JP2009053303A - Discussion knowledge graph constructing method, device, and program, and recording medium with program recorded thereon - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method and the like for constructing a discussion knowledge graph which can be utilized as an effective knowledge for solving a problem similar to the problem that has been under collaborative learning not only by the participants in the collaborative learning but also by other learners. <P>SOLUTION: The discussion knowledge graph constructing method is constituted of the steps of: (step S1) obtaining conversation history, annotation history indicating to which comment of each learner an annotation is added, and history of note strip indicating the difference by the prescribed time of a note recorded as his/her own solution; (step S2) extracting the note strip out of the history of note strip; (step S3) extracting a group of comments out of the conversation history on the basis of the annotation history; (step S4) correlating the extracted comment group to the extracted note strip on the basis of the annotation history; and (step S5) constructing node data and edge data of the discussion knowledge graph having an upper layer containing the utterance group as a hint to a problem and a lower layer containing the note strip as a solution to the hint. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a technology for supporting learning using a computer or a network, and in particular, visualizes knowledge for solving problems extracted from a learning history related to discussion of collaborative learning, which is group learning via a network. The present invention relates to a discussion knowledge graph construction method, a discussion knowledge graph construction device, a discussion knowledge graph construction program, and a recording medium on which the program is recorded.

  Collaborative learning (CSCL: Computer Supported Collaborative Learning) is a form of learning in which a plurality of learners cooperate to deal with one problem. FIG. 16 schematically shows a conventional collaborative learning system. The collaborative learning system 100 includes a server 102 and a plurality of (for example, three) clients 103 that are communicably connected to each other via a network NW. The server 102 includes a communication unit 111, a group data management unit 112, and a group learning data management unit 113. The server 102 stores information on learners participating in the collaborative learning in the group data storage unit 112a in advance. When the communication unit 111 receives a request for learner information from the client 103 via the network NW, group data management is performed. The requested learner information is retrieved by the unit 112 and returned. Further, when the server 102 receives a learner's statement (information input by the learner) from the client 103 by the communication unit 111, the server 102 transmits the received statement to all the clients 103 by the group learning data management unit 113. The history is stored in the group learning data storage means 113a.

  The client 103 is operated by a learner participating in collaborative learning, and includes a user interface 121, a communication unit 122, and a learner learning data management unit 123. The client 103 transmits a message input via the user interface 121 by the communication unit 122 to the server 102 and receives all messages on the network NW and displays them on the screen. That is, conversations are exchanged via the server 102. In addition, the client 103 uses the learner learning data management unit 123 to describe the learner's individual learning note (individual learning note history) input via the user interface 121 and the annotation (annotation history) added by the learner. ) Are displayed on the screen and stored in the learner learning data storage means 123a.

  FIG. 17 is a diagram schematically showing a window that is simultaneously displayed on the screen of the client. The client 103 displays a conversation window 1701 (see FIG. 18), an individual learning window 1702 (see FIG. 19), and an annotation window 1703 (see FIG. 20) on the screen. As shown in FIG. 18, the conversation window 1701 includes, for example, a conversation history display area having an ID 1801, a speaker 1802, and a message content 1803 as items for each message, a message input field 1804, and a message transmission button 1805. ing. As shown in FIG. 19, the individual learning window 1702 includes, for example, a text editor (individual learning note) 1901 that can freely describe a unique solution to a predetermined problem. Each learner inputs a question point or the like in the comment input field 1804 of the conversation window 1701 (see FIG. 18) and looks at the opinion displayed in the conversation history display area, and then the text editor (individual learning note shown in FIG. 19). ) 1901 describes the solution. As shown in FIG. 20, the annotation window 1703 includes, for example, a conversation history display area having an ID 2001, an annotation 2002, and an utterance content 2003 as an item for each utterance, an annotation input field 2004, and an annotation transmission button 2005. Yes. The learner gives annotations indicating the same viewpoint as his / her solution (for example, “a”) or annotations (for example, “b”) indicating a viewpoint different from his / her solution to the remarks that he / she finds useful. Can be added. Note that “2a” described in the annotation input field 2004 indicates that the ID is “2” and the annotation is the same viewpoint “a”.

  In the field of research on technologies that support collaborative learning, the development of technologies aimed at assisting during collaborative learning has been the main focus so far, but in recent years, technologies have also been developed to make use of the results of collaborative learning after collaborative learning. (For example, refer nonpatent literature 1 and nonpatent literature 2). The technology described in Non-Patent Document 1 and Non-Patent Document 2 enables the past collaborative learning to be reproduced from a specific time point by dividing the learning history into scenes according to specific criteria such as time intervals. Yes.

  On the other hand, not only collaborative learning technology, but if the target is limited to conversation history, in order to utilize the already obtained conversation history, the content of the discussion and the surface information of the speech (such as the type of speech and the context) There is also known a technique for analyzing the contents of all discussions from surface representations and keywords, extracting topics, and structuring the discussions for each topic (for example, see Non-Patent Document 3 and Non-Patent Document 4). . The technology described in Non-Patent Literature 3 and Non-Patent Literature 4 makes it easy to understand the content of discussions even when multiple topics are discussed at once or even when one topic is discretely discussed. .

Furthermore, in the collaborative learning technique, there is a technique for extracting a speech effective for problem solving for the learning participant based on the consciousness of the learning participant (see, for example, Non-Patent Document 5). In the technique described in Non-Patent Document 5, the remarks that the learning participants feel useful are based on the assumption that the remarks that are effective for the learning participants depend on the solution at the time of collaborative learning. On the other hand, let the learning participants add annotations with the same viewpoint as their own solution (for example, “a”) or annotations with a different viewpoint from their own solution (for example, “b”). From the above, comments with annotations are extracted and presented. This makes it possible to select comments that are useful for solving problems of learning participants by using the annotations of learning participants.
Mizue Kayama, Toshio Okamoto, “Study on Event Data Management for Learning Resource Sharing and Learning State Recovery in Collaborative Learning”, IEICE Transactions, January 2005, p.45-55 B. Goodman, M. Geier, L. Haverty, and R. McCready: “A Framework for Asynchronous Collaborative Learning and Problem Solving”, Artificial Intelligence in Education, JD Moore et al. (Eds), IOS Press (2001) J. Conklin and ML Begeman: “gIBIS: A Hypertext Tool for Exploratory Policy Discussion”, Proc. Of CSCW-1988, p.140-152 (1088) Masahiro Matsumura, Yuu Kato, Yukio Osawa, Mitsuru Ishizuka, "Discovery and Understanding of Issues by Visualizing the Structure of Discussion", Journal of the Japan Fuzzy Society, vol.15, no.5, p.554-564 (2003) M. Kakehi, T. Kojiri and T. Watanabe: “Annotation Interpretation of Collaborative Learning History for Self-learning”, Proc. Of Kes 2006, Part II, LNAI 4252, p.1062-1070 (2006)

  The technologies described in Non-Patent Document 1 and Non-Patent Document 2 focus only on the learning history in the shared space such as the conversation history and the operation history on the shared screen as the learning history utilized after the collaborative learning. Therefore, these conventional techniques cannot be used in a learning environment in which a learning participant derives a solution in the individual learning space of each individual learning participant based on the contents discussed in the shared space. This is because in such a learning environment, a conversation (conversation history) made in a shared space presents a hint for problem solving, and a solution is derived in each individual learning space. In other words, the conventional technique that focuses only on the learning history of the shared space only deals with problem solving hints. Therefore, it is difficult for these conventional techniques to collect sufficient problem-solving knowledge to support not only learning participants but also other learners.

  Further, for example, in the conventional technique for dividing a learning history into scenes as described in Non-Patent Document 1, a time interval or the like is used as a specific reference for dividing the learning history into scenes. Therefore, in this method, other learners who are trying to solve a problem similar to the content of the discussion (learning content) conducted by a collaborative learning participant about a certain problem will receive the result (learning history) of the collaborative learning. It cannot be used in a satisfactory manner. The reason is that the time interval is not directly related to the learning content. That is, it is difficult for other learners to search for a scene corresponding to a specific derivation step or knowledge from scenes divided according to criteria that do not depend on the learning content.

