JP2019028910A - Dialogue analysis system and dialogue analysis program - Google Patents

Abstract

To make it more efficient to search the contents of explanations to a customer for dialogues that need to be monitored and to do the monitoring.SOLUTION: The dialogue analysis system includes: a dialogue analysis unit 11 for performing natural language analysis processing on a sound recognition text 32 and outputting the analysis result; a dialogue attribute extraction unit 12 for giving a topic tag based on customer attribute data 42 to each dialogue and recording the topic tag as analyzed dialogue data 15; a user interface unit 14 for extracting dialogues from the analyzed dialogue data 15 on the basis of a specified condition and causing a user terminal 51 to display a list of the extracted dialogues; and a monitoring part extraction unit 13 for displaying a first region for displaying the words in a specified dialogue in time-series and a second region for displaying words that have been determined to describe a predetermined item.SELECTED DRAWING: Figure 1

Description

  TECHNICAL FIELD The present invention relates to a natural language analysis technique, and in particular, a dialog analysis system and a dialog analysis program for explaining matters to be explained by a sales person in business negotiations with a customer and checking that matters that should not be stated are not stated. It is related to technology effective when applied to.

  For example, when a sales person in charge of a financial institution conducts a sales activity on a financial product, it is necessary to comply with predetermined matters for compliance. This predetermined item explains, for example, the risks of financial products and other matters that must be explained (hereinafter sometimes referred to as “necessary items”), , “Do not speak up”, such as “I will definitely make a mistake”, not to speak (sometimes referred to as “prohibited expression” below), eligibility for customers such as elderly people and people with dementia, etc. This includes the fact that there is no problem with sex (with or without sales restrictions).

  For example, the manager of the sales department or the compliance check person in charge at the head office organization arouses and confirms the recorded data of the business negotiations (hereinafter referred to as “monitoring” in some cases). This takes a lot of work time. While the number of requests for compliance is increasing year by year, the number of cases to be monitored is increasing, but it is not realistic to monitor all customer interactions with limited personnel. Monitoring is based on sampling the dialogue. In addition, there are many veteran employees in charge of monitoring work, and the reduction of monitoring power due to the retirement of these personnel all at once in the near future is also an issue.

  Further, when a business negotiation is established as a result of a business activity such as explanation of financial products through a plurality of business negotiations, all necessary items are not necessarily explained in units of one business negotiation. In this case, it may be necessary to confirm the contents of a series of business negotiations that have been made in the past, which also increases the working time.

  On the other hand, it is considered to improve the efficiency of the monitoring work by adding the viewability and searchability by converting the recorded data of the negotiation contents into a text by using a voice recognition (dictation) technique. For example, JP-A-2006-85697 (Patent Document 1) discloses analyzed text information obtained by performing natural language analysis processing on text data obtained by converting the contents of each utterance of a sales representative into text, and necessary information. When the similarity with the predefined text data is calculated and the similarity exceeds a predetermined value, it is determined that the necessary items of the target are explained in the target utterance, and the prohibited expression information is included in the prohibited expression information. If there is something that matches the predefined text data, it is determined that the target prohibited expression is stated in the target utterance, so the presence or absence of “prohibited expression” and “necessary items” are included. Whether or not there is a check target is described.

JP-A-2006-85697

  According to the technology described in Patent Document 1, whether the contents explained or stated to the customer by the sales person in the course of business negotiations or business activities comply with the compliance to some extent. It is possible to check. Also, based on data such as explanations to customers during business negotiations and sales activities, whether or not “necessary items” are included and whether or not there are “prohibited expressions” may be included in the check. it can.

  However, for example, in a large-scale financial institution where a large number of sales representatives are active, the number of business negotiations and sales activities is enormous. The extraction work also requires a large work time. Further, the work time required for actual monitoring is also required to be shortened by further increasing efficiency. That is, there is room for further improvement in the technique described in Patent Document 1.

  Accordingly, an object of the present invention is to search for a dialog that needs to be actually monitored from data such as explanation contents to a customer during business negotiations or sales activities, and a dialog analysis system that makes the monitoring work efficient and It is to provide a dialogue analysis program.

