JP2003108589A - Home page navigator - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a home page navigator capable of easily and efficiently retrieving an intended home page by natural conversation input without relying keyword matching. SOLUTION: This home page navigator 2 of a home page navigator system 1 has a Web server 21 for providing a home page retrieving site on the Internet, a conversation control system 22, and a home page database 23 for storing various home page information. This conversation control system 22 can actively retrieve the home page related to the content of a conversation by utilizing a conversation type knowledge base, and an operator (a user) of a communication terminal 4 can arrive at the proper home page only by inputting one's own interest and a matter wanted to know in the form of a natural conversation. Time required for retrieving the home page having a complicated hierarchy is sharply shortened by using this conversation control system 22.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a homepage navigator capable of actively searching homepages through natural conversation. More specifically, a new conversation control system that greatly contributes to the improvement of human-machine interface technology, which is capable of drastically expanding the range of application of computer-based dialogue technology, is used for natural conversations or conversations such as homepage search. It relates to a home page navigator that can be performed by sentences.

[0002]

2. Description of the Related Art A home page search using a conventional search engine is forced to be a tree type search. For example, if you enter a keyword on the screen and the websites that have the keyword or are included in it are output as a list, you can enter another keyword on the screen to narrow them down and reach the target website. I am trying. Therefore, in order to find the homepage intended by the searcher, it is extremely important to select the keyword to be used, but in some cases, the appropriate keyword may not come to mind. As described above, the conventional methods are extremely fixed and passive, and in most cases, it is not possible to easily perform a search according to the searcher's intention.

[0003]

[Problems to be Solved by the Invention] Here, if the meaning of a natural conversational sentence is recognized and a homepage that matches the meaning can be retrieved, the homepage can be easily searched and a searcher can easily find it. It will be possible to reach the homepage intended by.

Therefore, a conversation control technology is incorporated in a search engine to interpret an input sentence for searching a home page, recognize the meaning contained therein, and perform a home page search based on this. It is possible to do so.

However, the current level of the conversation control technology is basically keyword matching as in the case of the current engine for home page search (for example, when answering a mountain, a river is answered), and conversation history and There is no technology such as topic tracing. This limits mechanization approaches only to the very limited area of responding to specific keywords with specific responses. Therefore, we cannot expect more results than the current home page search engine.

An object of the present invention is to drastically expand the application area of computer interactive technology by "conversation response with cohesiveness", "conversation history control" and "topic trace". The purpose is to propose a homepage navigator that enables users to search homepages through natural conversations using various conversation control systems.

[0007]

In order to solve the above problems, the home page navigator of the present invention has a receiving section for receiving an input sentence for searching a home page via the Internet, and information regarding various home pages. The home page database that recognizes the conversation meaning of the input sentence input through the receiving unit and the home page database that is suitable for the recognized conversation meaning are searched from the home page database, and the searched sentence corresponding to the input sentence is searched. A conversation control system that creates an answer including information about a home page and outputs the answer as an output sentence, and a transmission unit that provides the output sentence to the sender of the input sentence via the Internet. Is characterized by.

Here, in the above conversation control system, the input sentence is subjected to morphological analysis, syntactic analysis, case grammar analysis and vocabulary analysis, and the transition of the conversation state based on the past conversation history and the relevance of the topic. And the cohesiveness that indicates the relevance of the topic to the previous conversation, and based on this, the conversation meaning of the input sentence is recognized, the transition of the discriminated conversation state, and the cohesion that indicates the relevance of the topic. The input sentence having the set conversation state and the cohesiveness is set such that the conversation state is changed and the cohesiveness of the answer is set so that each of the sexes is appropriate as an answer to the recognized input sentence. It is characterized in that an answer is created and the answer is output as an output sentence.

In the above conversation control system, the basic structure of conversation is regarded as three hierarchies of a single utterance, a text, and a time-ordered list of texts, and the utterance is defined as one continuous utterance unit. In the case of chat, one line of input is regarded as one utterance, the text is regarded as a list of utterances connected by a certain cohesiveness, and the causal relationship, time, place, and impression are considered as the utterance cohesive elements. Using the supplementary explanation relation such as, a parallel strong order relation for explaining one case, and a metaphor sentence, the conversation is regarded as a list in which a plurality of utterances related to a topic are temporally formed, When the meaning of the text is expressed as a vocabulary by recognizing the meaning of the conversation by the text, the vocabulary of the text (Se
vocabulary is defined by eight elements: the subject of the action, the target of the action, the action, the action occurrence condition, the action occurrence time, the action occurrence place, the person who thought the action occurred, and the person's conclusion. The model is composed of the eight elements of the conversation described above, and F-Sentence (Formal
Sentence) is defined as follows, and the value of each element is defined as follows.