  Also, for example, the technology for structuring discussions (conversation history) for each topic as described in Non-Patent Document 3 and Non-Patent Document 4 structures all remarks for each topic. Other learners who are trying to solve the same problem as the content of the discussion (learning content) conducted by the learning participants cannot use the results of the collaborative learning (learning history) in a satisfactory manner. The reason is that other learners cannot distinguish between topics that are effective for problem solving and topics that are not necessary for problem solving.

  In addition, for example, the technology for structuring the discussion (conversation history) described in Non-Patent Document 5 for each topic is aimed at selecting effective remarks for the learning participants. It does not extract useful knowledge for problem solving.

  Therefore, the present invention has been made in view of the above-described problems, and solves the same problem as the problem that is not only collaborative learning participants but also other learners. The purpose is to provide a technology for constructing a discussion knowledge graph that can be used as effective knowledge.

  In order to solve the above-described problem, the discussion knowledge graph construction method according to the present invention is based on a statement that a plurality of learners input to each learner terminal about a predetermined problem and responds to other learners' remarks. When collaborative learning is performed while adding annotations indicating the same viewpoint or a different viewpoint from the solution method, information extracted from the discussion is obtained based on the information related to the discussion acquired from the respective learner terminals. A discussion knowledge graph construction method of a discussion knowledge graph construction device for constructing a discussion knowledge graph indicating a graph visualized as derivation knowledge of a solution to the predetermined problem, wherein the collaborative learning is performed as information related to the discussion by an information acquisition unit. A conversation history showing all the utterances entered during the acquisition, and for each learner, the learner Annotation history indicating whether the annotation has been added to the utterance of a note, and a history of note pieces indicating a difference in recorded contents for each predetermined time of a note recorded as a solution of the predetermined problem by the learner The information acquisition step of acquiring the message group, the message group extraction unit, the message group extraction step of extracting a message group including one or more messages from the acquired conversation history based on the annotation history, and the note piece extraction unit, Based on the annotation history, the note piece extracting step of extracting a predetermined note piece from the note piece history acquired for each learner, and the extracted comment group and the extracted note piece by the association means. The association step and the graph construction means, and Node data and edges for a graph having a higher layer that includes a group of statements associated with the hint as a hint to the predetermined problem and a lower layer that includes a note piece associated with the valid group of statements as a solution to the hint And a graph construction step for constructing data.

  According to such a procedure, the discussion knowledge graph construction device not only acquires the learning history (conversation history) of the shared space during the collaborative learning as in the conventional case, but also the annotation history and the learning history of the individual learning space for each learner. Also obtain (note piece history). Then, the discussion knowledge graph construction device associates the comment group extracted from the conversation history based on the annotation history and the note piece extracted from the note piece history based on the annotation history. Therefore, conventionally, only the hints included in the conversation during the collaborative learning can be extracted as knowledge by targeting the conversation history. By associating with individual learning histories as extraction targets, not only hints but also solutions derived by learning participants can be extracted as problem solving knowledge. The discussion knowledge graph construction apparatus constructs a graph having an upper layer including a comment group as a hint and a lower layer including a note piece as a solution to the hint as a discussion knowledge graph. As a result, the discussion knowledge graph becomes easy for the learner to search for a solution to the problem when visualized.

  Further, in the discussion knowledge graph construction method according to the present invention, the comment group extraction step extracts the comment added with the annotation from the acquired conversation history as an effective comment, and the extracted An utterance that calculates an average value of the number of utterances included in a group of utterances including valid utterances based on the number of utterances to which the annotation is added and the total number of utterances included in the acquired conversation history. A plurality of utterances each having the annotation added thereto in a utterance group including a number average value calculating step and a number of utterances including the extracted effective utterances and less than the calculated utterance number average value. When it is determined that there is no effective utterance determination step for determining whether or not there is a plurality of utterances each having the annotation added thereto, The effective speech group first extraction step for extracting the extracted effective speech as an effective speech group, and when it is determined that there are a plurality of speeches to which the annotations are respectively added, An effective speech group second extraction step of extracting, as an effective speech group, a speech group having the maximum number of speeches from the start to the end when two annotations each having an annotation added are used as start and end speeches, respectively. And a keyword extracting step of extracting a keyword predetermined for search from the extracted effective message group.

  According to the discussion knowledge graph construction method of this procedure, the discussion knowledge graph construction device extracts a comment with an annotation from a conversation history as an effective comment when extracting a comment group. The reason why an annotated annotation can be considered as an effective statement is that it was useful for solving the problem of the learning participant who added the annotation. This is because it can be considered as a remark that serves as a hint. However, in the conversation history, knowledge useful for problem solving (problem solving knowledge) is not necessarily expressed by only one utterance. One problem solving knowledge may be explained by a plurality of statements. In such a case, there is a possibility that the utterance to which the annotation is added differs for each learning participant. In order to deal with such a situation, the discussion knowledge graph construction device first calculates the average value of the number of utterances included in the utterance group including effective utterances, and includes the average number of utterances including effective utterances. It is determined whether or not a plurality of utterances each having an annotation added exist in a utterance group composed of utterances equal to or less than the value. This is because such a statement may be a statement related to the same problem solving knowledge. Then, the discussion knowledge graph construction device extracts a speech group estimated to be a series of speech related to the same problem solving knowledge as an effective speech group. In addition, since the discussion knowledge graph construction device extracts keywords from the effective remark group, including the keywords in the discussion knowledge graph makes it easy to perform the search process using the keywords when the discussion knowledge graph is visualized. It becomes possible.

  Further, in the discussion knowledge graph construction method according to the present invention, the information acquisition step acquires the recorded contents of the notes for each learner at a predetermined time interval during the collaborative learning, and continues for each learner. It is preferable that the difference between the recorded contents of the two notes acquired in this way is acquired as the note piece.

  According to the discussion knowledge graph construction method of such a procedure, the discussion knowledge graph construction device obtains a note piece from a note obtained for each learner, and therefore performs a process of obtaining the note piece from the note in the learner side device. There is no need. That is, it is possible to reduce the processing load of the apparatus without requiring a high level of capability from the apparatus on the learner side.

  In the discussion knowledge graph construction method according to the present invention, after the learner adds an annotation to another learner's comment, the association step includes an effective comment including the comment with the annotation added. A note piece number determining step for determining the number of note pieces recorded by the learner before adding an annotation to an utterance other than the group; and if it is determined that there are a plurality of note pieces, the plurality of notes A note piece creation step for connecting pieces to create a new note piece, and when it is determined that the number of note pieces is 1 or more, this indicates that the annotation is in the same viewpoint with respect to the statement A labeling step for giving the annotation a label for identifying both solutions according to whether it is a point of view or a different viewpoint, When it is determined that the number of note pieces is 1, the note pieces, a comment group including the comments to which the annotation is added, and a label for identifying the solution are associated, while a plurality of note pieces are provided. It is preferable to have an association control step of associating the created note piece, a comment group including the comment with the annotation added thereto, and a label for identifying the solution when it is determined that it exists.

  According to the discussion knowledge graph construction method of such a procedure, the discussion knowledge graph construction device, when associating an annotation with a note piece, when there are a plurality of note pieces recorded immediately after the target annotation is added. Connect those note pieces. Therefore, related knowledge can be combined into one knowledge. Further, when a note piece exists, the discussion knowledge graph construction device gives a label identifying the solution indicated by the target annotation to the annotation. Therefore, by including a label in the discussion knowledge graph, it is possible to easily perform a search process using the label when the discussion knowledge graph is visualized.

  Further, in the discussion knowledge graph construction method according to the present invention, the graph construction step includes a root node indicating the predetermined problem, an upper node indicating an effective message group associated with the note, and the effective message group. Node data for a discussion knowledge graph comprising a lower node indicating a note associated with the upper node, an upper edge connecting the root node and the upper node, and a lower edge connecting the upper node and the lower node In addition, as the edge data, it is preferable that the keyword is given as an attribute value to the upper node, and a label for identifying the solution is given to the lower edge.