  The above and other objects and novel features of the present invention will be apparent from the description of this specification and the accompanying drawings.

  Of the inventions disclosed in this application, the outline of typical ones will be briefly described as follows.

  A dialog analysis system according to a representative embodiment of the present invention is a dialog analysis system for determining whether or not a predetermined matter is explained during a dialog between an explainer and a counterpart, and the contents of the dialog are recorded. A dialog analysis unit that performs natural language analysis processing including morphological analysis on speech recognition text obtained by converting speech data into text using speech recognition technology and outputs an analysis result, and an analysis result output from the dialog analysis unit A dialogue attribute extraction unit that attaches a tag related to the topic based on the data including the attribute data of the other party and records the dialogue as analyzed dialogue data.

  A user interface unit that extracts dialogs that match the conditions from the analyzed dialog data based on conditions specified by a user via the information processing terminal, and displays the list on the information processing terminal; For dialogs designated by the user from the list display via the processing terminal, a first area for displaying each utterance of the presenter and the other party in the time series in the dialog, and the predetermined matter by the presenter A second area for extracting and displaying an utterance determined to be explained, and a monitoring location extracting section for displaying the utterance on the information processing terminal via the user interface section.

  The present invention can also be applied to a dialog analysis program that causes a computer to operate as the dialog analysis system as described above.

  Among the inventions disclosed in the present application, effects obtained by typical ones will be briefly described as follows.

  That is, according to a typical embodiment of the present invention, a task for searching for a dialogue that needs to be actually monitored from data such as explanation contents to a customer during business negotiations or sales activities, and a task for the monitoring It becomes possible to shorten the work time concerning monitoring work.

It is the figure which showed the outline | summary about the structural example of the dialog analysis system which is one embodiment of this invention. It is the figure which showed the outline | summary about the example of the dialog search screen displayed on the user terminal in one embodiment of this invention. It is the flowchart which showed the outline | summary about the example of the flow of the process which extracts and provides attribute information with respect to each dialog in one embodiment of this invention. It is the figure which showed the outline | summary about the example of the monitoring screen of the content of dialog displayed on the user terminal in one embodiment of this invention. It is the flowchart which showed the outline | summary about the example of the flow of a process which extracts the location which should be monitored from the dialogue of the object in one embodiment of this invention. It is the figure which showed the outline | summary about the example of the matching rule of the category provision information in one embodiment of this invention. It is the figure which showed the outline | summary about the setting example of the priority and appearance order of the category of speech in one embodiment of this invention. It is the figure which showed the outline | summary about the example of the learning data taken in into the machine learning engine in one embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. Note that components having the same function are denoted by the same reference symbols throughout the drawings for describing the embodiment, and the repetitive description thereof will be omitted. On the other hand, parts described with reference numerals in some drawings may be referred to with the same reference numerals although not illustrated again in the description of other drawings.

  The dialog analysis system according to an embodiment of the present invention records, for example, the contents of a dialog related to an explanation made by a sales person such as a financial institution to a customer during a business negotiation or sales activity as voice data. A natural language analysis is performed on the basis of the data converted into text, and it is automatically determined whether “necessary items” are explained and “prohibited expressions” are not issued, and the result is displayed. As a result, the sales staff and their managers can monitor the status of compliance in sales activities.

  At this time, in this embodiment, it is possible to shorten the time required for the monitoring work by providing a user interface that can easily identify the utterances that have issued “necessary items” and “prohibited expressions” during the conversation. . Furthermore, in the present embodiment, a topic tag (a predefined conversation type) is automatically assigned to each conversation based on the analyzed conversation content. As a result, it is possible to improve the efficiency of searching for a dialog that needs to be actually monitored from a large number of dialogs, and to shorten the time required for this.

  In the present embodiment, a description will be given of a compliance check for a business negotiation when a sales person in charge of a financial institution conducts a sales activity on a financial product, but the present invention is not limited to this. The dialogue analysis system according to an embodiment of the present invention determines whether or not a predetermined matter is stated in the dialogue between the presenter and the other party (whether a “necessary matter” is issued or a “prohibited expression” is set). It can be widely applied to tasks that need to be checked.

