WO2012098838A1 - Report document creation assistance system, report document creation assistance method, and report document creation assistance program - Google Patents

Abstract

A report document creation assistance system comprises: a conversation report pairing storage means for pre-storing a set of pairs of conversations and report documents which are texts which report the content of the conversations; and a relevance calculation means for calculating relevance for each utterance or each expression unit which are extracted from an inputted conversation, which represents the extent to which the utterance or the expression unit includes content which should be reported in the report document, based on the degree to which the content of each conversational utterance or expression unit in the set of the pairings of the conversations and report documents which are stored in the conversation report pairing storage means is included in the corresponding report documents.

Description

Report document creation support system, report document creation support method, and report document creation support program

The present invention relates to a report document creation support system, a report document creation support method, and a report document creation support program that generate a summary of dialog contents from an input dialog.

The call center responds to a huge number of inquiries from customers every day. The contents of the response are recorded and managed, and are used for customer response follow-up, response, product and service problem analysis and improvement. Response contents are often recorded and accumulated by voice data recording a call, report memo created by an operator, and the like.

The advantage of recording the response content by recording the call is that the dialogue between the operator and the customer is recorded without omission. However, when recording is performed by recording a voice, there is a problem that it takes time to reproduce the voice, which is a method for confirming the contents. For this reason, in many call centers, recording is performed using report memos. The recording of the report memo has the advantage that it is easy to check the contents and reuse easily because the main points of the contents of the response are summarized in a short text. However, there is a problem that a report memo preparation cost and an education cost are generated by the operator. In addition, there are problems such as leakage of contents to be reported and variations in quality depending on the reporter.

Therefore, there is a demand for automatically generating or supporting the generation of text that summarizes the main points of the contents of the response from the call voice.

For example, Non-Patent Documents 1 and 2 describe an example of a summarization system that generates a short text that summarizes the main points of content from a call content text generated as a result of speech recognition for the call speech.

In the summarization system described in Non-Patent Document 1, when the input means inputs the call content text, the sentence importance degree calculation means is based on the word weight calculated from the document set stored in the document data storage means. Calculate the importance of each sentence of the input text. Then, the sentence extraction unit generates summary text by extracting and arranging sentences with high importance, and outputs the summary text.

Further, the summarization system described in Non-Patent Document 2 is a standard response record model for each type of industry, which is stored in the storage means in advance by the summarization engine when the input means inputs the call content text. A response record sentence is automatically generated based on the “summary template”. The summarization engine automatically generates a summary sentence while ensuring unity of terms and style, completeness (there is no omission of necessary items), and conciseness (no unnecessary items are described) according to the template.

Iwano, Hirohata, Shinnaka, Furui, “Study on automatic speech summarization technique and its objective evaluation method by extracting important sentences”, IEICE Technical Report, Volume 105, 132 (SP2005 20-26), 2005 , P. 1-6 Hori, Takehara, “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, p. 10-13

However, in the summarization system described in Non-Patent Document 1, general contents in document data are judged to be less important and tend not to be included in summaries. For this reason, there is a problem that if the text that summarizes the main points of the response contents is automatically generated from the call voice, the main points are likely to be leaked particularly in the frequently used response contents.

In addition, the summarization system described in Non-Patent Document 2 needs to describe a “summary template” for ensuring the uniformity, completeness, and conciseness of terms and styles in advance. It is impossible to create a template that can ensure uniformity, completeness, and conciseness for all of the contents of customer service and various expressions of customers. In addition, for example, it is determined whether each sentence in the dialogue is a standard expression such as “Thank you for calling” or a greeting expression such as “Good morning”, and an unnecessary expression is obtained by analyzing such meaning. It is often the case that it is too verbose as a summary to simply remove typical expressions. Therefore, in the summarization system described in Non-Patent Document 2, when a variety of expressions are given to the customer, which is difficult to imagine, or when various expressions are given to the customer, the leakage of the main points is particularly noticed in the case of such a non-standard or new response. There is a problem that it is easy to occur, or there is a problem that a summary including many expressions that are not the main points is created as a result of trying to pick them up widely.

Accordingly, the present invention provides a report document creation support system, a report document creation support method, and a report document that can generate a high-quality summary covering the contents to be reported, which is suitable for a request for response management at a call center site or the like. The purpose is to provide a creation support program.