  According to the discussion knowledge graph construction method of such a procedure, the discussion knowledge graph construction device constructs a discussion knowledge graph including nodes indicating keywords, a group of comments, and notes, respectively, and keyword and label data. Therefore, when visualized, it is easy to search for a comment group as a hint by using the keyword in the discussion knowledge graph, and it becomes easy to understand the difference in the solution in the note by the label.

  In order to solve the above-mentioned problem, the discussion knowledge graph construction device according to the present invention is connected to a plurality of learner terminals so as to be communicable, and a plurality of learners input a predetermined problem to each learner terminal. Information related to the discussion acquired from each learner terminal when collaborative learning is performed while adding annotations indicating that the solution is the same viewpoint or a different viewpoint with respect to the speech of other learners. A discussion knowledge graph constructing apparatus for constructing a discussion knowledge graph showing a graph obtained by visualizing information extracted from the discussion as derivation knowledge of a solution to the predetermined problem, wherein the cooperation is used as the information about the discussion. Acquires a conversation history showing all the utterances input during learning, and for each learner, the learner determines which utterance of other learners An annotation history indicating whether or not the annotation has been added, and a note piece history indicating a difference between recorded contents of the notebook recorded by the learner as a solution to the predetermined problem every predetermined time are acquired. Information acquisition means, comment group extraction means for extracting a comment group including one or more comments from the acquired conversation history based on the annotation history, and a predetermined note from the history of note pieces acquired for each learner A note piece extracting means for extracting a piece, an association means for associating the extracted message group and the extracted note piece based on the annotation history, and the discussion knowledge graph associating with the note piece. An upper layer including a comment group as a hint for the predetermined problem, and a note piece associated with the valid comment group Characterized in that it comprises a graph construction means for constructing a node data and edge data for the graph and a lower layer containing as a solution to the cement.

  According to such a configuration, the discussion knowledge graph construction device acquires the learning history (conversation history) of the shared space during collaborative learning, and the annotation history and the learning history (note piece history) of the individual learning space for each learner. To do. Then, the discussion knowledge graph construction device associates the comment group extracted from the conversation history based on the annotation history and the note piece extracted from the note piece history based on the annotation history. Therefore, the discussion knowledge graph construction device can extract not only the hint but also the solution derived by the learning participant as problem solving knowledge by associating the conversation history with the individual learning history as an extraction target. . In addition, the discussion knowledge graph construction device constructs a graph having an upper layer including a comment group as a hint and a lower layer including a note piece as a solution to the hint as a discussion knowledge graph, so the discussion knowledge graph is visualized. Sometimes learners can easily search for solutions to problems.

  A discussion knowledge graph construction program according to claim 7 causes a computer to execute the discussion knowledge graph construction method according to any one of claims 1 to 5. By being configured in this way, a computer in which this program is installed can realize each function based on this program.

  The computer-readable recording medium according to claim 8 is characterized in that the discussion knowledge graph construction program according to claim 7 is recorded. By being configured in this way, a computer equipped with this recording medium can realize each function based on a program recorded on this recording medium.

  According to the present invention, a discussion knowledge graph that can be used as effective knowledge for solving a problem similar to a problem that has been collaboratively learned not only for collaborative learning participants but also for other learners is constructed. be able to.

  The best mode (hereinafter referred to as “embodiment”) for carrying out the discussion knowledge graph construction device and the discussion knowledge graph construction method of the present invention will be described in detail below with reference to the drawings.

[Configuration of collaborative learning system]
FIG. 1 is a configuration diagram schematically showing a collaborative learning system including a discussion knowledge graph construction device (server) according to an embodiment of the present invention. The collaborative learning system 1 includes a server (discussion knowledge graph construction device) 2 and a plurality of (for example, three) clients (learner terminals) 3a, 3b, 3c (3) connected to each other via a network NW. ). The network NW is, for example, a LAN (Local Area Network) or the Internet. Here, three clients (learner terminals) 3a, 3b, 3c (3) are shown as an example, but the number of clients is not limited to this, and is simply referred to as client 3 below.

  When collaborative learning is performed, the server 2 constructs a discussion knowledge graph showing a graph obtained by visualizing information extracted from the discussion as derivation knowledge of a solution to a predetermined problem based on information on the discussion acquired from each client 3. To do. The server 2 includes, for example, a central processing unit (CPU), a random access memory (RAM), a read only memory (ROM), a hard disk drive (HDD), and a network interface card (NIC) for communication. Etc.

  The client 3 is operated by a learner who participates in collaborative learning (hereinafter referred to as a “learning participant” when distinguished), and is composed of, for example, a personal computer. In this collaborative learning system 1, as a premise, a plurality of learners discuss a predetermined problem with the remarks input to their respective clients 3, and respond to the remarks of other learners who feel “effective”. While adding annotations indicating that they are the same viewpoint or different viewpoints from the solution method, opinions are exchanged through conversation, and learning (collaborative learning) is performed while deriving unique solutions to individual learning notes (Non-Patent Document 5). reference).

<Outline of server>
As shown in FIG. 1, the server 2 includes a communication unit 11, a group data management unit 12, a group learning data management unit 13, a graph construction unit (graph construction unit) 14, and a discussion knowledge graph management unit 15. I have.
The server 2 stores information on learners participating in the collaborative learning in the group data storage unit 12a such as HDD in advance, and when the communication unit 11 receives a request for learner information from the client 3 via the network NW, The group data management unit 12 searches for the requested learner information and sends it back.

  When the server 2 receives the speech of the learning participant (information input by the learning participant) from the client 3 by the communication unit 11, the group learning data management unit 13 transmits the received speech to all the clients 3 and the conversation. The history is stored in the group learning data storage means 13a such as an HDD. When the server 2 receives the individual learning notes (notes) and annotation history of the learner from the client 3 by the communication unit 11, the server 2 accumulates the received data as learner learning data for each learner in the group learning data storage unit 13a. To do. Here, the notes are stored as note pieces divided into small units. Details of the group learning data management unit 13 will be described later.

When the server 2 receives a message indicating the end of learning from the client 3 by the communication unit 11, the server 2 constructs node data and edge data of the discussion knowledge graph based on the received learner learning data by the graph construction unit 14. The knowledge graph management unit 15 stores the data in a graph data storage unit 15a such as an HDD. The details of the graph construction unit (graph construction means) 14 will be described later together with the discussion knowledge graph.
When the server 2 receives a search request for knowledge contained in the discussion knowledge graph from the client 3 by the communication unit 11, the server 2 converts the discussion knowledge graph data as knowledge retrieved from the graph data storage unit 15 a by the discussion knowledge graph management unit 15. 3 to send.

<Client configuration>
The client (learner terminal) 3 includes a user interface 21, a communication unit 22, a learner learning data management unit 23, and a discussion knowledge graph display control unit 24.
The user interface 21 includes input means and output means. The input means is composed of, for example, a keyboard, a mouse, a disk drive device, and the like. The output means is, for example, a monitor such as a graphic board (output interface) and a liquid crystal display connected thereto.

  The client 3 transmits a message input via the user interface 21 to the server 2 by the communication unit 22 and receives all messages on the network NW and displays them on the screen. In addition, the client 3 uses the learner learning data management unit 23 to describe the learner's individual learning note (individual learning note history) input via the user interface 21 and the annotation (annotation history) added by the learner. ) Are displayed on the screen, and these are stored as learner learning data in the learner learning data storage means 23a such as an HDD. The client 3 transmits these learner learning data to the server 2. The client 3 transmits the search request input via the user interface 21 to the server 2 by the communication unit 22 and the discussion knowledge acquired from the server 2 in response to the search request by the discussion knowledge graph display control unit 24. Based on the graph data, the discussion knowledge graph is reproducibly displayed (visualized) on the display.