F-Sentence ⊆ (subject, subject, action, condition, time, place, observer, impressionist conclusion) Subject, subject ε (“phrase”, null, interrogative, interrogative “phrase”, “phrase” "Question, F-Sentence phrase) Action ε (" phrase ", null, question, question verb" phrase "," phrase "question) Condition ε (F-Sentence, phrase, null, question) Time ε ( F-Sentence, phrase, null, interrogative place Location ε (F-Sentence, phrase, null, interrogative) Observer ε (null, phrase) In the above conversation control system, in controlling the conversation state, Agents can be defined as a humorous type (often mischievous), a self-centered type (often talking about themselves), or Sequential type (I often hear about other people,
Answers are straightforward agents) and cheeky types (agents who ask a question with a good-looking face). Agents have emotions such as happiness, anger, sorrow, and comfort as a definition of their emotional form. And
It is assumed that there are three states of conversation: question, answer, and statement. As an event related to the transition of the conversation state, the event that causes the transition of the conversation state is an external event by a message from the other party of the conversation and the agent itself. It is composed of internal events that occur, and for external events, it is characterized by defining questions, answers, and declarations according to the conversation state of the other party, and as internal events, "I want to hear" is defined.

Further, in the above conversation control system, the determination of the cohesiveness, which means the relation of the conversation, is related to the meaning between the conversation sentences exchanged between the agent and the human. This is the formal definition of cohesiveness, and the minimum unit of conversation is the F-S
It is formed and defined using entence and is defined as cohesion = implication (f-sentence 1) implication (f-sentence 2) implication (f-sentence 3) when the set is not an empty set. However, the cohesiveness of conversation is understood from both theoretical and illogical relationships existing between sentences, and the logical relationship is defined as the definition of the subject, subject, and action in the preceding clause. Questions and answers about, the time when the preamble occurs, supplementary explanation about the place, sentences showing causal relations with the preamble, counter papers against the preamble, and impression sentences to the preamble, and illogical The characteristic relevance is a topic-based relevance.

[0012]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of a home page navigator to which the present invention is applied will be described below with reference to the drawings.

(Overall Configuration) FIG. 1 is a schematic configuration diagram showing a home page navigator system of this example. As shown in this figure, the home page navigator system 1 of this example
Is composed of a homepage navigator 2 and a communication terminal 4 such as a personal computer having an internet communication function capable of accessing the homepage navigator 2 via the internet 3.

The homepage navigator 2 includes a web server 21 that provides a homepage search site on the Internet, a conversation control system 22, and a homepage database 23 that stores various homepage information. The web server 21 sends and receives data to and from the communication terminal 4.

The conversation control system 22 of this example can actively search the homepage related to the content of the conversation by utilizing the conversational knowledge base, and the operator (user) of the communication terminal 4 can: You can reach the appropriate homepage simply by entering your own interests and interests in the form of natural conversation. Also, this conversation control system 2
By using 2, the time required to search a home page with a complicated hierarchy is greatly reduced.

(Overall Structure of Conversation Control System) FIG.
It is a whole lineblock diagram of conversation control system 22 of this example. The conversation control system 22 of this example is a Japanese context analysis machine 102.
, The data modeling machine 103, and the conversation control machine 10
4 and. Upon receiving the input sentence, the Japanese context analysis machine 102 cooperates with the data modeling machine 103 to cause the morphological analysis unit 105, the syntax analysis unit 106, and the case grammar analysis unit 107 to perform various Japanese analysis dictionaries 108.
Refer to for Japanese context analysis. The data modeling machine 103 includes the analysis result by the Japanese context analysis machine 102, the conversation history data / talk model 109 accumulated so far, and the term / genre dictionary 11 of vocabulary.
Based on 0, the vocabulary model unit 111 models the input sentence according to a predetermined data expression method and judges the content. Based on the analysis result by the data modeling machine 103, the conversation control machine 104 controls the conversation state in the conversation state control unit 112, controls the cohesion of the response in the response cohesiveness control unit 113, and the semantic representation network of the knowledge representation. With reference to the discourse dictionary 114 and the home page database 23, an output sentence corresponding to the input sentence is created and output. That is, an answer including homepage information most suitable for the intended content of the input sentence is generated and output as an output sentence consisting of a natural conversation.

