JP2013190991A - Voice interactive summarization device, voice interactive summarization method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voice interactive summarization device in which in a support method of increasing efficiency of response monitoring (hearing a voice and evaluation thereof) in a call center, even to a set of responses which are out of an interactive flow of a talk script, efficiency of a monitoring work can be increased.SOLUTION: A voice interactive summarization device extracts a plurality of speeches out of an interactive text set, and extracts, as a speech cluster, speeches which are connected with each other in an implication relation out of the extracted plurality of speeches.

Description

  The present invention relates to a voice dialogue summarizing apparatus, a voice dialogue summarizing method, and a program.

  The call center has a role as a “corporate face” that connects customers to the company and expands their profits. As response quality has a major impact on customer loyalty, managers not only need to improve call center operational efficiency (response rate, availability, processing time), but also the need to improve and homogenize response quality. It has been.

  At the call center operation site, the basic response skills of the communicator's response, whether or not the dialogue can be performed according to the script, whether or not the customer speaks inappropriately or feels uncomfortable. It is necessary to confirm and evaluate the above problems and the quality of reception by listening to the actual call recording of each communicator.

  However, since the method of confirming and evaluating the content by playing back the call recording requires time, the call center has a demand for a support method that improves the efficiency of response monitoring (evaluation by listening to voice).

  Non-Patent Document 1 describes an example of a summary system. “TRUE TELLER VOICE digest” described in Non-Patent Document 1 provides a response quality evaluation report generation function. FIG. 11 is a block diagram showing the configuration of the summarization system described in Non-Patent Document 1. Referring to FIG. 11, the summarizing system includes voice recognition means 101, topic transition graph generation means 102, and script matching degree calculation means 103. The speech recognition means 101 performs speech recognition for telephone reception at a call center, and extracts a dialogue text set as a recognition result. The topic transition graph generation means 102 matches the conversation text set of telephone reception against a talk script (that is, a collection of expressions collected for each topic) operated in the call center, and generates a topic transition graph. FIG. 12 is a diagram illustrating a topic transition graph as an example. The script matching degree calculation means 103 performs an evaluation of the reception quality based on the matching rate with the expression in the talk script, and generates an evaluation report.

Hori Nobuo and Takehara Kazuaki, “Utilization of“ Voice of Voice ”Realized by Dialogue Summarization—Realization of Automatic Summarization of Telephone Response and Monitoring of All Items” ”, NRI IT Solution Frontier, September 2010, pp. 10-13.

  The following analysis was made by the present inventors.

  According to the summarization system described in Non-Patent Document 1, it is not possible to sort and arrange response cases that do not match the conversation flow of the talk script. Therefore, according to such a summarization system, there is a problem in that it is not possible to obtain an effect of improving the efficiency of the monitoring work for a response case that does not match the conversation flow of the talk script.

  Therefore, in the support method to improve the response monitoring (evaluation by listening to voice) in the call center, it becomes an issue to make the monitoring work more efficient even for the response set that is out of the conversation flow of the talk script. . An object of the present invention is to provide a voice dialogue summarizing apparatus, a voice dialogue summarizing method and a program for solving such a problem.

  The spoken dialogue summarizing apparatus according to the first aspect of the present invention extracts a plurality of utterances from a set of dialogue texts, and extracts a plurality of extracted utterances connected in an implication relationship as an utterance cluster.

  In the speech dialogue summarizing method according to the second aspect of the present invention, a computer extracts a plurality of utterances from a set of dialogue texts, and utterances of the extracted plurality of utterances connected in implication relation with each other. Extracting as a cluster.

  The program according to the third aspect of the present invention includes a process of extracting a plurality of utterances from a set of dialogue texts, a process of extracting a plurality of extracted utterances connected to each other by an implication relationship as an utterance cluster, , Execute on the computer. The program can be provided as a program product recorded on a non-transient computer-readable storage medium.

  According to the voice dialog summarizing apparatus, the voice dialog summarizing method and the program according to the present invention, in the support method for improving the response monitoring (evaluation by listening to voice) in the call center, the response set deviating from the dialog flow of the talk script. Can also improve the efficiency of monitoring work.