[Overview of collaborative learning system operation]
Next, the concept of the operation of the collaborative learning system shown in FIG. 1 will be described with reference to FIGS. FIG. 2 shows a conversation history and a discussion knowledge graph, respectively. Here, in order to simplify the description, as an example, as shown on the left side of FIG. 2, a plurality of learning participants use individual learning to solve a predetermined problem by collaborative learning (individual learning notes). It shall be derived in 201. Before each of these learning participants describes the derivation of the solution in their individual learning notes (notebooks), in response to other learners' comments that they found “effective” in the conversation during collaborative learning, Annotation 202 is added to indicate that the viewpoint is the same or different from that of the user's solution. Then, an annotation 202 is associated with a predetermined utterance in the conversation history 203 indicating a conversation during collaborative learning.

  The discussion knowledge graph 210 created after the end of the collaborative learning is a structured empirical knowledge discussed during the collaborative learning, as shown on the right side of FIG. Show. A root node (problem) 211 shown in FIG. 2 corresponds to the “problem” discussed in the collaborative learning. A node (upper node) 212 that is a child node of the root node 211 corresponds to “partial question hint (utterance group)”. A node (lower node) 213 that is a child node of the node (upper node) 212 corresponds to “description of solution to partial problem (individual learning history)”. The root node 211 and the node 212 are connected by an edge (upper edge), and the node 212 and the node 213 are connected by an edge (lower edge).

  FIG. 3 shows the relationship between the shared space and the individual learning space. Here, as an example, out of three learning participants, one person speaks a hint of a solution to a predetermined problem by collaborative learning, and the remaining two persons describe a solution to the predetermined problem in the individual learning spaces 310 and 320, respectively. It shall be derived in For example, in the individual learning space 310 that is the learning environment of one of the remaining two learners, the contents described in the individual learning notes are obtained by using three note pieces 311, 312, and 313 described at different timings. It is a combination. One learner added annotations 311a, 312a, and 313a before describing the three note pieces 311, 312, and 313, respectively. Then, the utterance input by the one learner, the three note pieces 311 to 313, and the three annotations 311a to 313a are transmitted to the server 2 and accumulated. These accumulated data are processed in the shared space 330 of the server 2.

  Similarly, in the individual learning space 320 which is the learning environment of the other learner, the content described in the individual learning note is a combination of three note pieces 321-323, and the other learner Annotations 321a to 323a were added. Then, the speech input by the other learner, the three note pieces 321 to 323, and the three annotations 321a to 323a are transmitted to the server 2 and accumulated.

  In the shared space 330, note pieces 311 to 313 and annotations 311a to 313a acquired from one learner are associated with each other in a storage area related to one learner. Similarly, in the shared space 330, note pieces 321 to 323 and annotations 321a to 323a acquired from the other learner are associated with each other in a storage area related to the other learner. In the shared space 330, in the storage area related to the conversation history, annotations acquired from both learners are associated with each utterance. For example, if the annotation histories acquired from both learners are for the same utterance, they are associated as annotations 331a, 332a, 333a, 334a, 335a, 336a for the same utterance as shown.

  In the present embodiment, the server 2 uses not only the learning history (conversation history) of the shared space 330 but also the learning history (individual learning note history) of the individual learning spaces 310 and 320, and from these, the problem of collaborative learning is determined. Extract solution knowledge. For this purpose, the server 2 associates the utterances included in the conversation history with a part (note piece) of the individual learning note history from the viewpoint of problem solving knowledge. Thus, attempts to associate the conversation history with the individual learning note history have not been made in the prior art. However, in this embodiment, the server 2 targets the conversation history and the individual learning history, so that not only the hints included in the conversation during the collaborative learning but also the solutions derived by the learning participants can be solved. It becomes possible to extract as. In addition, by using annotations, it is also possible to specify statements that are useful for solving problems from the conversation history.

  Here, the graph construction unit (graph construction means) 14 shown in FIG. 1 and the discussion knowledge graph to be constructed will be described. The graph construction unit 14 includes, as a discussion knowledge graph, an upper layer that includes a speech group associated with a note piece as a hint for a predetermined problem, and a lower layer that includes a note piece associated with a valid speech group as a solution to the hint. Node data and edge data for a graph having In the present embodiment, the graph construction unit 14 includes a root node indicating a predetermined problem, an upper node indicating a valid speech group associated with the note, and a lower node indicating a note associated with the effective speech group. And node data and edge data for the discussion knowledge graph comprising the upper edge connecting the root node and the upper node, and the lower edge connecting the upper node and the lower node. Along with the attribute value, a label identifying the solution is assigned to the lower edge.

Details of the discussion knowledge graph 210 shown in FIG. 2 are shown in FIG.
As shown in FIG. 4, the graph construction unit 14 constructs a discussion knowledge graph arranged for each subproblem using the effective speech group 401 and the note piece 411 extracted from the collaborative learning history and associated with each other. To do. In addition, before the graph construction unit 14 constructs the discussion knowledge graph, a difference in the solution of each note piece 411 is detected based on the annotation (see FIG. 3) given to the comment group 401 when associating in advance. Labels (solution 1, solution 2, etc.) corresponding to the difference in solution are given.

  In the discussion knowledge graph shown in FIG. 4, the extracted comment group 401 becomes a node 212 that is a child node of the root node (problem) 211 by the processing of the graph construction unit 14, and the keyword extracted from the comment group 401 402 is stored as an attribute (node data). A node 213 that is a child node of the node (partial question hint) 212 is a note piece 411 associated with the message group 401. Further, a label 412 indicating a difference in solution is added to an edge (lower edge) that connects the node 212 including the message group 401 and the node 213 indicating the note piece.

[Configuration of Server Group Learning Data Management Unit]
FIG. 5 is a block diagram showing the configuration of the server shown in FIG. As shown in FIG. 5, the group learning data management unit 13 of the server 2 includes an information acquisition unit 30, a comment group extraction unit 40, a note piece extraction unit 50, and an association unit 60.

The information acquisition means 30 acquires and stores information related to discussion in collaborative learning, and includes a conversation history acquisition means 31, an annotation history acquisition means 32, a note history acquisition means 33, and a note piece acquisition means 34. I have.
The conversation history acquisition means 31 acquires a conversation history indicating all comments input during the cooperative learning as information related to the discussion in the cooperative learning.
The annotation history acquisition unit 32 acquires, as information related to the discussion in collaborative learning, for each learner, an annotation history indicating which utterance of the other learner the learner has added an annotation to.

The note history acquisition unit 33 acquires, for each learner, the history of notes recorded by the learner as his / her solution to a predetermined problem as information related to the discussion in collaborative learning.
The note piece acquisition means 34 acquires the recorded contents of the notes for each learner at a predetermined time interval during the collaborative learning, and uses the difference between the recorded contents of the two notes acquired continuously for each learner as a note piece. To get. In the collaborative learning, the learner derives a solution based on the speech that is a hint displayed on the client 3 and describes it in the individual learning note. According to the note piece acquisition means 34, it is possible to extract necessary information from the individual learning notes for each solution knowledge derived by using the hint statement. In the present embodiment, the note piece acquisition unit 34 acquires the note piece history from the note history acquired by the note history acquisition unit 33. Note that the solution derived according to the hint may not be included in the final solution derived by the learner.

  The utterance group extraction unit 40 extracts a utterance group including one or more utterances from the conversation history acquired by the information acquisition unit 30 based on the annotation history. The effective utterance extraction unit 41 and the average number of utterances are calculated. Means 42, valid speech discrimination means 43, valid speech group extraction means 44, and keyword extraction means 45 are provided.

  The effective utterance extraction unit 41 extracts, from the conversation history acquired by the information acquisition unit 30, an utterance added with an annotation as an effective utterance. The reason why an annotated utterance can be considered as an effective utterance is that the annotated utterance is an utterance that has helped solve the problem of the learning participants who have added the annotation. This is because it can be regarded as a remark as a hint for problem solving. However, in the conversation history, knowledge useful for problem solving (hereinafter referred to as problem solving knowledge) is not necessarily expressed by only one statement. One problem solving knowledge may be explained by a plurality of statements. In such a case, there is a possibility that the utterance to which the annotation is added differs for each learning participant. In order to cope with such a situation, the speech group extraction unit 40 includes a speech number average value calculation unit 42, an effective speech determination unit 43, and an effective speech group extraction unit 44.

