JPH0922415A - Knowledge storage method and device and re-recognition method and device for knowledge data base - Google Patents

Abstract

PROBLEM TO BE SOLVED: To acquire a high and fast information retrieval function by constructing a knowledge data base based on a data structure where the recognition and storage unified together, i.e., a recognition module structure. SOLUTION: A knowledge data base includes a recognition module system where the perception, recognition and storage are unified together and includes an input part 11 such as a keyboard, etc., a processing part 12 including a CPU and a memory, a storage part 13 such as a RAM, a hard disk, etc., and a display part 20 such as a CRT display, etc. Then a data object containing the attribute value is generated by input of the object attribute value. A name is given to the attribute of the data object, and an instance object containing the attribute name and the attribute value is generated. Thus an object is recognized. Then the instance object is stored in a storage area to secure every storage step of the object. These input, recognition and storage operations are carried out in a single processing unit.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a storage method in a computer system, and more particularly to a knowledge acquisition method, a storage method, and a search method in a knowledge base of a knowledge system.

[0002]

2. Description of the Related Art In a knowledge system, knowledge about a certain field, that is, facts and rules are stored in a knowledge base, and a reasoning part is a new fact based on the facts and rules stored in this knowledge base. Guide Knowledge systems can thereby provide humans with assistance in making decisions and problem analysis.

In constructing a conventional knowledge system, a knowledge engineer (Knowledge Expert) plays an important role. The knowledge engineer contacts the expert, collects the necessary knowledge (facts and rules) in the field, analyzes the problem solving method of the expert, and converts them into software. In the process of constructing a knowledge system, interviews between knowledge engineers and experts are repeated, and a prototype based on a few basic facts and rules is gradually changed to a practical knowledge system. Development is made.

As a method of storing knowledge about facts and rules in a knowledge base, there is a method of storing physical objects or objects that are conceptual entities as a semantic network that is linked by links that express relationships between objects. . When using a semantic network, each object that is a node of the network is treated as having an attribute and an attribute value. For example, an object called an apple has attributes such as color, shape, and taste, and each attribute value takes a concrete value such as red, round, or sweet. Also, an object can inherit the attributes of a more general object. For example, an apple object can inherit the attributes of a fruit object. By this inheritance, objects can be represented in a hierarchical structure, and information stored as knowledge can be organized.

In constructing such a knowledge base, a knowledge engineer first decides a schema (information about how the data is stored in a structure), and then the knowledge engineer decides the knowledge based on the decided schema. Store it.

[0006]

The process of constructing a knowledge system is a complicated process that takes a long time, and it is extremely difficult for a user who does not have knowledge of knowledge engineering to independently develop the knowledge system. Also, the constructed knowledge system is a complex one with a huge amount of facts and rules stored in the knowledge base and needs to be maintained by a knowledge engineer.

Therefore, although it is easy to extend the existing knowledge in the knowledge base so as to add new knowledge within the same framework, a problem occurs when the user wants to extend beyond the existing framework. . In other words, in order to extend and change the knowledge system that requires a schema change beyond the existing framework, it is necessary for a knowledge engineer to reconstruct the knowledge system over a long period of time.

For example, there may be a case where the user desires new processing or expansion of the application range of the knowledge system as the usage environment of the knowledge system changes.
At this time, the information structure and data structure such as the existing hierarchical structure and network structure of the knowledge base cannot be used as they are for the newly required application. However, the users of the conventional knowledge bases are merely users, and the knowledge bases cannot be modified as needed. Therefore, while resorting to specialization, changing the knowledge base structure requires almost the same time and effort as rebuilding from scratch.

That is, the conventional knowledge system has a static structure based on the schema determined at the designing / development stage, and it is possible to expand the target field / processing target and the processing method according to the user's needs. It is not something that can be improved. In other words, it is difficult to change the conventional knowledge system according to the change of its operating environment.
Moreover, the knowledge system itself cannot expand itself by acquiring new knowledge from experience and learning.

Further, in the knowledge base of the conventional knowledge system, there is a problem that the information retrieval ability is limited and the retrieval speed is slow. Regarding the ability and speed of information retrieval, for example, in comparison with human memory retrieval ability, humans have high-speed information retrieval ability by ambiguous retrieval.
The ambiguous search means that the object can be evoked by using an ambiguous concept for a specific search object as a starting point. In this way, it is considered that human beings realize high-speed information retrieval by retrieving objects by gradually reconstructing concepts from ambiguous ones.

However, in the conventional database, ambiguous retrieval cannot be performed, and generally, as the amount of stored information increases, the retrieval performance decreases in proportion to the square of the information amount. In addition, it is difficult to incorporate a sophisticated search such as a horizontal thinking search in which an object is searched from various viewpoints of recognition in a conventional knowledge system.

[0012] Therefore, the general purpose of the present invention is constructed by acquiring knowledge from experience and learning, and therefore it is possible to cope with changes in the operating environment, and various advanced information retrieval functions and high speed It is an object of the present invention to provide a new storage method and device which is a base of a knowledge database and the like capable of realizing an information search function.

Another object of the present invention is to provide a knowledge acquisition method and apparatus for a knowledge database which is constructed by acquiring knowledge from experience and learning and therefore can cope with changes in the operating environment. A further object of the present invention is to provide a knowledge database search method capable of realizing various advanced information search functions and high-speed information search functions.

[0014]

According to the invention of claim 1, a data object having the attribute value is generated by inputting the attribute value of the object, and an attribute name is given to the attribute value of the data object. By generating an instance object composed of the attribute name and the attribute value, and memorizing the object by storing the instance object in a storage area. The recognition and the storage are executed as one processing unit.

The invention of claim 2 further includes the step of generating a class object consisting of the attribute name of the instance object and storing it as an experience in the storage area in the invention of claim 1. In the invention of claim 3, in the step of recognizing in the invention of claim 2, the class object stored in the storage area as experience is applied to the data object, and the class object is stored as necessary. Each step of generating the instance object by modifying, deleting, and adding the attribute name of the data object is included.

In the invention of claim 4, in the step of storing as the experience in the invention of claim 2 or 3, the similar objects having some attribute names the same as the attribute name of the instance object are stored. The same attribute name of the instance object and the similar object is searched for in the storage area, a superclass object having the same attribute name is generated, and the superclass as a superclass of the class object. The method further includes each step of storing the object in the storage area.

According to the invention of claim 5, in the invention of claim 4, the superclass object superordinate to the superclass object is repeatedly generated and stored in the storage area. The method further includes forming a hierarchical structure.

According to a sixth aspect of the invention, in the step of retrieving according to the fifth aspect of the invention, the step of retrieving the similar object from an upper superclass to a lower superclass in the hierarchical structure. including. In the invention of claim 7, in the step of searching in the invention of claim 6, the index for search is used.

In the invention of claim 8, claims 1 to 7 are provided.
In the invention, each of the attribute name and the attribute value is expressed as a further object consisting of at least one of the attribute name and the attribute value, and a link connecting each of the attribute name and the attribute value and the corresponding object , All the objects are linked to the mesh structure.

According to a ninth aspect of the present invention, in the object linked to the mesh structure according to the eighth aspect, the attribute name and the attribute value are expressed as further objects. Contains atomic objects that cannot. In the invention of claim 10, in the invention of claim 9, the atomic object exists in all real world models stored in the knowledge base.

According to the invention of claim 11, in the invention of claims 4 to 10, an object is designated, the designated object is searched in the hierarchical structure, and the superclass object of the designated object is searched. Each of the superclass objects, and each step of making each of the superclass objects and the objects that are subclasses of the superclass object as search results, the superclass corresponding to different recognition of the designated object. An object and the object are obtained as a search result.

