JP2009169840A - Knowledge structure configuration method of natural language interactive agent, method of creating automatic response using knowledge structure and automatic response creating device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method of generating a quick and flexible automatic response. <P>SOLUTION: A computer responds to a message from a user by using a graphic structure of an expanded semantic network that can represent a relationship between an optional state belonging to a set of knowledge structures and an operation to thereby be able to quickly perform a response corresponding to the message from the user and flexibly perform construction and addition of a knowledge structure. By constructing a discourse structure in the knowledge structure to handle a dialogue based on a natural language and further constructing a following structure, a new element can be generated from elements of a knowledge structure that has been already stored. By handling a relationship between elements of the knowledge structure by using a class, expandability is secured, and an operation in response generation can be securely maintained even through an entity is huge data. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a program that responds in response to a user message. In particular, the present invention relates to a program that extracts a keyword from a message input by a user and responds quickly.

2. Description of the Related Art Conventionally, there is a program called a chat bot that performs a natural language dialogue corresponding to a text input by a user. These are programs that extract a keyword from a message input by a user and perform a response that matches a word in the database. The situation to which the situation identification data belongs by judging the situation identification data corresponding to the entered information by comparing the information entered by the user with the situation identification data and adding the score to the situation identification data corresponding to the inputted information Summarize the evaluation score of the data, determine the situation data that should be responded to the information entered according to the status of the assessment score, and according to the output format associated with the situation data to be responded by the response output generation means A system that generates response data and performs output processing is disclosed (see Patent Document 1).
Also, in the technical field related to natural language processing, graph theory is used to handle the data structure of components included in natural language data or the relationship between natural language processing algorithms, and edges (edges) and vertices (nodes) are used. There is known a method of showing a relationship between components of natural language data by a graph structure (see Non-Patent Document 1).
JP 2006-172280 A Wikipedia Free Encyclopedia, “Graph Theory” [online], last update on Thursday, October 25, 2007 at 16:05, [Search December 18, 2007], Internet, <URL: http://en.wikipedia.org/wiki/%E3%82%B0%E3%83%A9%E3%83%95%E7%90%86%E8%AB%96>

  However, in the conventional technique or a combination of the conventional technique and a technique known in the art, in order to extract a response corresponding to the message input by the user, the database items in which the situation and the action are associated are matched one by one. Therefore, there is a problem that it takes time to extract information for response.

In order to solve the above-mentioned problems, the present invention provides a knowledge structure with self-creativity having a method such as taking in its components from the outside by system activities, thereby providing a user with a natural language dialogue. The purpose is to speed up the response.
It is another object of the present invention to provide a knowledge structure called an extended semantic network in which the semantic network is expanded in order to provide the knowledge structure of the dialog agent with the self-creating property. In addition to the well-known graph theory consisting of edges and vertices, the extended semantic network implements the knowledge structure of interactive agents with excellent flexibility and extensibility by providing features that represent edges that modify edges. With the goal.

  The present invention provides the following solutions.

(1) A knowledge structure forming method for creating a knowledge structure in a storage unit that is a set including a relationship between a situation and an action included in a computer-processable language, wherein the relationship includes a start point and an end point as elements. Storing a term for specifying the situation and a term for specifying the action in the storage unit, defining a class for each of the situation, the action, and the relationship, and the situation Including the relationship between the term for specifying the term and the term for specifying the action included in the knowledge structure as the element, and the membership or subordinate relationship between the class and the relationship as the element Storing in a knowledge structure, wherein the relationship is a relationship in which both the start point and the end point are empty elements, or the start point and the end point are shared. The knowledge is either a relation to an element of the structure, configuring the knowledge structure.

The knowledge structure construction method according to the present invention is used for a dialogue program or the like for executing an automatic response using a computer.
A storage unit for creating a knowledge structure according to the present invention includes an electrical storage unit such as a semiconductor memory that is included in computer resources and can be used as a main storage or a virtual storage, a magnetic storage unit such as a magnetic disk, and an optical unit such as a magneto-optical disk. A storage means and the like as appropriate.
The knowledge structure construction method according to the present invention also provides a knowledge structure, which is a set of languages having both situations and actions as elements, and typically has a relationship where a situation is a start point and an action is an end point. It has common points with the prior art in that it can be an element of the knowledge structure. That is, the method of describing the relationship by the start point and the end point according to the prior art using a known graph structure or the like can be similarly used in the present invention. Here, in the knowledge structure representation technique using the graph structure, the start point and the end point are called “vertices” or “nodes”, and the relationship from the start point to the end point is called “side” or “edge”.
In the graph theory according to the related art, a set of start points or end points is prepared, and a set of edges is prepared using a combination of start points and end points. That is, the edge is defined from the start point to the end point.
On the other hand, associating elements belonging to the knowledge structure according to the present invention, elements are prepared without particularly distinguishing vertices and sides, and in the combination of two elements, a combination in which both the start point and the end point are empty elements. Treats as a vertex, and treats a combination whose start and end are not empty elements as an edge. That is, by combining two elements belonging to the knowledge structure, the combination is treated as a vertex and an edge.
The extended semantic network is a technique for representing vertices or edges of a graph structure by combining two elements in this way.
In the knowledge structure according to the present invention, if two elements belonging to the knowledge structure are combined, it can be a vertex or an edge. Specifically, the end point may be a situation or an action, or a relationship between the situation and the action. In this respect, the association of elements belonging to the knowledge structure according to the present invention has a feature not found in the prior art. That is, it is possible to perform association between a situation and a situation, association between a situation and an action, association between an action and an action, and association between an action and a relation having these relations as one relation. Therefore, in the knowledge structure configuration method according to the present invention, by providing an extended semantic network in addition to the conventional semantic network, it is possible to flexibly associate elements and relationships between elements included in the knowledge structure. It becomes possible to carry out.
In expressing the knowledge structure according to the present invention using the graph structure, in addition to the association from the start point to the end point similar to the related art, an edge for modifying the edge representing the association can be represented. That is, the end point of a side can also be a side. Using such an extended semantic network that is an extension of the prior art, the knowledge structure construction method according to the present invention can construct a knowledge structure for representing a meaning or storage structure.

ここに、ＡＳＮは拡張意味ネットワーク（Augmanted Semantic Network）であり、Ｅ_ＡＳＮは拡張意味ネットワークの集合であり、ｅは集合Ｅの要素である二項組要素であり、ｅ_ｓはＥ_ＡＳＮの要素である始点（３０１）等であり、ｅ_ｄはＥ_ＡＳＮの要素である終点（３０２）等であり、nullは空要素であり、さらに、拡張意味ネットワークは次の要素集合を有し、

ここに、ＡＳＮ_ＶＥは拡張意味ネットワークＡＳＮの構成要素であり、Ｅ_ｖは拡張意味ネットワークにおける頂点の集合であり、Ｅ_ｅは拡張意味ネットワークにおける辺の集合である。 (2) The knowledge structure configuration method according to (1), wherein in the step of defining a class for each of the situation, the action, and the relationship, the following mathematical formula is used to define the class.

Here, ASN is an enhanced semantic network (Augmanted Semantic _{Network), E ASN} is a set of extended semantic network, e is a two-tuple element is an element of the set E, _{e s} is an element of _{E ASN} a certain starting point (301) or the like, _{e d} is the end point (302) or the like is an element of _{E ASN,} null is empty element further extended semantic network has the following element set,

Here, ASN _VE is a component of the extended semantic network ASN, E _v is a set of vertices in the extended semantic network, and E _e is a set of edges in the extended semantic network.

Equation 1 indicates that the _e s, _{e d} constituting a two-tuple or elements of the element set _{E ASN,} or an empty element null.
Formula 2 represents that the extended semantic network according to the present invention includes the vertex set E _v and the edge set E _e as components.
Equation 3 indicates that the vertex set E _v includes elements e ( _es , _ed ) when the binomial elements _es , _ed are both empty elements.
Equation 4 is a set _{E e} of the edge indicates that contain elements _{e (e} s, _{e d)} when the binary sets elements _e s, _{e d} belongs to the set _{E v} sides.
Equations 3 and 4 are commonly expressed using elements e ( _es , _ed ). That is, in the extended semantic network according to the present invention, the end point of the side included in the element can be not only the end point _ed but also the side e ( _es , _ed ). Have. Therefore, in the extended semantic network according to the present invention, the end point of the side e _i (305) starting from the start point e _c (304) can be set as the side e _e (303).