(Conversation Theory: Data Modeling Method / Data Representation Method in Data Modeling Machine 3) The data modeling method / data expression method, which is one of the features of the conversation control system 22 of this example, will be described.

First, the structure of conversation is regarded as three hierarchical structures. That is, the basic structure of the conversation is a single utterance (text), a text (text),
It is divided into three layers called a time-ordered list of texts. Utterance is defined as one continuous voicing unit. In the case of chat, one line of input is regarded as one utterance. The text has some strong cohesion (cohesio)
It is a list of utterances connected by n). In some cases, each utterance becomes one text. here,
The cohesive elements of utterances include causal relationships, supplemental explanatory relationships such as time, place, and impression, and parallel strong order relational metaphors for explaining one case. A conversation (Speech) is a list in which a plurality of utterances related to a topic are temporally formed.

Next, the recognition of the meaning of the conversation (Cogni
(Tion of Implication) will be described. The key to understanding the meaning of conversation is not utterance,
It is a text. Text vocabulary (Semantic
s) is defined by the following eight elements. 1. Subject of action (Subject) 2. Action target (Object) 3. Action 4. Condition of occurrence of action (Condition) 5. Action occurrence time (Time) 6. Location of action (Location) 7. Respondents (Observe) for the occurrence of action
r) 8. Reviewer's conclusion The cognitive approach in the conversation control system 22 of this example is that the Japanese context analysis machine 2 and the data modeling machine 3 first perform context analysis of utterances and then complete the vocabulary. The context analysis is performed in the order of morphological analysis, syntactic analysis, and lexical analysis. The completion of the vocabulary includes the use of implicit knowledge, the introduction of implicit knowledge, the history of utterances, and the conversation schema (background) in order to complete the vocabulary obtained from the expression of one utterance by the relation with the preamble (utterance). Knowledge).

Here, the "concept" that is the basis for input sentence analysis and vocabulary completion will be described. First, the concept (Concept) is formed from a noun and a compound of nouns, and is used to express the subject of the utterance and the target person. Concepts can be generally classified into concepts that abstract one kind, concrete things, in other words, concepts that refer to instances of abstract concepts, and concepts that represent actions.

Further, explaining the concept hierarchy, the concept abstraction becomes hierarchical. The superordinate concept means a higher level abstraction, and the subordinate concept reflects a more concrete or instance.

Further, explaining the attributes of the concept,
Concepts other than the behavioral concept can be decomposed into smaller components. Each component is defined as a concept attribute. Concept attributes are also concepts. The property that these concepts constitute a concept is called the concept Aggregation. Further, to explain the limitation of the concept, when the object expressed by the concept has a category, the expression limited to a part of the category is called the limitation of the concept. Express the qualification of a concept by putting a determiner in front of the noun indicating the concept.

Next, the concept combination includes hierarchical combination,
Includes affiliative and limited composites. Here, the compound type concept is expressed by compounding nouns. There are the following methods for the composite pattern. Noun + "no" + noun → Express a hierarchical and affiliative compound. Adjective + noun, adjective + "na" + noun, noun +
"No" + noun → Express a limited relationship.

(Vocabulary Model) Next, a vocabulary model (Semant) in the vocabulary model section 111 in the data modeling machine 103 in the conversation control system 22 of this example.
ic Model) will be described.

The conventional conversation control is not the vocabulary of sentences,
This is a method of issuing an answer sentence by matching keywords appearing in the sentence. Occasionally, there is a difference in meaning or the opposite sentence. In particular, it can be said that this method does not allow continuous conversation according to a certain meaning. On the other hand, in vocabulary-based conversation, it is important how to establish the vocabulary model. There are two basic approaches to designing a vocabulary model. One is an inclusive reasoning model and the other is a human vocabulary structure model. In the former case, a meta-predicate expression knowledge base that includes all natural language words is required, and it is considered impossible at present. In the conversation control system 22 of this example, an approach of constructing a vocabulary structure model is adopted.