According to a twelfth aspect of the present invention, the first object and the second object are specified in the eighth aspect to the eleventh aspect, and the first object and the second object are designated in the mesh structure from the attribute name and the attribute value. A second related object related to the second object is searched sequentially for the first related objects related to the first object in the downstream direction with the direction toward the object as the downstream direction. Sequentially in the mesh structure in the downstream direction to extract matching attribute names and attribute values between the first related object and the second related object,
Creating a new object consisting of the matching attribute name and attribute value and storing the new object, and intelligently by associating the first object and the second object with the new object Support activities.

In the invention of claim 13, a new object is input, a similar object similar to the new object is searched from existing objects, and a common part between the new object and the similar object is extracted,
Extracting a difference part between the new object and the similar object, recognizing the new object as a combination of the common part and the difference part, and storing the new object as the combination of the common part and the difference part Store in area.

In the fourteenth aspect of the invention, in the thirteenth aspect of the invention, the inputting step includes the step of inputting the attribute name and the attribute value of the new object. Claim 15
The present invention further includes the step of inputting a domain in the invention of claim 13 or 14, wherein the similar object searched by the designation of the domain is different.

The invention of claim 16 further includes the step of inputting an object in the inventions of claims 13 to 14, and the similar object retrieved by designating the object is different. In the invention of claim 17, in the step of searching in the invention of claim 16, the similarity determination method is determined based on the domain and the purpose, and the similarity determination method is performed using the similarity determination method. Includes each step of searching an object.

According to the invention of claim 18, in the method of determining similarity according to claim 17, the selected attribute name selected from the attribute name of the new object exists in the existing object. The similarity is calculated based on whether or not and the weighting factor assigned to the selected attribute name.

According to a nineteenth aspect of the invention, in the eighteenth aspect of the invention, the selected attribute name and the weighting factor depend on the domain and the purpose. In the invention of claim 20, in the step of extracting the common part in the inventions of claims 14 to 17, both the attribute name and the attribute value are set between the new object and the similar object. The method further includes generating a superclass having the same attribute name and the same attribute value as the common part when they are the same.

In the twenty-first aspect of the invention, the step of extracting the common part in the eighteenth or nineteenth aspect of the invention is
When both the selected attribute name and the corresponding attribute value are the same between the new object and the similar object, a superclass having the same selected attribute name and the corresponding attribute value is used as the common part. The method further includes a generating step.

In the invention of claim 22, in the step of extracting the difference part in the invention of claim 20, the attribute name is the same between the new object and the similar object. When the attribute values are different, the step of extracting the same attribute name in the is-a relation and generating a class of the is-a relation as the difference part is further included.

In the invention of claim 23, in the step of extracting the difference part in the invention of claim 21, the selection attribute name is the same between the new object and the similar object. When the corresponding attribute values are different, the step of extracting the same selected attribute name in the is-a relation and generating a class of the is-a relation as the difference portion is further included.

In the invention of claim 24, in the step of extracting the difference part in the invention of claim 22, when there is an attribute name existing in the new object but not in the similar object, The method further includes extracting the nonexistent attribute name and the corresponding attribute value in a part-of relationship and generating a class of the part-of relationship as the difference part.

In the invention of claim 25, in the step of extracting the difference part in the invention of claim 23, there is the selected attribute name which exists in the new object but does not exist in the similar object. At this time, the selected attribute name that does not exist and the corresponding attribute value are extracted in a part-of relationship, and pa
The method further includes generating a class of rt-of relationships as the difference part.

In the invention of claim 26, in the step of extracting the difference part in the invention of claim 24, when there is an attribute name existing in the similar object but not in the new object, The method further includes extracting the nonexistent attribute name and the corresponding attribute value in a part-of relationship and generating a class of the part-of relationship as the difference part.

In the invention of claim 27, in the step of extracting the difference part in the invention of claim 25, there is the selected attribute name which exists in the similar object but does not exist in the new object. At this time, the selected attribute name that does not exist and the corresponding attribute value are extracted in a part-of relationship, and pa
The method further includes generating a class of rt-of relationships as the difference part.

According to a twenty-eighth aspect of the invention, when the new object is stored in the thirteenth to twenty-seventh aspects, the new object and the similar object are divided into the common part and the different part. By associating with each other, the existing object is grown.

According to a twenty-ninth aspect of the invention, in the inventions of the twenty-sixth to twenty-eighth aspects, the superclass, the is-a relation class, and the part-of relation class are hierarchically constructed. It In the invention of claim 30, in the inventions of claims 26 to 29, the superclass, the class of the is-a relation, and the class of the part-of relation are sensory to show their meanings. You can add expressions.

According to a thirty-first aspect of the invention, in the thirtieth aspect of the invention, the hierarchical structure is a mesh structure including links connecting the objects in an oblique direction. In the invention of claim 32, the ambiguous concept corresponding to the target object in the invention of claim 31 is input, and one of the sensory expressions corresponding to the ambiguous concept is used as a starting point,
The method further includes searching the mesh structure using the sensory expression arranged in the mesh structure and searching for the target object.

In the thirty-third aspect of the present invention, the knowledge storage device has means for generating a data object consisting of the attribute value of the object by inputting the attribute value of the object, and assigning an attribute name to the attribute value of the data object. And a means for recognizing the object by generating an instance object consisting of the attribute name and the attribute value, and a means for storing the object by storing the instance object in a storage area. , The recognition and the storage are executed as one processing unit.

According to the thirty-fourth aspect of the invention, the knowledge storage device has means for inputting a new object, means for retrieving a similar object similar to the new object from existing objects, and the new object and the similarity. Means for extracting a common part with the object, means for extracting a difference part between the new object and the similar object, means for recognizing the new object as a combination of the common part and the difference part, and the common Means for storing the new object in the storage area as the combination of the portion and the difference portion.

According to the first aspect of the invention, the attribute value defining the object is input, the attribute name is set, and the
Recognition and memory are executed as one processing unit. Therefore, the data structure can be determined not at the time of designing the knowledge database but at the time of data input, recognition, and storage.

According to the second aspect of the present invention, experience is accumulated by generating a class object and storing it as an experience in a storage area. In the invention of claim 3, the accumulated experience is utilized by applying the class object to the data object and modifying, deleting, and adding the attribute name of the data object as necessary to generate the instance object. To do.

In the fourth aspect of the present invention, the superclass object is stored in the storage area to facilitate utilization of accumulated experience. In the invention of claim 5, by forming a hierarchical structure of superclasses, utilization of accumulated experience is facilitated.

According to the sixth aspect of the present invention, a high-speed search is realized by searching for the similar object from the upper superclass to the lower superclass. In the invention of claim 7, the search index is used to realize a higher speed search.

In the eighth aspect of the invention, since all the objects are linked to the mesh structure, complex recognition is possible. In the invention of claim 9, the object can include an atomic object.

In the tenth aspect of the invention, the atomic object can exist in all real world models stored in the knowledge base. According to the invention of claim 11,
A horizontal thinking search is realized by obtaining superclass objects and objects corresponding to different recognition of designated objects as search results.

In the twelfth aspect of the invention, the flash search is realized by associating the first object and the second object with a new object. Claim 13
In the invention, the new object is recognized as a combination of the common part and the different part, and the new object is stored in the storage area as the combination of the common part and the different part. Therefore, it is easy to reuse the object.

In the fourteenth aspect of the invention, the new object is defined by the attribute name and the attribute value. According to the fifteenth aspect of the present invention, the similar object retrieved by designating the domain can be dynamically changed.

In the sixteenth aspect of the present invention, it is possible to dynamically change the similar object retrieved by designating the purpose. According to the seventeenth aspect of the present invention, a similar object is quantitatively searched for by using the similarity determination method based on the domain and the purpose.

According to the eighteenth aspect of the present invention, the similarity determining method is based on whether the selected attribute name selected from the attribute names of the new objects exists in the existing object and the weight assigned to the selected attribute name. A similar object is quantitatively searched by calculating the similarity based on the coefficient.