(3) receiving a term for specifying the situation or a term for specifying the action, and selecting and storing the received term for specifying the situation or the term for specifying the action The knowledge structure constructing method according to (1) or (2).

The receiving means may be input by the user using a keyboard or the like, or may be input by appropriately converting the call into character information that can be processed by a computer using a call analysis program or the like.
The selecting means may be a known means such as excluding a specific term based on a predefined keyword or the like, and the storing means may be a computer readable medium or the like, and the means for selecting and storing is appropriately designed. Yes.
As a result, the knowledge structure construction method according to the present invention can add terms to the knowledge structure according to the present invention by receiving, selecting and storing terms, phrases, etc. transmitted from the outside. become.

(4) generating a discourse structure that is a set having a conversation place, conversation history, and conversation context as elements, storing the conversation place as one element in the knowledge structure, and generating the conversation (1) to (3), further comprising the steps of: storing in the knowledge structure a total ordered set as follows; and storing in the knowledge structure the occurrence of the conversation and the sequential adjacency relationship of the occurrence of the conversation. A method for constructing the knowledge structure according to any of the above.

The discourse structure includes, in particular, a set having, as elements, a conversation context expressed using a computer-processable character code.
Specifically, a chat room is a chat room where a plurality of users connected via a network terminal such as a personal computer or a mobile phone can exchange character information or voice information with each other substantially in real time. Includes services such as services and network conferencing systems.
Specifically, the total ordered set includes a set having, as elements, means for regularly storing the temporal order of conversations in association with the magnitudes of numbers for each occurrence of conversations.
Thus, when the occurrence of a conversation is detected, the knowledge structure configuration method according to the present invention identifies the occurrence of each conversation using means for regularly storing the temporal order, and creates the knowledge structure according to the present invention. In addition, the conversation context expressed using the character code can be taken into the knowledge structure.

(5) generating a tracing structure that is a set including a relation class, a direction, and a destination in the elements; extracting an element having a starting point and an ending point among elements included in the knowledge structure; Adding a combination of its own class, a flag indicating forward association to the extracted start point, and a combination of end points to the tracing structure, and its own class, reverse association to the extracted end point The method of constructing a knowledge structure according to any one of (1) to (4), further including a step of adding a flag indicating a combination of start points to the trace structure.

The step of generating the tracing structure is a step of generating a new element based on the already stored knowledge structure element.
The relation class is a class to which a situation or action relation belongs, which is an element included in the knowledge structure according to the present invention. Specifically, this class can indicate whether the relationship is from operation to operation, whether it represents a single situation, or the like. By generating a tracing structure using a class, for example, a plurality of relationships that can define the class of the same user can be newly generated. Accordingly, it is possible to provide a means for the user to share through the class for elements that have not been associated with each other in the past.
The knowledge structure according to the present invention generates a new element from the elements of the already stored knowledge structure by generating a trace structure and adding a forward or reverse association and a combination of start point or end point to the trace structure. Yes.
As a result, the knowledge structure according to the present invention can be flexibly configured by an extended semantic network, and in addition to recursively generating a trace structure, the knowledge structure can be expanded and self-modified. Can be shared.

(6) Using the knowledge structure construction method according to any one of (1) to (5), in response to receiving a term for specifying the situation, an operation is performed from the set of trace structures. Extracting a knowledge structure element in which a combination of a relation class and a flag indicating a forward association exists, and selecting a random selection rule, an element extracted first or last from the extracted at least one element Selecting an element based on a rule selected, a rule for selecting an element extracted earlier or most recently, a rule for selecting an element most relevant to the discourse structure, and selecting the element. How to create an automatic response that determines the terminology used to identify

The term for specifying the situation may be a phrase that can be stored at a vertex or an edge in the extended semantic network, such as a word, a clause, or a sentence.
Alternatively, the extended semantic network according to the present invention may treat an entity of the term as an instance, define a class to be treated as an instance, and store the class at a vertex or an edge. As a result, even if the substance of a term or the like is huge data, the vertex or edge in the extended semantic network can be data representing a class definition word or phrase. Therefore, the method for creating an automatic response according to the present invention can suppress the memory consumption of the computer program for creating the response and safely maintain the operation.
Elements belonging to the set of tracing structures have a forward or reverse association as described above in (5). In determining a term for specifying an action from a situation, a forward association included in the tracing structure is used. As a result, it is not necessary to create a response by examining each element belonging to all knowledge structures one by one, and it is only necessary to create a response based on a flag indicating the association in the forward direction, and automatic response creation can be speeded up. .
Moreover, the determination of the term for specifying the action includes a rule for selecting an element most relevant to the discourse structure, and can be appropriately selected from predetermined rules. As a result, it is possible to create an automatic response whose contents are accurate, and it is possible to create an automatic response including an illogical response, and avoid the monotony of the automatic response from the viewpoint of the user.

(7) A computer program for causing a computer to execute each step of (1) to (6).

  As a result, the knowledge structure construction method and automatic response creation method according to the present invention can be implemented using a computer.

(8) An automatic response creation apparatus using the knowledge structure construction method according to any one of (1) to (6), wherein the term for identifying the situation and the term for identifying the action Knowledge that there is a combination of a flag indicating an association between a class of motion relations and a forward direction from the set of tracing structures in response to a user inputting a term for specifying the situation. A means for extracting structural elements; a rule for selecting a random selection rule from the extracted at least one element; a rule for selecting an element extracted first or last; a rule for selecting an element extracted the oldest or most recently Means for selecting a term for identifying the action based on a rule selected from a rule for selecting an element most relevant to the discourse structure, and transmits the selected term to the user And means that the creation system of an automatic response.

As a means for inputting a term by the user, a network terminal or the like can be used as appropriate, but is not limited thereto.
As a result, as an embodiment of the knowledge structure configuration method according to the present invention, an apparatus for implementing the above-described automatic response creation method in (5) can be provided.

(9) A computer-readable medium that stores the knowledge structure according to any one of (1) to (6).

  Thereby, in the knowledge structure construction method according to the present invention, the knowledge structure to which an element is added or referenced can be used as a computer-readable medium. The computer readable medium may be removable from the computer. Thereby, the knowledge structure according to the present invention can be handled as a computer-readable medium.

According to the present invention, in order to create an automatic response, the computer responds to the message from the user using the knowledge structure of the extended semantic network that can express the relationship between the situation and the action, so that the user can quickly There is an effect that a response corresponding to the message can be made.
In addition, according to the present invention, the dialog agent operates with an extended semantic network that can express not only the vertices included in the knowledge structure but also the edges to modify the edges, thereby providing excellent flexibility and extensibility. There is an effect that the knowledge structure of the dialogue agent can be implemented.
Further, according to the present invention, it is possible to handle a conversation based on a natural language by forming a discourse structure in the knowledge structure. In addition, by forming a tracing structure in the knowledge structure, there is an effect that a new element can be generated from the elements of the already stored knowledge structure.
Furthermore, according to the present invention, a class is defined for an element included in the knowledge structure, and the mutual relationship between the elements is handled by using the class, thereby facilitating the expansion of the element and the instance that is the entity of the element. Even for huge data, there is an effect that the memory consumption of a program such as response generation can be suppressed and the operation can be safely maintained.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. Note that the knowledge structure according to the present invention is created in the storage unit of the computer device. The storage unit appropriately includes electrical storage means such as a semiconductor memory that is included in computer resources and can be used as main memory or virtual storage, magnetic storage means such as a magnetic disk, and optical storage means such as a magneto-optical disk.

[Structure of extended semantic network]
FIG. 1 is a diagram illustrating the relationship of elements in a basic structure of a knowledge structure according to an embodiment of the present invention and the prior art.

