WO2004066164A1 - Topic net generation method and apparatus - Google Patents

Topic net generation method and apparatus Download PDF

Info

Publication number
WO2004066164A1
WO2004066164A1 PCT/JP2003/000610 JP0300610W WO2004066164A1 WO 2004066164 A1 WO2004066164 A1 WO 2004066164A1 JP 0300610 W JP0300610 W JP 0300610W WO 2004066164 A1 WO2004066164 A1 WO 2004066164A1
Authority
WO
WIPO (PCT)
Prior art keywords
topic
node
data
generation
step
Prior art date
Application number
PCT/JP2003/000610
Other languages
French (fr)
Japanese (ja)
Inventor
Ryosuke Miyata
Toshiyuki Fukuoka
Eiji Kitagawa
Original Assignee
Fujitsu Limited
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Fujitsu Limited filed Critical Fujitsu Limited
Priority to PCT/JP2003/000610 priority Critical patent/WO2004066164A1/en
Publication of WO2004066164A1 publication Critical patent/WO2004066164A1/en
Priority claimed from US11/121,001 external-priority patent/US7756897B2/en

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06QDATA PROCESSING SYSTEMS OR METHODS, SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORY OR FORECASTING PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORY OR FORECASTING PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00Commerce, e.g. shopping or e-commerce

Abstract

In order to effectively generate a topic that can be used for a wide range of interaction, a topic to be used in the interaction is analyzed/extracted from various information sources, thereby automatically generating a topic net as a set of links to interaction patterns.

Description

 Description Book Topic Net Generation Method and Apparatus (Technical Field)

 The present invention relates to a response technology using a computer for selling, searching for, and making inquiries of products and services suitable for a user.

(Background technology)

 2. Description of the Related Art Conventionally, various interactive systems have been provided for performing sales, search, inquiries, and the like of goods and services using a computer. For example, Japanese Patent Application No. 2001-400675 describes a dialogue system that enables effective information exchange between a computer user and a computer while maintaining the context. More specifically, in the above-described dialogue system, a dialogue having a contextuality in consideration of the context is performed as performed by humans. Therefore, the exchange between the user and the interactive system does not end in a round trip. Also, even if the exchanges are repeated multiple times, they are not independent of each other but related to each other.

 In the dialogue system, a possible dialogue pattern is dynamically created based on a topic net in which a plurality of topics are linked. This is fundamentally different from the method in which all possible dialog scenarios are prepared in advance by combining the questions output by the dialog system and the options for answering the questions. In order for dialogue by such a dialogue system to be effective, the topic net must include a sufficient number and range of topics appearing in the dialogue, and have sufficient dialogue information to make transitions between topics. Need to be.

By the way, in order to create a topic net, it is necessary to identify the topics that will appear in the dialogue from various information sources, analyze their relations, and prepare messages to output about each topic. For example, consider the case of creating a topic net for an interactive system that sells travel packs. In this case, topics about a variety of travel purposes and destinations may be discussed, including geographic information, transportation information, tourist destination information, and event information. Must be gathered from various sources, such as information, and associated and reviewed.

 Creating a topic net in this way is more efficient than preparing a dialog scenario by assuming all dialog flows, but it is difficult to create it by hand alone. Also, even if it is created manually, the scope of the dialogue that the topic network is likely to become insufficient tends to be insufficient.

 Also, if the topic net includes a specific topic, the conversation using the topic net becomes very effective. For the user, it is a dialog with an actual feeling that is close to his or her own viewpoint. As a source for collecting individual topics, information on individual contents written from an individual's perspective, such as travelogues and product reviews, can be considered. Such information is now often handled on computers. However, they are usually just text data or image data, so it is difficult to process on a computer as it is.

 On the other hand, table-based databases, mainly relational databases, are used in many systems. Many types of data, including, for example, product catalog information, store information, geographic information, tourist information, travel information, and bulletin board information, are handled in table format. In recent years, data in XML (eXtensible Markup Language) format has been frequently used. XML format data is a kind of data in the form of a file. Each node of the tree has a “content” and a “type”. "Types" are called tags in XML. Since the data described in XML contains the description of the data content itself and its structure is flexible, the use of XML has been promoted mainly for the purpose of distributing data. In fact, ML format data formats have been defined in many fields, including e-commerce, publishing, travel, and maps. Data that has been handled in relational databases up to now is also being converted to XML for distribution.

 An object of the present invention is to provide a technology for efficiently creating a topic net that can respond to a wide range of conversations with a small amount of labor.

Another object of the present invention is to provide a method for expanding an information collection source for generating a topic net. (Disclosure of the Invention)

 In order to solve the above-described problem, one embodiment of the present invention provides a method for generating a topic net including nodes associated with topics and links connecting the nodes. The method includes the following steps.

 A reading step of reading one record of data from a data table in which field data described in a plurality of fields are stored in association with each other; a topic class corresponding to all or a part of the plurality of fields; An association between the classes and a definition step for defining

 And a node generating step of generating a node corresponding to field data having a corresponding topic class among the data read in the reading step, and a topic corresponding to the node,

 • A link generation step of linking each node generated in the node generation step according to the association between the topic classes defined in the definition step.

 For example, consider a data table that stores field data corresponding to the fields “region”, “country”, and “city” as one record. Topic classes “Region”, “Country”, and “City” with the same name as each field are defined as topic classes corresponding to each field. As an association between topic classes, it is defined that links are created from “region” to “country” and from “country” to “city”. It is assumed that the read data is, for example, “Europe” as a region, “Italy” as a country, and “Milan” as a city. Then, nodes corresponding to "Europe", "Italy", and "Milan" are generated. Topics corresponding to each node are "Europe", "Italy", and "Milan". Next, links are created from "Europe" to "Italy" and "Italy" to "Milan". This is repeated for all data in the data table to generate a topic net. Also, one topic net can be generated from multiple data tapes.