According to the nineteenth aspect of the present invention, since the selected attribute name and the weight coefficient depend on the domain and the purpose,
They can be changed dynamically. According to the twentieth aspect of the present invention, by generating a superclass having the same attribute name and attribute value as a common part, a new object and a similar object can be associated with this superclass.

According to the twenty-first aspect of the present invention, the new object and the similar object can be associated with each other by generating the superclass having the same selected attribute name and the corresponding attribute value as the common part. . In the invention of claim 22, when the attribute names are the same but the attribute values are different, the same attribute name is extracted in the is-a relation and the class in the is-a relation is generated as the difference part. This is-a
Can be related by a relationship class.

According to the invention of claim 23, when the selected attribute names are the same but the corresponding attribute values are different, the same selected attribute name is extracted in the is-a relation and the is-a relation is extracted. By generating a class as a difference part, a new object and a similar object can be related by this is-a relationship class.

According to the twenty-fourth aspect of the present invention, when there is an attribute name that exists in the new object but does not exist in the similar objects, the attribute name that does not exist and the corresponding attribute value are
Since the class of the part-of relationship is extracted as the difference part by extracting with the part-of relationship, the new object and the similar object can be related by the class of the part-of relationship.

According to the twenty-fifth aspect of the present invention, when there is a selection attribute name that exists in the new object but does not exist in the similar objects, the selection attribute name that does not exist and the corresponding attribute value are extracted in a part-of relationship. Then, since the class of the part-of relation is generated as the difference part, the new object and the similar object can be related by the class of the part-of relation.

In the twenty-sixth aspect of the present invention, when there is an attribute name that exists in the similar object but does not exist in the new object, the attribute name that does not exist and the corresponding attribute value are
Since the class of the part-of relationship is extracted as the difference part by extracting with the part-of relationship, the new object and the similar object can be related by the class of the part-of relationship.

In the twenty-seventh aspect of the present invention, when there is a selection attribute name that exists in the similar object but does not exist in the new object, the selection attribute name that does not exist and the corresponding attribute value are extracted in a part-of relationship. Then, since the class of the part-of relation is generated as the difference part, the new object and the similar object can be related by the class of the part-of relation.

In the twenty-eighth aspect of the invention, the existing object can be grown by associating the new object and the similar object with the common part and the different part. In the invention of claim 29, since the superclass, the is-a relation class, and the part-of relation class are constructed in a hierarchical manner, a search base is provided.

According to the thirtieth aspect of the present invention, the superclass, the is-a relation class, and the part-of relation class are provided with a sensory expression representing their meanings, so that they are used for flexible retrieval. The base of is given. In the thirty-first aspect of the invention, since the hierarchical structure is a mesh structure including links that connect objects in an oblique direction, a flexible search base is provided.

According to the thirty-second aspect of the invention, since the target object is searched with one of the sensory expressions corresponding to the ambiguous concept as the starting point, the ambiguous search can be realized. In the thirty-third aspect of the invention, the knowledge storage device sets an attribute name after inputting an attribute value defining an object, and executes input, recognition and storage as one processing unit. Therefore, the data structure can be determined not at the time of designing the knowledge database but at the time of data input, recognition, and storage.

In the thirty-fourth aspect of the present invention, the knowledge storage device recognizes the new object as a combination of the common part and the different part, and stores the new object in the storage area as the combination of the common part and the different part. . Therefore, it is easy to reuse the object.

[0061]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows a principle configuration diagram of a knowledge database by a recognition module system in which perception, recognition, and memory are integrated according to the present invention. In FIG. 1, the knowledge database 10 includes an input unit 11 such as a keyboard, a processing unit 12 including a CPU and a memory, a storage unit 13 such as a RAM and a hard disk, and a display unit 20 such as a CRT display.
including.

The input unit 11 and the display unit 20 enable interactive processing between the operator and the knowledge database 10. The input unit 11 receives data regarding an object input by an operator, and the processing unit 12 processes the data based on conversational processing to recognize the object. The recognized object is stored in the storage unit 13. Further, the processing unit 12 follows the instruction from the operator.
Various types of searches are performed on the data stored in the storage unit 13.

FIG. 2 is a block diagram of the processing unit 12 according to the first embodiment, and the configuration shown here is realized by a computer. In the first embodiment, the processing unit 12 includes an attribute recognition module set selection unit 14, an object recognition module dynamic configuration unit 15, an object recognition module recognition unit 16, and an attribute recognition module recognition unit 1.
7, superclass-like recognition module recognition unit 18,
And a superclass-like recognition module creation unit 19. Further, the processing unit 12 includes a search unit 21.

The knowledge database 10 recognizes and stores the input object by the respective parts of the processing unit 12 operating in cooperation with each other. In order to describe the operation of each part of the processing unit 12, first, the data structure used in the present invention will be described. FIG. 3 is a diagram for explaining a data structure in which recognition and storage by the recognition module system according to the present invention are integrated. As shown in FIG. 3, an object has attributes for recognizing the object, and these attributes are called a recognition module.

In the present invention, there are three cases in which the attributes constituting an object are attribute names only, attribute names and attribute values, or only attribute values. This will be explained as shown in FIG. In FIG. 4A, a photograph object has a combination of an attribute name and an attribute value as a person:
Ryu, Time: PM5: 20; 3/3, Place: In front of the Great Buddha statue in Kamakura, Object: Photo. Here, for example, by setting all attribute values other than the attribute "thing: photograph" to NULL (blank), a general object "photo" that is not a specific photograph, as shown in FIG. 4B. Can be formed. Also, in reality, FIG.
When inputting the data of the photograph as shown in (a), it is often the case that the attribute name is not given and only the attribute value is given. In other words, when entering the data, "Dr.
The first data is "Photo taken in front of the Great Buddha in Kamakura". In the present invention, this is decomposed into each attribute value and treated as an object having only the attribute value (called a data object). That is, in this case, an object having only attribute values as shown in FIG. 4C is generated.

As described above, the present invention is not limited to the conventional object-oriented framework in which an instance is created from a class. Then, a class object (an object having only an attribute name), an instance object (an object having an attribute name and an attribute value), and a data object (an object having only an attribute value) coexist, and each can generate another. It is composed.

In the present invention, each attribute is called a recognition module because each attribute is the element for recognizing an object. The object is then recognized when it creates an instance object with an attribute name and an attribute value. This will be described with reference to FIG. If you translate this instance object called "photo", you can read "The person named Ryu, 3: 5: 30 pm,
You can create a description of "a place in front of Kamakura, a photo". Here, this explanation can be considered as the recognition itself of this "photograph".
In other words, in other words, this "photograph" instance object is composed of information about the recognition itself of this object. Thus, creating and storing an instance object in this way corresponds to recognizing and storing this object.

In the human memory mechanism, unrecognized information cannot be stored, and all stored information is perceived and recognized. Human beings recognize what they perceive, and when they are recognized, they remember their recognition target. In other words, in the human brain, perception, recognition, and memory are integrally processed. The present invention corresponds to such a human information processing mechanism, and perceives (data input)
-Handling recognition and memory as one.

Further, in the present invention, as shown in FIG. 3, each object recognition module forms a corresponding object. Each of these objects then consists of a further recognition module.
For example, in the example of FIG. 4A, one recognition module of person: dragon forms an object describing a person of dragon, which has attributes (reception modules) such as age and occupation. It will be. Thus, in the present invention, each recognition module has a nested structure.

An explanatory view of the recognition module having such a nested structure is shown in FIG. As shown in FIG. 5, the nested structure is not limited to tree-like and network-like structures. For example, the object X has the recognition module c, and the object c has the recognition module X. The objects are linked in a mesh shape. A group of objects thus combined in a mesh shape can be regarded as one object (called a compound object).