It is a block diagram which shows the structure of the speech dialogue summary system which concerns on embodiment as an example. It is a flowchart which shows operation | movement of the speech dialogue summary system which concerns on embodiment as an example. It is a figure which shows the dialogue text set in an Example as an example. It is a figure which shows the dependency relationship between the morphemes in an Example. It is a figure which shows the relationship between the partial expression of the utterance in an Example, and whether they are implied by the utterance. It is a figure which shows the utterance partial expression of the 2nd sentence in an Example as an example. It is a figure which shows the implication relationship between each utterance partial expression of the 1st sentence in an Example, and a 2nd sentence. It is a figure which shows the directed graph which made the utterance partial expression in an Example a vertex, and connected with the directed edge which goes to the side which implied between it as an example. It is a figure which shows an example which extracts the partial graph which does not contain the vertex used as the starting point in the directed graph in an Example. It is a figure which shows the speech cluster in an Example as an example. It is a block diagram which shows the structure of the voice dialog summary (generation | generation of a topic transition graph) system described in the nonpatent literature 1. It is a figure which shows the output (topic transition graph) of the speech dialogue summary system described in the nonpatent literature 1 as an example.

  First, an outline of one embodiment will be described. Note that the reference numerals of the drawings attached to this summary are merely examples for facilitating understanding, and are not intended to limit the present invention to the illustrated embodiment.

  FIG. 1 is a block diagram illustrating, as an example, a configuration of a voice dialogue summarizing apparatus according to the present disclosure. Referring to FIG. 1, the speech dialogue summarizing apparatus includes an utterance partial expression extraction unit (11), an implication determination unit (12), a directed graph generation unit (13), and an utterance cluster generation unit (14). The utterance partial expression extraction unit (11) extracts a partial expression implied by the utterance as an utterance partial expression from each utterance in the conversation text set. When an implication relationship is established between utterance partial representations extracted from different utterances, the directed graph generation unit (13) uses the utterance partial representations as vertices and connects the directed graphs connected by directed edges toward the implication side. Generate. The utterance cluster generation unit (14) extracts a subgraph that can follow a directed edge in a forward direction from a certain vertex in the directed graph, and an utterance set that is an extraction source of the utterance partial expression that is each vertex in the subgraph, Output as utterance cluster.

  According to such a voice dialogue summarization apparatus, in a support method that improves the efficiency of response monitoring (evaluation by listening to voice) in a call center, it is possible to improve the efficiency of monitoring work even if the response case does not match the conversation flow of the talk script. It becomes possible. The reason is that even if the response case does not match the conversation flow of the talk script, by generating and presenting an utterance cluster, it is possible to view a set of utterances related to the same utterance content in a batch and confirm the voice. .