 The second embodiment of the present invention provides a topic net generation method including nodes associated with topics and links connecting the nodes. The method includes the following steps.

• Field data described in a plurality of fields and the field has Reading data for one tree from the data file defining the tree structure and

 A definition step for defining topic classes corresponding to all or some of the plurality of fields and associations between the topic classes;

 A node generation step of generating a node corresponding to field data having a corresponding topic class among the data read in the reading step, and a topic corresponding to the node;

 A link generation step of linking the nodes generated in the node generation step according to the association between the topics defined in the definition step;

 Examples of data storage means for defining tree-structured data include an XML file and a database for storing the files. For example, it is assumed that a tree structure is defined under the field “sightseeing place” and the fields “location city” and “type”. Topic classes corresponding to each field "Tourist", "City of location", and "Type" are defined to generate links from "City of location" to "Tourist" and "Tourist" to "Type". Is done. Here, it is assumed that the read data is, for example, “Dowomo” as a tourist destination, “Milano” as a location city, and “Historic building” as a type. Then, nodes corresponding to “Dowomo”, “Milan”, and “Historical building” are generated. Topics corresponding to each node are “Dowomo”, “Milan”, and “Historical building”. Next, links from “Milan” to “Dowomo J” and “Dowomo” to “歷 Historic Building” are generated. By repeating this process, topic nets are automatically generated based on existing data files.

 In the above two embodiments, it is preferable that the node generation step includes a step of stopping generation of a node having a duplicate topic.

 Further, the definition step may further define a dialog information generation rule for generating dialog information including a message used to progress a dialog based on the topic net. The dialog information generation rule can generate the dialog information based on field data described in fields corresponding to the two topic classes associated in the definition step.

As an example of the dialogue information generation rule, “Kukukoku is recommended. I can do it. This conversation information generation rule generates conversation information based on field data described in fields “country” and “city” corresponding to topic classes “country” and “city”. If "Country" is Italy and "City" is Milan, the message generated will be "Milan is recommended in Italy." The dialogue information containing this message is linked to a link to the topic "Italy" from "Milan".

 If the topic class corresponding to all the fields included in one record is not defined, it is acceptable to use the field data described in the field that does not have a corresponding topic class to generate the conversation information. For example, suppose that the above data table includes “famousness” in addition to “region”, “country”, and “city”. It is also assumed that the topic class corresponding to “famousness” has not been defined. Even in this case, for example, the dialog information generation rule can be defined so that the generated message changes depending on whether the reputation is 7 or less or more than 7.

 The dialogue information generation rule may further include a direction in which the generated dialogue information indicates a force used to progress the dialogue from any one of the nodes of the two topic classes to any one of the nodes. Good.

 For example, the above-mentioned dialogue information generation rule “<Country> City> is recommended.” Includes the topic class “Country” to “City”. As a result, the topic of conversation using topic nets progresses from “country” to more specific “city”.

 The interaction information generation rule may further include a message type indicating what property the message has.

 For example, a dialog information template for generating a question message includes the message type “question”. Further, for example, in the case of a dialogue information template for generating an explanation message, the message type “explanation” is included.

 A topic net generation apparatus, a program, and a computer-readable recording medium on which the program is executed are included in the scope of the present invention. Here, examples of the recording medium include a flexible disk, a hard disk, a semiconductor memory, a CD-ROM, a DVD ヽ magneto-optical disk (MO), which can be read and written by a computer, and others.

In a third embodiment of the present invention, a node associated with a topic is connected to each node. And a method of generating a topic net including a link to the topic. This method includes the following steps. In the method, the data unit includes a text message and a data type indicating a property of the text message.

 A read step for reading the data units stored in association with the start topic and the end topic,

 A node generation step of generating nodes corresponding to the read start topic and end topic, and a topic corresponding to the node;

 A link generating step of generating a link from the node corresponding to the start topic to the node corresponding to the end topic,

 Generating, on the basis of the data unit, conversation information including a message used to progress a dialogue based on the topic net and a message type indicating a property of the message based on the data unit; A conversation information generation step to associate with the generated link.

 By using this method, topic net 1 can be generated based on, for example, travel reports created by individuals.

 In the third embodiment, the link generation step includes generating a link when the node corresponding to the start topic and the node corresponding to the end topic are the same node. You may stop it. In the dialog information generating step, when the generation of the link is stopped, the generated dialog information may be associated with the node.

 In this case, the generated dialogue information is related not to links but to nodes, that is, topics.

 In the third embodiment, the data unit may further include additional information. The dialog information generating step includes, when there are a plurality of data units including the same data type for the same link or the same node, select one of the data units based on the additional information, and perform a dialog based on the selected data unit. Information can be generated.

 For example, if the additional information is the date and time when the data unit was created, the interactive information can be generated based on the latest data unit.

In the third embodiment, the interactive information generation step includes the step of adding The interactive information can be generated based on the interactive information.

 For example, using the creation date and time of the text data included in the data unit is used as the additional information. If the date and time of creation is old, you can add a preface to the beginning of the message created based on the text data, "It's a bit of an old story."

 In the third embodiment, it is preferable that the node generation step includes a step of stopping generation of a node whose topic is duplicated! / ,.

 A topic net generation device, a program, and a computer-readable recording medium on which the program is executed are included in the scope of the present invention.

(Brief description of drawings)

 FIG. 1 is an explanatory diagram showing the configuration of a dialogue device to which the present invention is applied.

 FIG. 2A is an explanatory diagram showing an example of a topic net.

 FIG. 2 (b) is an explanatory diagram showing an example of the dialog information in the topic net shown in FIG. 2 (a).

 FIG. 3A is an explanatory diagram showing an example of data in a table format that is a source of a topic net.

 FIG. 3 (b) is an explanatory diagram showing an example of a topic class definition and a definition example of an association between topic classes.

 FIG. 4A is an explanatory diagram showing that a dialogue information template is defined for a link that associates a topic class with a topic class.