The vocabulary model and formal expression will be described in detail. First, the vocabulary model is composed of eight elements of the above-mentioned conversation and is formally expressed by F-Sentence (described later). That is, F-Senten
ce (Formal Sentence) is defined as a set of the following eight elements. F-Sentence⊆ (subject, target, action, condition, time, place, observer, impressionist conclusion) The values of each element are as follows. Subject, Object ε (“phrase”, null, interrogative, interrogative “phrase”, “phrase” interrogative, F-Sentence phrase) Action ε (“phrase”, null, interrogative, interrogative “phrase”, “phrase” Phrase "question" condition ε (F-Sentence, phrase, null, question) time ε (F-Sentence, phrase, null, question) place ε (F-Sentence, phrase, null, question) observer ε (Null, phrase) Recipient's conclusion ε (“affirmation”, “denial” “question”) Here, “null” means that no input has been made. The expressions such as an interrogative “phrase” and “phrase” interrogative mean a question for a superordinate concept of a phrase or a question for an attribute of the phrase concept. Also, from the condition,
When the element between the observers becomes a question, (s
It is a question for the declarative sentence of (object, object, action). Furthermore, the result is o
If bserver is present, result indicates the result of the impression. If the Observer does not exist explicitly, the result shows the form of the statement from a neutral standpoint. Therefore, if it cannot be judged, all are regarded as positive sentences. Also, the fact that an element with F-Sentence is F-Sentence means that F-Sent
nce has non-linearity.

Next, in this example, the basis of conversation control is how to output an answer sentence related to the text from the other party. For this reason, elements included in a series of related texts are defined as "topics", and a formal expression thereof is imposed.

First, the topic is a conceptualization of a topic related to a portion where the meanings of a series of conversation texts overlap.
Using the definition of F-Sentence, the topic (to
pic) is: topic = f (F-Sentence_1∩F-Se
ntence_2∩F-Sentence_3. ． ． =
Since it can be simplified to> (subject, object, action), the topic can be expressed as follows in a code calculation method. (Topic? Subject? Object?
action) subject, object, action here
Will be defined as the three elements of the topic. And (s
object, object, action) is called a topic F-Sentence.

Next, the comprehensive characteristics of the topic will be described.
As for the topic (topic), the more the above-mentioned elements are lacking, the wider the range of speaking becomes. Thus, there is the following inclusiveness between the topics. (Topic? Subject? Object?
action) ⊂ (topic? subject?
object) ⊂ (topic? subject) where (subject object actio
n) is a third-class topic; (subject object)
Is defined as a secondary topic; and (subject) is defined as a primary topic.

The "basic topic" is the subject.
The concept corresponding to the object or the attribute of the concept is the subject or the subject of the topic. The “deepening of a topic” means that a concept corresponding to a subject or an object of a topic shifts to a lower level of a concept abstraction hierarchy (a more concrete concept or its own instance).
"Expansion of topic" means subject or object of topic
It means that the concept corresponding to ct shifts to a higher level in the abstraction hierarchy of the concept (more abstracting concept or one's own parent).
“Related topic” means that in the concept abstraction hierarchy, another concept in the same hierarchy as the concept that is the subject or object of the current topic is the subject or object of the topic.
A topic that overlaps with or becomes a related topic is defined as a related topic.

An object model of the topic dictionary is shown in FIGS. 3, 4 and 5.

(Cohesiveness of conversation) Next, cohesiveness of conversation refers to the relevance of a series of conversations. F-Sentence is a unit of conversation as one Utterance.
Formally define by. The cohesiveness of conversation is formally expressed as cohesion = implication (f-se
ntence_1) ∩ implication (fs
entence_2) ∩ implication (f-
sentence_3) ∩. ． ． Can be defined as A cohesive conversation means that the above-mentioned set is not empty. Here, “implication” refers to a set of meanings of sentences.

The cohesiveness of conversation can be understood from both the logical relationship and the illogical relationship existing between sentences. The logical relationships include the following: Questions and answers about the definition of the subject, object, and action in the preceding clause; Supplementary explanation about the time and place where the preceding sentence occurs; Sentences showing causal relations with the preceding sentence; Counter-article against the preceding sentence; Impression sentence.

On the other hand, illogical relationships are connected by the above-mentioned topics.

Next, the "Talk model" will be described. First, "talk" is a model that models the logical cohesion of cohesion. Specifically, it has the following structure. talk {f-sentence n-sentence (note 1) list (note 2) {(type: (?) word ∈ (subject object action condition negotiation effect)) (note 3) note 4) 1: Natural language sentence that expresses utterance Note 2: Form one list in the order of the talk related to this topic, the talk at the top is the latest talk. Note 3: Type is a meta-definition sentence of derived talk, and is a definition regarding the position of the derived talk in the higher-ranked talk. Before the type keyword? If is attached, this talk is a question sentence for the part specified by the keyword. Note 4: There are multiple pairs of (type, talk).

A topic related to the same talk is defined as a "derivative topic".

