based on a natural language in the multidimensional database system having the above structure will be described.
Here, the operation of this system will be described by taking as an example the case of examining data of products whose sales have increased by 10% or more from the previous month and analyzing which color of the products are being sold.
The operation of this system is the “operation command generation process” (FIG. 6) in which an operation command of the multidimensional database 110 is generated by the client 102 and transferred to the server 101 (FIG. 6). Then, it is divided into an “operation translation / execution process” (FIG. 7) for returning the result to the client, and a “result display process” (FIG. 8) for displaying the result on the client.
[0018]
An operation command generation process is shown in FIG.
First, in a text editor, an operation instruction of the multi-dimensional database 110 is input in a natural language format, for example, “The color of a product whose sales ratio has increased by 10% or more” as shown in FIG. 9 (601).
Next, the natural language analysis unit 114 decomposes the input operation command into each grammatical element using natural language analysis (602).
A conceptual representation of the analysis results is shown in FIG.
In FIG. 10, <noun><caseparticle> etc., indicated by <...> 1001 are parts of speech, an arrow 1002 indicates a modification relationship, and a dashed arrow 1003 indicates a main statement relationship.
In this example, for example, “Sales” modifies “Last Month”, and “Last Month” and “Increased” have a relationship between the main part and the predicate.
[0019]
When the part-of-speech decomposition is completed, text linkage that associates a natural language phrase with a component (dimension / member, term, etc.) of a multidimensional database is executed automatically or manually based on the result (603).
A user interface as shown in FIG. 11 is conceivable as a manual interface.
For example, when the cursor is moved to the target phrase “increased” 1101, the analysis result such as the part of speech is displayed in a pop-up 1102.
The user refers to this and designates the “increased” cooperation destination as the “motion dimension member” 1103 using a pull-down list or the like.
[0020]
In the text link execution (603), the link rule / addition information 115 is referred to (604), and the association information is added (605).
In this embodiment, XML is used as association information to be added. A document data format other than XML may be used.
[0021]
The linkage rules in the linkage rules / additional information 115 are shown below, and a table of additional information showing details such as tags is shown in FIG.
Coordination rules
(1) The components (general dimension, action dimension, relation dimension, action dimension, term) of the multi-dimensional database at the cooperation destination are represented by tags (<...>).
(2) To clarify the structure, if a phrase is formed, enclose it in <phrase> tags.
<Phrase> tags can be nested, thereby clarifying the semantic structure of the sentence.
(3) Set “ID” attribute for each tag. The “ID” attribute is used by the system to identify or qualify.
(4) The role in the statement or phrase is expressed by the “role” attribute.
(5) If there is a modifier relationship, set the value of the “role” attribute of the modifier to “modifier”, and set the “ID” attribute of the modifier to the “qualifier” attribute.
[0022]
Further, the additional information is set according to the additional information table.
Based on this cooperation rule and additional information, a final operation command is generated.
FIG. 15 shows a description example of the operation command in the case of the example “The color of the product whose sales ratio has increased by 10% or more”.
[0023]
The operation translation / execution process is shown in FIG.
First, an operation command received from the client 102 is read by the server 101 (701).
Next, the contents of the phrase in the operation command are analyzed (702).
In the analysis of the contents of the phrase, the main predicate relation and the modification relation between the constituent elements of each <phrase> tag (<phrase> tag, <general dimension> tag, etc.) are analyzed. The analysis contents are as follows.
(1) If the predicate dimension is modified, the child member is added to the linked member of the modified word according to the automatic child member generation algorithm.
(2) The relationship between the main part and the predicate is that the predicate dimension is modified, and the child member corresponding to “operation with main part” is added to the linked target member of the modified word. To do.
(3) If the general dimension is qualified, execute the corresponding operation or operation (corresponding to operations such as drill-down, slicing, filtering by conditions). Also, a new general dimension is generated and the result of the operation or operation is registered as its member.
(4) If the <phrase> tag or its component is modified to another phrase, the phrase is translated after the entire modified phrase has been translated.
[0024]
Next, the definition contents are interpreted and translated (703).
Based on the result of step 702, the definition contents such as the term, the general dimension, and the predicate dimension are interpreted with reference to the term definition table and the dimension / member definition table. Depending on the result, each part is translated into multidimensional database operations / calculations. Also, the translation result is registered as a member of a new general dimension or predicate dimension.
[0025]
Next, translation into the operation of the multidimensional database is performed (704).
Based on the result of step 703, the operation command is translated into the operation content of the multidimensional database according to the value of the “role” attribute of each tag, and a translation list is generated. At that time, reference is made to the linkage rule / additional information 109 (this is the same as the linkage rule / additional information 115).