  The average number of utterances calculating means 42 is the average value of the number of utterances included in the utterance group (one topic) including the valid utterances extracted by the effective utterance extraction means 41 (hereinafter referred to as the average number of utterances). Is calculated based on the number of utterances to which annotations are added and the total number of utterances included in the acquired conversation history. This statement number average value calculation means 42 corresponds to the fact that one problem solving knowledge is explained by a plurality of statements together with a series of processes of the effective statement determination means 43 and the effective statement group extraction means 44. Here, In order to simply define the average value to be calculated, the following two assumptions will be introduced. That is, first, it is assumed that all statements in the conversation history are statements related to problem solving knowledge. Secondly, it is assumed that all comments with annotations indicate independent knowledge. Therefore, in the present embodiment, the message number average value calculating means 42 calculates the message number average value q constituting one problem solving knowledge by the equation (1). Here, n indicates the total number of utterances, and p indicates the number of utterances with annotations added.

  q = n / p Formula (1)

  The effective speech discriminating unit 43 includes a plurality of annotations added to each of speech groups composed of a number of speeches equal to or less than the average number of speeches calculated including the effective speech extracted by the effective speech extracting unit 41. It is discriminate | determined whether the said utterance exists. This is because “q” utterances before and after the annotation with the annotation added may be utterances regarding the same problem solving knowledge. For example, if the average number of utterances is “4.5”, the valid utterance discriminating unit 43 sets a discriminating group of “4” or less utterance groups including the annotated utterances.

  The effective speech group extraction unit 44, when the effective speech discrimination unit 43 determines that there are a plurality of speeches to which annotations are added, the two speeches to which annotations are respectively added from the target speech group. Is a speech group in which the number of utterances from the start to the end is maximized as an effective speech group. The reason for processing in this way is that if an annotation is added to “q” utterances before and after the annotated utterance, the two utterances with the annotation and the existing utterances are the same. This is because it can be estimated as a series of knowledge-related utterances, and even if there are multiple combinations thereof, it is uniquely determined. Further, the effective speech group extraction unit 44 determines the effective speech extracted by the effective speech extraction unit 41 when the effective speech determination unit 43 determines that there are no plurality of speeches to which annotations are added. Extract as a valid speech group.

  The keyword extraction unit 45 extracts a keyword predetermined for search from the effective statement group extracted by the effective statement group extraction unit 44. For this reason, the keyword extraction means 45 includes means for morphological analysis of each comment included in the comment group. Here, the morpheme refers to the minimum character string that has no meaning if it is made finer than this. The morpheme analysis is to analyze a sentence by breaking it down to the morpheme level. This keyword is intended to enable other learners to easily search for effective speech groups. In the case of utterances related to the same problem solving knowledge, it is considered that words indicating knowledge are frequently included in a short utterance group. Therefore, in the present embodiment, the keyword extraction unit 45 extracts a word (excluding a single character) that appears twice or more from a predetermined group of comments as a keyword of the group of comments.

  The note piece extraction means 50 extracts a predetermined note piece from the history of note pieces acquired for each learner.

  The associating means 60 associates the comment group extracted by the comment group extracting means 40 and the note piece extracted by the note piece extracting means 50 based on the annotation history. A piece creating unit 62, a label providing unit 63, and an association control unit 64 are provided.

  After the learner adds an annotation to another learner's utterance, the note piece number discriminating means 61 adds the annotation to a utterance other than the valid utterance group including the utterance with the annotation added. Before, the number of note pieces recorded by the learner is determined. This note piece number discriminating means 61 is used for associating the note piece extracted by the note piece extracting means 50 with the annotation added by the learner describing the note piece immediately before describing the note piece. Determine the number of.

The note piece creating means 62 is for creating a new note piece by connecting a plurality of note pieces when the note piece number judging means 61 determines that there are a plurality of note pieces.
When a plurality of note pieces extracted from a client 3 (one learner) correspond to one utterance group, it is assumed that the descriptions of the plurality of note pieces are derived from one utterance. Can do. Therefore, the note piece creating means 62 combines a plurality of related note pieces in tandem to form a new note piece.

  The label assigning means 63 indicates whether or not the annotation is the same viewpoint with respect to the utterance when the number of note pieces is determined to be 1 or more by the note piece number determining means 61. A label for identifying both solutions is given to the annotation depending on whether or not it indicates that it is.

  When the number of note pieces is determined to be 1 by the note piece number determination means 61, the association control means 64 is a label that identifies a note piece, a comment group including a comment with the annotation added, and a solution. Are associated with each other. When the note piece number discriminating unit 61 discriminates that there are a plurality of note pieces, the association control unit 64 discriminates the created note piece, the utterance group including the utterance with the annotation added, and the solution. Associate a label.

  The information acquisition unit 30, the comment group extraction unit 40, the note piece extraction unit 50, the association unit 60, and the graph construction unit 14 are a predetermined program stored in an HDD or the like of the storage unit by the CPU. Is implemented in a RAM and executed.

[Server operation]
The operation of the server 2 shown in FIG. 5 will be described with reference to FIG. 6 (see FIG. 5 as appropriate). FIG. 6 is a flowchart showing the operation of the server shown in FIG. The server 2 acquires the conversation history, annotation history, and note history by the information acquisition means 30 of the group learning data management unit 13 (information acquisition step: step S1). And the server 2 extracts the note piece of a predetermined learner from the group learning data storage means 13a by the note piece extraction means 50 (note piece extraction step: step S2). Then, the server 2 executes a speech group extraction process for a predetermined learner (a speech group extraction step: step S3). At this time, effective speech groups and keywords are extracted. Here, the processing order of step S2 and step S3 is arbitrary, and the server 2 executes the processing of step S2 and step S3 for all learning participants.

  Subsequently, the server 2 uses the association unit 60 to execute an association process between the annotation and the note piece (association step: step S4). At this time, a label indicating the type of annotation is given. Then, the server 2 constructs a discussion knowledge graph by the graph construction unit 14 based on the associated annotations and note pieces, the extracted comment group and keywords, and the assigned labels (graph construction step: Step S5). The node data and edge data of the constructed discussion knowledge graph are stored in the graph data storage unit 15a by the discussion knowledge graph management unit 15.

  Next, the comment group extraction process in step S3 and the association process in step S4 will be described with reference to FIG. FIG. 7A is a flowchart showing a comment group extraction process, and FIG. 7B is a flowchart showing an association process. First, in the comment group extraction process of step S3 described above, as shown in FIG. 7A, the server 2 uses the effective comment extraction means 41 of the comment group extraction means 40 to add a comment to which an annotation has been added from the conversation history. (Effective remarks) are extracted (effective remark extraction step: step S31). Then, the server 2 calculates the average value of the number of utterances constituting one topic based on the above-described formula (1) by the utterance number average value calculating means 42 (utterance number average value calculating step: step S32). .

  Whether the server 2 has added annotations to other utterances included in the utterance group including the utterances that are valid and less than the utterance count average value by the valid utterance discrimination means 43. It is discriminate | determined (effective speech discrimination | determination step: step S33). When no annotation is added to other utterances (step S33: No), the server 2 uses the effective utterance group extraction unit 44 to make the utterance itself an effective utterance group (effective utterance group first extraction step: Step S34). Subsequently, the server 2 extracts keywords from the effective utterance group by the keyword extraction means 45 (keyword extraction step: step S35). On the other hand, when an annotation is added to other utterances (step S33: Yes), the server 2 uses the effective utterance group extraction unit 44 to determine the longest utterance group having two annotations as both ends as an effective utterance group. (Effective speech group second extraction step: step S36). Then, the server 2 proceeds to step S35.