Further, in such a nested structure, as shown in FIG.
As shown in, there exists an object (called an atomic object) as a minimum unit. This atomic object is for an entity that cannot be created as an object consisting of attributes, that is, a description cannot be considered. For example, when applied to human neurons, it is considered that neurons at the end of the nervous system that control the five senses present in the sensory organs such as the eyes and ears correspond to this atomic object. This atomic object exists in words, programs, files, screens, books, and so on.

In the present invention, the knowledge database is constructed using the data structure in which the recognition and the memory are integrated as described above, that is, the recognition module structure. FIG. 6 shows a flowchart of processing for realizing the storage system according to the first embodiment of the present invention. In the storage system of the present invention, a data object is first input, and a class object is applied to the data object to generate an instance object, whereby the object is recognized.

In step S1, some attribute recognition modules are selected. The selection of the attribute recognition module mentioned here means that the operator inputs the attribute name of the object to collect the attribute name. In other words, for the object stored in the database, the operator inputs the attribute name that seems appropriate for describing the object. This means that the database recognizes what kind of attribute the operator recognizes the object.

At step S2, the object recognition module is constructed by collecting the selected attribute recognition modules. The object recognition module referred to here is shown in FIG.
It is a class object represented by a group of recognition modules (only attribute names) described in (b). That is, in step S2, a class object including only the attribute name is generated. It should be noted that the steps up to this step S2 are the steps of
This is a preparation process for the recognition / storage process, and is equivalent to generating class objects (objects having only attribute names) for various objects.

In step S3, a data stream is input. That is, the attribute value is input for the object to be input. This will create a data object for that object. In step S4, the superclass-like recognition module recognizes the object. Superclass-like here is a superclass created by extracting attributes common to multiple objects, and is-a relations and pas in object thinking.
It is a super class that also includes rt-of relationships. In step S4, the class object corresponding to the data object input in step S3 is searched from the class objects stored in advance. In this search, "object-superclass-like" formed by extracting common attributes
In the hierarchical structure of, the target class object is searched from the upper superclass-like to the lower superclass-like. In this case, since the upper superclass is semantically ambiguous, it can be interpreted that this search realizes the ambiguous search. The fuzzy search will be described in detail later. Step S4
In, if the existing class object retrieved matches the input data object,
Apply the retrieved class object to the data object and generate an instance object. That is, the input object is recognized.

In step S5, the object recognition module recognizes the object. Here, when the class object searched in step S4 does not completely match the input data object, the operator changes or adds the attribute name of the mismatched part to generate an instance object. This recognizes the input object.
In addition, without performing the processing of step S4,
It is also possible to create an instance object by giving all the attribute names in 5.

In step S6, the recognized object, that is, the instance object is stored in the storage unit. In step S7, a new superclass-like recognition module is created. In step S7, first, when the instance object is formed in step S5, the attribute name of the instance object is taken out to create a new class object. Further, common attributes are extracted by comparing the newly created class object with existing class objects, and a super class-like having the common attributes is created. The creation of this superclass-like continues as long as a new superclass-like can be created.

In step S8, the created class object and superclass-like are stored in the storage unit. According to the process of the flowchart shown in FIG. 6, the knowledge database 10 of the present invention in FIG. 1 integrally performs object perception (input), recognition, and memory. FIG.
The operation of each part of the processing unit 12 will be described with reference to again.

In FIG. 2, the attribute recognition module set selection unit 14 collects attribute names (recognition modules) for recognizing an object, corresponding to step S1. The attribute names collected by the attribute recognition module set selection unit 14 are grouped into class objects by the object recognition module dynamic configuration unit 15. That is, the object recognition module dynamic configuration unit 15 configures a plurality of class objects configured by collecting some attribute names.

Object recognition module recognition unit 16
Receives a data stream input to the input unit 11 and generates a data object. Further, the superclass-like recognition module recognition unit 18 is operated to search for a class object to be applied to the data object. The superclass-like recognition module recognizing unit 18 searches for a target class object from an upper superclass-like stored in the storage unit 13 to a lower superclass-like based on the data stream from the input unit 11. When the superclass-like recognition module recognition unit 18 finds the target class object, it supplies it to the object recognition module recognition unit 16. Object recognition module recognition unit 1
6 applies the given class object to a data object to generate an instance object for recognition, and stores the instance object in the storage unit 13.

If the class object retrieved by the superclass-like recognition module recognition unit 18 does not match the data object, the object recognition module recognition unit 16 needs to correct the mismatched attribute name or acquire the missing attribute name. To do. In such a case,
The object recognition module recognition unit 16 operates the attribute recognition module recognition unit 17 to correct the attribute name or newly acquire the attribute name. The attribute recognition module recognition unit 17 displays the data object and the class object on the display unit 20. The operator corrects / adds the attribute name via the input unit 11 based on the information displayed on the display unit 20. The attribute recognition module recognition unit 17 receives the corrected / added attribute names from the input unit 11 and supplies them to the object recognition module recognition unit 16.
The object recognition module recognition unit 16 applies the class object to the data object using the corrected / added attribute name to generate an instance object for recognition, and stores the instance object in the storage unit 13. At this time, the superclass-like recognition module producing unit 19 stores the newly created class in the storage unit 13 and also generates a superclasslike and stores it in the storage unit 13.

In this way, the knowledge database 10
Applies the class object to the input data stream of the object to create an instance object, and recognizes and stores the object. FIG.
6 is a flowchart showing a fuzzy search process in the knowledge database 10 according to the first embodiment of the present invention. This ambiguous search process is executed by the superclass-like recognition module recognition unit 18 in FIG.

In step S11, the uppermost superclass-like is searched for the input data stream. In step S12, it is checked whether the target object exists. If the target object does not exist, the process proceeds to step S13. If the target object exists, the process proceeds to step S15.

In step S13, the superclass-like one level below is searched. In step S14,
Check if the desired object exists. If the target object does not exist, step S
Return to 13 and repeat the subsequent steps. If the target object exists, the process proceeds to step S15.

In step S15, the obtained object is output. The state of the above fuzzy search is shown in FIG.
As shown in FIG. 8, in the ambiguous search, a target object is searched and navigated from a more ambiguous superclass-like in an upper hierarchy to a more specific superclass-like in a lower hierarchy. Since the superclass-like of the upper hierarchy is semantically ambiguous and broad, this method can search for a target object at high speed. That is, a fast search using ambiguous concepts as implemented in human memory is provided by the knowledge database 10 of the present invention.

When the number of superclass-like layers becomes large and the search speed becomes a problem, an expansion index for navigating may be prepared. In this case, the index method, Hush
By using the method, B-tree method, etc., it is possible to shuffle up to the level that seems necessary and add an index.

FIG. 9 shows a flowchart of a process for improving the search speed by using the expansion index in response to a request from the operator. In FIG. 9, in step S21, it is determined whether or not there is a deployment request from the operator. If there is a development request, in step S22, the stored knowledge data is searched in full text to collect data of the content requested by the operator. After step S22,
Proceed to step S24. When it is determined in step S21 that there is no expansion request, it is determined in step S23 whether the expansion table has already been created. If the expansion table has not been created, the processing ends.

When the expansion table has already been created,
This means that there was a deployment request from the operator before. Therefore, it is necessary to create the expansion table for the newly updated hierarchical structure. Therefore, if it is determined in step S23 that the expansion table has already been created, the process proceeds to step S24.

In step S24, the expansion table is created and the process is terminated. In this way, by expanding using the index, the knowledge database 10 can perform ambiguous search at high speed even when the superclass-like hierarchical structure becomes deep.