 FIG. 4 (b) is an explanatory diagram showing an example of the dialog information template shown in FIG. 4 (a).

 FIG. 5 is a flowchart showing an example of a flow of a generation process for generating a topic net based on data in a table format.

 6 and 7 are explanatory diagrams of the generation processing for generating the topic net of FIG. 2 (a) according to the flowchart shown in FIG.

 FIG. 8A is an explanatory diagram showing an example of another topic net.

Fig. 8 (b) shows an example of dialog information in the topic net shown in Fig. 8 (b). FIG.

 FIG. 9 is an explanatory diagram showing an example of the data in the form of a file which is the source of a topic net. FIG. 10 (a) is an explanatory diagram showing the definition of the tree structure.

 Figure 10 (b) shows an example of a topic class definition, an example of the definition of association between topic classes, and an explanation showing that a dialogue information template is defined for the link that associates topic classes with topic classes. FIG.

 FIGS. 11A and 11B are explanatory diagrams showing an example of the interaction information template shown in FIG. 10B.

 FIG. 12 is a flowchart showing an example of a flow of a generation process for generating a topic net based on data in the format of a file.

 FIG. 13 is an explanatory diagram of the generation processing for generating the topic net of FIG. 8A according to the flowchart shown in FIG.

 Figure 14 (a) is a conceptual explanatory diagram of topic data associated with a topic on a topic net.

 FIG. 14 (b) is an explanatory diagram showing that the dialog information is generated using the dialog generation script.

 FIG. 15 (a) is an explanatory diagram showing a specific example of topic data in the topic net of FIG. 8 (a).

 FIG. 15 (b) is an explanatory diagram showing an example of the dialogue generation script for generating the topic net of FIG. 8 (a).

 FIG. 16 is an explanatory diagram showing an example of another data format serving as a source of a topic net. FIG. 17 is an explanatory diagram showing fragments of topic nets generated based on the data shown in FIG.

 FIG. 18 is a flowchart showing an example of the flow of a generation process for generating a topic net based on the data shown in FIG.

(Best mode for carrying out the invention)

 Overall configuration>

FIG. 1 shows a configuration of a dialogue system to which the present invention is applied. Dialogue system, dialogue It includes an apparatus 10 and a GUI (Graphical User Interface) 20. The interactive device 10 and the GUI 20 operate on a computer connected to input / output means such as a mouse, a display, and a touch panel.

 The dialogue device 10 includes a topic net management unit 1 that generates a topic net, and a dialog control unit 2 that progresses a dialog based on the topic net. The topic net management unit 1 can access databases 30 a, b, and c for generating a topic net. The databases 30 a, b, and c may be built in a hard disk of a computer on which the interactive device 10 is mounted. Further, the databases 30 a, b, and c may be constructed so that the interactive device 10 can be accessed via the network 40. The data stored in the databases 30a, 30b, 30c is not particularly limited, such as a table format, a tree format, and an XML format, as described later. One topic net may be generated from multiple types of data.

 The generation of the topic net by the topic net management unit 1 can be collectively performed before the user operating the GUI 20 and the dialogue device 10 start the dialogue. Also, topic nets may be generated little by little as the dialogue progresses. When generating a topic net as the dialogue progresses, it is necessary to generate a topical net that includes the topic needed as well as the surrounding topics as the dialogue progresses. In this way, unnecessary topic nets can be prevented from being generated, and computer resources are not wasted for storing unnecessary data. Further, the topic net management unit 1 may store the creation date and time of the topic net that has already been generated, and delete an old part after a certain period of time. In this way, the topic net can be kept up to date. Further, the topic net management unit 1 may store the frequency of use of topics included in the topic net and the frequency of use of dialog information described later. As a result, it is possible to detect a part of the topic net that is used infrequently and delete such a part from the topic net, so that the entire topic net can always be used effectively.

 <Topic Net>

Next, a topic net generated by the present invention and its use will be described. Figure 2 is a conceptual illustration of topic nets. The topic net includes nodes and links connecting the nodes, as shown in FIG. Each node has a topic and The topic ID is being spoken. The topic ID is an identifier that identifies the node of the topic net. For example, a node with topic “Rome” is identified by topic ID “TP111”. Dialog information is associated with each topic or link. Dialogue information is information that includes a text message for promoting a dialogue using a topic net. In dialogue using topic nets, the dialogue progresses as the topic changes along the links connecting the topics. In the example of Fig. 2, the links connecting the topics have directions. This direction indicates that the topic at the link destination is a more specific topic than the topic at the link source. However, depending on the topic net, the direction of the link may have a different meaning. In some cases, the link need not necessarily have a direction. .

 FIG. 2B shows an example of the dialog information associated with the link. This example shows the dialogue information associated with a link to topic "Italy" ^ topic "Rome". The text message included in the conversation information is specifically divided into a system message and a user message. The system message is a text message that the interactive device 10 outputs to the user. The user message is a text message that the user can select as his / her own utterance with respect to the interactive device 10. The conversation information includes either a message flag "s" indicating a system message or a message flag "u" indicating a user message. The conversation information may include not only text messages but also image files and audio files.

Also, as shown in this example, the conversation information includes a message type indicating the nature of the text message. In this figure, “message”, “recommended explanation”, “question”, “positive answer”, and “negative answer” are illustrated as message types. In addition, the dialogue information associated with the link includes a “direction” from which topic to which topic the text message transitions to. In other words, "direction" indicates which direction the text message is used to enter the link. The conversation information is used, for example, as follows. Assume that the dialogue device 10 outputs one of the system messages associated with the link from the topic "Italy" to the topic "Rome", "Do you want to go to Rome in Italy?" In contrast, If he chooses the affirmative answer “I want to go to Rome”, the topic goes from “Italy” to “Rome”. The conversation progresses as the topic changes in this way.