In the present invention, the following modes are possible.
[Form 1]
This is the same as the speech dialogue summarizing apparatus according to the first viewpoint.
[Form 2]
The speech dialog summarizing apparatus may include an utterance partial expression extraction unit that extracts a partial expression implied by the utterance as an utterance partial expression from each utterance extracted from the conversation text set.
[Form 3]
The speech dialogue summarization apparatus includes a first utterance partial expression extracted from a first utterance extracted from the dialogue text set, and a second utterance extracted from the second utterance extracted from the dialogue text set. When an implication relationship is established with the utterance partial representation, the first utterance partial representation is the first vertex, the second utterance partial representation is the second vertex, the first vertex and the second There may be provided a directed graph generation unit that generates a graph in which the vertices are connected by edges.
[Form 4]
The speech dialogue summarization apparatus, when an implication relationship is established between the first utterance partial representation and the second utterance partial representation, the speech set including the first utterance and the second utterance, You may provide the utterance cluster production | generation part extracted as an utterance cluster.
[Form 5]
When the first utterance partial expression implies the second utterance partial expression, the directed graph generation unit generates a directed graph connected by directed edges from the first vertex to the second vertex. May be generated as
[Form 6]
The utterance cluster generation unit extracts a subgraph obtained by tracing the directed graph in a forward direction, and extracts a set of utterances from which an utterance partial expression corresponding to a vertex included in the subgraph is extracted as the utterance cluster. May be.
[Form 7]
The utterance cluster generation unit excludes vertices shared with other subgraphs among the vertices included in the subgraph, and a set of utterances from which utterance partial expressions corresponding to the vertices included in the subgraph are extracted May be extracted as the utterance cluster.
[Form 8]
It is as the voice dialogue summarizing method according to the second viewpoint.
[Form 9]
In the spoken dialogue summarizing method, the computer may include a step of extracting, from each utterance extracted from the dialogue text set, a partial expression implied by the utterance as an utterance partial expression.
[Mode 10]
In the spoken dialogue summarizing method, a computer is extracted from a first utterance partial expression extracted from a first utterance extracted from the dialogue text set and a second utterance extracted from the dialogue text set. When an implication relationship holds with the second utterance partial representation, the first utterance partial representation is the first vertex, the second utterance partial representation is the second vertex, and the first vertex and A step of generating a graph connecting the second vertexes with edges may be included.
[Form 11]
In the spoken dialogue summarizing method, if the computer has an implication relationship between the first utterance partial representation and the second utterance partial representation, an utterance set including the first utterance and the second utterance May be extracted as the utterance cluster.
[Form 12]
In the spoken dialogue summarizing method, when the first utterance partial expression implies the second utterance partial expression, a directed graph connected by a computer at a directed edge from the first vertex to the second vertex May be generated as the graph.
[Form 13]
In the spoken dialogue summarizing method, the computer includes a step of extracting a subgraph obtained by tracing the directed graph in a forward direction,
A set of utterances from which utterance partial expressions corresponding to vertices included in the subgraph are extracted may be extracted as the utterance cluster.
[Form 14]
The program is related to the third viewpoint.
[Form 15]
The program may cause the computer to execute a process of extracting a partial expression implied by the utterance as an utterance partial expression from each utterance extracted from the conversation text set.
[Form 16]
The program includes: a first utterance partial expression extracted from a first utterance extracted from the conversation text set; and a second utterance partial expression extracted from a second utterance extracted from the conversation text set. The first utterance partial representation is a first vertex, the second utterance partial representation is a second vertex, and the first vertex and the second vertex are You may make it make a computer perform the process which produces | generates the graph connected with the edge.
[Form 17]
When the implication relationship is established between the first utterance partial expression and the second utterance partial expression, the program sets the utterance set including the first utterance and the second utterance as the utterance cluster. You may make it make a computer perform the process to extract.
[Form 18]
When the first utterance partial representation implies the second utterance partial representation, the program generates, as the graph, a directed graph connected by directed edges from the first vertex to the second vertex. You may make it make a computer perform the process to perform.
[Form 19]
The program extracts a partial graph obtained by tracing the directed graph in a forward direction;
You may make it make a computer perform the process which extracts the collection of utterances of the extraction part of the utterance partial expression corresponding to the vertex contained in the said partial graph as said utterance cluster.

(Embodiment)
Next, the speech dialogue summarizing apparatus according to the embodiment will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing, as an example, the configuration of the voice dialogue summarizing apparatus according to the present embodiment. Referring to FIG. 1, the speech dialogue summarizing apparatus includes a computer (central processing unit; processor; data processing unit) 10 that operates by program control, and a storage unit 20.

  The storage unit 20 includes a dialogue text set storage unit 21 and an utterance cluster storage unit 22. The dialog text set storage unit 21 stores a set of dialog texts. On the other hand, the utterance cluster storage unit 22 stores utterance clusters.

A computer (central processing unit; processor; data processing unit) 10 includes an utterance partial expression extraction unit 11, an implication determination unit 12, a directed graph generation unit 13, and an utterance cluster generation unit 14.
The utterance partial expression extraction unit 11 extracts, from each utterance in the dialog text set stored in the dialog text set storage unit 21, a partial expression implied by the utterance as an utterance partial expression.

  The implication determining unit 12 determines whether an implication relationship exists between the two utterance partial expressions. Further, when there is an implication relationship between the two utterance partial expressions, the implication determination unit 12 determines which of the two utterance partial expressions is on the side of implication.

  When the implication relationship is established between the utterance partial expressions extracted from different utterances, the directed graph generation unit 13 generates a directional graph in which the utterance partial expressions are used as vertices and connected between the directional edges toward the implication side. .

  The utterance cluster generation unit 14 extracts a subgraph obtained by tracing a directional edge in a forward direction from a certain vertex in the directed graph, and an utterance set from which the utterance partial expression that is each vertex in the subgraph is extracted Are output as utterance clusters.