  Next, in the association process of step S4 described above, as shown in FIG. 7B, the server 2 adds another annotation after adding the annotation by the note piece number discrimination unit 61 of the association unit 60. The number of previously recorded note pieces is discriminated (note piece number discriminating step: step S41). When it is determined in step S41 that the number of note pieces is “2 or more”, the server 2 creates a new note piece by connecting two or more note pieces by the note piece creation means 62 (note piece creation). Step: Step S42). Subsequent to step S42 or when it is determined in step S41 that the number of note pieces is “1”, the server 2 assigns a label for identifying the solution to the annotation by the label attaching unit 63 (label attaching step). : Step S43). Then, the server 2 associates the extracted note piece or the created note piece, the message group, and the label with the association control means 64 (association control step: step S44), and ends the process. If it is determined in step S41 that the number of note pieces is “0”, the server 2 skips the processes in steps S42 to S44 and ends this association process.

  The server 2 can also be realized by executing a discussion knowledge graph construction program that causes a general computer to execute each step described above. This program can be distributed via a communication line, or can be distributed by writing in a recording medium such as a CD-ROM.

[Concrete example]
<Examples of valid speech group extraction>
FIG. 8 is a diagram illustrating an example of the conversation history accumulated in the server. The conversation history shown in FIG. 8 is recorded when the problem “Create a program for searching for a character string” is collaboratively learned as an experiment. In this experiment, the total number of remarks n was “94”. FIG. 8 shows a part thereof. Of all the utterances, the number p of utterances with annotations added was “20”. In this case, the utterance number average value calculating means 42 (see FIG. 5) calculates “q = 4.7” from the equation (1). Therefore, the valid speech discriminating means 43 discriminates whether or not there is another speech with the annotation within “4 speeches” before and after the speech with the annotation added.

  In the example shown in FIG. 8, an annotation is added by the learner B to the utterance “ID = 54”. Within “4 utterances” before and after the utterance of “ID = 54”, the utterance of “ID = 52” exists as the utterance with the annotation added. Therefore, the statement of “ID = 52” and the statement of “ID = 54” are regarded as a statement related to the same problem solving knowledge, the statement of “ID = 52” is the start statement, and the statement of “ID = 54” is It can be regarded as a statement of termination. In this case, the valid speech group extraction unit 44 sets a speech group including a speech with “ID = 52”, a speech with “ID = 54”, and a speech with “ID = 53” between them as an effective speech group. Extract.

  Here, if an annotation is also added to the message of “ID = 51”, the message of “ID = 52” is included within “4 messages” before and after the message of “ID = 54”; The message “ID = 51” exists. In this case, it is assumed that the number of utterances is the maximum as the effective utterance group, and the utterance group having the utterance of “ID = 51” and the utterance of “ID = 54” as both ends is extracted as an effective utterance group. To do.

  An effective speech group refers to a speech group that includes at least one annotation. For example, an annotation is added by the learner C to the utterance “ID = 63”. There are no utterances with an annotation within “4 utterances” before and after the utterance of “ID = 63”. However, the effective speech group extraction unit 44 extracts the speech with “ID = 63” as an effective speech group.

  FIG. 9 shows an example of a speech group composed of 7 speeches as the speech group obtained in this experiment. When the utterances included in the utterance group shown in FIG. 9 are divided into words, the underlined words, that is, “argument”, “character”, “execute”, “image”, “command line”, “user”, “Encryption” appears twice and “Input” appears three times. Therefore, the keyword extraction means 45 extracts these characters as keywords of this comment group.

<Acquisition of note pieces>
The note history acquisition unit 33 (see FIG. 5) acquires an individual learning note shown in FIG. 10A from a certain client 3 at time “t”. Further, the note history acquisition unit 33 acquires the individual learning note shown in FIG. 10B from the same client 3 at the time “t + 1”. The note piece acquisition means 34 acquires the difference between the individual learning note shown in FIG. 10A and the individual learning note shown in FIG. 10B as the note piece of the learning participant using the client 3. That is, the note piece acquisition means 34 acquires the note piece shown in FIG.

<Example of association>
FIG. 11 shows an example of a note piece obtained from a client 3 obtained in this experiment and an example of a statement (another learner's statement) annotated by a learner describing the note piece. Note pieces 1111 and 1112 shown in FIG. 11 are described after the learner adds an annotation to the utterance 1101. A note piece 1113 shown in FIG. 11 is described after the learner adds an annotation to the utterance 1102.

  The note piece creating means 62 (see FIG. 5) connects the note pieces 1111 and 1112 shown in FIG. 11 to create a new note piece 1211 as shown in FIG. The association control unit 64 (see FIG. 5) associates a new note piece 1211 with the keyword (the keyword extracted by the keyword extraction unit 45) 1201 of the utterance group to which the utterance 1101 shown in FIG. Further, the association control means 64 associates the keyword of the comment group to which the comment 1102 shown in FIG. 11 belongs (the keyword extracted by the keyword extraction means 45) 1202 and the single note piece 1113 shown in FIG.

  In FIG. 11 and FIG. 12, the case where one learner's note piece exists with respect to one comment group was demonstrated. On the other hand, when a plurality of learner's note pieces exist for one utterance group, each learner does not always add an annotation having the same meaning. That is, each learner's note piece does not necessarily suggest the same solution.

  Here, two viewpoints can be considered. In the first aspect, the difference in each solution method described in each note piece is regarded as a difference in solution between the learner describing the note piece and the speaker of the utterance group corresponding to the note piece. It is. That is, a learner who adds an annotation (for example, represented by “a”) indicating the same viewpoint as his / her solution can be regarded as describing the same solution as that of the speaker in a note piece. Further, a learner who has added an annotation (for example, represented by “b”) indicating a different viewpoint from his / her solution method can be regarded as describing the solution method different from that of the speaker in a note piece.

  Further, in the second aspect, the difference in the respective solution methods described in the respective note pieces is regarded as the difference in the solution method between the learners describing the note pieces. That is, learners who have added an annotation (for example, represented by “a”) indicating the same viewpoint as their solution can be regarded as describing the same solution in a note piece. Also, learners who have added annotations (for example, “b”) indicating a different viewpoint from their own solution may describe the same solution in a note piece, or may describe a different solution in a note piece. Absent. Here, it is assumed that there is a high possibility that a different solution is described in a note piece. In the present embodiment, for the above two viewpoints, in the present embodiment, the label attaching unit 63 (see FIG. 5) assigns a label for identifying the difference in the solution method of the note piece to the annotation based on the type of annotation, Based on this, this problem is dealt with by constructing a discussion knowledge graph in the subsequent graph construction unit 14 (see FIG. 5).

<An example of the structure of the discussion knowledge graph>
The discussion knowledge graph shown in FIG. 13 is recorded when the problem “Create a program for encrypting a character string” is collaboratively learned as an experiment. Here, for reasons of space, in the discussion knowledge graph shown in FIG. 13, each node of the first layer 1301, second layer 1302, and third layer 1303 and attribute values (node data) for the second layer 1302 are used. Displays the assigned keyword.

  The discussion knowledge graph shown in FIG. 13 includes a root node to which the “character string encryption problem” is assigned in the first layer 1301. In this discussion knowledge graph, the second layer 1302 having nine child nodes (upper nodes) for the root node is an effective statement group (partial problem) extracted from the conversation history regarding the “character string encryption problem”. Constructed and corresponds to hints for deriving the solution of the problem. Here, the keyword extracted from the message group becomes the attribute value of the node. For example, the node 1302a includes a comment group including the keyword “charAt”, and the node 1302b includes a comment group including the keyword “backword”. The same applies hereinafter.

  FIG. 14 shows details of some of the nodes 1302a and 1302b and their lower nodes shown in FIG. As shown in FIG. 14, the discussion knowledge graph shows a specific solution of a problem in which the second layer 1302 is derived from the upper (second layer) hint. For example, the node 1302a includes an area 1401a for displaying a comment group and an area 1402a for displaying the keyword “charAt” extracted from the comment group. Similarly, the node 1302b includes an area 1401b for displaying a comment group and an area 1402b for displaying the keyword “backword”.