ここに、ＳＮ：意味ネットワーク（Ｓｅｍａｎｔｉｃ Ｎｅｔｗｏｒｋ）
Ｖ_ＳＮ：意味ネットワークに含まれる頂点（Ｖｅｒｔｅｘ）の集合
Ｅ_ＳＮ：意味ネットワークに含まれる辺（Ｅｄｇｅ）の集合
ｖ：頂点の集合Ｖの要素
ｅ：辺の集合Ｅの要素
ｖ_ｓ：始点（３１０）等、Ｖ_ＳＮの要素
ｖ_ｄ：終点（３１１）等、Ｖ_ＳＮの要素
である。
数式５は、意味ネットワークが頂点の集合Ｖ_ＳＮ及び辺の集合Ｅ_ＳＮで構成されることを表す。
数式６は、頂点の集合Ｖ_ＳＮ要素ｖを有することを表す。
数式７は、辺の集合Ｅ_ＳＮは辺の要素ｅを有し、辺の要素ｅは始点ｖ_ｓ及び終点ｖ_ｄで定義され、始点ｖ_ｓ及び終点ｖ_ｄは頂点の集合Ｖ_ＳＮに属することを表す。 First, a known semantic network format used for a model for representing a meaning or a memory structure will be described.
FIG. 1B is a diagram illustrating a graph structure representing a semantic network. The side e (312) is defined by the start point v _s (310) and the end point v _d (311), and is represented by an arrow from the start point to the end point. The semantic network has a format represented by the following equations 5 to 7.

Here, SN: Semantic Network
V _SN : a set of vertices (Vertex) included in the semantic network
E _SN : set of edges included in the semantic network
v: Element of vertex set V
e: Element of edge set E
v _s : element of V _SN such as start point (310)
v _d : an element of V _SN such as end point (311).
Equation 5 represents that the semantic network is composed of a vertex set V _SN and an edge set E _SN .
Equation 6 represents having a vertex set V _SN element v.
Equation 7, the set _{E SN} sides has a side of the element e, the element e of the side defined by the start _{v s} and the end point _{v d,} belong be the starting point _{v s} and the end point _{v d} is set _{V SN} of vertices Represents.

ここに、ＡＳＮ：拡張意味ネットワーク（Augmanted Semantic Network）
Ｅ_ＡＳＮ：拡張意味ネットワークの集合
ｅ：集合Ｅの要素である二項組要素
ｅ_ｓ：始点（３０１）等、Ｅ_ＡＳＮの要素
ｅ_ｄ：終点（３０２）等、Ｅ_ＡＳＮの要素
null：空要素
である。
数式８は、二項組を構成するｅ_ｓ、ｅ_ｄは要素集合Ｅ_ＡＳＮの要素か、あるいは空要素ｎｕｌｌであることを表す。さらに、拡張意味ネットワークは、次の要素集合を有する。

ここに、ＡＳＮ_ＶＥ：拡張意味ネットワークＡＳＮの構成要素
Ｅ_ｖ：拡張意味ネットワークにおける頂点の集合
Ｅ_ｅ：拡張意味ネットワークにおける辺の集合
である。
数式９は、本発明に係る拡張意味ネットワークが頂点の集合Ｅ_ｖ及び辺の集合Ｅ_ｅを構成要素とすることを表す。
数式１０は、頂点の集合Ｅ_ｖは二項組要素ｅ_ｓ、ｅ_ｄが両方とも空要素であるときの要素ｅ（ｅ_ｓ，ｅ_ｄ）を含むことを表す。
数式１１は、辺の集合Ｅ_ｅは二項組要素ｅ_ｓ、ｅ_ｄが辺の集合Ｅ_ｖに属するときの要素ｅ（ｅ_ｓ，ｅ_ｄ）を含むことを表す。
数式１０及び数式１１は、共通して要素ｅ（ｅ_ｓ，ｅ_ｄ）を用いて表現される。すなわち、本発明に係る拡張意味ネットワークにおいては、要素に含まれる辺の終点が終点ｅ_ｄだけではなく辺ｅ（ｅ_ｓ，ｅ_ｄ）でもありうることにおいて、公知の意味ネットワークとの相違点を有する。従って、本発明に係る拡張意味ネットワークにおいては、図１（ａ）に示したように、始点ｅ_ｃ（３０４）から始まる辺ｅ_ｉ（３０５）の終点を辺ｅ_ｅ（３０３）としうる。 Next, the format of the extended semantic network according to the present invention will be described.
FIG. 1A is a diagram illustrating a graph representing an extended semantic network. The side e _e (303) is defined by the start point e _s (301) and the end point e _d (302), and is represented by an arrow from the start point to the end point. The side e _i (305) has a start point e _c (304) and the end point is a side e _e (303). The extended semantic network according to the present invention has the following format.

Here, ASN: Augmented Semantic Network
E _ASN : Set of extended semantic networks
e: Binary element that is an element of set E
e _s: starting point (301), and the _like, the elements of _{E ASN
ed} : E _ASN element such as end point (302)
null: An empty element.
Equation 8 indicates that the _e s, _{e d} constituting a two-tuple or elements of the element set _{E ASN,} or an empty element null. Further, the extended semantic network has the following element set:

Where ASN _VE : component of the extended semantic network ASN
E _v : vertex set in the extended semantic network
E _e : A set of edges in the extended semantic network.
Expression 9 represents that the extended semantic network according to the present invention includes the vertex set E _v and the edge set E _e as components.
Expression 10 indicates that the vertex set E _v includes an element e ( _es , _ed ) when the binomial elements _es , _ed are both empty elements.
Equation 11 is a set _{E e} of the edge indicates that contain elements _{e (e} s, _{e d)} when the binary sets elements _e s, _{e d} belongs to the set _{E v} sides.
Equations 10 and 11 are commonly expressed using elements e ( _es , _ed ). That is, in the extended semantic network according to the present invention, the end point of the side included in the element can be not only the end point _ed but also the side e ( _es , _ed ). Have. Therefore, in the extended semantic network according to the present invention, as shown in FIG. 1A, the end point of the side e _i (305) starting from the start point e _c (304) can be set as the side e _e (303).

In the extended semantic network and the known semantic network according to the present invention, it is common that vertices are used to store concepts and events, and edges connecting the vertices are used to store the meaning of their relationship. ing. In addition, the specific meanings of vertices and edges are common in giving from the outside.
In the extended semantic network according to the present invention, it is possible to generate a degree of freedom that a side can be a connection target of the side.

[Knowledge storage example]
FIG. 2 is a diagram illustrating an example of knowledge stored using an extended semantic network according to an embodiment of the present invention and the related art.
For example, the user means an English word “car” meaning “car”, another word “is a” constituting one sentence in natural language, and “vehicle” which is a superordinate concept of the “car”. The case where the English word “vehicle” is given is shown.
FIG. 2A is a diagram illustrating an example of knowledge stored using the extended semantic network according to the embodiment of the present invention. The English word “car” is stored in the element 321, the word “is a” is stored in the element 323, and the English word “vehicle” is stored in the element 322. Element 321 and element 322 may be the start and end points of side 325, respectively. Element 323 and side 325 may be the start and end points of side 324, respectively.
FIG. 2B is a diagram illustrating an example of knowledge stored using the semantic network according to the related art. The English word “car” is stored at the start point 331, the word “is a” is stored at the side 333, and the English word “vehicle” is stored at the end point 332.

  In the extended semantic network according to the present invention, unlike the semantic network according to the prior art, a word that modifies an edge can be treated as an element. That is, an element that modifies a side can be treated as an object that modifies another element or is modified from another element in the same manner as an element that can be a start point or an end point. As a result, it is possible to provide a knowledge structure that is more flexible and more extensible than the prior art.

[Configuration of automatic response output creation system]
FIG. 3 is a diagram showing a configuration of an automatic response output creation system according to an embodiment of the present invention. The automatic response output creation system 20 appropriately includes an automatic response output creation device 10, a knowledge database (knowledge DB) 30, a user terminal 60, a network 38, and the like. The network 38 may be an intranet or the Internet. The user terminal 60 includes a terminal that can be connected to a network, such as a personal computer or a mobile phone terminal. The knowledge DB 30 includes an arbitrary word / phrase included in the word / phrase transmitted from the user, knowledge / phrase used by the automatic response generation apparatus 10 for a response, an arbitrary word / phrase included in the generated response, and a generation grammar in the field of natural language. Derived information such as the frequency of rules and phrases is stored as appropriate.
Although FIG. 3 shows a plurality of devices connected to each other via the network 38, the present invention is not limited to this, and the functions of the automatic response creation device 10, the knowledge DB 30, and the user interface (not shown) are provided in one device. May be.
By using the automatic response output generation system according to an embodiment of the present invention, a system that automatically generates a response to a user input can be provided.