A report document creation support system according to the present invention is stored in advance in a dialog report pair storage unit that stores a set of pairs of a dialog and a report document that is a sentence that reports the contents thereof, and the dialog report pair storage unit. For each utterance or each expression unit extracted from the input dialogue based on the degree to which the content of each utterance or expression unit in the dialogue is included in the corresponding report document in the set of dialogue and report document pairs, Report level calculating means for calculating a report level indicating the degree of whether or not the utterance or the expression unit includes the contents to be reported in the report document is provided.

Further, the report document creation support method according to the present invention stores in advance a set of pairs of a dialogue and a report document that is a sentence reporting the contents thereof in a storage means, and the dialogue stored in the storage means For each utterance or each expression unit extracted from the input dialogue based on the degree to which the content of each utterance or expression unit in the dialogue is included in the corresponding report document in the set of report document pairs Alternatively, the degree of reporting indicating the degree of whether or not the expression unit includes the content to be reported in the report document is calculated, and the calculated degree of reporting of each utterance or expression unit in the input dialogue is calculated together with the utterance or expression unit. A summary is generated by extracting an utterance or an expression unit from the input dialog based on the output or the reported degree of each utterance or expression unit in the input dialog.

In addition, the report document creation support program according to the present invention includes a computer that stores the contents of each utterance or expression unit in a dialogue in a set of pairs of dialogue documents stored in the storage means and a report document that reports the contents. Based on the degree of inclusion in the corresponding report document, for each utterance or each expression unit extracted from the input dialogue, whether the utterance or the expression unit contains the content to be reported in the report document The process of calculating the degree of reporting and the degree of reporting of each utterance or expression unit during the input dialogue is output together with the utterance or representation unit, or each utterance or expression during the calculated input dialogue It is characterized in that a process of generating a summary by extracting an utterance or an expression unit from an input dialogue based on the reporting degree of the unit is executed.

According to the present invention, it is possible to generate a high-quality summary covering the contents to be reported, which is suitable for the request for response management.

It is a block diagram which shows the structural example of 1st Embodiment. It is a flowchart which shows an example of operation | movement of 1st Embodiment. It is a block diagram which shows the structural example of 2nd Embodiment. It is a flowchart which shows an example of operation | movement of 2nd Embodiment. It is a block diagram which shows the structural example of 3rd Embodiment. It is a flowchart which shows an example of operation | movement of 3rd Embodiment. It is explanatory drawing which shows an example of a dialog text. It is explanatory drawing which shows an example of the report degree calculated with respect to each expression unit (morpheme) in a dialog text. It is explanatory drawing which shows an example of the report degree calculated with respect to each expression unit (utterance) in a dialog text. It is explanatory drawing which shows an example of the summary text produced | generated. It is explanatory drawing which shows an example of the report document with respect to the dialogue text shown in FIG. It is a block diagram which shows the outline | summary of this invention.

First, the features of the present invention will be described. In the present invention, the degree of reporting, which is the degree to which each input part in the dialogue (for example, by utterance or by a predetermined unit of expression) is a part related to the content that has been reported in the past reception, is displayed in the dialogue report. The calculation is based on the appearance tendency of the dialog text expression and the report text expression in the set of dialog and report document pairs stored in the pair storage means. Further, the present invention is characterized in that a summary text is generated by extracting a sentence including an expression part having a high reporting degree based on the degree of reporting of each expression part in the input dialogue thus calculated.

By adopting such a configuration and calculating the degree of reporting for each utterance or each expression unit of the input dialogue, which enables the high-priority content reported in the past response to be preferentially extracted into the summary, It is possible to generate a high-quality summary covering the contents to be reported, which is suitable for the request for response management at a call center site or the like.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

Embodiment 1. FIG.
FIG. 1 is a block diagram showing a configuration example of the first embodiment of the present invention. The system shown in FIG. 1 may be implemented, for example, as a reporting degree calculation system for supporting summary generation or report document creation. The system shown in FIG. 1 includes a computer (central processing unit; processor; data processing unit) 100 that operates under program control, a storage medium 200, an input unit 300, and an output unit 400. The computer 100 also includes a reporting degree calculation unit 101a. The storage unit 200 includes a dialogue report pair storage unit 201a.

The dialogue report pair storage unit 201a stores a set of pairs of dialogue and a report document that is a sentence that reports the content.