 Example of the first embodiment>

 Next, a method of generating a topic net by the topic net management unit 1 of the interactive device 10 will be specifically described. In the present embodiment, a method of generating a topic net based on data in a table format will be described. For ease of explanation, a case where the topic net shown in the above 囪 2 is generated will be described. A topic class is associated with each node of the topic net generated by the following method. Topic classes are "region", "country", and "city" in the example of this figure. Topic classes are defined in advance before generating topic nets. The details of the topic class will be described later.

 FIG. 3 (a) shows an example of tabular data from which the topic net of FIG. 2 is generated. Here, two tables are created. In one table, “region” and “country” are stored in one record in association with each other. In the other table, “country” and “city” are stored in one record in association with each other. FIG. 3 (b) illustrates the definition of a topic class associated with all or part of the table fields for such a table. This figure also illustrates the definition of association between topic classes. In this example, topic classes corresponding to all fields are defined. In this example, the field names “region”, “country”, and “city” are the same as the topic class names “region”, “country”, and “city”. In addition, the definition between topic classes is defined to generate a “link” from “region” to “country” and a “link” from “country” to “city”.

According to the definition of the topic class and the definition of the association between the topic classes, a node of the topic net, a topic of the node, and a link connecting the node are generated. For example, nodes corresponding to the topic classes “Country” and “City”, topics of each node, and a link from “Country” to “City” are generated. More specifically, nodes having topics "Italy" and topic "Rome" are generated from the field data "Italy" and "Rome" forming one record, respectively. The topic "Italy" Generates a link to the topic "Rome". Here, "Italy" and "Rome" correspond to the topic classes "Country" and "City", respectively.

 Figure 4 (a) shows that the dialog information template is further defined for each link that defines between topic classes. The dialogue information template generates dialogue information that associates each topic with a link connecting the topics. Figure 4 (b) is an example of a dialog information template defined for a link from the topic class “Country” to the topic class “City”. The dialog information template includes a “template”, a “message type”, a “direction”, and a “message flag” for generating a text message.

 “Template” is defined so that a text message is completed based on field data described in a field in one record. The field data corresponds to the topic class of the node connected by the link with which the generated dialogue information is associated. For example, the data is embedded in <Country> and <Tokyo> of the template “<Country> is famous for <City>”, and the text message is completed. For example, in <country>, data of the field “country” corresponding to the topic class “country”, for example, “Italy” is embedded. The data of the field “city” corresponding to the topic class “city”, for example, “Rome” is embedded in “kuto”. “Message type” specifies the message type of the dialog information generated by the template. “Direction” defines the direction of the dialog information generated by the template. “System flag” specifies the system flag r s / Uj of the text message generated by the template.

 FIG. 5 is a flowchart illustrating an example of the flow of a generation process performed by the topic net management unit 1. This generation process generates topic nets from tabular data.

 Step S1: First, the topic net management unit 1 reads data for one record from a set of data in the table format.

Step S2: Next, the topic net management unit 1 generates a node for the field data in which the corresponding topic class exists among the field data included in the read record, based on the definition of the language class. It also generates a topic corresponding to that node. In this embodiment, the topic is the same as the field data. Also, the topic net management unit 1 does not newly generate a node having an overlapping topic if a node having the same topic has already been generated.

 Step S3: The topic net manager 1 links the generated topics according to the definition between topic classes.

 Step S4: Further, the topic net management unit 1 uses the dialogue information template to generate dialogue information associated with the generated link.

 Step S5: The topic net management unit 1 performs the processes of steps S1 to S4 for all records included in the table format data.

 Step S 6: Topic net 1, when there are a plurality of tables for generating the topic net, the management unit 1 performs the processing of steps S 1 to S 5 for all tables. FIGS. 6 and 7 are explanatory diagrams showing a process in which the topic net management unit 1 generates a topic net based on the flowchart of FIG. For ease of explanation, table data, topic class definitions, associations between topic classes, and dialog information templates are as shown in FIGS.

 First, one record is read from one table. This record contains the region "Europe" and the country "Italy" (Figure 6 (a)). Next, nodes and topics are generated for the field data "Europe" and "Italy" corresponding to the topic classes "Region" and "Country" (Fig. 2 (b)).

 The generated nodes are then linked according to the definition of the link from “region” to “country”. In other words, a link from the topic "Europe" to the topic "Italy" is generated (Fig. 3 (c)). Dialog information is generated for the generated link based on the template for linking "region" to "country" ((d) in the same figure).

 The same process is repeated for the next record that includes “Just one bite” and “France”. However, since a node having the topic “Europe” has already been generated, a new node and topic based on the field data “Europe” are not generated. Similarly, for records containing the region “Asia”, especially “Japan”, the topics “Asia” and “Japan” are generated, and links connecting the topics and dialog information are generated (Fig. ) To Figure 7 (h)).

When all data in the table format that associates "region" and "country" is read, Next, data in a table format in which “country” and “city” are associated with each other are sequentially read (FIG. 7 (i)). For example, the first records in this table, "Italy" and "Rome," are read. Since a node with topic "Italy" has already been generated, a new node and topic based on field data "Italy" are not generated. Then, the node for the field data “Rome” and its topic “Rome” are newly generated because they do not yet exist (Fig. (J)). Then, according to the definition of the link from "country" to "city", the topic "Italy" and the topic "Rome" are linked (Fig. (K)). After that, conversation information is generated based on the template for linking “country” to “city” ((1) in the same figure). This process is repeated for all records to generate a topic net.

 As described above, if table data fields and topic classes and definitions between topic classes are defined, topics of topic nets can be automatically generated simply by preparing table records. it can. If there is table format data created for another purpose, topic nets can be automatically generated by defining topic classes accordingly. In this way, a topic net having a huge number of topics can be efficiently generated using an existing database. .

 <Second embodiment example>

 Next, a method in which the topic net management unit 1 of the interactive device 10 generates a topic net based on tree-form data will be described. For ease of explanation, consider the case of generating the topic net shown in Fig. 8. Figure 8 (a) shows an example of another topic net. This topic net includes nodes and directional links connecting the nodes. Each node is associated with a topic, a topic ID, and a topic class. Dialog information is associated with each link.