(1) If the attribute value is “predicate”, translate it into the corresponding “predicate dimension” slice.
(2) If the “phrase” (the content of the <phrase> tag) modifies another word / phrase, it is translated into an operation in the same manner as (4) of step 702 depending on the content of the modification.
[0026]
Note that, when a member is generated at a lower level of the link destination of the modified word, member generation processing is executed.
The member generation process is shown in FIG.
The member generation process starts upon receiving a command from the operation procedure list generation unit 105 during the translation of the modifier (1401). At that time, first, a modified word is searched (1402).
Next, the dimension / member definition table 108 is referenced to refer to the definition contents of the link destination of the modified word, and the dimension and level are checked (1403).
Next, a new member is generated under the modified word (1404), and the definition content is generated (1406).
Finally, the definition content regarding the added member is added to the dimension / member definition table 108 (1406), and the member generation process is terminated 1407.
A translation example in the case of “modification of predicate dimension” is shown in FIG.
In this example, a member that “increases by 10% or more” is generated under the member that “increases” in the motion dimension.
[0027]
Next, an operation list is generated (705).
Based on the translation result in step 704, the operation order is determined based on the nesting and modification relationships of the <phrase> tag, and an operation list of the multidimensional database is generated.
At this time, the order is determined by referring to the grammatical analysis information of the operation instruction information and the property information such as the analysis target.
For example, in the case of the example operation command “The color of a product with a sales ratio of 10% or more last month”, the “Sales last month ratio” qualifies “Product”, so refer to the term definition of “Last month ratio”. Then, it calculates that the monthly ratio of sales for all products is calculated, and a new general dimension of “comparison of sales of all products last month” is generated.
Further, the display method designation content is interpreted as it is, and in the case of the example, it is interpreted as “sorting the color of products in descending order of the number of sales and displaying it in a pie chart”.
[0028]
As a result of the above analysis, in the case of an example, the following operation list is generated.
(1) Calculate the ratio of sales last month for all products to be operated, and create a new general dimension “comparison of sales of each product last month”.
(2) Based on the term definition and dimension definition, a new motion dimension member “increased by 10% or more” is created at a level below the “increase” member.
(3) Slice the motion dimension member “increase by 10% or more”.
In other words, the sales of all products are evaluated, and only those with 10% or more are set as operation targets.
(4) Sort the product colors in descending order of product sales.
(5) Display the results in a pie chart in the order of the color with the most sales.
Finally, the multidimensional database 110 is operated 706 based on the operation list, and the result is transmitted 707 to the client 102.
[0029]
The result display process is shown in FIG.
First, the result information received from the server 101 is read (801).
The result display method 606 specified in the operation command generation process is confirmed (802).
Finally, it is displayed on the screen display unit according to the display method (803).
A display example of the result is shown in FIG.
[0030]
When qualitative analysis is performed with this system, for example, a natural language is input such as “What caused the sales increase of product A”.
Then, the factor can be analyzed using the operation of the action dimension corresponding to the “factor”.
[0031]
【The invention's effect】
According to the present invention, the operability of the multidimensional database is improved, the application range is expanded, and the utility value of the multidimensional database is improved. As a result, it is possible to perform more accurate analysis without missing any useful information in the analysis work that leads to decision making.
[Brief description of the drawings]
FIG. 1 is a diagram showing an overall configuration of a system using a multidimensional database.
FIG. 2 is a diagram for explaining a general dimension, which is a constituent element of a multidimensional database for reflecting a noun semantic structure of a natural language.
FIG. 3 is a diagram for explaining predicate dimensions and components of a multidimensional database for reflecting a predicate semantic structure of a natural language.
FIG. 4 is a diagram showing a dimension / member definition table for giving semantic contents to each dimension and its constituent members of a multidimensional database.
FIG. 5 is a diagram illustrating a term definition table for translating natural language terms into components, operations, and operations of a multidimensional database.
FIG. 6 is a diagram illustrating an operation command generation process until an operation command is generated in a natural language.
FIG. 7 is a diagram showing an operation translation / execution process in which an operation command is translated into an operation of a multidimensional database and executed.
FIG. 8 is a diagram illustrating a result display process for displaying an operation result.
FIG. 9 is a diagram illustrating an input example of an operation command.
FIG. 10 is a diagram illustrating an example of natural language analysis of an operation instruction.
FIG. 11 is a diagram illustrating an example of an interface for realizing text cooperation.
FIG. 12 is a diagram showing a table of additional information used for converting the contents of cooperation into information.
FIG. 13 is a diagram showing a member generation process for reflecting the modification relationship in the structure of the multidimensional database.