[Knowledge DB data structure]
FIG. 4 is a diagram illustrating a data structure of a knowledge database (knowledge DB) according to an embodiment of the present invention.
The knowledge DB data structure 80 includes a start point 84, an end point 85, a label 86, a pointer 88, and the like that are associated with one of the data individually identified by the identifier 82. Not limited to these, information included in and associated with the knowledge DB data structure 80 can be designed as appropriate.
The identifier 82 may be a number or a character as long as it can identify stored data, and can be designed as appropriate.
The start point 84 and the end point 85 indicate a modification relationship between the data stored in the knowledge DB data structure 80 and other data. One start point 84 and one end point 85 may be defined for data having one identifier, a plurality may be defined, or a blank space may be defined. Since both the start point 84 and the end point 85 are blank, this indicates that the data is a vertex in the extended semantic network, and is included in the data that can be handled in the extended semantic network.
The label 86 is an area that is reserved in advance in the extended semantic network for one piece of data individually identified by the identifier 82 or is defined by the user. More specifically, the reservation is for defining a class for an element handled in the extended semantic network and predefining a relationship between elements generated in the operation of the extended semantic network such as an element being added. It is.
For example, the following can be used to define a reserved class:
(Class 1) v _root : starting point of all knowledge elements (class 2) v _usr : user class (class 3) v _act : action class (class 4) v _ins : instantiation-related class v _root is all It is the starting point of the knowledge element and can be the starting point for defining any other element. v _usr is a class for organizing operation classes described below according to purpose and user. v _act is a class for including, in a knowledge element, operations such as a dialogue response performed by the automatic dialogue response generation system according to the embodiment of the present invention. v _ins is a class for indicating an entity such as a set of data arranged on a memory in the automatic dialog response generation system and the like according to the embodiment of the present invention. For example, a vertex u _br that organizes basic response operations is defined by being connected from v _root by a branch u _i that is an instance of the class v _usr . The basic operations such as keyword, relation, message, and the like can be defined by being connected from the vertex u _br by branches that are instances of the class v _act . The apex label _br and the apex label 86 representing the basic operation may include explanatory characters that are useful when the knowledge DB 30 is managed by a person.
The pointer 88 is data having one identifier, and is used for storing and handling data including information other than the above-described start point 84, end point 85, or label 86 in the knowledge DB 30. . For example, a character string processing function known to those skilled in the art can be imported into the knowledge DB 30 by associating the function with the pointer 88.

  The automatic dialog generation system or the like according to the embodiment of the present invention includes the knowledge DB data structure 80, so that data that can be handled in the extended semantic network and general character string processing that is not particularly related to the extended semantic network. Any function or the like can be taken into the knowledge DB 30 and used.

ここで、Ｅ_ｃｇ（ｅ_ｖ）：要素ｅ_ｖからの連結グラフ
である。
本発明に係る知識構造の要素は、ｖ_ｒｏｏｔを始点とする連結グラフに含まれる。
数式１２の右辺に｛｝を用いて囲まれた４個の集合は、それぞれ、要素ｅ_ｖの集合、始点ｅ’及び終点ｅ’’を有する辺ｅ_ｅ（ｅ，ｅ’’）の集合、辺ｅ_ｅ（ｅ’，ｅ）の終点である要素ｅの集合、辺ｅ_ｅ（ｅ，ｅ’）の始点である要素ｅの集合を表し、これらを連結グラフとして取り扱う。
図５に、これらの連結グラフの例を示す。連結グラフ３４０は、要素ｅ_６、ｅ_７、ｅ_１及び、要素ｅ_６から要素ｅ_７への辺、要素ｅ_７から要素ｅ_８への辺を含む。
連結グラフ３５０についても同様である。数式１２の右辺第３項に含まれる終点である要素ｅは、数式１１を用いて前述のように、辺により修飾される辺であってもよい。例えば、連結グラフ３５０に、要素ｅ_５を始点とし、要素ｅ_９から要素ｅ_１０への辺を終点とする辺を含むことができる。
本発明の一実施形態に係る知識構造の構成方法において、拡張意味ネットワークの要素を含む連結グラフを用いることにより、辺を修飾する辺等を知識構造の要素に含めることが可能になる。 [Knowledge structure elements]
FIG. 5 is a diagram illustrating a connected graph including elements of an extended semantic network according to an embodiment of the present invention.
Elements of the extended semantic network are used to construct knowledge structures. Each element has a relationship with _vroot which is the starting point of all the knowledge elements described above. This relationship can be expressed using, for example, a connected graph defined as the following equation.

_{Here, E} cg _(e v): is a connected graph from element _{e v.}
Elements of knowledge structure according to the present invention, v is included in the connection graph starting from the _root.
Four sets surrounded by {} on the right side of Equation 12 are a set of elements _ev, a set of sides e _e (e, e ″) having a start point e ′ and an end point e ″, A set of elements e that is the end point of the edge e _e (e ′, e) and a set of elements e that are the start point of the edge e _e (e, e ′) are represented, and these are handled as a connected graph.
FIG. 5 shows examples of these connection graphs. The connection graph 340 includes elements e ₆ , e ₇ , e ₁ , an edge from the element e ₆ to the element e _7, and an edge from the element e ₇ to the element e ₈ .
The same applies to the connection graph 350. The element e which is the end point included in the third term on the right side of Expression 12 may be an edge that is modified by an edge as described above using Expression 11. For example, the connected graph 350 can include an edge starting from the element e ₅ and ending at the edge from the element e ₉ to the element e ₁₀ .
In the method for constructing a knowledge structure according to an embodiment of the present invention, by using a connected graph including elements of an extended semantic network, it is possible to include edges that modify edges in the elements of the knowledge structure.

The user class v _usr takes the user relationship of the knowledge structure as an instance. The user relationship always starts from the _{root vroot} .
The user relationship includes a reaching point u as a vertex. The knowledge structure is divided by this reaching point u. _That, u _1, u 2, ..., when a plurality of arrival points represented by _{u n, etc., v} knowledge structure starting from the _root, each goal _{u 1,} u 2, ..., _{u n} In the associating process, the number of arrival points u is divided. In other words, the vertex u that can divide the knowledge structure is the user.

The action class v _act takes an action relation of the knowledge structure as an instance. The operational relationship always starts with user u.
The motion relationship includes a reaching point a as a vertex. This reaching point a means the operation of the dialogue agent or the like. In other words, the arrival point a starting from the user u is an action.
The operation of the dialogue agent can be implemented by a programming language. An action may be associated with a pointer, such as a function coded in a programming language. The action is a class for the actual data. When the data instance is accessed, it acts as an interaction agent. The types of operations to be performed include utterances, system resource operations, knowledge structure changes, and the like.

The instantiation relation class v _ins takes the instance relation of the knowledge structure as an instance. The instance relation always starts from the vertex c.
The arrival point of the instance relationship is an arbitrary element i, which means that it belongs to the class that the vertex c means. Element i is called an instance of vertex c, and vertex c is called a class of element i. An instance may belong to multiple classes. Almost every element is an instance of at least one class.

[Configuration of automatic response output creation device]
FIG. 6 is a diagram showing a hardware configuration of an automatic response output creation device according to an embodiment of the present invention. Hereinafter, the automatic response output creation apparatus will be described as a server. However, the same applies to the case where a personal computer is used as the automatic response output creation apparatus.

  The automatic response output generation apparatus 10 includes a CPU 110 (a plurality of CPUs such as the CPU 120 may be added in a multiprocessor configuration), a bus line 107, a communication I / F 140, a main memory 150, a BIOS (Basic Input). An output system 160, a USB port 190, an I / O controller 170, an input unit such as a keyboard and mouse 180, and a display device 122. The I / O controller 170 can be connected to a storage unit 109 such as a tape drive 172, a hard disk 174, an optical disk drive 176, and a semiconductor memory 178. The BIOS 160 stores a boot program executed by the CPU 110 when the server is started, a program depending on the server hardware, and the like. The hard disk 174 stores various programs for functioning as a server and programs for executing the functions of the present invention. As the optical disk drive 176, for example, a DVD-ROM drive, a CD-ROM drive, a DVD-RAM drive, a CD-RAM drive, or the like can be used. In this case, the optical disk 177 corresponding to each drive is used. A program or data can be read from the optical disk 177 by the optical disk drive 176 and provided to the main memory 150 or the hard disk 174 via the I / O controller 170. Similarly, the tape medium 171 corresponding to the tape drive 172 can be used mainly for backup.