Next, "List of topics (Topic Li
st) ”is a structure in which talks having a basic topic, a deepening topic of the topic, an expanding topic, and a related topic are listed in time order, and is a model that summarizes logical and illogical contents of cohesiveness. Is.

The "conversation history" is recorded by a topic stack to form a history model of a plurality of conversation histories in time order.

(Discourse Dic
Next, the discourse dictionary provided in the conversation control machine 104 will be described. The “agent” created by the conversation control system 22 of the present example basically prepares his / her conversation content centering on a topic when talking with the other party. A dictionary that summarizes the conversation contents of agents by topic is called a discourse dictionary. In a discourse dictionary, it is called a dictionary item discourse, and the subject of the topic F-sentence and the character strings of the target and the verb are used as search keys. It can be a compound word. In addition, for each item that is the subject, all related answers are patterned and prepared. Furthermore, when the subject is a concept, a hierarchical relation of the concept is defined. In addition to this, the similar expressions of the subject are listed.

Next, the topic dictionary system has the following contents. Dictionary database (Dictionary DB): This is a basic dictionary,
It is a set of files of a specialized dictionary group, a corporate questionnaire dictionary, and a personal dictionary. Dictionary definition language DDML (Discourse Dic)
tionary Markup Language-X
Script language extended on ML) Dictionary generation tool (Basically, this tool is implemented as one function of a conversation server.) Here, the topic definition language DDML will be described. 6 to 9 show the structure of the DDML language.

Next, the structure of the topic dictionary group will be described. The topic dictionary group includes a character basic dictionary specialized dictionary (for example, sports and music), a corporate questionnaire dictionary, an individual customized dictionary, and the like. In it, a basic dictionary, a specialized dictionary, and a corporate questionnaire dictionary are generated by DDML.
The personal dictionary shall be automatically created online. In addition, in order to perform context analysis, prepare a practical dictionary. FIG. 10 shows the structure of the dictionary group file.

Here, the topic dictionary has items used for some special purposes. The following search keys are provided to search the contents related to these items. Interrogative: As a content, it is a discourse with no topic in which words for all question sentences such as "where" and "what" are arranged as similar words. Connecting particle: A discourse in which connecting particles that connect nouns such as "no" and "na" are stored as similar words. General negative sentence: This is a discourse that aggregates general negative sentences that are not related to the content. It is possible to classify according to the structure of the denied sentence. General affirmative sentence: This is a discourse that aggregates general affirmative sentences that are not related to the content. It is possible to classify according to the structure of the affirmative sentence. Agent's desire: The contents of the agent's own question are put under this discourse as the desire of the agent. Contents can be divided into topics and stored. These desires serve as triggers for more specific specialized dictionaries.

Next, a method of constructing a questionnaire dictionary of companies will be described.
Complete as a topic dictionary. Specifically, each company will enter one phrase in the dictionary.

[0044]     discourse (company name)             attribute (products, sales, events, etc.)             category (by company type such as automobile manufacturer and home appliance manufacturer)             synonym                               concept (various calls to the company)             topic (subject object action)                                 question Question sentence… (Note 1)                                 impression impressions (Note 2)             topic (subject object action)                                 question Question sentence… (Note 1)                                 impression impressions (Note 2)     end Note 1: Questionnaire text for the questionnaire, multiple allowed Note 2: Word-of-mouth (sent to a conversation partner), multiple allowed Note 3: Preparations for corporate questionnaires are questions and impressions. The answer from the other party is Put it in a text file with the question.

(Control of Conversation State) Next, the conversation state in the conversation state control, which is a feature of the conversation control system 22 of this example, will be described.

First, in this example, the agent has the following four personalities as the "agent personality definition". Playful: Playful naughty agent self-centered:
Agents who talk about themselves well Obedient type (meek): Listen to others well,
Agents who are also responsive to answers Cheeky type: Agents who ask questions with a look that looks great, where messy and obedient agents often learn the other person's expressions and learn the other person's words without knowing it. . The self-centered type and the cheeky type memorize the expressions of the other person, but emphasize the one's existence without bothering to use the saying in the conversation.

Further, to describe the "definition of emotional form", it is assumed that the emotion of the agent has emotions. In order for the agent to understand the conversation with the human being, a common emotional form is provided between them. At this stage, the following emotional forms are defined. Pleasure, Anger, Sorrow, Joy Next, the state of the conversation scenario will be described. First, as the state of conversation, there are a question, an answer, and a declaration.