The input means 300 inputs target conversation data. The dialogue data may be in the form of audio data or text data. When the dialogue data is input in the form of voice data, the dialogue is provided by voice recognition means (not shown) for performing voice recognition on the input voice data and outputting text data as a voice recognition result. It is also possible to treat each utterance or each expression unit extracted from as character string data.

The report degree calculation means 101a is input based on the appearance tendency of the dialog text expression and the report text expression in the set of the past response dialog text and report text stored in the dialog report pair storage means 201c. Calculate the degree of reporting of each expression part in the dialogue text. Here, the degree of reporting is the degree of how much the targeted expression part is related to the content that has been reported in the past reception. In other words, it is an index representing the degree of whether or not the expression part includes the content to be reported in the report document.

The output unit 400 outputs the reporting level of each utterance or each expression unit during the input dialogue calculated by the reporting level calculation unit 101a. The output unit 400 may output, for example, the degree of reporting of each utterance or each expression unit together with the speech data or text data (character string data) of each utterance or each expression unit included in the input dialogue. The output data format of each utterance or each expression unit during the input dialogue may be voice data or text data.

As an example, when dialogue data is input as text data, the output unit 400 may output each utterance or each expression unit extracted from the dialogue as character string data. For example, when dialog data is input as voice data, the output unit 400 may output each utterance or each expression unit extracted from the dialog as voice data as it is, or input voice data It is also possible to output as character string data based on text data generated by voice recognition.

Next, the operation of the reporting degree calculation system of this embodiment will be described. FIG. 2 is a flowchart showing an example of the operation of the reporting degree calculation system of the present embodiment. In the example shown in FIG. 2, first, the input unit 300 inputs a dialogue for which a reporting degree is calculated for each utterance or each expression unit (step A1). In this example, it is assumed that the dialogue is input in a text format. The summary generated by the system is also generated in text format and output. In the case of input as voice data, after step C1, the voice recognition means may convert the input dialogue voice into text data and output it as dialogue text to the reporting degree calculation means 101a.

Next, the report degree calculation unit 101a reports a report document corresponding to the content of each utterance or each expression unit in the dialogue in the set of past response dialogue and report document pairs stored in the dialogue report pair storage unit 201a. Based on the degree (for example, appearance frequency) included in the input dialogue, the reporting degree of each utterance or each expression unit during the input dialogue is calculated (step A2). It is assumed that the reporting degree calculation means 101a has predetermined criteria for expression parts to be calculated for the reporting degree and how to extract them from the input dialogue. The reporting degree calculation unit 101a extracts each utterance or each expression unit included in the input dialogue according to such setting, and calculates a reporting degree for each extracted utterance or each expression unit.

Finally, the output means 400 outputs each utterance or expression unit in the input dialogue together with the report level (step A3).

As described above, in the present embodiment, the reporting degree calculation system is the degree of whether or not each utterance or each expression unit extracted from the inputted dialogue includes the contents to be reported in the report document. The degree of reporting is calculated, and each utterance or expression unit is output together with the degree of reporting. Therefore, the user can grasp the content portion to be reported in the input call.

Embodiment 2. FIG.
Next, a second embodiment of the present invention will be described with reference to the drawings. FIG. 3 is a block diagram showing a configuration example of the second embodiment of the present invention. The system shown in FIG. 3 is implemented as an interactive summary system, for example. The system shown in FIG. 3 includes a computer 100 that operates under program control, a storage medium 200, input means 300, and output means 400. The computer 100 also includes a reporting degree calculation unit 101b and a summary generation unit 102b. The storage unit 200 includes a dialogue report pair storage unit 201b.

The configuration of the present embodiment is different from the configuration of the first embodiment shown in FIG. 1 in that the computer 100 includes summary generation means 102b. Note that the reporting degree calculation unit 101b and the dialogue report pair storage unit 201b of the present embodiment are the same as the reporting degree calculation unit 101a and the dialogue report pair storage unit 201a in the first embodiment. Hereinafter, description of the same components as those in the first embodiment will be omitted.

The summary generation unit 102b selects an expression portion having a high reporting level from the input dialogue data based on the reporting level of each utterance or each expression unit included in the input dialogue calculated by the reporting level calculation unit 101b. Extract the sentence containing it and generate summary text.

In the present embodiment, the output unit 400 outputs the summary text generated by the summary generation unit 102b.

Next, the operation of the reporting degree calculation system of this embodiment will be described. FIG. 4 is a flowchart showing an example of the operation of the dialog summary system of the present embodiment. Note that the operations in steps B1 and B2 shown in FIG. 4 are the same as those in steps A1 and A2 in the first embodiment shown in FIG.