  The program provided to the server is provided by being stored in a recording medium such as the hard disk 174, the optical disk 177, or a memory card. This program may be installed in the server and executed by being read from the recording medium via the I / O controller 170 or downloaded via the communication I / F 140.

  The above-described program may be stored in an internal or external storage medium. Here, examples of the storage medium include, but are not limited to, a magnetic recording medium, a magneto-optical recording medium, an optical recording medium, and an electronic medium. The magnetic recording medium includes a magnetic tape, a floppy (registered trademark) disk, a hard disk 174, a magnetic random access memory (MRAM), etc., and the magneto-optical recording medium includes a magneto-optical disk (MO and MD), etc. Recording media include write-once compact discs (CDs), write-once digital multipurpose discs (DVDs), write-once Blu-ray discs, write-once holographic discs, etc., and electronic media are electrically writable read-only memory, random access memory , Flash memory, silicon storage disk, etc. In addition, a storage device such as a hard disk 174 or an optical disk library provided in a server system connected to a communication line such as a dedicated communication line or the Internet may be used as a recording medium, and the program may be provided to the server via the communication line. .

  Here, the display device 122 displays a screen for accepting data input by the server administrator, or displays a screen of calculation processing results. The display device 122 is a cathode ray tube display device (CRT) or a liquid crystal display device (LCD). Display devices. The input means accepts input by the user, and may be configured with a keyboard and mouse 180 or the like. The communication I / F 140 is a network adapter for enabling the server to be connected to a terminal via a dedicated network or a public network. The communication I / F 140 may include a modem, a cable modem, and an Ethernet (registered trademark) adapter.

  The means for storing the knowledge structure for the automatic response output creating apparatus 10 according to the embodiment of the present invention may be the main memory 150 described above, and a storage medium such as the hard disk 174 or a memory disk is used as a part of the main memory. A virtual memory may be used, and any details can be used as long as it can be used as a memory. Alternatively, the knowledge structure may be stored in a built-in storage medium or a network-accessible storage medium. The hardware resource for storing the knowledge structure may be any means capable of storing computer-processable character information.

  In the above example, the hardware configuration of the automatic response output creating apparatus 10 has been mainly described. However, the functions described above are performed by installing a program in a computer and operating the computer as the automatic response output creating apparatus 10. Can be realized. Therefore, the functions realized by the automatic response output creation apparatus 10 described as one embodiment in the present invention are executed by executing the above-described method by the computer or by introducing the above-described program into the computer. Can also be realized.

[Operation of knowledge structure and discourse structure]
FIG. 7 is a diagram illustrating an example of operation of a knowledge structure according to an embodiment of the present invention.
The operation of all knowledge structures can be described using an extended semantic network starting from the start point _Vroot (360) of all knowledge elements described above.
The operation that is an element of the knowledge structure according to the present invention starts from the user. That is, in FIG. 7, the user u ₁ (361) has an association with a side starting from V _root (360), and the actions a ₁ (362), a ₂ (363)... _An (364) On the other hand, it can be the starting point. Further, operations a ₁ (362), a ₂ (363)... A _n (364) that can be associated with user u ₁ (361) are function pointers f ₁ (365), f ₂ (366) associated with the respective operations. ) ... f _n (367) is included as appropriate.
Specifically, in response to the user starting input such as the start of utterance, the implementation means of the knowledge structure construction method according to the present invention can start operation. For example, exemplary means on how to configure the knowledge structure of the present invention, in response to receiving the text information that the user sends "Hello" from the operation which has an association with the situation, with a predetermined rule Based on the selected action, a response such as “How are you?”, Which is the context of the conversation included in the selected action, may be generated and sent to the user. Thereby, the operation | movement which produces | generates a response with respect to the message from a user can be produced as a procedure for implementing using a computer.
Here, the predetermined rule includes a random selection rule, a rule for selecting an element extracted first or last, a rule for selecting an element extracted earlier or most recently, an element most relevant to the discourse structure Contains rules for selecting. As a result, it is possible to create an automatic response whose contents are accurate, and it is possible to create an automatic response including an illogical response, and avoid the monotony of the automatic response from the viewpoint of the user.

In the operation of the above knowledge structure, even if the operation a ₁ (362), a ₂ (363)... A _n (364) includes a discourse structure that is a set including a conversation place, a conversation history, and a conversation context as elements. Good.
The discourse structure includes, in particular, a set having, as elements, a conversation context expressed using a computer-processable character code.
User _u 1 (361) may utters context of the conversation, which can represent a discourse structure at each operation _a 1 ... _{a n.}
Specifically, a chat room service where a plurality of users connected via a network terminal such as a personal computer or a mobile phone can exchange character information or voice information with each other in real time. Including services such as network conferencing systems. The place of conversation can be stored in the knowledge structure as an element.
In each operation a ₁ ... _An , a fully ordered set whose conversation is an element may be stored in the knowledge structure. The total ordered set includes, for example, a set having elements as means for regularly storing the temporal order of conversations in association with the magnitude of each number of conversations. Further, the occurrence of conversation and the sequential adjacency relationship of the occurrence of conversation may be stored as elements of the knowledge structure.
Thus, when the occurrence of a conversation is detected, the knowledge structure configuration method according to the present invention identifies the occurrence of each conversation using means for regularly storing the temporal order, and creates the knowledge structure according to the present invention. In addition, the conversation context expressed using the character code can be taken into the knowledge structure.

ここに、ｅ_ｕｒ１：ユーザ関係を表す辺
ｉ_１：ｅｕｒ１がユーザ関係であることを表す辺
ｉ_２：ｉ_１がインスタンス関係であることを表す辺
である。ユーザ関係を示す辺ｅ_ｕｒ１はルートｖ_ｒｏｏｔから頂点ｕ_ｋｒを結ぶ。辺ｅ_ｕｒ１はユーザクラスｖ_ｕｓｒのインスタンスである。辺ｉ_１は辺ｅ_ｕｒ１がユーザ関係であることを示し、辺ｉ_２は辺ｉ_１がインスタンス関係であることを示す。辺ｉ_２もインスタンス関係を意味するが、これを説明する辺は無限ネストを避けるために省略される。
次いで、動作キーワード、メッセージ、関係をそれぞれ表す要素ａ_ｋｗ、ａ_ｍｓｇ、ａ_ｒｅｌを格納する。要素ａ_ｋｗは応答ルールのキーワード動作のクラスを、ａ_ｍｓｇは応答メッセージ動作のクラスを、ａ_ｒｅｌはキーワードから応答メッセージ間への対応関係動作のクラスを、それぞれ意味する。本発明に係る知識構造においては、これらの要素は次式で表される。

ここに、ｅ_ａｒ１、ｅ_ａｒ２、ｅ_ａｒ３は動作関係を示す辺である。
動作関係を示す辺ｅ_ａｒ１、ｅ_ａｒ２、ｅ_ａｒ３はユーザｕ_ｋｒからそれぞれ頂点ａ_ｋｗ、ａ_ｍｓｇ、ａ_ｒｅｌを結ぶ。辺ｅ_ａｒ１、ｅ_ａｒ２、ｅ_ａｒ３は動作クラスｖ_ａｃｔのインスタンスである。辺ｉ_３、ｉ_５、ｉ_７はそれぞれ辺ｅ_ａｒ１、ｅ_ａｒ２、ｅ_ａｒ３が動作関係であることを示し、辺ｉ_４、ｉ_６、ｉ_８はそれぞれ辺ｉ_３、ｉ_５、ｉ_７がインスタンス関係であることを示す。辺ｉ_４、ｉ_６、ｉ_８もインスタンス関係を意味するが、これを説明する辺は無限ネストを避けるために省略される。
これらのルールを格納する準備の後に、ルールデータを格納する。
一例として、「カレー」というキーワードに対して、「おいしいですね！」という応答を行うルールを示す。キーワードインスタンスを意味する要素をｋ_１、応答メッセージインスタンスを意味する要素をｍ_１とする場合に、「カレー」というキーワードで「おいしいですね！」という応答を行うルールは、本発明に係る知識構造において、次式で表される。

データ要素ｋ_１、ｍ_１、ｒ_１はそれぞれ動作ａ_ｋｗ、ａ_ｍｓｇ、ａ_ｒｅｌのインスタンスである。辺ｉ_１０、ｉ_１２、ｉ_１４はそれぞれ辺ｉ_９、ｉ_１１、ｉ_１３がインスタンス関係であることを示す。辺ｉ_１０、ｉ_１２、ｉ_１４もインスタンス関係を意味するが、これを説明する辺は無限ネストを避けるために省略される。 [Example of interactive response]
A keyword response is illustrated as an automatic response creation method according to an embodiment of the present invention. The keyword response is a function for executing a response associated with a keyword when a predefined keyword appears during the user's utterance. In order to realize this function, a rule that associates a keyword with a response is stored in a knowledge structure as shown below.
First, as a preparation for storing a rule, an element u _kr representing a response to a user-defined keyword is stored. The element u _kr is the starting point in the keyword response rule in the knowledge structure. In the knowledge structure according to the present invention, the starting point of the keyword response rule including the element u _kr is expressed by the following equation.