FIG. 8 (b) is an explanatory diagram showing an example of the dialog information included in the topic net of FIG. 8 (a). In this example, the dialogue information associated with the link to the topic "Milan" power ^ the topic "Dowomo" and the dialogue information associated with the link from the topic "Dowomo" to the topic "Historic building" are shown. ing. Similar dialog information is associated with other links. FIG. 9 shows an example of the data in the form of a file which is the source of the topic net shown in FIG. In this example, data in a tree format is described in an XML file. In the XML file, you can define what tags and the values of the enclosing attributes are located in the tree. In this figure, the part between the tag and the sightseeing spot is equivalent to one record in table format data, and has a clear structure. Note that a set of data in the tree format can be described not only in an XML file but also in an ordinary text file, an SGML (Standard Generalized Markup Language) file, an HTML (Hyper Text Markup Language) file, and the like. The format is not particularly limited.

 FIG. 10A is an explanatory diagram showing a tree structure described by the XML file of FIG. The tree structure of the XML file is based on the tag “Tourist” in the XML file,

"City of location", "type", "fee", and "famousness" are formed by having a hierarchy. In the tree of the XML file in Fig. 9, the tags "City of location", "Type", "Fare", and "Famousness" are located under the tag "Tourist". Figure 10 (b) illustrates the definition of topic classes and their associations. In this figure, the definition of the topic classes and their associations is the definition for the tree structure in FIG. In this example, topic classes corresponding to three of the five types of tags included in the XML file are defined. More specifically, the topic class “Tourist” corresponds to the tag tourist area “Tourist” The topic class “City” corresponds to the city located in the tag area, and the topic class “Type” corresponds to the tag type 1S Each is defined. Topic classes corresponding to the tag rates> and famous names> in the XML file are not defined. The link between the three topic classes, “City” to “Tourist Spot” and “Tourist Spot” to “Type”, is defined as the topic class. The links connecting topic classes are preferably defined to correspond to the tree structure, but they do not necessarily need to correspond.

According to the definition of the topic class and the definition of the association between the topic classes, the nodes of the topic net, the topics of the nodes, and the links connecting the nodes are generated. For example, nodes corresponding to the topic classes “city” and “tourist spot”, topics of each node, and links from “city” to “tourist spot” are generated. More specifically, one From the field data “Dowomo” and “Milan” that form the tree, nodes having the topic “Dowomo” and the topic “Milan” are generated, respectively. In addition, a link from the topic “Mirano” to the topic “Dowomo” is generated. Here, “Dowomo” and “Mirano” correspond to the topic classes “Tourist” and “City”, respectively.

 Figure 10 (b) further shows the definition of the dialogue information template in addition to the topic classes and the definitions between topic classes. In this figure, the dialogue information template 1 is defined for the link from the topic class “city” to “sightseeing spot”. The dialogue information template 2 is defined for the link to the topic class “sightseeing spot” or “type”. FIGS. 11 (a) and (b) show examples of the dialog information templates 1 and 2, respectively. Like the dialog information template in the table format data, the dialog information template includes a template, a message type, a direction, and a message flag.

 FIG. 12 is a flowchart illustrating an example of a flow of a generation process in which the topic net management unit 1 generates a topic net from the data in the clear format.

 Step S11: First, the topic net management unit 1 reads data for one tree from a tree-type data set such as an XML file (S11).

 Step S12: Next, the topic net management unit 1 determines the nodes of the topic net and their topics for each node forming the tree, for example, for each tag of the XML file for which the corresponding topic class exists. Generate. The topic of a node is the genus of the node in the tree. If a node having the same topic as the attribute value of the node in the tree has already been generated, it is not necessary to generate a new node and topic.

 Step S13: The topic net manager 1 links the generated topics according to the definition of the association between the topic classes.

 Step S14: Further, the topic net manager 1 uses the dialog information template defined for each link to generate the dialog information associated with the generated link.

Step S15: Next, the topic net manager 1 determines whether there is any data that has not been read out of the data set, for example, the XML file, and , The processes of steps S11 to S14 are repeated.

 Step S16: The topic net management unit 1 executes the processing of steps S11 to S15 for all tree-based data sets that are the source of the topic net. For example, one topic net can be generated based on multiple XML files, or one topic net can be generated based on an XML file and an HTML file.

 FIG. 13 is an explanatory diagram showing a process in which the topic net management unit 1 generates a topic net based on the flowchart of FIG. For ease of explanation, it is assumed that the tree format data, the definition of topic classes, the definition of association between topic classes, and the definition of the dialogue information template are as shown in FIGS.

 First, the data for one byte is read from the XML file (Fig. 13 (a)). For example, it reads out the tourist destination "Dowomo", the city of the location "Milan", the type "Historic building", the fee "None", and the reputation "10". Next, nodes and their topics are generated for the attribute values of the tags that have a corresponding topic class among the tags that make up the tree (Fig. 3 (b)). The attribute value itself of the tag corresponding to the topic class becomes the topic of the node corresponding to the tag. For example, the topic of the node generated for the tag tourist attraction> corresponding to the topic class “tourism” is “domo”. Also, the topic of the node generated for the tag <city> corresponding to the topic class “city” is “Milan”.

Next, the generated topics are linked according to the definition of the association between the topic classes (FIG. 9C). As a result, a link from the topic “Milan” to the topic “Dowomo” and a link and power from the topic “Dowomo” to the topic “Historic building” are generated. Conversation information is generated for the generated link according to the conversation information template (Fig. (D)). This is repeated for all data in the XML format contained in the XML file. When one topic net is generated based on a plurality of tree format data files, this process is repeated for each file. As a result, a topic net can be generated based on the data in the format. As described above, by defining topic classes and topic classes in accordance with the set of data in the form of the form, it is possible to form a topic net containing a huge number of topics. Automatically generated from formula data.