Next, the horizontal thinking retrieval in the knowledge database 10 will be described. Humans can make multiple interpretations of an entity as if they were unrelated. For example, consider the entity "apple". The apple can be interpreted as a food, but it can also be used as a ball, or as a weight instead of a paperweight. Furthermore, the apple can be used as a subject when writing a painting, or as a vase for a vase. That is, for the entity "apple", food,
Multiple different interpretations of the ball, paperweight, subject, vase, etc. will be possible. This is called human horizontal thinking.
Such a horizontal thinking mechanism can be realized by the recognition module configuration of the present invention.

That is, it is assumed that the object “apple” has, for example, “shape: round” as one attribute and “property: thrown” as another attribute. Further, when this object “apple” is created, another object “ball” exists in the knowledge database 10, and similarly, “shape: round” and “property:
It is assumed that it has "thrown" as an attribute. In this case, the two objects “apple” and “ball” have the attributes “shape: round” and “property: thrown” as common attributes, so a superclass-like having these common attributes is generated. This is shown in FIG. The superclass-like generated in this way is stored in the knowledge database as an object "thing like a ball", for example.

Similarly, the object "apple" will have a superclass-like called fruit (or food), and will be superclass-like such as paperweight (like), subject, vase (like). Can have. Therefore, a horizontal thinking search can be realized by using the hierarchical structure configured in this way. Such a hierarchical structure is shown in FIG. In FIG. 11, the lower layer objects have superclass-like as upper objects, which represent different recognitions for each. Therefore, in this hierarchical structure, for example, "apple"
Starting from the object called "," discover the object called "ball" through "thing like a ball,"
It can be output as a search result.

FIG. 12 shows a flowchart of the process of horizontal thinking retrieval. This process is performed by the knowledge database 1 of FIG.
It is executed by the search unit 21 of 0. In step S31, the operator receives information regarding what kind of data should be collected. In step S32, it is determined whether or not the data thus collected already exists. If it already exists, the process ends. If such data does not exist, in step S33, the full-text search is performed on the data stored in the storage unit 13, and the data matching the data collection condition is collected in the table. Step S34
Then, it is determined whether the full text search is completed. If not completed, the process returns to step S33 to perform further full-text search. When the full text search is completed, the entire process is completed.

Since such horizontal thinking search processing is a full-text search, it takes a long time to complete data collection. However, since this processing may be executed while the knowledge database system 10 is performing other processing or during idle time, it does not matter that the processing takes a long time.

Next, the flash search in the knowledge database system 10 will be described. Humans can make a flash that mimics horizontal thinking, but this flash is generally difficult to define. Inspiration is defined here as finding some association between an object and objects that are not related to it. Applying this definition to the human thought process, when thinking about one thing, it is equivalent to another thing that seems to have nothing to do with that thing with some kind of connection.

In other words, in the present invention, a flash is defined as finding some relation between the expanded objects by expanding and interpreting one object while expanding the other object. This can be achieved by using the nested recognition module structure of the present invention. A diagram for explaining this is shown in FIG.

In FIG. 13, with respect to two objects A and B which are not related (have no common attribute), the recognition modules having a nested structure are searched one after another, starting from A. Look for a common recognition module between the search and the search starting from B. When a common recognition module (attribute) is found, a new object C having the attribute as a property is generated. By this processing, the objects A and B which seem to be unrelated at first glance are related by the object C.

FIG. 14 shows a flowchart of the flash search process. In step S41, a plurality of objects having few relations are input by the operator. In step S42, each recognition module is further subjected to nesting processing by nesting recognition to search for a match in the contents of a plurality of objects. In step S43, it is determined whether there is a match. If there is no match, step S
Go to 44. In step S44, it is determined whether or not all the recognition modules having the nested structure have been searched.
If all have been searched, the process is terminated, and if there are more search targets, the process returns to step S42 and the subsequent steps are repeated.

If it is determined in step S43 that there is a match, the process proceeds to step S45. Step S4
At 5, an object with matching attributes is created. After the object is created, the process ends. As described above, the first embodiment of the present invention can realize the storage system in which the perception (data input), the recognition, and the storage are integrated by using the recognition module structure. Furthermore, the first embodiment of the present invention can assist the intellectual activity of the operator by performing high-speed fuzzy search, high-performance horizontal thinking search, and flash search search.

The second embodiment of the present invention relates to a knowledge acquisition method, that is, a re-recognition method, for acquiring new knowledge as compared with the knowledge experienced in the past. FIG. 15 is a block diagram of the second embodiment of the processing unit 12 of the intelligent database 10 shown in FIG. 1, and the configuration shown here is realized by a computer. The second embodiment shown in FIG. 15 is
3 is a detailed diagram of a portion related to a knowledge re-recognition method in the portion of the first embodiment of FIG. 2.

In the storage system of the first embodiment shown in FIG. 2, a class object experienced in the past is applied to a data object to generate an instance object, which is recognized and stored. In this way, when storing (storing) a new object, associating the new object with an already existing object and recognizing and storing the common part and the difference part is called re-recognition. The second embodiment shown in FIG. 15 realizes such a re-recognition method.

In FIG. 15, the processing section 12 includes a difference degree determination method selection section 31, a similarity search section 32, a superclass-like search section 33, a search result output section 34, and a common part detection section 3.
5, a difference portion detection unit 36, a superclass-like definition registration unit 37, a modified portion detection unit 38, a deleted portion detection unit 39,
An additional portion extraction unit 40 and a relation registration unit 41 are included. Recognizing an object is realized by the cooperation of each of these parts. Before explaining the operation of each part, the concept of rerecognition in the present invention will be described.

FIG. 16 is a diagram for explaining the concept of re-recognition. Consider a case where a program for copying a record as shown in FIG. 16B is newly input when a program for copying a file as shown in FIG. 16A exists. At this time, as shown in FIGS. 16A and 16B, the disc read function, the disc write function, and the file recognition function are common to both programs. In addition, in the file copy program, "Target:
The attribute "file" is the difference. Furthermore, the attributes of "function: record recognition function" and "target: record" are different parts in the record copy program.
Therefore, when the record copy program shown in FIG. 16B is input, the file copy program is recognized as “common part + target: file”. The record copy program is recognized as "common part + (function: record recognition function + target: record)".
The recognition in the present invention is to generate and store an instance object as described in the first embodiment. That is, it is a re-recognition that the new instance object is stored in association with the existing instance object as described above. By this re-recognition, as will be described later, it is possible to reuse the knowledge accumulated as experience.

The determination of the common part and the difference part at the time of such re-recognition is not fixed, but dynamically changes depending on the way of designating the domain (application field). For example, consider the example shown in FIG. FIG. 17A shows the same file copy program as FIG. 16A,
FIG. 17B is the same record copy program as FIG. 16B. When these two programs exist, it is assumed that a record display program as shown in FIG. 17C is newly input. At this time, for example, if a domain called file operation is set, all programs in this area are targeted, and therefore, in FIG. 17 (d), (e),
The recognition is performed as shown in (f). If a domain called record operation is set, FIG.
The recognition is performed as shown in (g) and (h). In this way, the search can be efficiently performed by selecting the group view as to in which group the re-recognition is performed.

Furthermore, the recognition of the common portion and the difference portion can be dynamically changed according to the purpose. FIG. 18 is a diagram for explaining this. , Figure 1
Consider a case where a file copy program as shown in FIG. 8A exists in advance and the program for record editing shown in FIG. 18B is created. In the knowledge database 10, the substance of the program (source code or executable code) is stored as actual data,
The program attributes as shown in FIGS. 18A and 18B are stored as metadata for managing actual data. When a new object (metadata) having the attributes shown in FIG. 18B is newly input, the knowledge database 10 searches for objects having similar attributes (FIG. 18A) and compares the objects to recognize them again. . Based on the information of the common part and the different part recognized at this time, the operator can change / add the program of the actual data to obtain the program of record display. That is, the knowledge can be reused.