Where e _ur1 : side representing user relationship
i ₁ : edge indicating that eur ₁ is a user relationship
i ₂ : An edge indicating that i ₁ is an instance relationship. The edge e _ur1 indicating the user relationship connects the vertex u _kr from the route v _root . The edge e _ur1 is an instance of the user class v _usr . The side i ₁ indicates that the side e _ur1 has a user relationship, and the side i ₂ indicates that the side i ₁ has an instance relationship. Side i ₂ also means an instance relationship, but the side describing this is omitted to avoid infinite nesting.
Next, elements a _kw , a _msg , and a _rel representing the operation keyword, message, and relationship are stored. The element a _kw means the class of the response rule keyword action, a _msg means the response message action class, and a _rel means the correspondence action class from the keyword to the response message. In the knowledge structure according to the present invention, these elements are represented by the following equations.

Here, e _ar1 , e _ar2 , and e _ar3 are sides indicating operational relationships.
The sides e _ar1 , e _ar2 , and e _ar3 indicating the operation relationship _connect vertices a _kw , a _msg , and a _rel from the user u _kr , respectively. The sides e _ar1 , e _ar2 , and e _ar3 are instances of the action class v _act . _Sides i ₃ , i ₅ , and i ₇ indicate that the sides e _ar1 , e _ar2 , and e _ar3 are in an operational relationship, respectively, and sides i ₄ , i ₆ , and i ₈ are sides i ₃ , i ₅ , and i ₇ , respectively. Indicates an instance relationship. The sides i ₄ , i ₆ , and i ₈ also mean instance relationships, but the sides that explain this are omitted to avoid infinite nesting.
After preparing to store these rules, the rule data is stored.
As an example, a rule for making a response “delicious!” To the keyword “curry” is shown. When the element meaning a keyword instance is k ₁ and the element meaning a response message instance is m ₁ , the rule for making a response “delicious!” With the keyword “curry” is the knowledge structure according to the present invention. Is expressed by the following equation.

Data elements k ₁ , m ₁ , r ₁ are instances of operations a _kw , a _msg , a _rel , respectively. Sides i ₁₀ , i ₁₂ , and i ₁₄ indicate that the sides i ₉ , i ₁₁ , and i ₁₃ are in an instance relationship, respectively. The sides i ₁₀ , i ₁₂ , and i ₁₄ also mean instance relationships, but the sides that explain this are omitted to avoid infinite nesting.

By using the data stored in the knowledge structure in this way, the implementation means of the automatic response generation method according to the present invention generates and transmits responses such as the following (R1) to (R9), for example. Action can be taken.
(R1) The user's speech “Today is hot but curry is indispensable” is received.
(R2) The user relationship is traced from the route v _root and the user u _kr is specified.
(R3) Follow the operation relationship and specify the keyword operation class a _kw .
(R4) The instance data elements of class a _kw are checked in order, and it is checked whether or not the character string associated with the element appears in the user utterance.
(R5) It is detected that the character string “curry” associated with the element k ₁ has appeared.
(R6) from the element _{k 1} following the instance sides _{r 1} relationship operation class _{a rel} reached elements _{m 1,} it attempts to execute.
(R7) It is detected that the class of the element m ₁ is a _msg .
(R8) The action class a _msg is associated with a function pointer that outputs a character string associated with the instance data element as a response message. Invoke this function.
(R9) The character string “is delicious!” Linked to the element m ₁ is transmitted as a response.
The function class function pointer may be undefined, and may not provide instance data of an action class that can be a specific action target, and may provide attribute information or the like that is referred to by another action. A plurality of _{a rel} instance may be connected to the element _{k 1.} If there are multiple elements that can be the target of a specific action at the connection destination, one of them may be selected at random, and the rules for selecting a specific action from the multiple elements can be designed as appropriate .

  By constructing the dialogue structure using the knowledge structure construction method according to the present invention, an automatic dialogue such as a keyword response can be generated using a procedure for tracing the association of elements. That is, in the present invention, it is not necessary to examine every element included in the knowledge structure one by one, and a quick response generation can be realized.

[Example of dialogue script tracking and tracing structure]
FIG. 8 is used to show an example in which the interactive script tracking and tracing structure according to an embodiment of the present invention is represented as a graph structure.
Here, the tracing structure is a set including relationships between elements that can be illustrated using the knowledge structure of the extended semantic network according to the present invention. Specifically, the tracing structure can be a set including a relation class, a direction, and a destination as elements.
FIG. 8 is a diagram showing data for script tracking in the knowledge structure construction method according to an embodiment of the present invention.
Both the element u _dst (390), which is the starting point in the dialogue script in the knowledge structure, and the element u _ca (392), which is the starting point when accessing the script pointer in the knowledge structure in the dialogue channel, are the user. _Is associated with the starting point _vroot (380) of all knowledge elements by class v _usr (381).
Here, the user class v _usr (381) is selected (a _sel ) (410), the selection message (a _sm ) (412), and the selection rule (a _sr ) (a) by the operation belonging to the operation class v _act (382). 414). Similarly, interaction channel access is associated with context (a _c ) (420), final selection (a _ls ) (422), final response (a _lr ) (424), and further by the instance class (384): Interaction channel access is associated with the elements (430, 432, 434) associated with the interaction.
As described above, the dialogue script according to the embodiment of the present invention can be represented as a graph structure according to the present invention. That is, the dialogue script according to an embodiment of the present invention can be tracked using the graph structure according to the present invention.

各要素の詳細説明はキーワード応答と同様であり省略する。
数式１６は、図８を用いて説明したグラフ構造の、ｖ_ｒｏｏｔ（３８０）、ｖ_ｕｓｒ（３８１）、ｕ_ｄｓｔ（３９０）、及びｕ_ｃａ（３９２）を含む。さらに、数式１６はインスタンスのクラスｖ_ｉｎｓ（３８４）が式中の関連付けに含まれうることも表す。
次いで、スクリプトの動作の集合に次の要素が含まれると仮定する。すなわち要素として、選択（ａ_ｓｅｌ）、選択メッセージ（ａ_ｓｍ）、選択ルール（ａ_ｓｒ）のそれぞれを、ユーザｕ_ｄｓｔに関連付けて格納する。
さらに、スクリプトの動作の集合に次の要素が含まれると仮定する。すなわち要素として、コンテキスト（ａ_ｃ）、最終選択（ａ_ｌｓ）、最終応答（ａ_ｌｒ）のそれぞれの初期値を、ユーザｕ_ｃａに関連付けて格納する。
要素ａ_ｓｅｌはスクリプト内選択肢動作のクラスを、ａ_ｓｍは選択肢メッセージ動作のクラスを、ａ_ｓｒはスクリプト構造の開始点を、ａ_ｃは文脈ポインタ動作のクラスを、ａ_ｌｓは最終選択肢ポインタ動作のクラスを、ａ_ｌｒは最終応答ポインタ動作のクラスを、それぞれ表す。
これらのスクリプトの動作に含まれる要素は、次式で表される。

次いで、ポインタのインスタンスを表す要素を格納する。例えば、文脈ポインタ動作のインスタンス（ｐｉ_ｃ）、最終選択肢ポインタ動作のインスタンス（ｐｉ_ｌｓ）、及び最終応答ポインタ動作のインスタンス（ｐｉ_ｌｒ）を用いる。これらのインスタンスと他の要素との関係は次式で表される。

数式１７及び数式１８は、図８にグラフ構造として示した要素と要素の関連付けを含む。これにより、スクリプトに含まれるユーザのクラス、動作のクラス、インスタンスのクラス、ポインタのインスタンスを表す要素を、本発明に係る知識構造の構成方法に格納し、スクリプトの動作を追跡しうる。 The dialogue script tracking function is a function for performing dialogue according to a predefined script. Scripts may be branched. This function maintains a pointer, which remembers where in the script the dialog has progressed.
First, u _dst that is a user element representing script tracking and u _ca that represents a channel number on which the script is operated are stored. The element u _dst is the starting point in the dialogue script in the knowledge structure. The element u _ca is a starting point when accessing a script pointer or the like in the knowledge structure in the dialogue channel. In the knowledge structure according to the present invention, the start point of the script including the element u _dst is expressed by the following equation.

