JP2018045639A - Dialog log analyzer, dialog log analysis method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable the flow of a large dialog in a dialog log and the structural relation of a minute dialog inserted in somewhere of the large dialog to be conveyed to a user in a way easily understood at a glance.SOLUTION: A dialog log analyzer of an embodiment comprises: an estimation unit for adding a dialog action tag to a spoken sentence; a first determination unit for determining the scope and label of a first dialog block in the dialog log on the basis of a first rule for detecting the boundary position of the first dialog block, with a combination of a speaker tag and a dialog action tag as a condition; a second determination unit for determining the scope and label of a second dialog block in the dialog log on the basis of a second rule for detecting the boundary position of the second dialog block, with a combination of a speaker tag and a dialog action tag as a condition; and a display control unit for exercising control so that first display information that indicates the scope and label of the first dialog block and second display information that indicates the scope and label of the second dialog block are added to the dialog log as it is displayed.SELECTED DRAWING: Figure 1

Description

  Embodiments described herein relate generally to a dialog log analysis device, a dialog log analysis method, and a program.

  With the improvement of speech recognition technology, it is now possible to convert person-to-person dialogues such as interactions between operators and customers in call centers into text and store them as logs. A log in which such dialogue is recorded is called a dialogue log. For example, in the case of a call center, the dialogue log is used for various purposes such as evaluation of customer response by an operator and feedback to product development.

  When a user views a dialogue log, there is a need to easily grasp the content of the dialogue from the dialogue log. To assist in grasping the content of the dialogue, the dialogue log is summarized or a turning point of the topic from the dialogue log. Techniques have been proposed to find out. However, conventional approaches such as summarizing dialogue logs and finding turning points are not enough to easily grasp the flow of large dialogues. In addition, for example, in the case of a call center, a small dialogue that often exchanges individual contents in the flow of a large dialogue is often inserted. Cannot be clearly communicated to the user.

Japanese Patent No. 5728374 Japanese Patent No. 5346327 Japanese Patent No. 2849341 Japanese Patent No. 2802602

  The problem to be solved by the present invention is that a dialog log that can easily convey to a user at a glance the flow of a large dialog in the dialog log and the structural relationship of detailed dialogs inserted in a large dialog An analysis apparatus, a dialog log analysis method, and a program are provided.

  The dialogue log analysis apparatus according to the embodiment is a dialogue log analysis device that analyzes a dialogue log including a plurality of utterances constituting a dialogue and a speaker tag indicating a speaker of each utterance, and includes an estimation unit, 1 determination part, 2nd determination part, and a display control part are provided. The estimation unit estimates a dialogue action for an utterance sentence included in the dialogue log and assigns a dialogue action tag. The first determination unit is a rule for detecting a boundary position of a first dialogue block composed of a plurality of continuous utterance sentences on the condition of at least a combination of a speaker tag and a dialogue action tag. Based on a first rule that specifies a label to be given to the block, a range of the first dialogue block in the dialogue log and a label to be given to the first dialogue block are determined. The second determination unit is a rule for detecting a boundary position of a second dialogue block that is larger than the first dialogue block on the condition that at least a combination of a speaker tag and a dialogue action tag is provided. Based on the second rule designating the label to be assigned, the range of the second dialogue block in the dialogue log and the label to be given to the second dialogue block are determined. The display control unit displays, for the dialog log, first display information indicating a range of the first dialog block and a label of the first dialog block, a range of the second dialog block, and a label of the second dialog block. Control is performed so that the second display information is displayed.

The block diagram which shows the functional structural example of the dialog log analyzer of 1st Embodiment. The figure which shows an example of a dialogue log. The figure which shows an example of the dialogue log by which the utterance sentence was divided | segmented into the unit utterance sentence. The figure which shows the specific example of a dialogue action tag. The flowchart which shows an example of the process sequence of a dialogue action estimation part. The figure which shows an example of the dialog log with a tag. The figure which shows the structural example of a 1st rule database. The flowchart which shows an example of the process sequence of the interruption dialog block determination part. The figure which shows an example of the data structure of the determination result of the interruption dialog block determination part. The figure which shows the structural example of a 2nd rule database. The flowchart which shows an example of the process sequence of a dialogue block determination part. The figure which shows an example of the data structure of the determination result of a dialogue block determination part. The figure which shows an example of a display screen. The figure which shows the other example of a display screen. The block diagram which shows the functional structural example of the dialog log analyzer of 2nd Embodiment. The flowchart which shows an example of the process sequence of a keyword extraction part. The figure which shows an example of a concept class. The figure which shows an example of the dialogue log with a keyword tag. The figure which shows the structural example of a 1st rule database. The figure which shows the structural example of a 2nd rule database. The figure which shows an example of a display screen. The figure which shows the structural example of a 1st rule database. The block diagram which shows the functional structural example of the dialog log analyzer of 4th Embodiment. The figure which shows an example of a display screen. The block diagram which shows the hardware structural example of a dialog log analyzer.

  Hereinafter, a dialog log analysis device, a dialog log analysis method, and a program according to embodiments will be described in detail with reference to the accompanying drawings.

<Outline of Embodiment>
The dialog log analysis apparatus according to the embodiment analyzes a dialog log by natural language processing, a dialog block (second dialog block) that is a block of a large dialog in the dialog log, and a detailed dialog inserted in the large dialog The range and the label are determined for each of the interrupt dialog blocks (first dialog blocks) which are the blocks. Then, the dialog log analysis apparatus according to the embodiment controls the dialog log so that display information that can recognize the range and label of the dialog block and the interrupt dialog block in the dialog log is added and displayed. Let the user browse. As a result, the flow of a large dialogue in the dialogue log and the structural relationship of fine dialogues inserted in the big dialogue can be communicated to the user at a glance.

<First Embodiment>
FIG. 1 is a block diagram illustrating a functional configuration example of the dialog log analysis apparatus according to the first embodiment. As shown in FIG. 1, the dialogue log analysis apparatus according to the present embodiment includes an utterance sentence dividing unit 11 (dividing unit), a dialogue action estimating unit 12 (estimating unit), and an interrupting dialogue block determining unit 13 (first determining unit). ), A first rule database 14, a dialogue block determination unit 15 (second determination unit), a second rule database 16, and a display screen generation unit 17 (display control unit). The dialog log analysis device according to the present embodiment allows the user to browse the display screen DS through the processing by the above-described units with respect to the dialog log D1 input from, for example, an external device.

  FIG. 2 is a diagram showing an example of the dialogue log D1, and is an example on the subject of the exchange between the operator and the customer in the call center. The dialogue log D1 shown in FIG. 2 shows only a part of the first half of the dialogue. As shown in FIG. 2, the dialogue log D1 includes a plurality of utterances constituting a dialogue arranged in time series in the order of utterances, and each utterance shows the speaker of the utterance. Speaker tags are associated. The speaker tag may be added inside the dialogue log analysis apparatus of this embodiment. Hereinafter, the above-described units in the dialogue log analysis apparatus according to the present embodiment will be described together with specific examples while appropriately using the dialogue log D1 shown in FIG. 2 as an example.