That is, first, the input means 300 inputs a dialogue (step B1). Next, as in the first embodiment, the reporting degree calculating unit 101b performs each utterance or each expression in the dialogue in a set of past response dialogue and report document pairs stored in the dialogue report pair storage unit 201b. Based on the degree to which the content of the unit is included in the corresponding report document, the degree of reporting of each utterance or each expression unit during the input dialogue is calculated (step B2).

When the degree of reporting is calculated for each utterance or each expression unit in the input dialogue, the summary generation unit 102b performs the input during the dialogue based on the reporting degree of each utterance or each expression unit included in the inputted dialogue. Among the utterances or expression units, utterances or expression units having a high reporting level are extracted and connected to generate a summary (step B3). Note that the summary generation method not only connects the sentences containing the extracted expression units, but also, for example, a simple arrangement like a bulleted list, or the original sentence that supplements the subject or adds or corrects the sentence end expression. It is also possible to use a method such as connecting after performing auxiliary editing work that does not change the context of.

Finally, the output unit 400 outputs the summary generated by the summary generation unit 102b (step B4).

As described above, in the present embodiment, the reporting degree calculation system extracts utterances or expression parts having a high reporting degree during conversation and connects them to generate a summary. Therefore, the reporting degree calculation system can generate a high-quality summary that suppresses the points to be reported, which is suitable for the requirements of the site of reception management.

Embodiment 3. FIG.
Next, a third embodiment of the present invention will be described with reference to the drawings. FIG. 5 is a block diagram showing a configuration example of the third embodiment of the present invention. The system shown in FIG. 5 is implemented as a report creation support system, for example. The system shown in FIG. 5 includes a computer 100 that operates under program control, a storage medium 200, input means 300, and output means 400. The computer 100 also includes a report degree calculation unit 101c, a summary generation unit 102c, and a dialogue report pair registration unit 103c. The storage unit 200 includes a dialogue report pair storage unit 201c.

The configuration of this embodiment is different from the configuration of the second embodiment shown in FIG. 3 in that the computer 100 includes a dialog report pair registration unit 103c. Note that the reporting degree calculation unit 101c, summary generation unit 102c, and dialogue report pair storage unit 201c of the present embodiment are the same as the reporting degree calculation unit 101b, summary generation unit 102b, and dialogue report pair storage unit 201b in the second embodiment. It is. Hereinafter, description of the same components as those in the second embodiment will be omitted.

When the report document is input from the input unit 300, the dialog report pair registration unit 103c additionally stores the dialog report pair storage unit 201c in pairs with the dialog corresponding to the input report document.

For example, when a report document having the same identification number as the dialog identification number input in the past is input, the dialog report pair registration unit 103c registers them in the dialog report pair storage unit 201c as a pair. Also good. For example, when the identification number of the dialogue and the identification number of the report document are different, the dialogue report pair registration unit 103c inputs the report document including information indicating which dialogue corresponds to the report document. A correspondence relationship may be identified by providing an interface. Further, for example, when the dialog report pair registration unit 103c outputs the summary generated by the summary generation unit 102b, the dialog report pair registration unit 103c edits the dialog with the dialog identification number assigned by the system (the user edits the summary). Screen to create a report document), and when the user inputs an instruction to create and save the report document via the edit screen, the report document is input together with the identification number of the dialog. A simple user interface may be provided.

Next, the operation of the report creation support system of this embodiment will be described. FIG. 6 is a flowchart showing an example of the operation of the report creation support system of this embodiment. The operations in steps C1 to C3 shown in FIG. 6 are basically the same as those in steps B1 to B3 in the second embodiment shown in FIG.

In this example, first, the input means 300 inputs a dialog text (step C1). Next, the reporting degree calculation unit 101c stores a reporting degree, which is a degree of how much each expression part in the inputted dialogue text is related to the contents reported in the past reception, in the dialogue report pair storage unit 201c. Calculation is performed based on the appearance tendency of the expression of the dialog text and the expression of the report text in the set of the stored pairs of the dialog text and the report text of the past response (step C2).

Next, the summary generation means 102c extracts sentences including an expression part with a high degree of report in the dialogue text, and generates a summary text by arranging them (step C3).

Next, the output unit 400 outputs the generated summary text (step C4).