FIG. 14 is a diagram showing an example of a result display in a system using a multidimensional database system.
FIG. 15 is a diagram illustrating a description example of an operation command.
FIG. 16 is a diagram illustrating a translation example in the case of “modification of predicate dimension”;
[Explanation of symbols]
101 Server 102 Client 103 Network 104 Operation Command Analysis Unit 105 Operation Procedure List Generation Unit 106 Operation Control Unit 107 Term Definition Table 108 Dimension / Member Definition Table 109 Linkage Information / Additional Information 110 Multidimensional Database 111 Operation Command Generation Unit 112 Operation Result Display Control unit 113 Screen display unit 114 Natural language analysis unit 115 Cooperation information / additional information

Claims (6)

In a multidimensional database system,
General dimension information having a hierarchical structure with natural language noun elements as members, action dimension information having a hierarchical structure with natural language predicate elements as members, relation dimension information, and action dimension information Predicate dimension information with information, dimension / member definition information that defines the semantic content of each dimension, and term definition information that defines terms in natural language as components, operations, operations, etc. of a multidimensional database Is set in the storage device,
Translating operation instructions written in natural language into multidimensional database operations using general dimension information, predicate dimension information, dimension / member definition information, and term definition information set in the storage device Have translation means,
The translating means interprets the definition contents of terms and each dimension with reference to the dimension / member definition information and the term definition information when translating the operation instruction described in the natural language into the operation of the multidimensional database. Depending on the result of the interpretation, the word of the operation instruction described in the natural language is translated into the operation / operation of the multidimensional database, and the translation result is registered as a member of the general dimension information or the predicate dimension information. A multi-dimensional database system using natural language characterized by In a multidimensional database system,
General dimension information having a hierarchical structure with natural language noun elements as members, action dimension information having a hierarchical structure with natural language predicate elements as members, relation dimension information, and action dimension information Information on predicate dimensions that contain information, general and purpose-specific dimension / member definition information that defines the semantic content of each dimension, and terms in natural language for multidimensional database components, operations, and operations General purpose and purpose-specific term definition information defined as
Translating operation instructions written in natural language into multidimensional database operations using general dimension information, predicate dimension information, dimension / member definition information, and term definition information set in the storage device Have translation means,
When translating the operation instruction described in the natural language into the operation of the multidimensional database, the translating means translates the terms and definitions of each dimension into the general and purpose-specific dimension / member definition information and the general purpose. A multi-dimensional database system using natural language characterized by referring to and interpreting term definition information for each purpose. In the multidimensional database system using the natural language according to claim 1 or 2,
When the members in the general dimension information and the predicate dimension information are modified by a modifier, the general dimension information and the predicate dimension in which the member with the modifier is added to a lower level of the member A multi-dimensional database system using natural language, characterized in that the information is set in a storage device. In a multidimensional database system,
General dimension information having a hierarchical structure with natural language noun elements as members, action dimension information having a hierarchical structure with natural language predicate elements as members, relation dimension information, and action dimension information Predicate dimension information with information, dimension / member definition information that defines the semantic content of each dimension, and term definition information that defines terms in natural language as components, operations, operations, etc. of a multidimensional database Is preset in the storage device,
When the processing device of the multidimensional database system translates the operation instruction described in the natural language input from the input means into the operation of the multidimensional database,
Performs a process of interpreting the definition contents of terms and each dimension with reference to dimension / member definition information and term definition information set in the storage device,
Based on the interpretation result of the process, a process of translating the words of the operation instruction described in the natural language into the operation / operation of the multidimensional database,
An operation instruction translation method, wherein the translation result of the processing is registered as a member of general dimension information or predicate dimension information set in the storage device. In a multidimensional database system,
General dimension information having a hierarchical structure with natural language noun elements as members, action dimension information having a hierarchical structure with natural language predicate elements as members, relation dimension information, and action dimension information Information on predicate dimensions that contain information, general and purpose-specific dimension / member definition information that defines the semantic content of each dimension, and terms in natural language for multidimensional database components, operations, and operations General purpose and purpose-specific term definition information defined as
When the processing device of the multidimensional database system translates the operation instruction described in the natural language input from the input means into the operation of the multidimensional database,
Processing to interpret terms and definitions of each dimension, etc. with reference to general and purpose-specific dimension / member definition information and general-purpose and purpose-specific term definition information set in the storage device A featured operation command translation method. The operation command translation method according to claim 4 or 5,
When the members in the general dimension information and the predicate dimension information are modified by a modifier, the general dimension information and the predicate dimension in which the member with the modifier is added to a lower level of the member Information is set in a storage device.

Images