Detailed description of each element is the same as the keyword response, and is omitted.
Equation 16 includes v _root (380), v _usr (381), u _dst (390), and u _ca (392) of the graph structure described with reference to FIG. Furthermore, Equation 16 also indicates that the instance class v _ins (384) can be included in the association in the equation.
Then, assume that the set of script actions includes the following elements: That is, each of selection (a _sel ), selection message (a _sm ), and selection rule (a _sr ) is stored as an element in association with user u _dst .
Assume further that the set of script actions includes the following elements: That is, the initial values of the context (a _c ), final selection (a _ls ), and final response (a _lr ) are stored in association with the user u _ca as elements.
The element a _sel is the class of the choice action in the script, a _sm is the class of the choice message action, a _sr is the starting point of the script structure, a _c is the class of the context pointer action, and a _ls is the final choice pointer action A _lr represents the class of the final response pointer operation.
Elements included in the operations of these scripts are represented by the following expressions.

Then, an element representing an instance of the pointer is stored. For example, an instance of context pointer action (pi _c ), an instance of final choice pointer action (pi _ls ), and an instance of final response pointer action (pi _lr ) are used. The relationship between these instances and other elements is expressed by the following equation.

Equations 17 and 18 include the association of elements shown in FIG. 8 as a graph structure. Thus, the elements representing the user class, the action class, the instance class, and the pointer instance included in the script can be stored in the knowledge structure construction method according to the present invention, and the action of the script can be traced.

Expression 17 represents elements of the knowledge structure according to the present invention, and includes a tracing structure that is a set including a relation class, a direction, and a destination. For example, in Equation 17, since the side element i ₂₀ includes the side element i ₁₉ at the end point and the side element i ₁₉ includes the element _ear4 at the end point, the relation class belongs to these elements, and the direction is is reversed, the destinations from i ₂₀ may constitute the follow structure etc. is i _19. That is, the tracing structure can generate a new element based on the already stored knowledge structure element and include it in the tracing structure as a set.
Further, the tracing structure includes a relation class in the element, and by generating the tracing structure using the class, for example, a plurality of relations that can define a class of the same user can be newly generated. Accordingly, it is possible to provide a means for the user to share through the class for elements that have not been associated with each other in the past.
The knowledge structure according to the present invention generates a new element from the elements of the already stored knowledge structure by generating a trace structure and adding a forward or reverse association and a combination of start point or end point to the trace structure. Yes.
As a result, the knowledge structure according to the present invention can be flexibly configured by an extended semantic network, and in addition to recursively generating a trace structure, the knowledge structure can be expanded and self-modified. Can be shared.
Further, by using the trace structure, it is possible to efficiently acquire the destination of the graph without restricting the graph update operation. When the trace structure is not used, calculation time proportional to the scale of the graph is required. However, when the trace structure is used, a certain calculation time is sufficient.

まず、本発明に係る知識構造に、選択キーワードインスタンスを意味する要素ｓｒ_ｋ（ｋ＝１、２、…、ｎ、以下同様）、選択肢インスタンスを意味する要素ｓ_ｋ、応答メッセージインスタンスを意味する要素ｍ_ｋ、スクリプト順序関係インスタンスを意味する要素ｒ_ｋ、選択肢メッセージインスタンスを意味する要素ｓｍ_ｋ、を格納する。
表１に示すように、要素ｓｒ_ｋは文字列データと結び付いているとする。ここで、要素ｍ_４の「×××」には具体的なレストランの名称等が入る。簡便のため、選択ルールａ_ｓｒのインスタンスｓｒ_１、…、ｓｒ_８、及び、選択ａ_ｓｅｌのインスタンスｓ_１、ｓ_２、ｓ_３、メッセージａ_ｍｓｇのインスタンスｍ_１、…、ｍ_９は省略する。
次に辺要素の具体例を、インスタンスに関する辺を省略して、次式で示す。

図９に、数式１９、数式２０、数式２１を用いて示した、レストラン名を提案するスクリプトにおける各辺要素の関係をグラフ構造として示す。
選択ルールａ_ｓｌ（４４０）は選択ルールのインスタンスｓｒ_１（４４２）と関連付けられ、選択ｓ_１により次の選択ルールのインスタンスｓｒ_２（４４６）及びメッセージｍ_２（４６０）と関連付けられる。これ以降についても同様であり、本発明に係る知識構造の構成方法に、スクリプトに含まれる選択ルール、選択、メッセージを格納することができる。 <Example 1>
As an example, for a keyword “curry”, a script for asking a curry type and a restaurant area and proposing a restaurant name is shown.
FIG. 9 shows an example of the relationship between the elements of the knowledge structure in the interactive script tracking according to the present invention, and each element included in FIG. 9 will be described first using a table.
Table 1 shows an example of correspondence with the character string of the interactive script tracking example data.

First, in the knowledge structure according to the present invention, an element sr _k (k = 1, 2,..., N, the same shall apply hereinafter) meaning a selected keyword instance, an element s _k meaning an option instance, and an element meaning a response message instance m _k , an element r _k indicating a script order relation instance, and an element sm _k indicating a choice message instance are stored.
As shown in Table 1, and the elements sr _k is associated with the character string data. Here, specific restaurants such as the name is entered in the "×××" of the element m _4. For convenience, the instance _sr 1 selection rule _{a sr,} ..., sr _8, and select _a instance _s 1 of _sel, s 2, _{s 3,} instance _m 1 message _{a msg,} ..., _{m 9} will be omitted.
Next, a specific example of a side element is represented by the following formula, omitting sides related to instances.

FIG. 9 shows, as a graph structure, the relationship between each side element in the script for proposing a restaurant name, which is shown using Equations 19, 20, and 21.
Selection rule a _sl (440) is associated with selection rule instance sr ₁ (442), and selection s ₁ is associated with the next selection rule instance sr ₂ (446) and message m ₂ (460). The same applies to the subsequent steps, and the selection rule, selection, and message included in the script can be stored in the knowledge structure configuration method according to the present invention.