  The utterance sentence dividing unit 11 divides the utterance sentence included in the dialogue log D1 into unit utterance sentences of a predetermined unit. For example, the utterance sentence dividing unit 11 divides the utterance sentence included in the dialogue log D1 into unit utterance sentences according to a predetermined division rule. An example of the dialogue log D1 'in which the utterance sentence is divided into unit utterance sentences is shown in FIG. In the dialogue log D1 'shown in FIG. 3, the utterance sentence on the first line and the utterance sentence on the eleventh line of the dialogue log D1 shown in FIG. 2 are each divided into two unit utterance sentences. If each utterance sentence included in the input dialogue log D1 is already the unit utterance sentence of the dialogue log D1 'shown in FIG. 3, the utterance sentence dividing unit 11 is not necessary. That is, as long as the dialog log analysis apparatus receives the dialog log D1 in the format of the dialog log D1 'shown in FIG.

  The dialogue act estimation unit 12 estimates a dialogue act on a unit utterance sentence included in the dialogue log D1 'and assigns a dialogue act tag. The dialogue action is that the speaker speaks with some intention, and a tag indicating what kind of intention is the speech action tag. A specific example of the dialogue action tag is shown in FIG. As the dialogue act tags, for example, as shown in FIG. 4, “information presentation”, “request”, “question”, “confirmation”, “selection”, “affirmation”, “denial”, “thank you”, “consideration” , “OK” etc. Dialogue action tags are used particularly in the technical field of natural language processing.

  FIG. 5 is a flowchart illustrating an example of a processing procedure of the dialogue action estimation unit 12. For example, the dialogue action estimation unit 12 receives the dialogue log D1 'output from the utterance sentence division unit 11 (step S101). If each utterance sentence included in the dialogue log D1 is a unit utterance sentence, the dialogue log D1 may be input. When the dialogue log D1 'is input, the dialogue action estimation unit 12 first extracts one unit utterance sentence from the dialogue log D1' (step S102). And the dialogue action estimation part 12 performs the process which specifies the word division | segmentation and the part of speech of each word by a morphological analysis with respect to the unit utterance sentence taken out by step S102 (step S103).

  Next, the dialogue action estimation unit 12 performs dialogue action on the unit utterance sentence extracted in step S102 from the information obtained by the process in step S103 based on, for example, a statistical model learned in advance or a predetermined estimation rule. Estimation is performed (step S104), and it is determined whether or not there is a corresponding dialogue action (step S105). Then, if there is a corresponding dialogue act (step S105: Yes), the dialogue act estimation unit 12 assigns a dialogue act tag indicating the dialogue act to the unit utterance sentence extracted in step S102 (step S106). If there is no corresponding interactive action (step S105: No), step S106 is skipped.

  Thereafter, the dialogue action estimation unit 12 determines whether or not all unit utterances have been extracted from the dialogue log D1 ′ input in step S101 (step S107), and if there is a unit utterance that has not been extracted (step S107). : No), returning to step S102, the subsequent processing is repeated. On the other hand, if all unit utterance sentences have been extracted (step S107: Yes), the dialogue action estimation unit 12 has a dialogue log D1 ′ to which a dialogue action tag is attached (hereinafter referred to as a tagged dialogue log D2). Is output (step S108), and the series of processing ends.

  As the statistical model used for the estimation of the dialogue action in step S104 in FIG. 5, for example, an SVM (Support Vector Machine) generally used in machine learning or a maximum entropy method can be used. In addition, as information used for estimating the dialogue action, a speaker tag, unit utterance sentences before and after included in the dialogue log D1 ', and the like may be used.

  An example of the tagged dialogue log D2 output by the dialogue action estimation unit 12 is shown in FIG. In the tagged dialogue log D2 shown in FIG. 6, for example, a unit utterance sentence “Z900 in the fifth line of the dialogue log D1 ′ shown in FIG. The dialogue act tag “Question” is added to the unit utterance sentence such as “Is it?” And the dialogue act tag “Confirm” is added to the unit utterance sentence such as “It was Z9000” on the 12th line. Is granted. Note that, as shown in the example of the tagged dialogue log D2 in FIG. 6, some unit utterances included in the dialogue log D1 'include unit utterances to which no dialogue log is assigned.

  The interrupt dialog block determination unit 13 refers to the first rule database 14 and determines the range of the interrupt dialog block in the tagged dialog log D2 and the label to be given to the interrupt dialog block. As described above, the interrupt dialogue block is a detailed dialogue block inserted in a large dialogue, and is composed of a plurality of unit utterance sentences continuous in time series. The first rule database 14 stores a set of interrupted dialogue block determination rules. The interrupt dialog block determination rule is a rule for detecting the boundary position of an interrupt dialog block on the condition of a combination of a speaker tag and a dialog action tag, and a label to be given to the interrupt dialog block is designated.

  FIG. 7 is a diagram illustrating a structure example of the first rule database 14. As shown in FIG. 7, each interrupting dialogue block determination rule constituting the first rule database 14 includes a condition unit 21 including a combination of a speaker tag and a dialogue action tag, and a result unit 22 including a boundary type and a label. And the unit utterance sentence suitable for the combination of the speaker tag and the dialogue action tag shown in the condition part 21 is a candidate for the boundary of the interruption dialogue block indicated by the boundary type of the result part 22, and the range of the interruption dialogue block Indicates that the label shown in the result section 22 is assigned to the interrupt dialog block.

  For example, the interrupt dialogue block determination rule in the first line of FIG. 7 is an interrupt in which a unit utterance sentence with a speaker tag “operator” and a dialogue action tag “question” is given a label “confirm customer status”. When the end position of the interrupt dialog block is detected and the range of the interrupt dialog block is determined, the interrupt dialog block is labeled “Check Customer Status”. Show. Further, in the interrupt dialogue block determination rule in the second line of FIG. 7, the unit utterance with the speaker tag “operator” and the dialogue action tag “confirmation” or “acceptance” is labeled “customer status confirmation”. When the interrupt dialog block is assigned, the unit utterance sentence before the unit utterance sentence is a candidate for the start position of the interrupt dialog block. This indicates that a label “confirm user status” is given to the dialogue block.

  FIG. 8 is a flowchart illustrating an example of a processing procedure of the interrupt dialog block determination unit 13. The interrupted dialogue block determination unit 13 receives the tagged dialogue log D2 output by the dialogue action estimation unit 12 (step S201). When the tagged conversation log D2 is input, the interrupted dialogue block determination unit 13 refers to the first rule database 14 (step S202), and based on the interrupted dialogue block determination rule of the first rule database 14 in step S201. The boundary position of the interrupt dialog block is detected from the input dialog log D2 with tag (step S203). Specifically, the interrupt dialog block determination unit 13 searches for a unit utterance sentence that matches the condition part 21 of the interrupt dialog block determination rule from the tagged dialog log D2 input in step S201, and starts the interrupt dialog block. When the position and end position can be uniquely identified, the process of determining the range of the interrupt dialog block is repeated.

  For example, in the example of the tagged dialogue log D2 shown in FIG. 6, since the speaker tag of the unit utterance sentence on the eighth line is “operator” and the dialogue action tag is “question”, the first tag shown in FIG. It matches the condition part 21 of the interrupt dialog block determination rule on the first line of the rule database 14. Therefore, the interrupt dialog block determination unit 13 sets the position of the unit utterance sentence in the tagged dialog log D2 as a candidate of the start position of the interrupt dialog block to which the label “confirm customer status” is given. Further, in the example of the tagged dialogue log D2 shown in FIG. 6, the speaker tag of the unit utterance sentence on the 12th line is “operator”, and the dialogue action tag is “confirmation”. It matches the condition part 21 of the interrupt dialogue block determination rule on the second line of the rule database 14. Therefore, the interrupt dialog block determination unit 13 sets the position of the unit utterance sentence in the tagged dialog log D2 as a candidate end position of the interrupt dialog block to which the label “confirm customer status” is given. Then, when the candidate for the start position and the end position of the interrupt dialog block to which the same label is assigned are found in this order, for example, the condition part 21 of another interrupt dialog block determination rule between them is displayed. If the starting position and the ending position can be uniquely specified without a matching unit utterance sentence, the interval between them is determined as the range of the interrupt dialog block.