  FIG. 2 is a flowchart showing an example of the operation of the voice dialogue summarizing apparatus of this embodiment. With reference to the block diagram of FIG. 1 and the flowchart of FIG. 2, the operation of the speech dialogue summarizing apparatus of this embodiment will be described in detail.

  First, the utterance partial expression extraction unit 11 extracts a plurality of partial expressions as utterance partial expression candidates from each utterance in the conversation text set stored in the dialog text set storage unit 21, and the utterance implies the contents of each partial expression. The implication determining unit 12 determines whether or not to do so. The utterance partial expression extraction unit 11 extracts the partial expression determined to imply the utterance as the utterance partial expression for the utterance (step A1).

  Next, the directed graph generation unit 13 determines whether or not an implication relationship is established between utterance partial expressions extracted from different utterances. When it is determined that the utterance is implied, the directed graph generation unit 13 generates a directional graph in which the utterance partial expressions are used as vertices and connected between the directional edges toward the side that implies the utterance (step A2). .

  Next, the utterance cluster generation unit 14 extracts a subgraph obtained by tracing a directional edge in a forward direction from a certain vertex in the directed graph, and extracts the utterance partial expression that is each vertex in the subgraph. A certain utterance set is stored in the utterance cluster storage unit 22 as an utterance cluster (step A3).

  The voice dialog summarization apparatus of this embodiment can browse and collect a set of utterances related to the same utterance content at once by generating and presenting an utterance cluster even if the response case does not match the conversation flow of the talk script. It is configured to be possible. According to such a voice dialogue summarizing apparatus, it is possible to improve the efficiency of the monitoring work even in the case of a response that does not match the dialogue flow of the talk script.

  Next, with reference to the block diagram of FIG. 1 and the flowchart of FIG. 2, the operation of the speech dialogue summarizing apparatus of this embodiment will be described based on a specific example.

  First, the utterance partial expression extraction unit 11 extracts a plurality of partial expressions as utterance partial expression candidates from each utterance in the conversation text set as shown in FIG. 3 stored in the dialog text set storage unit 21.

  For example, when the utterance is “Is it always the same place that is clogged?”, The utterance partial expression extraction unit 11 performs morphological analysis and dependency (syntax) structure analysis, as shown in FIG. A dependency relationship between morphemes is generated, and a plurality of partial expressions as shown in the right column of FIG. 5 are extracted from a subset of the dependency relationship.

  Next, the utterance partial expression extraction unit 11 determines whether or not the utterance “Is it always the same place that is clogged” with the content of each of these partial expressions is determined by the implication determination unit 12, and FIG. Obtain the implications of utterances as shown in the left column.

  Therefore, the utterance partial expression extraction unit 11 determines that the partial expressions determined to imply the utterance “Is it the same place that is clogged?” And “Is it always the same place that is clogged?” Is extracted as an utterance partial expression for (step A1 in FIG. 2).

  Next, the directed graph generation unit 13 determines whether or not an implication relationship is established between utterance partial expressions extracted from different utterances.

  Here, it is assumed that different utterances are the first sentence “Is it always the same place that is jammed?” And the second sentence “Is the location of the paper jam every time together?”. Although two sentences are used for convenience of explanation, the same applies to three or more sentences. Similarly, the utterance partial expression extraction unit 11 obtains an utterance partial expression as shown in FIG. 6 for the second sentence.

  The directed graph generation unit 13 obtains an implication relationship between the utterance partial expressions of the first sentence and the second sentence as shown in FIG. Further, when it is determined that an implication relationship is established between the utterance partial representations, the directed graph generation unit 13 generates a directed graph in which the utterance partial representations are vertices and connected between the directional edges toward the implication side. (Step A2). FIG. 8 shows a directed graph generated by the directed graph generation unit 13 as an example.

  Next, the utterance cluster generation unit 14 extracts a subgraph that can follow a directed edge in a forward direction from a certain vertex in the directed graph. When the directed graph has the structure shown in FIG. 8, the utterance cluster generation unit 14 can trace the directed edge in the forward direction from the vertex corresponding to the utterance partial expression “Is the place together?” To extract. Here, at the time of subgraph extraction, since the vertex that is the starting point when following a directed edge in the forward direction may be shared with another subgraph, the utterance cluster generation unit 14 As shown, a subgraph that does not include a starting point may be extracted.