 <Third embodiment example>

 The method for generating the dialog information is not limited to the generation using the dialog information template. In the present embodiment, the topic net management unit 1 generates the conversation information based on the dialog generation stabilization. For ease of explanation, generation of conversation information of the topic net shown in FIG. 8 will be described as an example.

 First, as shown in Fig. 14 (a), nodes that make up the topic net are associated not only with the topic ID, topic, and topic class, but also with attribute information. Then, as shown in Fig. 14 (b), a dialogue generation script is defined for each link. This dialogue generation script generates dialogue information based on topic, topic class, and attribute information. The attribute information can be extracted from a data set in a table format.

 FIG. 15 (a) is an explanatory diagram showing a specific example of the attribute information. The node TP 101 and the node TP 112 each have “type”, “fee”, and “famousness” as attribute information. In this example, the tags and attribute values that do not have a corresponding topic class among the tags and their attribute values in the XML file shown in Fig. 9 correspond to the tags at the top of the library structure. It is used for the attribute information of the node. In other words, tag type>, price>, and famousness> and their attribute values are attribute information of each node corresponding to tag sightseeing spot> at the top of the tree structure.

 FIG. 15 (b) is an explanatory diagram showing a specific example of the dialogue generation script. This example shows a dialogue generation script for generating dialogue information based on the data shown in FIG. 15 (a). This dialog generation script is defined for the Vunk from the topic class "city" to "sightseeing spot". The dialog script also generates a text message based on the value of the topic class “Tourist spot” and “City” and the attribute information “Famousness”.

The dialogue generation script can generate the dialogue information more flexibly than the dialogue information template. The generation of dialogue information using the dialogue generation script can be applied to the case where a topic net is generated from the data in the above-mentioned table format and the case where a topic net is generated from data in the clear format. Also, the format of both It can also be applied to the case of generating a topic net from a combination of the above data. <Fourth embodiment example>

 The topic net management unit 1 can generate a topic net based on a set of data other than data in a table format data format. In order to conduct an effective dialogue using the topic net, it is preferable to give the topic net specific topics based on individual experiences. The data that provides such a topic is, for example, if a topic net for product search is generated, a review article of the product or a comment on a bulletin board. Also, if you are creating a topic net for deciding a travel place, you can use a travel journal. In addition, if you are generating a topical net for searching for restaurants, you can mention restaurants. Therefore, a method of generating a topic net based on such data will be described. In the following, for the sake of simplicity, an example will be described in which a travel net is used to generate a topic net.

 FIG. 16 is an explanatory diagram showing another example of a data set serving as a source of a topic net. In this data set, “data unit”, “start topic”, and “end topic” are stored in association with each other. The data unit contains a text message generated by dividing an individual's travel diary into appropriate lengths as a single message during a conversation. Rather than generating a data unit by dividing a travel book that has already been written, it is also possible to use a data unit of a written travel book while creating a data set. In this example, the data unit is text data, but may be image data or audio data.

The starting topic is a topic that triggers the output of a data unit text message. The ending topic is the topic that is reached according to the contents of the data unit text message. As shown in FIG. 16, one of the start topic and the end topic may be plural. If there are multiple start topics, the same text message is output from multiple topics, which means that the dialogue shifts to the same topic. Conversely, if there are multiple ending topics, the same text message is output from the same topic, which means that the conversation shifts to one of the multiple topics. The data unit includes a message type indicating the classification of the text message. Have been. The message type is set by selecting one of the messages ¾¾ [J included in the conversation information. Further, the data unit may include other additional information. For example, the date and time of creation of the text message and the creator of the text message may be included in the report.

 FIG. 17 is an explanatory diagram showing a method of generating a topic net fragment based on the data set shown in FIG. First, one set of the start topic, data unit, and end topic is read. If there is more than one start topic or end topic, multiple start topics or end topics are read together. Next, a node corresponding to the read start topic and end topic and a topic of the node are generated. If there is already a node with the same topic as the new node to be generated, no new node with the same topic is generated. The topic is the read start topic or end topic itself. In addition, a link that connects nodes from the start topic to the end topic is generated. Furthermore, dialog information to be associated with the generated link is generated based on the data unit. Here, the message type of the conversation information is the message type of the data unit. In addition, the text message of the conversation information is the text message included in the data unit. If the start topic and the end topic are the same, the conversation information generated based on the data unit is associated with the node that has that topic. If there are multiple start topics or end topics, there will be multiple links originating from the start topic or links to the end topic. Such multiple links are associated with dialog information generated from the same data unit. In this way, fragments of topic nets are generated, and each fragment is connected to generate one topic net.

By providing the data unit with additional information, it is possible to change the generated dialog information, not the text message itself contained in the data unit. For example, consider the case where the data unit includes the date and time of creation of the text message. In this case, the creation date and time are checked at the stage of generating the conversation information message. If the information is older than a certain amount of time, it is possible to create a conversational information message that connects the introductory text and the text message. Also, for the same topic or link, If there is more than one data unit including the message type, one of the data units can be selected based on the additional information. Accordingly, the selected Detayuni: may generate the interactive information based on Tsu bets. For example, if the date and time of creation are included in the data unit, it is possible to generate conversation information based on the latest data unit. For example, if the creator of the text message in the data unit is provided as additional information, a text message of a popular creator can be preferentially used as a source of conversation information.

 FIG. 18 is a flowchart illustrating an example of a flow of a generation process in which the topic management unit 1 generates a topic net based on text data.

 Step S21: First, the topic net manager 1 reads one set of the start topic, data unit, and end topic.

 Step S22: Next, the topic net management unit 1 generates a node whose topics are the start topic and the end topic. If the topic node has already been generated, no new node with the same topic is generated.

 Step S23: The topic net manager 1 generates a link from the node corresponding to the start topic to the node corresponding to the end topic.

 Step S24: The topic net management unit 1 further generates conversation information based on the text message / message type included in the data unit.