Among the events related to the conversation state transition, the event that causes the conversation state transition is composed of an external event generated by a message from a conversation partner and an internal event generated by the agent itself. For external events, the following types are defined according to the conversation state of the other party. Question (asked) Answered (replied) Declared (declared) On the other hand, the following types are defined for internal events. I want to ask (ask) FIG. 11 shows the state transition of the conversation scenario.

Next, the sub-states of the above Declaration conversation will be described. Declaration
In the state, consider the transition form in which the agent shifts from the "willingness to speak" state to remove. The "willingness to speak" state is divided into the following three substates. Somewhat (favorable): talks with the other person (begins to have a good feeling) quite (likeable): has a lot of hobbies with the other person (likely to be friends) very much (friendly): liked the other person (Become friends) At the same time, consider the following sub-states when you are "not willing to talk". A little disagreeable: I feel uncomfortable with the other person (dislikeable): I hate the other person (unbearable): I can not stand the other person, and more I no longer want to talk.

This intermediate state is normal: a state in which neither one likes nor dislikes the other party. As the event that induces the transition of the sub-state,
Two types of events occur, which are "please" and "display". The transition of sub-states is shown in FIG. Note that all conversations start from the normal state in the Declaration state.

Next, the sub-state of the Query conversation will be described. In the Query state, the agent makes a state transition between four substates. In other words, listening: Asking the other party (preparing a sentence) Waiting: Waiting for a reply from the other Learning (learning): Understanding and remembering the other party's words There are the following: issue: Send a question sentence to the other party learn: From the other party to unknown content (satisfied)
Entered in the customized dictionary after getting the answer unknown: It is found that the received sentence has unknown contents. The following contents can be considered as a scenario of state transition. When you receive an ask event from another state, first ask
Enter the ing state, create a question sentence there, and send it to the other party. When the send event issuance is issued, it waits for a reply from the other party, waiting, and as soon as a reply is received, if the original question text is a mischievous text or a questionnaire, it immediately replies.
An external event d is generated. Otherwise, issue a learn event to remember the other party's reply. Issuing a learn event causes learning
Enter the state and learn. Then, if the reply still contains an unknown sentence, and if the number of questions is still less than 2, ask unkno to ask another question.
Issue a wn event and enter the asking state.

FIG. 13 shows the transition of the query substates.

Next, the sub-states of the Answer conversation will be described. In the Answer state, the agent makes a state transition between four sub-states. That is, listening (replying): Asking the other person (preparing a sentence)
State waiting (waiting): Waiting for reply from the other party Internal event in state substate issue: Sending a question sentence to the other party Also, the following contents can be considered as a state transition scenario. When receiving the external event asked, first reply
Enter the ing state. Here, according to the designated answer form, it is determined whether a decent answer sentence or a mischievous question sentence is to be issued and transmitted. Before issuing the answer sentence, check the other person's question sentence for unknown contents, and if there is,
Put the answer sentence on hold and ask a question about the unknown content. Therefore, the external event ask is issued. When the internal event issue or the external event ask is issued, it enters the waiting state and waits for a reply from the other party. Here, when the reply is received, if the reply from the partner is the reply to the question of the user himself / herself, a reply sentence is created using the reply contents of the partner and the reply is issued. Also, when the other person asks a new question (asked event is received), it returns to the replying state again. When receiving a declaration from the other party, issue the external event declare and change the declaration from the answer state.
Enter the state.

FIG. 14 shows the sub-state transition of Answer. In addition, FIG. 15 shows a classification table of response methods.

(Basic Algorithm of Each Operation of Conversation Control System) FIGS. 16 to 43 show a basic algorithm of each operation in the conversation control system 22 of the present example.

(Example of conversation scenario) FIGS. 44 to 47
In the figure, various patterns of conversation scenarios realized by the conversation control system 22 of this example are shown. As described above, by using the system 1 of this example, a high level conversation can be realized in consideration of the conversation history and the like.

[0057]

As described above, the home page navigator of the present invention is equipped with the conversation control system capable of grasping the contents of the input sentence based on the vocabulary model by tracing the topic in consideration of the conversation history. And
It is possible to interpret the meaning of the input sentence by the conversation control system, search the most appropriate homepage, and create and output the search result in the same natural conversation sentence form.

Therefore, according to the present invention, unlike the conventional engine for home page search which is basically based on keyword matching, since it is based on the conversation history of the input sentence and the transition of the topic, it is appropriate for the input sentence. You can make a response. That is, an appropriate homepage search can be performed.