  Thereafter, the interrupt dialog block determination unit 13 outputs a determination result D3 indicating the range of the interrupt dialog block whose range is determined by the process of step S203 and the label to be given to the interrupt dialog block (step S204). The process ends.

  Note that the above-described interrupt dialog block determination rule is an example, and the present invention is not limited to this. For example, a rule that also considers unit utterance sentences immediately before and after may be used. Specifically, in addition to the condition part 21 of the interrupt dialogue block determination rule on the first line of the first rule database 14 shown in FIG. 7, the speaker tag of the unit utterance immediately after is “user” and the dialogue action tag is “ Using the rule that the condition “Yes” is added, and the unit utterance sentence that matches the condition indicated by the condition part 21 is set as a candidate for the end position of the interrupt dialog block to which the label “confirmation of customer status” is given. Also good. Further, the boundary type of the result section 22 is not limited to “start” and “end”, and other expressions that can indicate the boundary position of the interrupt dialog block may be used.

  Also, the interrupt dialog block determination rule may be added with information such as a priority order with respect to another interrupt dialog block determination rule. In this case, for example, the interrupt dialog block determination unit 13 detects another candidate for the start position of the interrupt dialog block to which a certain label is assigned, and does not detect the end position candidate for the interrupt dialog block. When a candidate for the start position of an interrupt dialog block to which a label is attached is detected, it is possible to determine which interrupt dialog block start position has priority according to the priority order of the interrupt dialog block determination rule used. It becomes like this.

  An example of the data structure of the determination result D3 of the interrupt dialog block determination unit 13 is shown in FIG. The determination result D3 output by the interrupt dialog block determination unit 13 is, for example, as shown in FIG. 9, an interrupt that is identification information uniquely assigned to the interrupt dialog block whose range is determined by the process of step S203 of FIG. It includes a dialog block ID, a start pointer indicating the start position of the interrupt dialog block, an end pointer indicating the end position of the interrupt dialog block, and a label assigned to the interrupt dialog block. As the start pointer and the end pointer, for example, an offset indicating how many characters the start position and the end position are deviated from the first character of the dialog log D1 and the tagged dialog log D2 can be used. In addition, when each utterance sentence (unit utterance sentence) included in the dialogue log D1 or the tagged conversation log D2 is managed by a line number, the line number of the utterance sentence (unit utterance sentence) serving as a start position or an end position. May be used as a start pointer or an end pointer. The range of the interrupt dialog block is between the start pointer and the end pointer.

  The interaction block determination unit 15 refers to the second rule database 16 to determine the range of interaction blocks in the tagged interaction log D2 and the label to be assigned to the interaction block. As described above, the dialogue block is a large dialogue block in the dialogue log D1, and is a unit block larger than the interruption dialogue block. In other words, the dialog block and the interrupt dialog block have a structural relationship in which the interrupt dialog block is included in the dialog block as part of a larger unit dialog block. The second rule database 16 stores a set of dialogue block determination rules. Similar to the interrupting dialog block determination rule, the dialog block determination rule is a rule for detecting the boundary position of a dialog block on the condition of a combination of a speaker tag and a dialog action tag, and a label to be given to the dialog block is designated. ing.

  FIG. 10 is a diagram illustrating a structure example of the second rule database 16. As shown in FIG. 10, each dialogue block determination rule constituting the second rule database 16 includes a condition unit 31 including a combination of a speaker tag and a dialogue act tag, a result unit 32 including a boundary type and a label, And the unit utterance sentence that matches the combination of the speaker tag and the dialogue action tag shown in the condition part 31 is a candidate for the boundary of the dialogue block indicated by the boundary type of the result part 32, and the range of the dialogue block is determined. This indicates that the label shown in the result section 32 is given to the dialogue block.

  For example, in the dialogue block determination rule on the first line in FIG. 10, the unit utterance with the speaker tag “customer” and the dialogue action tag “information presentation” or “question” If the dialog block is a candidate for the end position of a dialog block that is labeled "Start", and the boundary of another dialog block has not been detected before that, the dialog block is labeled "Start dialog" It is shown that. Further, in the dialogue block determination rule in the second line of FIG. 10, the unit utterance with the speaker tag “customer” and the dialogue action tag “information presentation” or “question” is labeled “inquiry from customer”. It becomes a candidate for the start position of the given dialog block, and when the end position of the dialog block is detected and the range of the dialog block is determined, the label “inquiry from customer” is given to the dialog block. Show.

  Further, in the dialogue block determination rule on the third line in FIG. 10, the unit utterance with the speaker tag “operator” and the dialogue action tag “information presentation” is the unit utterance immediately before the “utterance from customer”. When the dialogue block end position is given as a candidate, the unit utterance sentence before the previous unit utterance sentence is a candidate for the dialogue block start position, and the dialogue block range is determined, This indicates that the label “inquiry from customer” is given to the dialogue block. Further, the dialogue block determination rule on the fourth line in FIG. 10 is that the unit utterance sentence with the speaker tag “operator” and the dialogue action tag “information presentation” is given the label “answer from operator”. It becomes a candidate for the start position of the block, and when the end position of the dialog block is detected and the range of the dialog block is determined, the label “answer from operator” is given to the dialog block.

  FIG. 11 is a flowchart illustrating an example of a processing procedure of the dialogue block determination unit 15. The dialogue block determination unit 15 receives the tagged dialogue log D2 output from the dialogue action estimation unit 12 and the determination result D3 output from the interruption dialogue block determination unit 13 (step S301). The dialogue block determination unit 15 refers to the second rule database 16 when the tagged dialogue log D2 and the determination result D3 of the interruption dialogue block determination unit 13 are input (step S302), and the dialogue block of the second rule database 16 is referred to. Based on the determination rule and the determination result D3 of the interrupt interaction block determination unit 13 input in step S301, the boundary position of the interaction block is detected from the tagged interaction log D2 input in step S301 (step S303).

  Specifically, the dialogue block determination unit 15 searches for a unit utterance sentence that matches the condition unit 31 of the dialogue block determination rule from unit utterance sentences included in the tagged dialogue log D2 input in step S301, and performs dialogue. When the block start position and end position can be uniquely identified, the process of determining the dialog block range is repeated. At this time, the dialogue block determination unit 15 determines the unit utterance sentence within the range of the interruption dialogue block determined by the interruption dialogue block determination unit 13 so that the range of the interruption dialogue block is not divided by the boundary of the dialogue block. Among them, the unit utterance sentence excluding the start position and end position of the interrupt dialog block is excluded from the search target, and the unit utterance sentence that matches the condition part 21 of the interrupt dialog block determination rule is searched.