Step S 2 5: topic network management unit 1, the processing of the step S 2 1 to S 2 4, Repetitive all set; return Ri.

 Through the above processing, a topic net is generated based on data that does not have a table format, for example, data such as travel reports and product reviews created by individuals, and a topic net that includes individual and specific topics is generated. can do. (Industrial applicability)

 According to the present invention, a topic net including a wide range of topics can be efficiently created. Further, the present invention can generate a topic net based on a wide range of information sources. If the topic net generated by using the present invention is applied to a dialogue system for interacting with humans, a dialogue having abundant topics and natural context can be executed.

Claims

The scope of the claims
1. A topic net generation method including nodes associated with topics and links connecting the nodes,
 A reading step of reading one record of data from a data table storing field data described in a plurality of fields in association with each other; a topic class corresponding to all or a part of the plurality of fields; And a definition step for defining
 A node for field data in which a corresponding topic class exists among the data read in the reading step, and a topic corresponding to the node;
 A link generation step of linking each node generated in the node generation step in accordance with the association between the topic classes defined in the definition step.
2. The topic net generation method according to claim 1, wherein the node generation step includes a step of stopping generation of a node having a duplicate topic.
 3. The defining step further defines a dialogue information generation rule for generating dialogue information including a message used for advancing the dialogue based on the topic net,
 The dialog information generation rule generates the dialog information based on field data described in fields corresponding to the two topic classes associated in the defining step.
 A method for generating a topic net according to claim 1.
4. The dialog information generation rule further includes a direction indicating which of the nodes of the two topic classes the generated dialog information is used for the progress of the dialog from which node. The topic net generation method according to claim 3.
5. The topic net generation method according to claim 3, wherein the interaction information generation rule further includes a message type indicating what kind of property the message has.
6. A generating apparatus for generating a topic net including nodes associated with topics and links connecting the nodes,
 Reading means for reading one record of data from a data table storing field data described in a plurality of fields in association with each other;
 Definition means for defining topic classes corresponding to all or a part of the plurality of fields, and association between the topic classes;
 A node for field data in which a corresponding topic class exists in the data read by the reading unit; and a node corresponding to the node;
 Link generation means for linking each node generated by the node generation means according to the association between the topic classes defined in the definition step;
 Topic net generation device provided with.
7. A generation program for generating a topic net including nodes associated with topics and links connecting the nodes,
 A reading step of reading one record of data from a data table storing field data described in a plurality of fields in association with each other; a topic class corresponding to all or a part of the plurality of fields; And a definition step for defining
 A node for field data in which a corresponding topic class exists among the data read in the reading step, and a topic corresponding to the node;
 A link generation step of linking each node generated in the node generation step according to the association between the topic classes defined in the definition step; and a computer for executing a topic net generation program.
8. A method for generating a topic net including nodes associated with topics and links connecting the nodes,
Reads one tree of data from the data file that defines the field data described in a plurality of fields and the tree structure of the field. A definition step of defining topic classes corresponding to all or a part of the plurality of fields, and association between the topic classes;
 A node for field data in which a corresponding topic class exists among the data read in the reading step, and a topic corresponding to the node;
 A link generation step of linking each node generated in the node generation step according to the association between the topics defined in the definition step;
 Topic net generation method including.
 9. The topic net generation method according to claim 8, wherein the node generation step includes a step of stopping generation of a node having a duplicate topic.
 10. The defining step further defines a dialogue information generation rule for generating dialogue information including a message used to advance a dialogue based on the topic net,
 The dialogue information generation rule generates the dialogue information based on field data described in fields corresponding to the two topic classes associated in the definition step.
 9. The topic net generation method according to claim 8.
 11. The dialogue information generation rule includes a direction indicating which of the nodes of the two topic classes the generated dialogue information is to be used for the progress of the dialogue from which node. 10. A method for generating a topic net according to claim 10.
12. The topic net generation method according to claim 10, wherein the interaction information generation rule further includes a message type indicating what kind of property the message has.
 13. A generating device for generating a topic net including nodes associated with topics and links connecting the nodes,
 Reading means for reading one minute of data from a data file defining field data described in a plurality of fields, a tree structure of the field, and
A topic class corresponding to all or a part of the plurality of fields; And a defining means for defining
 A node for field data in which a corresponding topic class exists in the data read by the reading unit; and a node corresponding to the node;
 Link generation means for linking each node generated by the node generation means according to the association between topics defined by the definition means;
 Topic net generation device provided with.
14 4. A generation program for generating a topic net including nodes associated with topics and links connecting the nodes,
 Reading data for one tree from a data file defining field data described in a plurality of fields and a tree structure of the field;
 A definition step of defining topic classes corresponding to all or a part of the plurality of fields, and association between the topic classes;
 A node for field data for which a corresponding topic class exists among the data read in the reading step, and a topic corresponding to the node;
 A link generation step of linking each node generated in the node generation step according to the association between the topics defined in the definition step;
 A program for generating topic nets that causes a computer to execute
 15 5. A method of generating a topic net including nodes associated with topics and links connecting the nodes,
 A reading step of reading the data unit stored in association with the start topic and the end topic;
 A node corresponding to the read start topic and end topic, a node corresponding to the node, and a node generation step of generating
 A link generation step of generating a link from the node corresponding to the start topic to the node corresponding to the end topic,
A message used to advance a dialogue based on the topic net and the message Generating conversation information including a message type indicating the nature of the message based on the data unit, and associating the link information with the link generated in the link generation step.
 The data cut includes a text message and a data type indicating a property of the text message.
 How to generate topic nets.
 16. The link generation step stops the generation of the link when the node corresponding to the start topic and the node corresponding to the end topic are the same node, and the interaction information generation step includes: If the generation is stopped, associate the generated conversation information with the node;
 A method for generating a topic net according to claim 15.
 1 7. The data unit further includes additional information,
 The dialog information generating step includes, when there are a plurality of data units including the same data type for the same link or the same node, selecting one of the data units based on the additional information, and generating the dialog information based on the selected data unit. Generate information,
 A method for generating a topic net according to claim 16.
 18. The data unit further includes additional information,
 16. The topic net generation method according to claim 15, wherein said dialogue information generating step generates said dialogue information based on said additional information.
 19. The topic net generation method according to claim 15, wherein said node generation step includes a step of stopping generation of a node whose topic is duplicated.
 20. A generation device for generating a topic net including nodes associated with topics and links connecting the nodes,
 Reading means for reading out the data unit stored in association with the start topic and the end topic,
 Node generation means for generating nodes corresponding to the read start topic and end topic, and topics corresponding to the nodes;
A link from a node corresponding to the start topic to a node corresponding to the end topic Link generating means for generating a link;
 Generating, based on the data unit, dialog information including a message used to progress a dialog based on the topic net and a message type indicating a property of the message, and associating the link information with the link generated by the link generating unit; And a dialogue information generating means.
 The data unit includes: a text message; and a data type indicating a property of the text message.
 Topic net generation device.
21. A generation program for generating a topic net including nodes associated with topics and links connecting the nodes,
 A reading step of reading out a text message, a data unit including a data type indicating the property of the text message, a start topic, and an end topic in association with
 A node corresponding to the read start topic and end topic, and a topic corresponding to the node;
 A link generating step of generating a link from a node corresponding to the start topic to a node corresponding to the end topic,
 Generating, based on the data cut, dialog information including a message used to advance a dialog based on the topic net and a message type indicating the nature of the message; and generating the link generated in the link generating step. A conversation information generation step relating to
 Topic net generation program that causes a computer to execute
PCT/JP2003/000610 2003-01-23 2003-01-23 Topic net generation method and apparatus WO2004066164A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/JP2003/000610 WO2004066164A1 (en) 2003-01-23 2003-01-23 Topic net generation method and apparatus