Further, by controlling the conversation state, agents having different characters can respond to the same input sentence in the same manner as persons having different personalities can respond. It is possible to realize a dialogue close to a conversation between humans. Therefore, unlike conventional machine search, while having a natural conversation with people with different personalities,
You can search the homepage.

Furthermore, by controlling the cohesiveness,
By determining the relevance of the conversation and creating and outputting an output sentence that takes into account the current topic, the relevance to the previous topic, and so on, the dialogue will have a form that depends on the content that has been spoken so far. Become. Therefore, since it is possible to narrow down the search that reflects the user's intention to the previous search response, it becomes possible to easily reach the appropriate homepage in a short time.

[Brief description of drawings]

FIG. 1 is a schematic configuration diagram showing an example of a home page navigator system to which the present invention is applied.

FIG. 2 is a configuration diagram of the conversation control system in FIG.

FIG. 3 is an explanatory diagram showing an object model of a topic dictionary.

FIG. 4 is an explanatory diagram showing an object model of a topic dictionary.

FIG. 5 is an explanatory diagram showing an object model of a topic dictionary.

FIG. 6 is an explanatory diagram showing a definition language DDML of a topic.

FIG. 7 is an explanatory diagram showing a definition language DDML of a topic.

FIG. 8 is an explanatory diagram showing a definition language DDML of a topic.

FIG. 9 is an explanatory diagram showing a definition language DDML of a topic.

FIG. 10 is an explanatory diagram showing the structure of a dictionary group file.

FIG. 11 is a state transition diagram showing state transition of a conversation scenario.

FIG. 12 is a state transition diagram showing substates of a Declaration conversation.

FIG. 13 is a state transition diagram showing substates of a Query conversation.

FIG. 14 is a state transition diagram showing sub states of a Query conversation.

FIG. 15 is an explanatory diagram showing a classification table of response methods.

FIG. 16 is a flowchart showing an algorithm regarding talk.

FIG. 17 is a flowchart showing an algorithm regarding talk.

FIG. 18 is a flowchart showing an algorithm regarding talk.

FIG. 19 is a flowchart showing an algorithm regarding talk.

FIG. 20 is a flowchart showing a preliminary process.

FIG. 21 is a flowchart showing a preliminary process.

FIG. 22 is a flowchart showing an operation in a query state.

FIG. 23 is a flowchart showing an operation in a query state.

FIG. 24 is a flowchart showing an operation in a query state.

FIG. 25 is a flowchart showing an operation in a query state.

FIG. 26 is a flowchart showing an operation in a query state.

FIG. 27 is a flowchart showing an operation in a query state.

FIG. 28 is a flowchart showing an operation in the answer state.

FIG. 29 is a flowchart showing an operation in the answer state.

FIG. 30 is a flowchart showing an operation in the answer state.

FIG. 31 is a flowchart showing an operation in a declaration state.

FIG. 32 is a flowchart showing an operation in a declaration state.

FIG. 33 is a flowchart showing an operation in a declaration state.

FIG. 34 is a flowchart showing an operation of question sentence processing.

FIG. 35 is a flowchart showing a search operation of a dictionary discourse by a search key.

FIG. 36 is a flowchart showing a processing operation of a confirmation question sentence.

FIG. 37 is a flowchart showing a collation operation of F-Sentence.

FIG. 38 is a flowchart showing the processing operation of a question sentence.

FIG. 39 is an explanatory diagram showing an initial state of control regarding a declaration statement.

[Fig. 40] Fig. 40 is a flowchart illustrating an algorithm of a declaration-type attached answer.

FIG. 41 is a flowchart showing a (selection sentence) algorithm for topic selection.

FIG. 42 is a flowchart showing an algorithm of topic shift.

FIG. 43 is a flowchart showing an algorithm of topic shift.

FIG. 44 is an explanatory diagram showing an example of a conversation scenario.

FIG. 45 is an explanatory diagram showing an example of a conversation scenario.

FIG. 46 is an explanatory diagram showing an example of a conversation scenario.

FIG. 47 is an explanatory diagram showing an example of a conversation scenario.