  For example, in the example of the tagged dialogue log D2 shown in FIG. 6, since the speaker tag of the unit utterance sentence on the fifth line is “customer” and the dialogue action tag is “question”, the second tag shown in FIG. It matches the condition part 31 of the dialogue block determination rule on the first and second lines of the rule database 16. Therefore, the dialogue block determination unit 15 sets the position of the unit utterance immediately before the unit utterance in the tagged dialogue log D2 as a candidate of the end position of the dialogue block to which the label of “dialogue start” is given. The position of the unit utterance is set as a candidate for the start position of the dialogue block to which the label “inquiry from the customer” is given. Here, in the example of the tagged dialogue log D2 shown in FIG. 6, since there is no unit utterance sentence that matches the condition part 31 of the dialogue block determination rule before the fourth line, the top of the tagged dialogue log D2 The position of the unit utterance sentence is set as the start position of the dialog block to which the label “start dialog” is given, and the first to fourth lines from the tagged dialog log D2 are determined as the range of the dialog block. Further, the tagged dialogue log D2 shown in FIG. 6 omits the 15th and subsequent lines in accordance with the example of the dialogue log D1 of FIG. 2, but at a position where the range of the interrupted dialogue block is not divided after the 15th line. For example, if there is a unit utterance sentence in which the speaker tag is “operator” and the dialogue action tag is “information presentation”, the dialogue block determination on the third and fourth lines of the second rule database 16 shown in FIG. Since it matches the condition part 31 of the rule, the position of the unit utterance immediately before the unit utterance becomes a candidate for the end position of the dialogue block to which the label “inquiry from the customer” is given, and the position from the fifth line Is determined as the range of the dialogue block to which the label “inquiry from the customer” is given.

  Thereafter, the dialogue block determination unit 15 outputs a determination result D4 indicating the range of the dialogue block whose range has been determined by the processing in step S303 and the label to be given to the dialogue block (step S304), and ends the series of processing. .

  The above-described dialog block determination rule is an example, and the present invention is not limited to this. For example, as in the case of the interrupt dialog block determination rule, a rule that also considers the unit utterances immediately before and immediately after may be used. Further, the boundary type of the result section 32 is not limited to “start” and “end immediately before”, and other expressions that can indicate the boundary position of the dialogue block may be used.

  The dialogue block determination rule may be added with information such as a priority order with respect to another dialogue block determination rule. In this case, for example, the dialog block determination unit 15 detects another candidate of the start position of the dialog block to which a certain label is assigned, and assigns another label in a state where the candidate of the end position of the dialog block is not detected. When a candidate for the start position of the interaction block to be detected is detected, it is possible to determine which interaction block start position is given priority according to the priority order of the used interaction block determination rule.

  In the present embodiment, the dialogue block determination unit 15 detects the boundary position of the dialogue block using the determination result D3 of the interruption dialogue block determination unit 13 in addition to the dialogue block determination rule of the second rule database 16. However, the boundary position of the dialogue block may be detected without using the determination result D3 of the interruption dialogue block determination unit 13. In this case, there is a concern that an inconsistency occurs between the interrupt dialog block range and the dialog block range. When such an inconsistency occurs, for example, when the display screen generation unit 17 to be described later generates the display screen DS. In addition, the display position may be adjusted so as to eliminate the mismatch.

  An example of the data structure of the determination result D4 of the dialogue block determination unit 15 is shown in FIG. The determination result D4 output by the dialog block determination unit 15 is an identification uniquely assigned to the dialog block whose range is determined by the process of step S303 in FIG. 11, similarly to the determination result D3 of the interrupt dialog block determination unit 13. The dialogue block ID which is information, a start pointer indicating the start position of the dialog block, an end pointer indicating the end position of the dialog block, and a label assigned to the dialog block are included. As the start pointer and the end pointer, an offset, a line number, or the like can be used as in the determination result D3 of the interrupt dialog block determination unit 13. The range of the dialogue block is between the start pointer and the end pointer.

  The display screen generation unit 17 controls the dialog log D1 to add and display the first display information indicating the range and label of the interrupted dialog block and the second display information indicating the range and label of the dialog block. To do. For example, the display screen generation unit 17 uses the determination result D3 of the interrupt dialog block determination unit 13 and the determination result D4 of the dialog block determination unit 15 to display the first display information indicating the range and label of the interrupt dialog block, A display screen DS in which the second display information indicating the range and label of the dialog block is added to the dialog log D1 is generated.

  An example of the display screen DS generated by the display screen generation unit 17 is shown in FIG. The display screen DS shown in FIG. 13 includes first display information in which the label L of the interrupt dialog block B1 is clearly shown in the frame F surrounding the range of the interrupt dialog block B1 in the dialog log D1, and the dialog block B2 in the dialog log D1. In the frame F surrounding the range, the second display information in which the label L of the dialogue block B2 is clearly displayed is superimposed on the dialogue log D1. The frame F surrounding the range of the interrupt dialog block B1 and the frame F surrounding the range of the dialog block B2, for example, are displayed in different colors, so that the ranges of the interrupt dialog block B1 and the dialog block B2 can be viewed at a glance. It can be recognized. Further, the frame F enclosing the range of the interrupt dialog block B1 has a shorter width (length in a direction orthogonal to the direction in which the spoken sentences are arranged) than the frame F enclosing the range of the dialog block B2, and the interrupt dialog block B1. And the dialogue block B2 can be grasped at a glance.

  The display screen DS generated by the display screen generation unit 17 is provided so as to be viewable to the user. For example, if the dialog log analysis device of the present embodiment includes a display device, the user can view the display screen DS by displaying the display screen DS on the display device. Further, the display screen DS may be displayed on the display device of the external device via a browser installed in the external device used by the user, thereby allowing the user to view the display screen DS. By viewing this display screen DS, the user can easily grasp the flow of a large dialogue in the dialogue log D1 and the structural relationship of fine dialogues inserted in the large dialogue.

  The display screen DS shown in FIG. 13 is an example, and the present invention is not limited to this. For example, as shown in FIG. 14, the display screen DS may be configured such that the label L of the interrupt dialog block B1 or the label L of the dialog block B2 is used as a heading and displayed in association with each frame F. At this time, the display position of the label L displayed as a heading may be configured to be different between the interrupt dialog block B1 and the dialog block B2. Also, the shape of the frame F may be different between the interrupting dialogue block B1 and the dialogue block B2. Further, the display screen DS may be configured to display the dialog log D1 by adding first display information indicating the range and label of the interrupted dialog block and second display information indicating the range and label of the dialog block. The display form is not limited to the image form described above, and any display form may be used.

  As described above in detail with reference to specific examples, the dialog log analysis apparatus according to the present embodiment has the speaker tag of each utterance sentence included in the dialog log D1 and the dialog action given by the dialog action estimation. Using the combination with the tag as a clue, the range and label of the interrupted dialogue block and the range and label of the dialogue block in the dialogue log D1 are determined. Then, a display screen DS to which the first display information indicating the range and label of the interrupting dialog block and the second display information indicating the range and label of the dialog block is added is generated for the dialog log D1, and viewed by the user We try to provide it as possible. Therefore, according to the dialog log analysis device of the present embodiment, the flow of a large dialog in the dialog log D1 and the structural relationship of the detailed dialog inserted in the large dialog are communicated to the user at a glance. be able to.

Second Embodiment
Next, a second embodiment will be described. The present embodiment further uses a keyword extracted from an utterance sentence included in the dialog log D1 (unit utterance sentence included in the dialog log D1 ′) or a concept class corresponding to a superordinate concept of the keyword, It is an example which detects the boundary of a dialogue block. In the following description, the same components as those in the first embodiment are denoted by the same reference numerals, and redundant description is omitted as appropriate. Only the characteristic features of the present embodiment will be described.

  FIG. 15 is a block diagram illustrating a functional configuration example of the dialog log analysis apparatus according to the second embodiment. As shown in FIG. 15, the dialog log analysis device of the present embodiment includes a keyword extraction unit 18 (first extraction unit) in addition to the configuration of the dialog log analysis device of the first embodiment (see FIG. 1).