  In the discussion knowledge graph shown in FIG. 14, each node of the second layer 1302 includes individual note pieces associated with the comment group as child nodes. That is, the third layer 1303 shows a specific solution of the problem derived from the upper (second layer) hint. Also, as shown in FIG. 14, when there are a plurality of lower nodes with respect to the upper node, in order to clarify the difference between the lower nodes, an attribute (lower edge) is assigned to the edge (lower edge) based on the difference in the solution of the note pieces. Label) is added. For example, the node 1302a has four child nodes, and a label 1403a indicating which of three types of solutions (solution 1, solution 2, and solution 3) is given. Similarly, the node 1302b has three child nodes and is given a label 1403b indicating that the solution is the same solution (solution 1). Note that “Solution 1” in the label 1403a and “Solution 1” in the label 1403b do not necessarily mean the same thing.

<Visualization example of discussion knowledge graph>
The discussion knowledge graph acquired and displayed by the client 3 from the server 2 not only visualizes solution knowledge for the problem being discussed, but also is solved by the operator (learning participant or other learner) of the client 3 It is desirable to make it visible so that it can be easily searched. An example of visualization of a discussion knowledge graph that satisfies such a requirement is shown in FIG.

  The discussion knowledge graph shown in FIG. 15 displays the title of the problem at the root node of the first layer 1301 and the keyword at the upper node of the second layer 1302. When the operator of the client 3 performs an operation of selecting a predetermined keyword, the discussion knowledge graph can display the details of the message group as shown in FIG. It has become. In addition, in this discussion knowledge graph, the type of the solution is displayed at the lower node of the third layer 1303 so that the difference in the viewpoint of the solution can be understood. When the operator of the client 3 performs an operation of selecting a predetermined lower node, the discussion knowledge graph can display in detail, for example, a note piece as shown in FIG. 14 as detailed information of the upper node. ing.

  Here, in the third layer, the same character is assigned to the solution of the same viewpoint, and different characters are assigned to the solution of the different viewpoints to the lower nodes having the same upper node. Note that the same viewpoint does not mean the same viewpoint for lower nodes having different upper nodes. For example, the solution “a” given to the lower node of the upper node 1302a and the solution “a” given to the lower node of the upper node 1302b do not necessarily mean the same viewpoint.

<Experiments with other learners>
As a result of an experiment in which two learners (other learners who are not learning participants) solve the same problem using the discussion knowledge graph shown in FIG. 15, “it is easy to search for an unknown part”, “1 You can compare various pieces of notebook with respect to one knowledge. " Both of these learners had the ability to derive the solution of the problem to some extent by themselves, and used the discussion knowledge graph instead of the dictionary. That is, two learners identified a node of a step whose understanding is ambiguous from a keyword, and observed only the note piece. Therefore, it can be presumed that the discussion knowledge graph is effective for learners who have some knowledge about the problem and the knowledge to be acquired is clear.

  According to the present embodiment, the server 2 can construct a discussion knowledge graph in which empirical problem solving knowledge derived by collaborative learning is organized for each partial problem. Therefore, by presenting the discussion knowledge graph constructed by the server 2 at the client 3, the learner searches for and uses appropriate knowledge according to his / her problem solving state from the empirical knowledge derived by collaborative learning. It becomes possible to do. In other words, the constructed discussion knowledge graph can support the problem solving of learners in various understanding states who intend to deal with the same problem as the presented discussion knowledge graph.

  As mentioned above, although each embodiment of this invention was described, this invention is not limited to this, It can implement in the range which does not change the meaning. For example, the discussion knowledge graph construction device (server 2) further includes means for extracting appropriate knowledge from the discussion knowledge graph according to the type of learner requesting the discussion knowledge graph based on the constructed discussion knowledge graph. You may comprise. In this case, the discussion knowledge graph construction device may be preset to a mode corresponding to one of the learner types, or the discussion knowledge graph construction device receives a request from the operator of the client 3 Sometimes, the type of learner is determined, and the mode may be automatically switched each time.

  The method of dividing the learner type is arbitrary. For example, the learner can be divided as follows, and can be operated as follows in each mode according to each learner. The knowledge during collaborative learning necessary for the learner depends on the learner's current learning objective and whether or not the content discussed in the collaborative learning is known. The learners can be broadly classified into learners who have participated in collaborative learning (learning participants) and other learners who are not. Among these, learning objectives for learning participants can include review of their derivation process and acquisition of different knowledge based on their own learning experience. For example, a learning participant who wishes to review his / her derivation process needs to present knowledge derived by the learning participant himself / herself in order to remind the learning participant's own derivation error. In this case, the discussion knowledge graph construction device presents the solution (knowledge) derived by the learning participant from the knowledge extracted from the collaborative learning history or the knowledge of the same solution as the learning participant. .

  Further, among learning participants, those who wish to acquire different knowledge can amplify knowledge if they understand knowledge different from the solution (knowledge) derived by the learning participant, that is, another solution. Therefore, in this case, the discussion knowledge graph constructing device has an effective remark group (an upper node corresponding to the discussion knowledge graph as a hint for a subproblem having a solution different from the solution derived by the learning participant itself). And the solution (note pieces included in the corresponding lower nodes of the discussion knowledge graph).

  Further, among learners, other learners who did not participate in collaborative learning can be classified into two types according to their abilities. The first learner is a learner who has the ability to solve problems to some extent by himself / herself and whose knowledge to be acquired is clear. The second learner is a learner who is difficult to solve the problem by himself and needs to strongly support the derivation of the solution including the problem solving procedure.

  Appropriate knowledge may be presented to the first learner in response to a search request for knowledge by the learner. It can be assumed that the knowledge requested to be searched is a partial problem that the learner is stuck with, or a partial problem that the learner has already derived. Therefore, in this case, the discussion knowledge graph construction device searches the knowledge required by the learner from the nodes of the discussion knowledge graph, and presents a comment group or a note piece corresponding to the corresponding partial problem.

  For the second learner, the discussion knowledge graph construction device presents knowledge along the structure of the discussion knowledge graph. Here, the discussion knowledge graph construction device further selects an effective statement group (a statement group included in the corresponding upper node of the discussion knowledge graph) as a hint for deriving the correct answer according to the understanding state of the second learner. A mode for presenting and a mode for presenting an extracted note piece that is an actual solution can be provided.

  In addition, when the discussion knowledge graph construction device operates in the operation mode according to the type of learner as described above, the history of use of the discussion knowledge graph of the learner (regardless of whether it is a learning participant or another learner) And knowledge corresponding to the operator (learner) may be extracted based on the accumulation result. For example, the discussion knowledge graph construction device searches for other learners who have the same usage tendency as the learner who requested the knowledge search, and gives priority to the problem solving knowledge used by the other learners searched. By presenting it, it is possible to recommend more appropriate knowledge to the requested learner. As described above, by operating in the operation mode according to the type of the learner, the discussion knowledge graph construction device can navigate the problem solving to the learner. As a result, the knowledge constructed as a discussion knowledge graph can be used more effectively.

  Furthermore, when other learners who are not learning participants derive the solution while looking at the hints of the discussion knowledge graph acquired from the discussion knowledge graph construction device, it is determined that the derived solution is derived from the hints. It is possible. In this case, the discussion knowledge graph construction device acquires the derived solution from the client 3 that has presented the hint of the discussion knowledge graph, adds the acquired solution to the discussion knowledge graph in association with the presented hint. accumulate. Thereby, the discussion knowledge graph construction device can increase the knowledge by combining the knowledge about the problem acquired by the collaborative learning with the knowledge newly acquired after the collaborative learning. In other words, the results of collaborative learning performed simultaneously by multiple learning participants (referred to as collaborative learning in a narrow sense for the sake of convenience) are accompanied by time-lapse delays and other learners who have obtained solutions from hints (convenient The result of collaborative learning in a broad sense) can be combined as a time lag.