  The utterance set that is the source of the utterance subexpression that is each vertex in the extracted subgraph is the first sentence “Is it always the same place that is jammed?” And the second sentence “The location of the paper jam every time is "Is it together?" Therefore, the utterance cluster generation unit 14 stores these utterances in the utterance cluster storage unit 22 as utterance clusters as shown in FIG. 10 (step A3).

  The speech dialogue summarization apparatus according to the present embodiment generates and presents an utterance cluster even when the response case does not match the dialogue flow of the talk script, thereby collectively browsing a set of utterances related to the same utterance content and It is configured so that it can be confirmed. Therefore, according to the speech dialogue summarizing apparatus of the present embodiment, it is possible to improve the efficiency of the monitoring work even in the case of a response that does not match the dialogue flow of the talk script.

  The voice dialogue summarizing apparatus according to the present embodiment can be applied to uses such as response quality evaluation from call voice in a contact center, text mining, search, and FAQ (Frequently Asked Questions) creation support.

  It should be noted that the disclosure of the above non-patent document is incorporated herein by reference. Within the scope of the entire disclosure (including claims) of the present invention, the embodiment can be changed and adjusted based on the basic technical concept. Further, various combinations or selections of various disclosed elements (including each element of each claim, each element of each embodiment, each element of each drawing, etc.) are possible within the scope of the claims of the present invention. It is. That is, the present invention of course includes various variations and modifications that could be made by those skilled in the art according to the entire disclosure including the claims and the technical idea.

DESCRIPTION OF SYMBOLS 10 Computer 11 Utterance partial expression extraction part 12 Implication determination part 13 Directed graph generation part 14 Utterance cluster generation part 20 Storage part 21 Dialogue text set storage part 22 Utterance cluster storage part 101 Speech recognition means 102 Topic transition graph generation means 103 Script coincidence calculation means

Claims (10)

  A speech dialogue summarization apparatus that extracts a plurality of utterances from a set of dialogue texts and extracts a plurality of extracted utterances connected in an implication relationship as an utterance cluster.   The speech dialogue summarization apparatus according to claim 1, further comprising: an utterance partial expression extraction unit that extracts a partial expression implied by the utterance as an utterance partial expression from each utterance extracted from the dialogue text set.   Between the first utterance partial expression extracted from the first utterance extracted from the conversation text set and the second utterance partial expression extracted from the second utterance extracted from the conversation text set. When an implication relationship holds, the first utterance partial representation is a first vertex, the second utterance partial representation is a second vertex, and the first vertex and the second vertex are connected by an edge. The spoken dialogue summarization apparatus according to claim 2, further comprising a directed graph generation unit configured to generate a graph.   An utterance cluster that extracts an utterance set including the first utterance and the second utterance as the utterance cluster when an implication relationship is established between the first utterance partial expression and the second utterance partial expression The speech dialogue summarization apparatus according to claim 3, further comprising a generation unit.   When the first utterance partial expression implies the second utterance partial expression, the directed graph generation unit generates a directed graph connected by directed edges from the first vertex to the second vertex. The speech dialogue summarization apparatus according to claim 4, wherein   The utterance cluster generation unit extracts a subgraph obtained by tracing the directed graph in a forward direction, and extracts a set of utterances from which an utterance partial expression corresponding to a vertex included in the subgraph is extracted as the utterance cluster. The voice dialogue summarizing apparatus according to claim 5.   The utterance cluster generation unit excludes vertices shared with other subgraphs among the vertices included in the subgraph, and a set of utterances from which utterance partial expressions corresponding to the vertices included in the subgraph are extracted Is extracted as the utterance cluster. The spoken dialogue summarization apparatus according to claim 6. A computer extracting a plurality of utterances from a set of dialogue text;
Extracting a plurality of extracted utterances connected in an implication relationship as an utterance cluster.   The speech dialogue summarizing method according to claim 8, further comprising: extracting, from each utterance extracted from the dialogue text set, a partial expression implied by the utterance as an utterance partial expression. Processing to extract multiple utterances from a set of dialogue texts;
A program that causes a computer to execute processing for extracting, as an utterance cluster, a plurality of extracted utterances connected in an implication relationship with each other.

Images