  The keyword extraction unit 18 performs a process of extracting a keyword from the unit utterance sentence included in the dialogue log D <b> 1 ′, and assigns a keyword to the unit utterance sentence when the keyword can be extracted.

  FIG. 16 is a flowchart illustrating an example of a processing procedure of the keyword extraction unit 18. For example, the dialogue log D1 'output from the utterance sentence dividing unit 11 is input to the keyword extracting unit 18 (step S401). If each utterance sentence included in the dialogue log D1 is a unit utterance sentence, the dialogue log D1 may be input. When the dialogue log D1 'is input, the keyword extraction unit 18 first extracts one unit utterance sentence from the dialogue log D1' (step S402). Then, the keyword extraction unit 18 performs a process of specifying the word division and the part of speech of each word by morphological analysis on the unit utterance sentence extracted in step S402 (step S403).

  Next, the keyword extraction unit 18 refers to, for example, a keyword list in which a predetermined keyword is registered (step S404), and uses the unit utterance extracted in step S402 using the information obtained in step S403. It is determined whether there is a word (or word string) that matches the keyword registered in the keyword list in the sentence (step S405). Then, if there is a word (or word string) that matches the keyword registered in the keyword list (step S405: Yes), the keyword extracting unit 18 extracts the word (or word string) as a keyword in step S402. It is given to the unit utterance sentence (step S406). If there is no word that matches the keyword registered in the keyword list (step S405: No), step S406 is skipped.

  Thereafter, the keyword extraction unit 18 determines whether or not all unit utterance sentences have been extracted from the dialogue log D1 ′ input in step S401 (step S407). If there is a unit utterance sentence that has not been extracted (step S407: No), it returns to step S402 and the subsequent processing is repeated. On the other hand, if all the unit utterance sentences have been extracted (step S407: Yes), the keyword extraction unit 18 outputs the dialogue log D1 ′ (the dialogue log with keywords) to which the keyword is assigned (step S408), The process ends.

  In the above example, the keyword is extracted from the unit utterance sentence with reference to the keyword list. However, the keyword extraction unit 18 does not refer to the keyword list, but uses the part of speech or the clue word, etc. The structure which extracts a keyword from a sentence may be sufficient. The keyword extracting unit 18 extracts a word or a word string that matches a keyword registered in the keyword list and is included in the unit utterance sentence as a keyword without performing morphological analysis on the unit utterance sentence. There may be.

  Further, the keyword extraction unit 18 may be configured to convert the extracted keyword into a concept class and to assign the keyword concept class to the unit utterance sentence. An example of the concept class is shown in FIG. As shown in FIG. 17, the concept class indicates a superordinate concept class to which a vocabulary (word or word string) serving as a keyword belongs. For example, in the example shown in FIG. 17, “product name” is assigned as the concept class to the vocabulary that is the keyword “Z9000”. When the keyword “Z9000” is extracted from the unit utterance sentence, the keyword extraction unit 18 converts the keyword into a concept class “product name” and assigns it to the unit utterance sentence. A concept class may be assigned together with a keyword. As shown in FIG. 17, keywords corresponding to each concept class may be defined in terms of words as vocabulary, or a vocabulary pattern such as “Z + (numerical value)” may be described by a rule. A keyword and a concept class may be used together, and in the case of a specific concept class, the concept class may be used without using the keyword. In other cases, the keyword may be used as it is. In the following description, it is assumed that a concept class is used.

  The keyword-added dialogue log output from the keyword extraction unit 18 is merged with the tagged dialogue log D2 output from the dialogue action estimation unit 12 to obtain a keyword / tagged dialogue log D5. An example of the interaction log D5 with keyword / tag is shown in FIG. In the dialogue log D5 with keyword / tag shown in FIG. 18, the unit utterance sentence “Z900 or Z9000?” On the 10th line of the dialogue log D1 ′ shown in FIG. The unit utterance sentence “Oh, it is Z9000” and the unit utterance sentence “Let ’s go to Z9000” on the 12th line are each assigned a concept class of the keyword “product name”. .

  The interrupt dialog block determination unit 13 of the present embodiment refers to the first rule database 14 ′ and determines the range of the interrupt dialog block in the keyword-tagged dialog log D 5 and the label assigned to the interrupt dialog block. The first rule database 14 ′ stores a set of interrupted dialogue block determination rules in the same manner as the first rule database 14 of the first embodiment. The interrupted dialogue block determination rule includes a speaker tag, an interactive action tag, and a keyword. Alternatively, a rule for detecting the boundary position of the interrupt dialog block is included on condition that the keyword is combined with the concept class.

  FIG. 19 is a diagram illustrating a structure example of the first rule database 14 ′ of the present embodiment. As shown in FIG. 19, each interrupt dialogue block determination rule constituting the first rule database 14 ′ of the present embodiment includes a speaker tag, a dialogue action tag, and a keyword (concept class) added to the condition as necessary. And a result part 42 including a boundary type and a label, and is adapted to a combination of a speaker tag and a dialogue action tag shown in the condition part 41 and a keyword (concept class) as necessary. The unit utterance sentence to be a candidate for the boundary of the interrupt dialog block indicated by the boundary type of the result part 42, and when the range of the interrupt dialog block is fixed, the label shown in the result part 42 is given to the interrupt dialog block ing.

  For example, the interrupt dialogue block determination rule on the fourth line in FIG. 19 is a unit utterance sentence in which the speaker tag is “operator”, the dialogue action tag is “confirmation”, and the keyword (concept class) is “model number” or “product name”. Is a candidate for the end position of the interrupt dialog block with the label “Product name confirmation”, and the unit utterance sentence before the unit utterance sentence is a candidate for the start position of the interrupt dialog block. When the range of the dialogue block is determined, the label “confirm product name” is given to the interrupt dialogue block. Further, the interrupt dialogue block determination rule on the sixth line in FIG. 19 is that the unit utterance sentence with the speaker tag “operator”, the dialogue action tag “confirmation”, and the keyword (concept class) “product type” is “product”. The interrupt dialog block end position candidate with the “type check” label is assigned, and the unit utterance sentence before the unit utterance sentence is a candidate for the start position of the interrupt dialog block, and the range of the interrupt dialog block Is confirmed, a label “confirm product type” is assigned to the interrupt dialogue block.

  The interrupt dialogue block determination unit 13 of the present embodiment refers to the first rule database 14 ′ as shown in FIG. 19, and unit utterances that match the condition unit 41 of the interrupt dialogue block determination rule from the keyword-tagged dialogue log D5. Processing for searching for a sentence, determining the range of the interrupt dialog block when the start position and end position of the interrupt dialog block can be uniquely identified, and assigning the label indicated by the result section 42 to the interrupt dialog block repeat.

  For example, in the example of the dialogue log D5 with keyword / tag shown in FIG. 18, the speaker tag of the unit utterance sentence on the eighth line is “operator” and the dialogue action tag is “question”. It conforms to the condition part 41 of the interrupt dialogue block determination rule on the third and fifth lines of the first rule database 14 ′. Therefore, the interrupt dialog block determination unit 13 is given the position of the unit utterance sentence in the keyword-tagged dialog log D5 by the interrupt dialog block to which the “product name confirmation” label is given or “product type confirmation”. As a candidate for the start position of the interrupt dialog block. In the example of the dialogue log D5 with keyword / tag shown in FIG. 18, the speaker tag of the unit utterance sentence on the 12th line is “operator”, the dialogue action tag is “confirmation”, and the keyword (concept class) is “product name”. Therefore, it conforms to the condition part 41 of the interrupt dialog block determination rule on the fourth line of the first rule database 14 ′ shown in FIG. Therefore, the interrupt dialog block determination unit 13 determines the range from the 8th line to the 12th line in the keyword / tagged dialog log D5 as the range of the interrupt dialog block to which the “product name confirmation” label is assigned, and the interrupt dialog Label the block with “Product name confirmation”.