Applications Claiming Priority (3)

Application Number Priority Date Filing Date Title
JP2004567138A JPWO2004066164A1 (en) 2003-01-23 2003-01-23 Topic net generation method and apparatus
PCT/JP2003/000610 WO2004066164A1 (en) 2003-01-23 2003-01-23 Topic net generation method and apparatus
US11/121,001 US7756897B2 (en) 2003-01-23 2005-05-04 Topic net generation method and apparatus

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US11/121,001 Continuation US7756897B2 (en) 2003-01-23 2005-05-04 Topic net generation method and apparatus

Publications (1)

Publication Number Publication Date
WO2004066164A1 true WO2004066164A1 (en) 2004-08-05

Family

ID=32750592

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2003/000610 WO2004066164A1 (en) 2003-01-23 2003-01-23 Topic net generation method and apparatus

Country Status (2)

Country Link
JP (1) JPWO2004066164A1 (en)
WO (1) WO2004066164A1 (en)

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2001084394A1 (en) * 2000-04-28 2001-11-08 Fujitsu Limited Interactive control system

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2001084394A1 (en) * 2000-04-28 2001-11-08 Fujitsu Limited Interactive control system

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Shigeo TERABE et al., "Hasso Shien Taiwa ni okeru Kyochoteki Oto no tameno Hatsuwa Planning", Information Processing Society of Japan Kenkyu Hokoku (98-NL-124), 13 March 1998, Vol. 98, No. 21, pages 111 to 118 *
Tadahiro KITAHASHI et al., "User no Kyomi o Koryo shi Hasso Shien o Mezashita Oto Seisei Shuho ni tsuite", The Institute of Electronics, Information and Communication Engineers Kenkyu Hokoku (NLC96-4), 17 May 1996, Vol. 96, No. 46, pages 21 to 26 *

Also Published As

Publication number Publication date
JPWO2004066164A1 (en) 2006-05-18

Similar Documents

Publication Publication Date Title
Newcomb et al. The" HyTime" hypermedia/time-based document structuring language
Friedrich et al. Process model generation from natural language text
Bieber et al. Fourth generation hypermedia: some missing links for the World Wide Web
Josuttis SOA in practice: the art of distributed system design
US7712024B2 (en) Application program interfaces for semantically labeling strings and providing actions based on semantically labeled strings
US7260579B2 (en) Method and apparatus for accessing data within an electronic system by an external system
US7257574B2 (en) Navigational learning in a structured transaction processing system
US6275852B1 (en) Interactive computer network and method of operation
US5655130A (en) Method and apparatus for document production using a common document database
RU2439680C2 (en) Real-time xml data synchronisation between applications
CA2671284C (en) Generating end-user presentations from structured data
JP5879260B2 (en) Method and apparatus for analyzing content of microblog message
US9330144B2 (en) Tagging of facet elements in a facet tree
US7283988B1 (en) Code generator for a distributed processing system
US6338053B2 (en) Inventory managing method for automatic inventory retrieval and apparatus thereof
US7865560B2 (en) System for summarization of threads in electronic mail
Toye et al. SHARE: A methodology and environment for collaborative product development
US20030101065A1 (en) Method and apparatus for maintaining conversation threads in electronic mail
CN1542657B (en) Method for ensuring data compatibility when storing data item in database
US20020087530A1 (en) System and method for publishing, updating, navigating, and searching documents containing digital video data
US20040093217A1 (en) Method and system for automatically creating voice XML file
JP4339554B2 (en) System and method for creating and displaying a user interface for displaying hierarchical data
US7210096B2 (en) Methods and apparatus for constructing semantic models for document authoring
TW571204B (en) Content publication system for supporting real-time integration and processing of multimedia content including dynamic data, and method thereof
US20050120021A1 (en) Metadata driven intelligent data navigation

Legal Events

Date Code Title Description
AK Designated states

Kind code of ref document: A1

Designated state(s): JP US

WWE Wipo information: entry into national phase

Ref document number: 2004567138

Country of ref document: JP

WWE Wipo information: entry into national phase

Ref document number: 11121001

Country of ref document: US