It is a block diagram which shows typically the collaborative learning system containing the discussion knowledge graph construction | assembly apparatus (server) which concerns on embodiment of this invention. It is a conceptual diagram which shows the outline | summary of the collaborative learning system shown in FIG. 1, Comprising: The conversation history and the argument knowledge graph are each shown. It is a conceptual diagram which shows the outline | summary of the collaborative learning system shown in FIG. 1, Comprising: The relationship between shared space and an individual learning space is shown. It is a figure which shows typically the structure of the argument knowledge graph shown in FIG. It is a block diagram which shows the structure of the server shown in FIG. It is a flowchart which shows operation | movement of the server shown in FIG. (A) is a flowchart showing a comment group extraction process, and (b) is a flowchart showing an association process. It is a figure which shows an example of the conversation log | history accumulate | stored in the server. It is a figure which shows an example of the message accumulate | stored in the server. It is a figure which shows an example of an individual learning note. It is a figure which shows an example of the comment and the note piece to which the annotation was added. It is a figure which shows an example of correlation with the comment to which the annotation shown in FIG. 11 was added, and a note piece. It is a figure which shows an example of the structure of a discussion knowledge graph. It is a figure which shows the detail of a part of discussion knowledge graph shown in FIG. It is a figure which shows the visualization example of the argument knowledge graph shown in FIG. It is a block diagram which shows the conventional collaborative learning system typically. It is a figure which shows typically the window simultaneously displayed on a client's screen. It is a figure which shows an example of the conversation window shown in FIG. It is a figure which shows an example of the individual learning window shown in FIG. It is a figure which shows an example of the annotation window shown in FIG.

Explanation of symbols

1 Collaborative learning system 2 Server (Discussion knowledge graph construction device)
3 (3a-3c) Client (learner terminal)
DESCRIPTION OF SYMBOLS 11 Communication part 12 Group data management part 12a Group data storage means 13 Group learning data management part 13a Group learning data storage means 14 Graph construction part (graph construction means)
DESCRIPTION OF SYMBOLS 15 Discussion knowledge graph management part 15a Graph data storage part 21 User interface 22 Communication part 23 Learner learning data management part 23a Learner learning data management part 24 Discussion knowledge graph display control part 30 Information acquisition means 31 Conversation history acquisition means 32 Annotation History acquisition means 33 Note history acquisition means 34 Note piece acquisition means 40 Speech group extraction means 41 Effective speech extraction means 42 Speech number average value calculation means 43 Effective speech discrimination means 44 Effective speech group extraction means 45 Keyword extraction means 50 Note piece extraction means 60 Association means 61 Note piece number discrimination means 62 Notebook piece creation means 63 Label assignment means 64 Association control means NW network

Claims (8)

While a plurality of learners discuss about a given problem by remarks input to each learner's terminal and add annotations indicating that other learners' remarks are the same or different viewpoints as their solution A discussion knowledge graph showing a graph obtained by visualizing information extracted from the discussion as derivation knowledge of the solution to the predetermined problem based on information on the discussion acquired from each learner terminal when performing collaborative learning A discussion knowledge graph construction method for constructing a discussion knowledge graph construction device,
The information acquisition means acquires a conversation history indicating all the utterances input during the collaborative learning as information relating to the discussion, and for each learner, the learner is responsive to which of other learners Information for acquiring a note history indicating a difference in recorded contents per predetermined time of a note recorded by the learner as his / her solution to the predetermined problem An acquisition step;
A speech group extraction step of extracting a speech group including one or more speeches from the acquired conversation history based on the annotation history by the speech group extraction means;
A note piece extraction step for extracting a predetermined note piece from a history of note pieces obtained for each learner by the note piece extraction means;
An associating unit for associating the extracted message group and the extracted note piece based on the annotation history;
By means of a graph construction means, as the discussion knowledge graph, an upper layer including a speech group associated with the note piece as a hint for the predetermined problem, and a note piece associated with the valid speech group as a solution to the hint A descriptive knowledge graph construction method comprising: a graph construction step of constructing node data and edge data for a graph having a lower layer. The speech group extraction step includes:
An effective utterance extraction step of extracting, as an effective utterance, the utterance added with the annotation from the acquired conversation history;
The average value of the number of utterances included in the utterance group including the extracted effective utterance is based on the number of utterances to which the annotation is added and the total number of utterances included in the acquired conversation history. A step of calculating the average number of utterances to be calculated;
It is determined whether or not there are a plurality of utterances each having the annotation added thereto in a utterance group including the extracted effective utterances and including the number of utterances equal to or less than the calculated average number of utterances. A valid speech discrimination step to perform,
An effective speech group first extraction step of extracting the extracted effective speech as an effective speech group when it is determined that there are no plurality of speeches to which the annotations are respectively added;
When it is determined that there are a plurality of utterances to which the annotations are respectively added, the two utterances to which the annotations are respectively added are set as start and end utterances in the utterance group. An effective speech group second extraction step for extracting the speech group having the maximum number of speeches as an effective speech group;
The discussion knowledge graph construction method according to claim 1, further comprising: a keyword extraction step of extracting a predetermined keyword for retrieval from the extracted effective utterance group. The information acquisition step includes
During the collaborative learning, the recorded content of the note is acquired at a predetermined time interval for each learner, and the difference between the recorded content of two notes acquired continuously for each learner is acquired as the note piece. The discussion knowledge graph construction method according to claim 1 or claim 2, characterized in that The association step includes
After the learner adds an annotation to another learner's utterance, before adding the annotation to a utterance other than the effective utterance group including the utterance added with the annotation, the learner A note piece number determining step for determining the number of recorded note pieces;
When it is determined that there are a plurality of note pieces, a note piece creating step of creating a new note piece by connecting the plurality of note pieces;
When it is determined that the number of note pieces is 1 or more, both depending on whether the annotation indicates the same viewpoint or a different viewpoint with respect to the utterance A labeling step for giving a label for identifying the solution of the annotation to the annotation,
When it is determined that the number of note pieces is 1, the note pieces are associated with a utterance group including the utterances to which the annotation is added and a label for identifying the solution, while there are a plurality of note pieces. And an association control step of associating the generated note piece, a comment group including the comment with the annotation added thereto, and a label for identifying the solution when it is determined. 3. The discussion knowledge graph construction method according to 3. The graph construction step includes
A root node indicating the predetermined problem; an upper node indicating a valid speech group associated with the note; a lower node indicating a note associated with the valid speech group; the root node and the upper node; The keyword is assigned as an attribute value to the upper node as node data and edge data for the discussion knowledge graph including the upper edge connecting the upper node and the lower edge connecting the upper node and the lower node. The discussion knowledge graph construction method according to claim 4, wherein a label for identifying the solution is given to the lower edge. It is connected to a plurality of learner terminals in a communicable manner, and a plurality of learners discuss a predetermined problem by the remarks input to each learner terminal, and the same viewpoint as their own solution for the remarks of other learners or When collaborative learning is performed while adding annotations indicating different viewpoints, the information extracted from the discussion is solved based on the information about the discussion acquired from the respective learner terminals. A discussion knowledge graph construction device for constructing a discussion knowledge graph showing a graph visualized as derived knowledge,
As the information related to the discussion, a conversation history indicating all the utterances input during the collaborative learning is acquired, and for each learner, the learner adds the annotation to any utterance of other learners. Information acquisition means for acquiring an annotation history indicating whether or not a note piece history indicating a difference in recorded contents every predetermined time of a note recorded as a solution to the predetermined problem by the learner;
A comment group extracting means for extracting a comment group including one or more comments based on the annotation history from the acquired conversation history;
Note piece extraction means for extracting a predetermined note piece from the history of note pieces obtained for each learner;
Association means for associating the extracted message group and the extracted note piece based on the annotation history;
As the discussion knowledge graph, an upper layer including a speech group associated with the note piece as a hint for the predetermined problem, and a lower layer including a note piece associated with the valid speech group as a solution to the hint. An apparatus for constructing a discussion knowledge graph, comprising: graph construction means for constructing node data and edge data for a given graph.   A discussion knowledge graph construction program that causes a computer to execute the discussion knowledge graph construction method according to any one of claims 1 to 5.   A computer-readable recording medium on which the discussion knowledge graph construction program according to claim 7 is recorded.

Images