  It is to be noted that the unit utterance sentence in the keyword-tagged dialogue log D5 includes an interrupt dialogue block determination rule including a keyword (concept class) in the condition portion 41 and an interrupt dialogue block determination rule including no keyword (concept class) in the condition portion 41. In both cases, it is desirable to preferentially apply the interrupt dialog block determination rule including the keyword (concept class) in the condition part 41.

  In addition, the dialogue block determination unit 15 of the present embodiment refers to the second rule database 16 'to determine the range of dialogue blocks in the keyword-tagged dialogue log D5 and the label to be given to the dialogue block. Similar to the second rule database 16 of the first embodiment, the second rule database 16 ′ stores a set of dialogue block determination rules. The dialogue block determination rule includes a speaker tag, a dialogue action tag, and a keyword or keyword. The rule for detecting the boundary position of the dialogue block is included under the condition of combination with the above concept class.

  FIG. 20 is a diagram illustrating a structure example of the second rule database 16 ′ of the present embodiment. As shown in FIG. 20, each dialogue block determination rule constituting the second rule database 16 ′ of the present embodiment includes a speaker tag, a dialogue action tag, and a keyword (concept class) added to the condition as necessary. It has a condition part 51 composed of a combination and a result part 52 including a boundary type and a label, and is adapted to a combination of a speaker tag and a dialogue action tag shown in the condition part 51 and a keyword (concept class) as necessary. The unit utterance sentence becomes a candidate for the boundary of the dialogue block indicated by the boundary type of the result unit 52, and when the range of the dialogue block is determined, the label shown in the result unit 52 is given to the dialogue block.

  For example, the dialogue block determination rule on the fifth line in FIG. 20 is that the unit utterance sentence with the speaker tag “operator”, the dialogue action tag “information presentation”, and the keyword (concept class) “operator name” The unit utterance sentence is a candidate for the end position of the dialogue block to which the label “Answer from operator” is given, and the candidate for the start position of the dialogue block to which the label “Answer from operator” is given before is detected If it is, it indicates that the label “answer from operator” is given to the dialogue block. Further, the dialogue block determination rule on the sixth line in FIG. 20 is that the unit utterance sentence with the speaker tag “operator”, the dialogue action tag “information presentation”, and the keyword (concept class) “operator name” is “dialogue”. If the dialog block is a candidate for the start position of a dialog block that is labeled "End", and no other dialog block boundary has been detected after that, the dialog block is labeled "End Dialog" It is shown that.

  The dialogue block determination unit 15 of the present embodiment refers to the second rule database 16 ′ as shown in FIG. 20 and selects a unit utterance sentence that matches the condition block 51 of the dialogue block determination rule from the keyword-tagged dialogue log D5. By searching, when the start position and end position of the dialog block can be uniquely identified, the range of the dialog block is determined, and the process of assigning the label indicated by the result unit 52 to the dialog block is repeated.

  The display screen generation unit 17 indicates the range and label of the interrupt dialog block using the determination result D3 of the interrupt dialog block determination unit 13 and the determination result D4 of the dialog block determination unit 15 as in the first embodiment. A display screen DS in which the first display information and the second display information indicating the range and label of the dialog block are added to the dialog log D1 is generated. An example of the display screen DS generated by the display screen generation unit 17 of the present embodiment is shown in FIG. In the present embodiment, since the first rule database 14 ′ shown in FIG. 19 is used, it can be seen that a label L of “product name confirmation” is given to the interrupt dialogue block B 1.

  As described above, the dialogue log analysis apparatus according to the present embodiment uses the keyword extracted from the utterance sentence (unit utterance sentence) or the concept class of the keyword as a condition in addition to the speaker tag and the dialogue action tag. The boundaries of blocks and dialogue blocks are detected. Therefore, according to the dialog log analysis apparatus of the present embodiment, it is possible to perform interrupt dialog block and dialog block determination more accurately and in detail than in the first embodiment described above, and a large dialog flow in the dialog log D1 It is easy to understand at a glance and accurately to the user the structural relationship of the detailed dialogue inserted in a large dialogue.

<Third Embodiment>
Next, a third embodiment will be described. In the present embodiment, the interrupt dialog block determination rule designates the upper label of the label in addition to the label given to the interrupt dialog block. The higher level label is a label of a higher level concept of the label to be assigned to the interrupting dialogue block, and is a label common to the label specified in any of the dialogue block determination rules. Then, the interrupt dialog block determination unit 13 determines not only the label of the interrupt dialog block but also its upper label, and the dialog block determination unit 15 determines the upper label determined by the interrupt dialog block determination unit 13 among the dialog block determination rules. And a dialog block range including the interrupt dialog block range determined by the interrupt dialog block determination unit 13 is determined based on a dialog block determination rule that specifies a common label.

  In addition, since the outline | summary of the structure of a dialog log analyzer and the whole process is common with 1st Embodiment or 2nd Embodiment mentioned above, below, about the same component as 1st Embodiment, the same code | symbol is attached | subjected below. A duplicate description will be omitted as appropriate, and only the characteristic features of this embodiment will be described.

  FIG. 22 is a diagram illustrating a structure example of the first rule database 14 ″ according to the present embodiment. As shown in FIG. 22, each interrupt dialogue block determination rule constituting the first rule database 14 ″ of the present embodiment includes a speaker tag, a dialogue action tag, and a keyword (concept class) that is added to the condition as necessary. And a result part 62 including a boundary type, a label, and a higher level label. A speaker tag and a dialogue act tag shown in the condition part 61 and keywords (conceptual classes) as necessary The unit utterance sentence that matches the combination of the above becomes a candidate for the boundary of the interrupt dialog block indicated by the boundary type of the result section 62, and when the range of the interrupt dialog block is determined, the label shown in the result section 62 is given to the interrupt dialog block. In addition, the higher level label is determined. In the example of FIG. 22, the upper label of the interrupt dialogue block to which the label “product name confirmation” is given is “inquiry from customer”.

  The interrupt dialogue block determination unit 13 of the present embodiment refers to the first rule database 14 '' as shown in FIG. 22, and is a unit suitable for the interrupt dialogue block determination rule condition unit 61 from the keyword-tagged dialogue log D5. Searching the utterance sentence, if the start position and end position of the interrupt dialog block can be uniquely identified, the range of the interrupt dialog block is determined, and a label indicated by the result unit 62 is assigned to the interrupt dialog block. At the same time, the process of determining the upper label is repeated. In this embodiment, as in the second embodiment, an example in which a boundary candidate of an interrupt dialog block is detected using a keyword or a keyword conceptual class is shown. However, as in the first embodiment, a keyword or a keyword is detected. A configuration that does not use the concept class of.