When the purpose is "creation of program of record edit", for example, each attribute of "language used", "operating machine" and "external interface" in FIG. 18B is selected. Then, the knowledge database 10 operates so as to recognize the common part and the different part with respect to these selected attributes. As a result of re-recognition by searching and comparing similar objects, the common part is recognized as "used language: C" and "operating machine: SUN4", and the different part is recognized as "external interface:". The selection of each of such attributes is dynamically set according to the purpose. For example, if a record edit program already exists, and the user wants to use the record edit program, the purpose is "use record edit program". in this case,
For example, the attribute of "function: file copy, record edit" is simply specified.

By changing the recognition of the common part and the difference part according to the purpose in this way, only the items suitable for the purpose can be selectively searched, so that the efficiency of the search can be improved. FIG. 19 is a flowchart showing the process of re-recognition according to the second embodiment of the present invention. This processing is shown in FIG.
This is realized by the second embodiment of the processing unit 12 shown in FIG.

In step S51, the domain is input and the way of looking at the group is determined. In step S52, the purpose is input, and the similarity determination method is selected according to the domain and the purpose. An example of the similarity determination method is shown in FIG. FIG.
(A) shows a parameter table that defines which attributes are selected according to purpose and what weighting is given to each attribute. This parameter table is accessed according to the purpose input in step S52 to obtain the weighting coefficient for each attribute. Then, for example, the similarity is calculated using the similarity determination formula as shown in FIG. 20B, and the object similar to the new object is searched from the storage unit 13.

In step S53, a new object is input. In step S54, similarity search processing is executed. In step S54, it is first determined whether or not each attribute as a search condition exists as attribute data (metadata) of an existing object. Then, the presence / absence of each attribute of the search condition is substituted into the determination formula of FIG. 20B to calculate the similarity. Objects whose similarity is higher than a certain value are extracted as objects satisfying the search condition.

At step S55, the attributes of the new object are compared with the attributes of the retrieved object,
It is determined whether the attribute name and the attribute value match each other. For attributes having the same attribute name and attribute value, the process proceeds to step S56. Step S
At 56, the attributes having the same attribute name and the same attribute value are extracted as a group of common parts. Next, step S5
At 7, a new super class or super class-like is created and registered, and an upper link and a lower link are attached. Then, the process proceeds to step S64.

In the second embodiment, the definition of superclass-like is used as being different from that in the first embodiment. In the first embodiment, the class object in which common attributes are extracted and summarized is defined as superclass-like. However, in the second embodiment, the object in which the common attributes are extracted and put together is defined as a super class, not like a super class. The superclass-like is an attribute that the operator intuitively defines for the superclass.

For attributes having the same attribute name but different attribute values, the process proceeds to step S58. In step S58, attributes having the same attribute name but different attribute values are extracted as is-a. Next, in step S59, a new is-a class or class-like is created and registered. Here, class-like is an attribute that the operator intuitively defines for the target class. afterwards,
The process proceeds to step S64.

For attributes that are in the new object but not in the existing object, the process proceeds to step S60. In step S60, the attributes that are present in the new object but not in the existing object are extracted as part-of. Next, in step S61, a new part-of class or class-like is created and registered. Then, the process proceeds to step S64.

For attributes not existing in the new object but existing in the existing object, the process proceeds to step S62. In step S62, the attributes that are not in the new object but are in the existing object are extracted as part-of. Next, in step S63, a new part-of class or class-like is created and registered. Then, the process proceeds to step S64.

In step S64, the metadata of the new object is created, the actual data, the class, the link to the class-like, and the writing of the actual data are performed. This ends the processing. According to the process of the flowchart shown in FIG. 19, the knowledge database 10 of the present invention in FIG. 1 re-recognizes the object. The operation of each part of the processing unit 12 according to the second embodiment will be described with reference to FIG. 15 again.

When the domain and the purpose are input from the input unit 11, the dissimilarity determination method selection unit 31 selects the similarity determination method based on the domain and the purpose. When new data is input from the input unit 11, the similarity search unit 32 executes a similarity search. In this similarity search, similar objects are searched for in the superclass-like hierarchical structure stored in the storage unit 13 via the superclass-like search unit 33. When the similar object is searched, the search result is displayed on the search result output unit 34. Further, the common part between the new object and the searched similar object is detected by the common part detection unit 35 and stored in the storage unit 13. Further, the difference between the two is detected by the difference detecting unit 36 and stored in the storage unit 13. Further, if the superclass-like is registered, the superclass-like registration unit 37 registers the superclass-like in the storage unit 13.

Based on the output of the different portion detecting portion 36, the reconstructed portion detecting portion 38 detects the portion to be reconstructed among the different portions. Further, based on the output of the different portion detecting unit 36, the deleted portion detecting unit 39 detects the portion to be deleted among the different portions. Further, the additional portion extracting unit 40 detects the added portion of the different portions based on the output of the different portion detecting unit 36. Based on the outputs of the modified portion detecting unit 38, the deleted portion detecting unit 40, and the additional portion extracting unit 40,
The association registration unit 41 associates a new object and an existing object, and stores the association between them in the storage unit 13.

FIG. 21 shows how the new object and the existing object are recognized in association with each other by the re-recognition process shown in the flowchart of FIG. FIG.
Again uses the file copy program and record edit program shown in FIGS. 18A and 18B as examples.

As shown in FIG. 21, the attributes of the creator, the language used, and the operating machine are the same in both program names and attribute values (see FIG. 18). It is extracted as a class. Also,
The attributes such as creation time, function, external interface, program size, processing time, and structure have the same attribute name but different attribute values between the programs, so they are extracted as is-a related classes. Furthermore, since the attribute of the edit input device exists only in the record edit program, it is extracted as a part-of relation class. Here, for attributes with different attribute values, it is further checked which attribute is the same and extracted as an is-a relationship or a part-of relationship. In this way, each program is recognized as a common part and a different part.

Further, for each program, FIG.
You can add superclass-like attributes as shown in. This superclass-like attribute is an attribute that the operator intuitively gives, as described above. Further, as shown in FIG. 23, both objects can be given a superclass-like "program for manipulating data" as a superordinate concept.

In the hierarchical structure constructed in the super class like manner as described above, not only the structure on the tree but also the existence of diagonal links as shown in FIG. 24 is allowed. Therefore, when a search for an object called “file copy” is executed in the hierarchical structure of FIG. 24, three “file-unit copy” objects are extracted as being associated with each other, as shown by shading. To be done. Further, since each object is provided with a superclass-like attribute that represents a sensory concept of the operator, a target object can be efficiently extracted even from an ambiguous concept. Further, as shown in FIG. 25, when the object “copy unlabeled file” is searched, the knowledge database 10 may ask the operator which one of the magnetic tape file and the disk file is the target. In this case, when the operator indicates that the magnetic tape is the target,
As shown in, an object of “copy by file (without label)” is selected.

Thus, according to the re-recognition method of the second embodiment of the present invention, the new object is expressed and recognized as the common part and the different part in the comparison with the existing object. Therefore, in the second embodiment of the present invention, it is possible to grow the incomplete knowledge of the existing object and suppress the increase of the data amount due to the increase of the knowledge. Furthermore, it is possible to reuse the existing data (actual data) based on the information regarding the common part and the different part. Further, in the second embodiment of the present invention, since the super class like based on the sense of the operator can be given as an attribute, an ambiguous search can be performed.

[0123]

According to the invention of claim 1, since the data structure is determined at the time of inputting, recognizing, and storing, it is not necessary to determine the data schema at the time of constructing the knowledge database, and the knowledge database capable of coping with changes in operating environment. Can be built.