Next, after the output unit 400 outputs the summary text, the dialogue report pair registration unit 103c sequentially inputs report documents created by the user, and additionally stores them in the dialogue report pair storage unit 201c in pairs with the corresponding dialogue ( Step C5).

The interactive report pair registration unit 103c may provide a user interface for inputting a report document created by the user referring to or editing the summary text based on the output summary text, for example. In addition, when a report document is input via such a user interface, the dialog report pair registration unit 103c determines the dialog and the report document based on the dialog identification number attached to the report document. A pair may be additionally stored in the dialogue report pair storage unit 201c.

When creating a report document from a dialogue input as a series of operations, the dialogue report pair registration unit 103c pairs the dialogue text input in step C1 with the report document input in step C5. And stored in the dialogue report pair storage unit 201c.

Thereafter, step C1 to step C5 are sequentially repeated according to the input of the dialogue and the report document.

As described above, in the present embodiment, the report creation support system presents the user with a summary of the points to be reported that are suitable for the site management requirements. Therefore, the report creation support system can improve the quality such as the reduction of report omission in the creation of the report document of the user and shorten the creation time. Further, the report creation support system is configured to input a report document created by the user with reference to the summary and reflect it in the calculation of the report degree from the next time. Therefore, the report creation support system can generate a high-quality summary in accordance with the latest judgment in selecting the report contents of the user.

Next, embodiments of the present invention will be described using specific examples. This example corresponds to the first embodiment described above.

First, the input means 300 inputs a dialog text for which a summary is to be created (step A1 in FIG. 4). FIG. 7 is an explanatory diagram illustrating an example of the dialog text input to the input unit 300. In the present embodiment, the case where the dialogue is input as text data will be described as an example, but the dialogue may be input as voice data. In that case, the report creation support system performs the subsequent processing on the dialog voice and the dialog text that is the voice recognition result.

Next, the reporting degree calculation means 101a stores in the dialogue report pair storage means 201a the degree of reporting, which is the degree to which each expression part in the inputted dialogue text is related to the content reported in the past response. A calculation is made based on the appearance tendency of the dialogue text expression and the report text expression in the set of the past dialogue text and report text pairs stored (step A2).

Here, the degree of reporting for each expression in the dialog text can be specifically calculated as follows. First, let w be the expression in the dialog text and w 'be the expression in the report text. Also, a virtual report text is introduced and this is set as _Rv . Here, the reporting degree calculation means 101a calculates the reporting degree for the expression w in the input dialog text as a probability P (w | R _v ) that the report text R _v implies the content of the expression w. Specifically, the reporting degree calculation unit 101a regards (w | R _v ) as the appearance probability of the dialogue expression w in the report content, and estimates it as in the following equation (1).

P (w | R _v ) = 1 / (response number including w) × Σ _{{report memo R corresponding to response including w}} P (w | R)
... Formula (1)

Here, P (w | R) is estimated by the following equation (2), and P (w | w ') is estimated by the following equation (3).

P (w | R) = 1−Π _{w′∈R} [1-P (w | w ′)] (2)
P (w | w ′) = P (w, w ′) / P (w ′) (3)

However, P (w, w ′) represents the co-occurrence probability that the expression w appears during the dialogue and the expression w ′ appears in the corresponding report text.

As a result of the above calculation, the reporting degree calculating means 101a obtains the dialogue text expression and the reporting degree. Note that even if the expression w in the dialog text is described using a different expression w ′ in the report text, it is taken into account for the calculation of the reporting degree of the expression w in the dialog text. is required.

That is, according to this calculation method, even if the expression w itself in the dialog text is not reported in the same expression, if it is reported as the content, the report is evaluated as having a high degree of reporting. Have.

Further, when calculating the reporting degree, in the evaluation of P (w | w ′), a new combination in which the combination of the expression w and the expression w ′ does not exist in the past response data (dialog text and report text pair). In the case of an expression pair, it is also possible to give an appropriate probability value distribution that is not zero. Such a probability value distribution can be given by using, for example, a “Mathed C” smoothing technique as described in the following document.

Literature: Witthen, I.H. & Bell, T.C., "The zero-fequency probulem: Estimating the probabilities of novel events in adaptive text compression.", IEEE Transaction on Information Theory 37 (4), 1991, p.85.

Specifically, P (w | w ′) can be estimated as in the following equation (4).