  Further, the dialogue block determination unit 15 of the present embodiment refers to the second rule database 16 ′ as shown in FIG. 20 (for example, the second rule database 16 as shown in FIG. 10 is used as in the second embodiment). If a unit utterance sentence matching the condition block 51 of the dialogue block determination rule is searched from the keyword-tagged dialogue log D5, and the dialogue block start position and end position can be uniquely identified. The process of determining the range of the dialogue block and assigning the label indicated by the result unit 52 to the dialogue block is repeated. At this time, when the dialog block determining unit 15 of the present embodiment determines the range of dialog blocks including the interrupt dialog block whose range has already been determined by the interrupt dialog block determining unit 13, the dialog block of the second rule database 16 ′ is used. Of the judgment rules, only the dialogue block judgment rule that specifies a label common to the upper label of the interrupt dialogue block is used.

  For example, when the interrupt dialog block determination unit 13 determines the range of the interrupt dialog block to which the “product name confirmation” label is assigned, the upper label of the interrupt dialog block is “inquiry from the customer”. The block determination unit 15 determines the range of the dialogue block including the range of the interrupt dialogue block to which the “product name confirmation” label is assigned, using only the dialogue block determination rule that specifies the label “inquiry from the customer”. To do. That is, when the rule A that specifies “inquiry from the customer” as a label as a dialog block determination rule and the rule B that specifies other labels are likely to match, the dialog determined by these rules A and B If there is an interrupt dialog block within the block range and the upper label is “inquiry from customer”, rule B is rejected and the range of the dialog block is determined using rule A.

  As described above, the dialog log analysis apparatus according to the present embodiment, when determining the dialog block range including the interrupt dialog block range, specifies the dialog block determination rule that specifies a label common to the upper label of the interrupt dialog block. Only to use. Therefore, according to the dialog log analysis apparatus of the present embodiment, it is possible to more easily and accurately perform the dialog block determination than the first embodiment and the second embodiment described above. It is easy to understand at a glance and accurately to the user the flow and the structural relationship of the detailed dialogue inserted in a large dialogue.

<Fourth embodiment>
Next, a fourth embodiment will be described. In this embodiment, information to be extracted from the dialog log D1 is determined in advance for each interrupted dialog block and dialog block label, and a display screen DS for displaying the information extracted from the dialog log D1 is generated to generate a user. It is an example provided to be viewable. In the following description, the same components as those in the first to third embodiments described above are denoted by the same reference numerals, and redundant description will be omitted as appropriate, and only the characteristic features of this embodiment will be described.

  FIG. 23 is a block diagram illustrating a functional configuration example of the dialog log analysis apparatus according to the fourth embodiment. As shown in FIG. 23, the dialog log analysis device of this embodiment includes an information extraction unit 19 (second extraction unit) in addition to the configuration of the dialog log analysis device of the second embodiment (see FIG. 15). In the example shown in FIG. 23, the information extraction unit 19 is added to the configuration of the second embodiment, but the information extraction unit 19 is added to the configuration of the first embodiment (see FIG. 1). It may be a configuration.

  The information extraction unit 19 receives the interaction log D1, the determination result D3 of the interruption interaction block determination unit 13, and the determination result D4 of the interaction block determination unit 15. Then, the information extraction unit 19 includes, among the utterances included in the interaction log D1, the utterances included in the range of the interrupt interaction block determined by the interrupt interaction block determination unit 13, that is, a plurality of utterances constituting the interrupt interaction block. To extract information corresponding to the label of the interrupt dialog block. In addition, the information extraction unit 19 includes, from the utterances included in the dialog log D1, the utterances included in the range of the dialog block determined by the dialog block determination unit 15, that is, the plurality of utterances constituting the dialog block, Information corresponding to the label of the dialogue block is extracted.

  It is assumed that what information is extracted in accordance with the interrupt dialog block and the label of the dialog block is predetermined. For example, in “start of dialogue”, since an operator generally expresses his / her affiliation and the like, “inquiry destination” is defined as information to be extracted. At this time, since “inquiry destination” is information included in the utterance of the operator, the information of “inquiry destination” may be extracted only for the utterance sentence whose speaker tag is “operator”. Also, in “Product Name Confirmation”, it is assumed that the operator specifically asks what product such as “model number” or “product name” is pointing to, so such information is “Product Name Confirmation”. Is previously determined as information to be extracted.

  The information extraction unit 19 can extract information corresponding to the interrupted dialogue block and the dialogue block label from the dialogue log D1 by the same method as the keyword extraction unit 18 described above, for example. That is, for example, the information extraction unit 19 can refer to a list in which information to be extracted in advance is registered, and extract information that matches the information registered in the list from the utterance sentence. Or the structure which extracts information from an utterance sentence using the part of speech, the clue word, etc. which were obtained by carrying out the morphological analysis of the utterance sentence may be sufficient. In addition, the information extraction unit 19 may extract information from the utterance sentence by a method different from the keyword extraction unit 18 described above. Information extracted by the information extraction unit 19 is passed to the display screen generation unit 17.

  The display screen generation unit 17 of the present embodiment uses the determination result D3 of the interrupt dialog block determination unit 13, the determination result D4 of the dialog block determination unit 15, and the information extracted by the information extraction unit 19 to use the interrupt dialog block. The first display information indicating the range and label of information and the information extracted by the information extraction unit 19 according to the label, and the information extracted by the information extraction unit 19 according to the range and label of the dialogue block and the label A display screen DS in which the second display information is added to the dialogue log D1 is generated.

  An example of the display screen DS generated by the display screen generation unit 17 of the present embodiment is shown in FIG. The display screen DS shown in FIG. 24 is a first display that clearly shows the label L of the interrupt dialog block B1 and the information C corresponding to the label L within the frame F surrounding the range of the interrupt dialog block B1 in the dialog log D1. Information and second display information in which a label L of the dialog block B2 and information C corresponding to the label L are clearly displayed in the frame F surrounding the range of the dialog block B2 in the dialog log D1 are displayed on the dialog log D1. The display form is superimposed. The configuration of the display screen DS is not limited to the example shown in FIG. 24, and various modifications can be made as in the first embodiment described above.

  As described above, the dialog log analysis apparatus according to the present embodiment extracts information corresponding to the interrupt dialog block and the dialog block label from the dialog log D1, and clearly indicates the extracted information together with the interrupt dialog block and the dialog block label. A display screen DS is generated and provided to the user so that it can be viewed. Therefore, according to the dialog log analysis apparatus of the present embodiment, the flow of a large dialog in the dialog log D1 and the structural relationship of fine dialogs inserted in the large dialog are considered important in those dialogs. Along with the information that can be transmitted, it can be easily communicated to the user at a glance.

  According to at least one of the embodiments described above, it is possible to convey to a user an easy-to-understand understanding of a large dialogue flow in a dialogue log and a detailed relation between detailed dialogues inserted in a large dialogue. .

<Supplementary explanation>
The dialog log analysis apparatus according to the above-described embodiment can be realized by, for example, cooperation between hardware configuring a general computer and a program (software) executed on the computer. For example, when the computer executes a predetermined program, the above-described utterance sentence division unit 11, dialogue action estimation unit 12, interruption dialogue block determination unit 13, dialogue block determination unit 15, display screen generation unit 17, keyword extraction unit 18 Each unit such as the information extraction unit 19 can be realized. In addition, the first rule database 14 and the second rule database 16 described above can be realized using a mass storage device provided in the computer.

  FIG. 25 is a block diagram illustrating a hardware configuration example of the dialogue log analysis apparatus according to the embodiment. For example, as shown in FIG. 25, the dialog log analysis apparatus communicates with a hardware processor such as a CPU 101, a storage device such as a RAM 102 and a ROM 103, a large-capacity storage device such as an HDD 104, and the outside via a network. It has a hardware configuration as a normal computer including an I / F 105 and a device I / F 106 for connecting peripheral devices.