According to the second aspect of the invention, since the experience is accumulated by generating the class object and storing it in the storage area, it is possible to construct a knowledge database that accumulates the experience as knowledge. According to the third aspect of the invention, the accumulated experience can be utilized to build a knowledge database capable of coping with changes in the operating environment.

According to the fourth aspect of the invention, it is possible to construct a knowledge database in which accumulated experience can be easily utilized. In the invention of claim 5, it is possible to construct a knowledge database that can easily utilize accumulated experience.

According to the sixth aspect of the invention, it is possible to construct a knowledge database that realizes high-speed retrieval. According to the seventh aspect of the present invention, by using the search index, it is possible to construct a knowledge database that realizes a higher speed search.

In the eighth aspect of the invention, since all the objects are linked to the mesh structure, it is possible to construct a knowledge database capable of complex recognition. In the invention of claim 9, the object can include an atomic object.

According to the tenth aspect of the invention, the atomic object can exist in all real world models stored in the knowledge base. According to the invention of claim 11,
It is possible to build a knowledge database with an advanced search function that can realize horizontal thinking search.

According to the twelfth aspect of the present invention, it is possible to construct a knowledge database having an advanced search function capable of realizing flash search. According to the thirteenth aspect of the invention, it is possible to construct a knowledge database in which objects can be easily reused.

In the fourteenth aspect of the invention, the new object can be defined by the attribute name and the attribute value. According to the fifteenth aspect of the invention, since the similar object searched for is dynamically changed by designating the domain, an efficient similarity search can be performed.

According to the sixteenth aspect of the present invention, since the similar object searched for is dynamically changed according to the purpose designation, efficient similarity search can be performed. According to the invention of claim 17, the similar object can be quantitatively searched.

In the eighteenth aspect of the invention, similar objects can be searched quantitatively. In the nineteenth aspect of the present invention, since the selected attribute name and the weighting coefficient are dynamically changed, efficient similarity search can be performed.

According to the twentieth aspect of the invention, since the new object and the similar object can be associated with each other by the superclass, it is possible to construct a knowledge database for incomplete knowledge growth. According to the twenty-first aspect of the invention, since the new object and the similar object can be associated with each other by the superclass, it is possible to construct a knowledge database for growing incomplete knowledge.

According to the twenty-second aspect of the present invention, since the new object and the similar object can be associated with each other in the class of is-a relationship, it is possible to construct a knowledge database for growing incomplete knowledge. In the invention of claim 23, the new object and the similar object are
-a Since it is possible to relate by class of relation, it is possible to build a knowledge database that aims to grow incomplete knowledge.

According to the twenty-fourth aspect of the present invention, since a new object and a similar object can be associated with each other by a class having a part-of relationship, it is possible to construct a knowledge database for growing incomplete knowledge. Claim 25
In the invention, since the new object and the similar object can be associated with each other by the class of the part-of relation, it is possible to construct the knowledge database for the incomplete growth of knowledge.

According to the twenty-sixth aspect of the present invention, since a new object and a similar object can be associated with each other by a class having a part-of relationship, it is possible to construct a knowledge database for growing incomplete knowledge. Claim 27
In the invention, since the new object and the similar object can be associated with each other by the class of the part-of relation, it is possible to construct the knowledge database for the incomplete growth of knowledge.

According to the twenty-eighth aspect of the present invention, it is possible to build a knowledge database capable of growing existing objects. According to the twenty-ninth aspect of the invention, it is possible to construct a knowledge database provided with a base for high-speed search.

According to the thirtieth aspect of the present invention, it is possible to construct a knowledge database provided with a flexible search base. According to the thirty-first aspect of the invention, it is possible to build a knowledge database provided with a flexible search base.

According to the thirty-second aspect of the present invention, it is possible to construct a knowledge database for realizing ambiguous retrieval. In the invention of claim 33, since the knowledge storage device determines at the time of inputting, recognizing and storing the data structure, it is not necessary to determine the data schema at the time of constructing the knowledge database, and the knowledge database capable of coping with changes in operating environment. Can be built.

In the thirty-fourth aspect of the present invention, the knowledge storage device can construct a knowledge database in which objects can be easily reused.

[Brief description of drawings]

FIG. 1 is a block diagram of a knowledge database system according to the present invention.

2 is a block diagram according to a first embodiment of the processing unit of FIG. 1. FIG.

FIG. 3 is a conceptual diagram showing a data structure of an object in the present invention.

4A, 4B, and 4C are diagrams showing examples of an object consisting of an attribute name and an attribute value, an object consisting of an attribute name, and an object consisting of an attribute value in the present invention, respectively. .

FIG. 5 is a conceptual diagram showing a mesh structure between objects in the present invention.

FIG. 6 is a flow chart showing processing of a storage system according to the present invention.

FIG. 7 is a flowchart showing a process of ambiguous search according to the present invention.

FIG. 8 is a conceptual diagram showing how a fuzzy search is performed in the present invention.

FIG. 9 is a flow chart showing a development table creation process for high-speed ambiguous search in the present invention.

FIG. 10 is a conceptual diagram for explaining horizontal thinking retrieval in the present invention.

FIG. 11 is a conceptual diagram for explaining a horizontal thinking search according to the present invention.

FIG. 12 is a flowchart showing the processing of horizontal thinking search according to the present invention.

FIG. 13 is a conceptual diagram for explaining flash search in the present invention.

FIG. 14 is a flowchart showing a flash search process according to the present invention.

FIG. 15 is a block diagram according to a second embodiment of the processing unit of FIG.

16 (a) and 16 (b) are conceptual diagrams for explaining re-recognition processing in the present invention.

17 (a) to 17 (h) are diagrams showing an example of how the view of a group changes depending on the domain in the present invention.

18 (a) and 18 (b) are diagrams showing an example in which determination of a common part and a different part is performed for a selected attribute name in the present invention.

FIG. 19 is a flowchart showing a re-recognition process according to the present invention.

FIG. 20 is a diagram showing an example of a similarity determination formula used in the re-recognition process in the present invention.

FIG. 21 is a diagram showing an example of extraction of a common part and a different part in the re-recognition process of the present invention.

FIG. 22 is a diagram showing an example of superclass-like attributes given to an object in the re-recognition processing of the present invention.

FIG. 23 is a diagram showing an example of a superclass-like attribute given to an object in the re-recognition processing of the present invention.

FIG. 24 is a diagram showing diagonal links in the superclass-like hierarchy according to the present invention.

FIG. 25 is a conceptual diagram showing a search based on an ambiguous concept in the superclass-like hierarchy according to the present invention.

[Explanation of symbols]

 10 knowledge database system 11 input unit 12 processing unit 13 storage unit 14 attribute recognition module set selection unit 15 object recognition module dynamic configuration unit 16 object recognition module recognition unit 17 attribute recognition module recognition unit 18 superclass-like recognition module recognition unit 19 super Class-like recognition module creation unit 20 Display unit 21 Search unit 31 Dissimilarity determination method selection unit 32 Similarity search unit 33 Superclass-like search unit 34 Search result output unit 35 Common part detection unit 36 Difference part detection unit 37 Superclass-like definition Registration part 38 Modified part detection part 39 Deleted part detection part 40 Additional part extraction part 41 Relationship registration part

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Hiroyuki Izumi, inventor Hiroyuki Izumi 1015, Kamiodanaka, Nakahara-ku, Kanagawa Prefecture, Fujitsu Limited (72) Inventor, Masahiko Murakawa, 1015, Kamikodanaka, Nakahara-ku, Kawasaki, Kanagawa Prefecture, Fujitsu Limited ( 72) Inventor Shoichi Nakamura 1-4-4 Hakataekimae, Hakata-ku, Fukuoka City, Fukuoka Prefecture, Fujitsu Kyushu Communication Systems Co., Ltd.