<Example 2>
The following (K1) to (K11) show examples of operations that can be performed by the implementation means of the method for creating an automatic response according to the present invention using the data stored in the knowledge structure as described above. In the following, “U” at the beginning of the line indicates the user's action, and “A” indicates the operation of the implementation means of the automatic response creation method according to the present invention.
(K1) (U) Says “I want to eat curry, though it ’s hot today”.
(K2) (A) follows the user relationship from the start point _{v root,} it found user _{u dst,} further follow operation relationship, to find the starting point _{a sr} script structure. Further, following the user relationship from the start point _{v root,} it found user _{u ca,} further follow operational relationship to discover context pointer pi _c via the class _{a c.} Like the elements pi _c, the final choice pointers _{pi ls} via a class _{a ls,} via the class _{a lr} finding a final response pointer _{pi lr.}
(K3) (A) participant elements pi _c is checked to ensure that it is null. The elements a _sr to the previous element traced in the a _rel relationship are accessed in order, and it is checked whether the character string associated with the element appears in the user utterance. The character string “curry” associated with the element sr ₁ is Detect the appearance.
(K4) (A) An attempt is made to execute the preceding element s ₁ traced from the element sr _{1 in the} a _rel relationship. Class a _sel of element s ₁ is associated with a pointer to a function that performs the choice operation. It responds with the character string “What is the type of curry?” _Associated with the previous element m ₂ traced from the element s ₁ to the a _sm relation, and the previous elements sr ₂ and sr ₃ traced from the element s ₁ to the a _rel relation The character strings “India” and “South India” respectively associated with are responded as option candidates. The _{s 1} to participant context pointer pi _c, the sr ₁ to participant final choice pointers _{pi ls,} sets the _{m 2} in relation destination final response pointer _{pi lr.}
(K5) (U) Say “India”.
(K6) (A) elements from related parties s ₁ of pi _c examine the string tied to the earlier of the elements followed by a _rel relationship, that the string tied to the element sr ₂ "India" has emerged Detect.
(K7) (A) The class of the previous element s ₂ traced from the element sr _{2 in the} a _rel relation is also an option action a _sel and is associated with the previous element m ₃ traced in the a _sm relation from the element s ₂ Responds with the character string “What is the area?” And the character strings “Roppongi”, “Ginza”, “Ochanomizu” associated with the previous elements sr ₄ , sr ₅ , sr ₆ traced from the element s _{2 in the} a _rel relationship, respectively. As a choice candidate. The _{s 2} in relation destination context pointer pi _c, the sr ₂ to participant final choice pointers _{pi ls,} sets the _{m 3} in relation destination final response pointer _{pi lr.}
(K8) (U) Say “Roppongi”.
(K9) (A) element pi from related parties s ₂ of _c examine the string tied to the earlier of the elements followed by a _rel relationship, that the string tied to the element sr ₄ "India" has emerged Detect.
(K10) (A) Since there are a plurality of elements m ₄ and m ₅ that can be executed from the element sr ₄ to the destination of the a _rel relationship, random selection is performed, and execution of m ₄ is attempted.
(K11) (A) The class of the element m ₄ is a _msg which performs a response message, and responds with the character string “xxx is recommended” linked to the m ₄ . Participant the context pointer pi _c does not change, the sr ₄ to related parties of the final choice pointers _{pi ls,} sets the _{m 4} to related parties of the final response pointer _{pi lr.}
Table 2 exemplifies a dialog that is a dialogue including the operation of the above-described automatic response creation method according to the present invention as a character string that can be displayed on a terminal device or the like excluding the internal operation of the program.

As described above, by using the method for creating an automatic response according to the present invention, a flexible response can be automatically created quickly and in response to a message from a user including a natural language.
A message created by an automatic response can also be an element of the knowledge structure according to the present invention. Therefore, the execution means of the method for creating an automatic response according to the present invention repeats the operation of the automatic response to successively add selection rules, response messages, etc. having associations to the knowledge structure and store more associations. Can build their own knowledge structure. Thereby, the knowledge structure according to the present invention can be self-creating.

  As mentioned above, although embodiment of this invention was described, this invention is not restricted to embodiment mentioned above. The effects described in the embodiments of the present invention are only the most preferable effects resulting from the present invention, and the effects of the present invention are limited to those described in the embodiments of the present invention. is not.

It is a figure which shows the relationship of the element in the basic structure of a knowledge structure based on one Embodiment of this invention and a prior art. It is a figure which shows the example of the knowledge stored using the semantic network which concerns on one Embodiment of this invention and a prior art. It is a figure which shows the structure of the automatic response output production system based on one Embodiment of this invention. It is a figure showing the data structure of knowledge DB based on one Embodiment of this invention. FIG. 6 is a diagram illustrating a connected graph including elements of an extended semantic network according to an embodiment of the present invention. It is a figure which shows the hardware constitutions of the automatic response output production apparatus based on one Embodiment of this invention. FIG. 4 is a diagram illustrating an example of operation of a knowledge structure according to an embodiment of the present invention. FIG. 4 is a diagram representing a dialogue script as a graph structure according to an embodiment of the present invention. FIG. 4 is a diagram illustrating an example of a relationship between elements of a knowledge structure in dialogue script tracking according to an embodiment of the present invention.

Explanation of symbols

10 Automatic response output creation device 20 Automatic response output creation system 30 Knowledge DB
38 Network 60 User terminal 80 Knowledge DB data structure 82 Identifier 84 Start point 85 End point 86 Label 88 Pointer 301, 302, 304, 321, 322, 323 Element 303, 305, 312, 324, 325, 333 Side 310, 331 Start point 311, 332 End point 340, 350 Connected graph 361 User 362, 363, 364 Operation 365, 366, 367 Function pointer 360, 380 Start point of all knowledge structure 381 User class 382 Action class 384 Instance class 390, 392, User 410 selection 412 Selection message 414 Selection rule 420 Context 422 Final selection 424 Final response 440 Selection rule 442, 446, 447, 450, 452, 454, 456, 458 Instance 460,462,464,466,468,470,472,474 instance 444,448,449 selection of messages

Claims (8)

A knowledge structure configuration method for creating a knowledge structure in a storage unit that includes a relationship between a situation and an action included in a computer-processable language, wherein the relationship is a set including a start point and an end point as elements,
Storing a term for specifying the situation and a term for specifying the action in the storage unit;
Defining a class for each of the situation and the action and the relationship;
Storing the relationship between the term for specifying the situation and the term for specifying the action included in the knowledge structure as the element;
Storing the affiliation or subordinate relationship between the class and the relationship as an element in the knowledge structure;
Including
Here, the relationship is either a relationship in which both the start point and the end point are empty elements, or a relationship in which both the start point and the end point are elements of the knowledge structure. Method. The method for constructing a knowledge structure according to claim 1, wherein, in the step of defining a class for each of the situation, the action, and the relationship, the following mathematical formula is used to define the class.

Here, ASN is an enhanced semantic network (Augmanted Semantic _{Network), E ASN} is a set of extended semantic network, e is a two-tuple element is an element of the set E, _{e s} is an element of _{E ASN} a certain starting point (301) or the like, _{e d} is the end point (302) or the like is an element of _{E ASN,} null is empty element further extended semantic network has the following element set,

Here, ASN _VE is a component of the extended semantic network ASN, E _v is a set of vertices in the extended semantic network, and E _e is a set of edges in the extended semantic network. Receiving a term to identify the situation or a term to identify the action;
Selecting and storing the received term for specifying the situation or the term for specifying the action;
The method for constructing a knowledge structure according to claim 1, further comprising: Generating a discourse structure that is a set having a conversation place, a conversation history, and a conversation context as elements;
Storing the conversation field as one element in the knowledge structure;
Storing in the knowledge structure a fully ordered set whose elements are conversation occurrences;
Storing the occurrence of the conversation and the sequential adjacency relationship of the occurrence of the conversation in the knowledge structure;
The method for constructing a knowledge structure according to any one of claims 1 to 3, further comprising: Generating a tracing structure that is a set including a relation class, a direction, and a destination as elements;
Extracting an element having a start point and an end point among elements included in the knowledge structure;
Adding a combination of its own class, a forward link flag, and an end point to the extracted start point to the tracing structure;
Adding its own class, a flag indicating reverse association to the extracted endpoint, and a combination of starting points to the tracing structure;
Further including
The method for constructing a knowledge structure according to any one of claims 1 to 4. Using the method for constructing the knowledge structure according to any one of claims 1 to 5,
In response to receiving a term to identify the situation,
Extracting a knowledge structure element in which a combination of a flag indicating an association of a motion relation class and a forward direction exists from the set of trace structures;
Random selection rule from the extracted at least one element, rule for selecting the first or last extracted element, rule for selecting the oldest or most recently extracted element, most relevant to discourse structure Selecting an element based on a rule selected from rules for selecting an element;
A method for creating an automatic response, wherein a term for specifying the action is determined using the method.   The computer program for making a computer perform the step in any one of Claims 1-6. An automatic response creation device using the knowledge structure construction method according to any one of claims 1 to 6,
Means for storing a term for identifying the situation and a term for identifying the action;
In response to the user entering a term to identify the situation,
Means for extracting an element of a knowledge structure in which a combination of flags indicating behavioral relation classes and forward association exists from the set of trace structures;
Random selection rule from the extracted at least one element, rule for selecting the first or last extracted element, rule for selecting the oldest or most recently extracted element, most relevant to discourse structure Means for selecting a term for identifying the action based on a rule selected from rules for selecting an element;
Means for transmitting the selected term to the user;
A device for creating an automatic response.

Images