When C (w, w ′)> 0,
P (w, w ′) = C (w, w ′) / (C (w ′) + r (w ′))
When C (w, w ′) = 0
P (w, w ′) = r (w ′) / (C (w ′) + r (w ′))
... Formula (4)

Here, C (w, w ′) represents the frequency in which the expression w appears in the dialog text and the expression w ′ appears in the corresponding report text in the pair set of the dialog text and the report text. Yes. C (w ′) represents the frequency of occurrence of the expression w ′ in the report text. Also, r (w ′) represents the number of different expressions w ′ that have appeared so far in the report text relative to the expression w in the dialog text.

By evaluating P (w, w ′) in this way, even when the combination of the expression w and the expression w ′ is a new expression pair that does not exist in the past reception data, non-zero probability value distribution is performed. As a result, even if it is a new expression pair, the same expression in the summary can be output. This is because, in the past response data, it is possible to obtain a higher degree of reporting with a new expression than an expression that appears in the dialog text but does not appear in the report text at all. is there.

FIG. 11 is an explanatory diagram showing an example of a report document for the dialog text shown in FIG. For example, “Increase” in the utterance “I want to increase the number of email accounts” during the dialogue shown in FIG. 7 is another expression “Increase” in the sentence “I want to increase the number of email accounts” in the report document shown in FIG. It can be appropriately evaluated that it is reported using.

In step A2, for example, when the expression unit in the dialog text is a morpheme, the degree of reporting is obtained for each morpheme (utilized form is converted to the original form) as shown in FIG. FIG. 8 is an explanatory diagram showing an example of the degree of reporting calculated for each expression unit (morpheme) in the dialog text. FIG. 8 shows that, for example, the reporting degree of the morpheme “mail account” is calculated as 0.81, and the reporting degree of the morpheme “increase” is calculated as 0.81.

Finally, the output unit 400 outputs each expression unit in the dialogue together with the report degree (step A3).

The output unit 400 may output each expression unit in the input dialogue together with the report degree in the form of a list as shown in FIG. At that time, the output unit 400 may output only the expression unit whose reporting degree is equal to or greater than a predetermined threshold together with the reporting degree. In addition, for example, when the reporting degree is calculated in units of utterances, the output unit 400 may output each utterance during the input dialogue or an utterance of a predetermined threshold or more together with the reporting degree. Note that the method for calculating the degree of reporting of each utterance during the input dialogue is described in the second embodiment.

As described above, in this embodiment, the report creation support system extracts each expression unit included in the input dialogue, and the reporting degree is such that each expression unit includes the content to be reported in the report document. , And each expression unit is output together with the report level. For this reason, the user can easily grasp the content portion to be reported in the input call.

Next, a second embodiment of the present invention will be described. This example corresponds to the second embodiment described above. The operation from step B1 to B2 is the same as the operation from step A1 to A2 in the first embodiment described in the first example.

In the present embodiment, the summary generation unit 102b further extracts utterances or expression units including an expression part having a high reporting degree from the expression parts in the dialog text, and generates the summary text side by side.

Here, in order to extract utterances with a high degree of reporting, the degree of reporting for the utterance is calculated, and the sum of the degree of reporting for each utterance extracted in the summary text is within a range that does not exceed the predetermined summary rate or summary length. What is necessary is just to select the combination of the utterances extracted so that a value may become the maximum.

The degree of reporting for an utterance is obtained, for example, as a value obtained by normalizing the sum of the degree of reporting for each morpheme during the utterance as shown in the following formula (5).

(Reporting degree for utterance) = (Total sum of reporting degree for each morpheme during utterance) / (Number of characters in utterance) (5)

Then, the degree of reporting for each utterance as shown in FIG. 9 is obtained. FIG. 9 is an explanatory diagram showing an example of the reporting degree calculated for each expression unit (utterance) in the dialog text. In FIG. 9, for example, the degree of reporting of the utterance “Thank you for a long time” is calculated as 0.52, and the degree of reporting of the utterance “I want to increase the number of email accounts” is calculated as 0.46. It has been shown.

Next, the summary generation means 102b allows the utterances to be extracted so that the sum of the reporting levels of the utterances extracted in the summary text is maximized within a range not exceeding the predetermined summary rate or summary length. Select a combination. Then, the summary generation means 102b generates a summary text by arranging the selected combinations of utterances (step B3).