Claims (34)

[Claims] 1. A data object having the attribute value is generated by inputting the attribute value of the object; an attribute name is given to the attribute value of the data object to generate an instance object having the attribute name and the attribute value. By recognizing the object, the step of storing the object by storing the instance object in a storage area is included, and the input, the recognition, and the storage are executed as one processing unit. A knowledge storage method for storing knowledge in a characteristic knowledge database. 2. The knowledge storage method according to claim 1, further comprising the step of generating a class object consisting of the attribute name of the instance object and storing it as an experience in the storage area. 3. The step of performing the recognition includes fitting the class object stored in the storage area to the data object as experience; modifying or deleting the attribute name of the data object as necessary.
The knowledge storage method according to claim 2, further comprising each step of generating the instance object by adding. 4. The storing as the experience includes searching the storage area for similar objects having several attribute names that are the same as the attribute names of the instance object; Extracting the same attribute name; generating a superclass object having the same attribute name; and further including steps of storing the superclass object in the storage area as an upper class of the class object The knowledge storage method according to claim 2 or 3. 5. The method further comprising the step of forming a hierarchical structure of the superclass by repeatedly generating an upper superclass object for the superclass object and storing the superclass object in the storage area. 4. Knowledge storage method described in 4. 6. The knowledge storage method according to claim 5, wherein the step of searching includes the step of searching the similar object from a superclass of a higher rank to a superclass of a lower rank in the hierarchical structure. 7. The knowledge storage method according to claim 6, wherein the searching step uses an index for searching. 8. Each of the attribute name and the attribute value is represented as a further object including at least one of the attribute name and the attribute value, and a link connecting each of the attribute name and the attribute value and the corresponding object. 8. The knowledge storage method according to any one of claims 1 to 7, wherein all the objects are linked to a mesh structure according to. 9. The knowledge of claim 8, wherein the objects linked to the mesh structure include atomic objects whose attribute names and attribute values cannot be represented as further objects. Memory method. 10. The knowledge storage method according to claim 9, wherein the atomic object exists in all real world models stored in the knowledge base. 11. An object is designated; the designated object is searched in the hierarchical structure; each of the superclass objects of the designated object is searched; each of the superclass objects and of the superclass object 11. The superclass object and the object corresponding to different recognition of the designated object are obtained as search results, including each step in which the objects that are subordinate classes are used as search results. The knowledge storage method according to any one of claims. 12. A first object and a second object are designated; a first object associated with the first object, wherein a direction from the attribute name and the attribute value toward the object in the mesh structure is a downstream direction. Retrieving one related object in the mesh structure in the downstream direction; retrieving a second related object in relation to the second object in the mesh structure in the downstream direction; Extracting the matching attribute name and attribute value between the related object and the second related object;
Creating a new object consisting of said matching attribute name and attribute value; including each step of storing said new object, and intelligently by associating said first object and said second object with said new object The knowledge storage method according to any one of claims 8 to 11, wherein the activity support is provided. 13. A new object is input; a similar object similar to the new object is searched from existing objects; a common part between the new object and the similar object is extracted; Recognizing the new object as a combination of the common part and the different part; storing the new object in the storage area as the combination of the common part and the different part Knowledge storage method. 14. The knowledge storage method according to claim 13, wherein the inputting step includes the step of inputting an attribute name and an attribute value of the new object. 15. Further comprising the step of entering a domain,
15. The knowledge storage method according to claim 13, wherein the similar objects retrieved by designating the domain are different. 16. The knowledge storage method according to claim 13, further comprising the step of inputting a purpose, wherein the similar objects searched by designating the purpose are different. 17. The method according to claim 16, wherein the searching step includes determining a similarity determination method based on the domain and the purpose; and searching the similar object using the similarity determination method. Knowledge storage method. 18. The similarity determination method determines whether a selected attribute name selected from the attribute name of the new object exists in the existing object and a weighting factor assigned to the selected attribute name. 18. The knowledge storage method according to claim 17, wherein the similarity is calculated based on the similarity. 19. The selected attribute name and the weighting factor are:
The knowledge storage method according to claim 18, wherein the knowledge storage method depends on the domain and the purpose. 20. In the step of extracting the common part, when both the attribute name and the attribute value are the same between the new object and the similar object, the same attribute name and the same attribute value are used. 18. The knowledge storage method according to any one of claims 14 to 17, further comprising the step of generating a superclass having as a common part. 21. In the step of extracting the common part, when both the selected attribute name and the corresponding attribute value are the same between the new object and the similar object, the same selected attribute name and The knowledge storage method according to claim 18 or 19, further comprising the step of generating a superclass having the corresponding attribute value as the common part. 22. In the step of extracting the different portion, when the new object and the similar object have the same attribute name but different attribute values, the same attribute name is-a 21. The knowledge storage method according to claim 20, further comprising the step of extracting with a relation and generating a class of an is-a relation as the difference portion. 23. In the step of extracting the difference part, when the selected attribute name is the same between the new object and the similar object but the corresponding attribute values are different, the same selected attribute name is selected. 22. The knowledge storage method according to claim 21, further comprising the step of extracting an is-a relation and generating a class of the is-a relation as the difference portion. 24. In the step of extracting the difference part, when there is an attribute name that exists in the new object but does not exist in the similar object, the attribute name that does not exist and the corresponding attribute value are in a part-of relationship. Extract and part-of
23. The knowledge storage method according to claim 22, further comprising the step of generating a class of relation as the difference portion. 25. In the step of extracting the difference part, when there is the selected attribute name which exists in the new object but does not exist in the similar object, the selected attribute name which does not exist and the corresponding attribute value are part- 24. The knowledge storage method according to claim 23, further comprising the step of extracting with a of relation and generating a class of a part-of relation as the difference portion. 26. In the step of extracting the difference portion, when there is an attribute name that exists in the similar object but does not exist in the new object, the attribute name that does not exist and the corresponding attribute value are in a part-of relationship. Extract and part-of
The knowledge storage method according to claim 24, further comprising the step of generating a class of relation as the difference portion. 27. In the step of extracting the difference part, when there is the selected attribute name that exists in the similar object but does not exist in the new object, the selected attribute name that does not exist and the corresponding attribute value are part- 26. The knowledge storage method according to claim 25, further comprising the step of extracting with a relation of and generating a class of a part-of relation as the difference portion. 28. When the new object is stored, by associating the new object and the similar object with the common part and the different part,
28. The knowledge storage method according to claim 13, wherein the existing object is grown. 29. The knowledge according to claim 26, wherein the superclass, the is-a relation class, and the part-of relation class are hierarchically constructed. Acquisition method. 30. The sensory expressions expressing their meanings can be given to the superclass, the is-a relation class, and the part-of relation class. 29. The knowledge acquisition method according to any one of 29. 31. The knowledge acquisition method according to claim 30, wherein the hierarchical structure is a mesh structure including links that connect the objects in an oblique direction. 32. Inputting an ambiguous concept corresponding to a target object; using one of the sensory expressions corresponding to the ambiguous concept as a starting point, and using the sensory expression arranged in the mesh structure. 32. The knowledge acquisition method according to claim 31, further comprising the steps of searching the mesh structure; searching the target object. 33. A means for generating a data object having the attribute value by inputting the attribute value of the object;
Means for recognizing the object by assigning an attribute name to the attribute value of the data object and generating an instance object consisting of the attribute name and the attribute value; by storing the instance object in a storage area Including means for storing the object, the input,
A knowledge storage device for storing knowledge in a knowledge database, characterized in that the recognition and the storage are executed as one processing unit. 34. means for inputting a new object;
Means for retrieving similar objects similar to the new object from existing objects; Means for extracting a common part between the new object and the similar object; Means for extracting a difference part between the new object and the similar object And; a knowledge storage device including means for recognizing the new object as a combination of the common portion and the different portion; and means for storing the new object in the storage area as the combination of the common portion and the different portion.