[Explanation of symbols]

1 Homepage navigator system 2 Homepage navigator 3 Internet 4 communication terminals 21 Web server 22 Conversation control system 23 Homepage database 102 Japanese context analysis machine 103 Data modeling machine 104 Conversation control machine 105 Morphological analysis unit 106 parser 107 Case Grammar Analysis Department 108 Japanese analysis dictionaries 109 Conversation history / talk model 110 vocabulary / genre dictionary 111 Vocabulary model part 112 Conversation state control unit 113 Response cohesion control unit 114 discourse dictionary

Front page continuation (51) Int.Cl. ⁷ Identification code FI theme code (reference) G06F 17/28 G06F 17/28 XF terms (reference) 5B075 KK07 ND20 NK32 PP02 PP24 UU40 5B091 AA15 BA15 CA02 CA05 CA06 CA12 CB02 CC04

Claims (5)

[Claims] 1. A receiving unit that receives an input sentence for searching a homepage via the Internet, a homepage database that stores information about various homepages, and a conversation meaning of the input sentence input through the receiving unit. Recognize and search the homepage database for a suitable homepage for the recognized conversation meaning,
A conversation control system that creates an answer including information about the searched homepage for the input sentence and outputs the answer as an output sentence, and a transmission that provides the output sentence to the sender of the input sentence via the Internet A homepage navigator characterized by having a section. 2. The conversation control system according to claim 1, wherein the input sentence is subjected to morphological analysis, syntactic analysis, case grammar analysis, and vocabulary analysis, and based on the past conversation history and topic relevance. The state transition and the cohesiveness indicating the relevance of the topic to the previous conversation are determined. Based on this, the conversation meaning of the input sentence is recognized, and the transition of the determined conversation state and the relevance of the topic are determined. The conversational state is changed and the cohesiveness of the answer is set so that the cohesiveness indicating each becomes appropriate as an answer to the recognized input sentence. A home page navigator characterized by creating an answer to an input sentence and outputting the answer as an output sentence. 3. The conversation control system according to claim 2, wherein the basic structure of conversation is regarded as three layers of a single utterance, a text, and a time-ordered list of texts, and the utterance is defined as one continuous utterance unit. In the case of chat, one line of input is regarded as one utterance, the text is regarded as a list of utterances bound by a certain strong cohesiveness, and the causal relationship, time Multiple utterances related to a topic were temporally formed in the conversation by using supplementary explanation relations such as place, place, and impression, parallel strong order relations for explaining one case, and metaphor sentences. When the meaning of the text is expressed as a vocabulary by recognizing the meaning of the conversation as a list and recognizing the conversation meaning by the text, the vocabulary of the text (Seman tics) is the subject of the action, the target of the action, the action, the condition under which the action occurs, the time when the action occurs, the place where the action occurs,
The vocabulary model is composed of the eight elements of the above-mentioned conversation, which are defined by the eight elements of the thinker and the conclusion of the thinker about the occurrence of the action.
-Sentence (Formal Sentence
A home page navigator characterized in that the value of each element is defined as follows by e). F-Sentence ⊆ (subject, target, action, condition, time, place, observer, impressionist conclusion) Subject, target ε (“phrase”, null, interrogative, interrogative “phrase”, “phrase” interrogative , F-Sentence phrase) Action ε (“phrase”, null, interrogative, interrogative “phrase”, “phrase” interrogative) Condition ε (F-Sentence, phrase, null, interrogative) Time ε (F-Sentence , Phrase, null, question) location ε (F-Sentence, phrase, null, question) observer ε (null, phrase) 4. The conversation control system according to claim 3, wherein, in the control of the conversation state, an agent has a messy type (a well mischievous agent) and a self-centered type (well The agent's emotional form is classified into four types: an agent who speaks about people), an obedient type (an agent who often listens to others and responds honestly), and a cheeky type (an agent who often asks a question with a big face). The agent has the emotions of happiness, anger, sadness, and comfort, and the conversational state has three questions: answer, statement, and statement. The conversational state is an event related to the transition of the conversational state. The event that causes the transition of the The home page is characterized by internal events that are generated by the users, questions, answers, and declarations are defined according to the conversation state of the other party for external events, and "I want to hear" is defined as an internal event. Navigator. 5. The determination according to claim 4, wherein the cohesiveness, which means the relevance of the conversation, is related to the meaning between the conversation sentences exchanged between the agent and the human. The formal definition of cohesion is that the smallest unit of conversation is F
-Formation is defined using Sentence, and cohesion = implication (f-sentence 1) implication (f-sentence 2) implication (f-sentence 3) When the set is not an empty set, "cohesion exists" We define the cohesiveness of conversation in terms of both theoretical and illogical relationships that exist between sentences, and the logical relationship is the relationship between the subject, subject, and action in the preceding clause. Includes questions and answers about definitions, time when the preamble occurs, supplementary explanation about the place, sentences showing causal relations with the preamble, counter papers against the preamble, and impressions about the preamble. A homepage navigator that is characterized by topic-based relationships.