  At this time, the program is provided by being recorded on, for example, a magnetic disk, an optical disk, a semiconductor memory, or a similar recording medium. As long as the recording medium which records a program is a recording medium which a computer system can read, the storage format may be any form. The program may be configured to be installed in advance in the computer, or the program distributed via a network may be installed in the computer as appropriate.

  The program executed by the computer includes the above-described utterance sentence division unit 11, dialogue action estimation unit 12, interrupt dialogue block judgment unit 13, dialogue block judgment unit 15, display screen generation unit 17, keyword extraction unit 18, information extraction. The module configuration includes functional units such as the unit 19, and the above-described units are generated on the main memory such as the RAM 102 when the processor appropriately reads and executes the program.

  The dialog log analysis apparatus according to the embodiment is configured to realize part or all of the functional units described above by dedicated hardware such as an application specific integrated circuit (ASIC) or a field-programmable gate array (FPGA). It may be.

  Further, the dialog log analysis apparatus according to the embodiment may be configured as a network system in which a plurality of computers are connected so as to be communicable, and the above-described units may be realized by being distributed to the plurality of computers. In addition, the dialogue log analysis apparatus of the embodiment may be a virtual machine that operates on a cloud system.

  As mentioned above, although embodiment of this invention was described, this embodiment is shown as an example and is not intending limiting the range of invention. The novel embodiment can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the scope of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are included in the invention described in the claims and the equivalents thereof.

DESCRIPTION OF SYMBOLS 11 Speech sentence division part 12 Dialogue action estimation part 13 Interruption dialogue block determination part 14 1st rule database 15 Dialogue block determination part 16 2nd rule database 17 Display screen generation part 18 Keyword extraction part 19 Information extraction part D1 Dialogue log DS display screen

Claims (10)

A dialog log analysis device that analyzes a dialog log including a plurality of utterances constituting a dialog and a speaker tag indicating a speaker of each utterance,
An estimation unit that estimates a dialogue action for an utterance sentence included in the dialogue log and assigns a dialogue action tag;
A rule for detecting a boundary position of a first dialogue block composed of a plurality of continuous utterance sentences on the condition that at least a combination of a speaker tag and a dialogue action tag is provided, and a label to be given to the first dialogue block A first determination unit that determines a range of the first interaction block in the interaction log and a label to be assigned to the first interaction block based on a first rule to be specified;
A rule for detecting a boundary position of a second dialogue block that is larger than the first dialogue block on the condition that at least a combination of a speaker tag and a dialogue action tag is specified, and specifies a label to be given to the second dialogue block A second determination unit for determining a range of the second dialogue block in the dialogue log and a label to be given to the second dialogue block based on a second rule;
First display information indicating the range of the first dialog block and the label of the first dialog block, and second display information indicating the range of the second dialog block and the label of the second dialog block for the dialog log And a display control unit for controlling the display to be added and displayed,
A dialog log analyzer comprising: The display control unit includes the first display information in which a label of the first dialog block is associated with a frame surrounding the range of the first dialog block in the dialog log, and the range of the second dialog block in the dialog log. Control to superimpose and display the second display information in which the label of the second dialog block is associated with the surrounding frame on the dialog log;
The dialog log analyzer according to claim 1. The frame surrounding the range of the first dialog block has a shorter length in the direction orthogonal to the direction in which the spoken sentences are arranged, than the frame surrounding the range of the second dialog block.
The dialog log analyzer according to claim 2. The second determination unit further determines the range of the second interaction block in the interaction log based on the range of the first interaction block determined by the first determination unit.
The dialog log analyzer according to any one of claims 1 to 3. A division unit that divides an utterance sentence included in the dialogue log into unit utterance sentences of a predetermined unit;
The estimation unit performs a dialogue act estimation for each of the divided unit utterances and assigns a dialogue act tag,
The dialogue log analyzer according to any one of claims 1 to 4. A first extraction unit that extracts a keyword from an utterance sentence included in the dialogue log and assigns the keyword to the utterance sentence, or converts the extracted keyword into a concept class and assigns it to the utterance sentence;
The first rule includes a rule for detecting a boundary position of the first dialogue block on condition of a combination of a speaker tag, a dialogue action tag, and a keyword or a concept class,
The second rule includes a rule for detecting a boundary position of the second dialogue block on condition of a combination of a speaker tag, a dialogue action tag, and a keyword or a concept class.
The dialog log analyzer according to any one of claims 1 to 5. The first rule specifies a label to be given to the first interaction block, a label that is a superordinate concept of the label, and is one of the labels specified by the second rule and a common superordinate label,
The first determination unit further determines an upper label of the first interaction block in the interaction log based on the first rule,
The second determination unit is determined by the first determination unit based on a rule for designating a label common to the upper label of the first dialogue block determined by the first determination unit among the second rules. Determining a range of the second interaction block that includes a range of the first interaction block;
The dialog log analyzer according to any one of claims 1 to 6. Information corresponding to the label of the first dialog block determined by the first determination unit is extracted from a plurality of utterances constituting the first dialog block, and the second dialog determined by the second determination unit A second extraction unit that extracts information corresponding to the label of the block from a plurality of utterances constituting the second dialogue block;
The first display information is display information indicating a range of the first dialog block, a label of the first dialog block, and information extracted by the second extraction unit,
The second display information is display information indicating a range of the second dialog block, a label of the second dialog block, and information extracted by the second extraction unit.
The dialog log analyzer according to any one of claims 1 to 7. A dialogue log analysis method for analyzing a dialogue log including a plurality of utterances constituting a dialogue and a speaker tag indicating a speaker of each utterance,
Providing a dialogue act tag by estimating a dialogue act on an utterance sentence included in the dialogue log; and
A rule for detecting a boundary position of a first dialogue block composed of a plurality of continuous utterance sentences on the condition that at least a combination of a speaker tag and a dialogue action tag is provided, and a label to be given to the first dialogue block Determining a range of the first interaction block in the interaction log and a label to be assigned to the first interaction block based on a first rule to be specified;
A rule for detecting a boundary position of a second dialogue block that is larger than the first dialogue block on the condition that at least a combination of a speaker tag and a dialogue action tag is specified, and specifies a label to be given to the second dialogue block Determining a range of the second dialogue block in the dialogue log and a label to be given to the second dialogue block based on a second rule;
First display information indicating the range of the first dialog block and the label of the first dialog block, and second display information indicating the range of the second dialog block and the label of the second dialog block for the dialog log And a step of controlling to add and display,
Interactive log analysis method including On the computer,
A function of estimating a dialogue action and giving a dialogue action tag to a utterance sentence included in a dialogue log including a plurality of utterance sentences constituting a dialogue and a speaker tag indicating a speaker of each utterance sentence;
A rule for detecting a boundary position of a first dialogue block composed of a plurality of continuous utterance sentences on the condition that at least a combination of a speaker tag and a dialogue action tag is provided, and a label to be given to the first dialogue block A function of determining a range of the first dialog block in the dialog log and a label to be given to the first dialog block based on a first rule to be specified;
A rule for detecting a boundary position of a second dialogue block that is larger than the first dialogue block on the condition that at least a combination of a speaker tag and a dialogue action tag is specified, and specifies a label to be given to the second dialogue block A function of determining a range of the second dialogue block in the dialogue log and a label to be given to the second dialogue block based on a second rule;
First display information indicating the range of the first dialog block and the label of the first dialog block, and second display information indicating the range of the second dialog block and the label of the second dialog block for the dialog log And a function to control the display to be added and
A program to realize