Here, for example, when the summary length is determined to be 120 characters or less, the summary generation means 102b may generate 119 character summary text not exceeding that as shown in FIG. FIG. 10 is an explanatory diagram illustrating an example of the generated summary text. As shown in FIG. 10, the summary generation unit 102b may also output information indicating who each utterance is from (in this example, from a client or an operator). .

Finally, the output unit 400 outputs the generated summary text (step B4).

In the present embodiment, the report creation support system extracts, from the summary, utterances or expression points with a high reporting degree during the actual dialogue based on the degree of reporting calculated for each expression unit included in the input dialogue. For this reason, the report creation support system can generate a high-quality summary with the points to be reported, which is suitable for the requirements of the site of reception management. In addition, the report creation support system outputs the generated summary so that the user can easily create a report document by referring to the summary. It is possible to improve quality such as reduction and shorten creation time.

Next, a third embodiment of the present invention will be described. This example corresponds to the third embodiment described above. The operations from step C1 to C3 are the same as the operations from step A1 to A2 and step B3 described in the first and second embodiments.

In this embodiment, for example, when the user creates a report document based on the summary text output in step B3 in the second embodiment, the user inputs this report document from the input means 300, and the dialogue report pair. The registration unit 103c additionally stores this report document in the dialogue report pair storage unit 201c as a pair together with the corresponding dialogue (step C5).

In this embodiment, it is assumed that the report document shown in FIG. 11 is input as the report document for the dialogue shown in FIG. Then, the dialogue report pair registration unit 103c additionally stores the report document together with the corresponding dialogue (dialog shown in FIG. 7) in the dialogue report pair storage unit 201c. In this example, it is assumed that the corresponding dialogue has been input at a stage (in this example, step C1) before the report document is input. In step C1, the report creation support system stores the input dialogue together with the identification number in a storage unit that temporarily holds the dialogue. The dialogue report pair storage unit 201c may also serve as a storage unit for temporarily holding the storage unit.

In this embodiment, the report creation support system presents the user with a summary of the points to be reported that are suitable for the site management requirements, and immediately reports the report document created by the user who referred to it. It can be used to calculate the degree of subsequent reporting. As a result, the report creation support system also reflects the latest conversion methods such as the latest judgment of the selection of report contents in the report creation by the user and how the expression during dialogue is paraphrased in the report document. And a higher quality summary can be generated.

Next, the outline of the present invention will be described. FIG. 12 is a block diagram showing an outline of the present invention. As shown in FIG. 12, the report document creation support system according to the present invention includes a report degree calculation unit 11 and a dialogue report pair storage unit 12.

The dialogue report pair storage unit 12 (for example, dialogue report pair storage unit 201a, 201b, 201c) stores a set of pairs of a dialogue and a report document that is a sentence reporting the contents in advance.

The reporting degree calculating means 11 (for example, the reporting degree calculating means 101a, 101b, 101c) is the contents of each utterance or expression unit in dialogue in the set of dialogue and report document pairs stored in the dialogue report pair storage means 12. For each utterance or each expression unit extracted from the input dialogue based on the degree to which is included in the corresponding report document, whether the utterance or the expression unit contains the content to be reported in the report document Calculate the degree of reporting that represents the degree of.

The report degree calculation means 11 uses the input report degree of each utterance or expression unit in the dialogue as a probability of the occurrence of the expression w in the dialogue as a different representation w ′ in the report document. You may calculate based on the appearance probability to the report document of the same expression in the past dialog calculated using w ').

Although not shown, the report document creation support system outputs the report level of each utterance or expression unit in the input dialogue calculated by the report level calculation means 11 together with the utterance or expression unit. Means may be provided.

In addition, based on the degree of reporting of each utterance or expression unit in the input dialogue calculated by the reporting degree calculation unit 11, there may be provided summary generation means for extracting the utterance or expression unit from the input dialogue and generating a summary. Good.

Also, summary output means for outputting the summary generated by the summary generation means, report documents created based on the output summary are sequentially input, and the input report document and the dialog corresponding to the report document are provided. A dialog report pair registering unit that additionally stores the pair in the dialog report pair storing unit may be provided.

Further, a dialog input means for inputting a dialog as text data may be provided, and each utterance or each expression unit extracted from the dialog may be character string data.

In addition, a dialogue input means for inputting a dialogue as voice data and a voice recognition means for generating text data from the inputted voice data, each utterance or each expression unit extracted from the dialogue is voice data or character string data. There may be.

Further, the summary generation means may generate a summary of the text data by using the speech data or character string data utterance or expression unit extracted from the dialogue input based on the report level.

As mentioned above, although this invention was demonstrated with reference to embodiment and an Example, this invention is not limited to the said embodiment and Example. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

This application claims priority based on Japanese Patent Application 2011-007170 filed on January 17, 2011, the entire disclosure of which is incorporated herein.

The present invention can be suitably applied to uses such as report creation support from call voice in a contact center, cleansing of call text for text mining and search, and FAQ creation support.

100 computer (central processing unit; processor; data processing unit)
11, 101a, 101b, 101c Report degree calculation means 102b, 102c Summary generation means 12, 103c Dialog report pair registration means 200 Storage medium 201a, 201b, 201c Dialog report pair storage means 300 Input means 400 Output means

Claims (10)

Dialog report pair storage means for storing a set of pairs of a dialogue and a report document that is a sentence that reports the contents in advance;
Based on the degree to which the contents of each utterance or expression unit in dialogue in the set of dialogue and report document pairs stored in the dialogue report pair storage means are included in the corresponding report document, from the inputted dialogue For each utterance or each expression unit to be extracted, there is provided a reporting degree calculation means for calculating a degree of reporting indicating whether or not the utterance or the expression unit includes contents to be reported in the report document. Report document creation support system. The report document creation support system according to claim 1, further comprising reporting degree output means for outputting the degree of reporting of each utterance or expression unit in the input dialogue calculated by the reporting degree calculation means together with the utterance or expression unit. The summarization generating means for extracting the utterance or expression unit from the input dialog and generating a summary based on the reporting degree of each utterance or expression unit in the input dialog calculated by the reporting degree calculation means. Item 3. The report document creation support system according to item 2. Summary output means for outputting the summary generated by the summary generation means;
Dialog report pair registration means for sequentially inputting report documents created based on the output summary, and additionally storing the input report document and dialog corresponding to the report document in the dialog report pair storage means A report document creation support system according to claim 3. The reporting degree calculation means uses the input reporting degree of each utterance or expression unit during dialogue as a co-occurrence probability P (w, w, which is a probability that the expression w during dialogue appears as a different expression w ′ in the report document. The report document creation support system according to any one of claims 1 to 4, wherein the report document creation support system calculates the occurrence probability of the same expression in a report document during a past dialogue calculated using '). A dialog input means for inputting a dialog as text data is provided.
The report document creation support system according to any one of claims 1 to 5, wherein each utterance or each expression unit extracted from the dialogue is character string data. A dialog input means for inputting a dialog by voice data;
Voice recognition means for generating text data from the input voice data,
The report document creation support system according to any one of claims 1 to 5, wherein each utterance or each expression unit extracted from the dialogue is voice data or character string data. A report document creation support system comprising speech recognition means for generating text data from input speech data,
8. The report according to claim 6, wherein the summary generation means generates a summary of the text data using a speech or expression unit of speech data or character string data extracted from the dialogue input based on the degree of reporting. Document creation support system. In advance, the storage means stores a set of pairs of dialogue and a report document, which is a sentence that reports the contents,
Based on the degree to which the content of each utterance or expression unit in the dialogue in the set of dialogue and report document pairs stored in the storage means is included in the corresponding report document, it is extracted from the inputted dialogue. For each utterance or each expression unit, calculate a degree of reporting that indicates the extent to which the utterance or the expression unit contains content to be reported in the report document,
Output the degree of reporting of each utterance or expression unit in the calculated input dialogue together with the utterance or expression unit, or utterance from the input dialogue based on the degree of reporting of each utterance or expression unit in the calculated input dialogue Alternatively, a report document creation support method characterized by extracting a representation unit and generating a summary. On the computer,
Input based on the degree to which the content of each utterance or expression unit in the dialogue is included in the corresponding report document in the set of pairs of dialogue documents and the sentences that report the contents stored in the storage means. For each utterance or each expression unit extracted from the spoken dialogue, a process for calculating the degree of reporting that represents the extent to which the utterance or the expression unit contains content to be reported in the report document, and the calculated input Output the degree of reporting of each utterance or expression unit in the dialogue together with the utterance or expression unit, or select the utterance or expression unit from the input dialogue based on the calculated degree of reporting of each utterance or expression unit in the input dialogue. A report document creation support program for executing the process of extracting and generating a summary.

Images