<System configuration>
FIG. 1 is a diagram showing an outline of a configuration example of a dialog analysis system 1 according to an embodiment of the present invention. In the dialog analysis system 1, for example, a sales representative or the like is provided to a user such as a sales representative of a financial institution or a manager thereof (more specifically, a user terminal 51 which is an information processing terminal used by the user). This is an information processing system that supports compliance check work for business negotiations.

  The dialog analysis system 1 is configured by, for example, a server device or a virtual server constructed on a cloud computing service. Then, a middle processing such as an OS (Operating System), a DBMS (DataBase Management System), and a Web server program developed on a memory from a recording device such as an HDD (Hard Disk Drive) by a CPU (Central Processing Unit) (not shown) By executing the software that operates on the above, various functions to be described later relating to the support of the compliance check work are realized.

  The dialog analysis system 1 includes various units such as a dialog analysis unit 11, a dialog attribute extraction unit 12, a monitoring location extraction unit 13, and a user interface unit 14 implemented by software. In addition, the analyzed dialogue data 15 implemented by a database or the like, and other files, tables, etc. (not shown) are held.

  The dialogue analysis unit 11 performs natural language analysis including morphological analysis based on the speech recognition text 32 obtained by converting the speech data of the negotiation into text, for example, checking for the presence of “prohibited expressions” by keyword matching, and calculating similarity. It has a basic function of checking whether or not it includes “necessary items”. Each basic function of the dialog analysis unit 11 is, for example, each unit (for example, “text analysis unit 210”, “similarity calculation unit 220”, “keyword matching unit 230” of the compliance check system described in Patent Document 1 described above. , Etc.), a detailed description thereof is omitted here.

  Note that the voice recognition text 32 is recorded, for example, as dialogue recording data 21 by the dialogue recording system 2 by recording the dialogue contents between the sales representative and the customer involved in the business negotiation, and the contents of the dialogue recording data 21 are voice recognition. It is obtained by converting into text by the speech recognition unit 31 of the system 3.

  The dialogue recording system 2 may record, for example, a direct dialogue between a sales person and a customer using a microphone (not shown) or may record the contents of a telephone call. The recorded dialog recording data 21 is transmitted to the dialog analysis system 1 or can be accessed from the dialog analysis system 1 so that the dialog analysis system 1 can reproduce it as voice. In addition, the technology for converting the content of the dialog recording data 21 into text in the voice recognition unit 31 of the voice recognition system 3 is not particularly limited, and a known system, software, library, or the like can be used as appropriate.

  The dialog attribute extracting unit 12 includes, for example, information including customer attribute data 42 related to the target customer acquired from the CRM data 41 held by the CRM system 4 of the financial institution for the dialog data of the analysis result by the dialog analyzing unit 11. Based on the above, a topic tag is assigned to each dialogue, a plurality of dialogues handled as a series of dialogues are grouped and classified, and dialogue data as a processing result is recorded as analyzed dialogue data 15. As a result, it is possible to improve the efficiency of searching for a dialog that needs to be actually monitored from a large number of dialogs. Detailed processing contents in the dialog attribute extracting unit 12 will be described later.

  In addition, while performing the process in each part of the speech recognition part 31 of the speech recognition system 3, the dialog analysis part 11 of the dialog analysis system 1, and the dialog attribute extraction part 12, for example, while performing as a batch process at timings, such as daily, respectively, You may make it perform at any time based on the instruction | indication from a user.

  The monitoring part extraction unit 13 assigns a category to each utterance in the dialogue and adds a category to the dialogue specified by the user via the user interface unit 14 described later in the dialogue of the analyzed dialogue data 15. Has a function of extracting utterances that the user seems to be explaining "necessary items" to customers. This makes it possible to present to the user a high-priority location that should actually be monitored in the target dialogue. Detailed processing contents in the monitoring location extraction unit 13 will be described later.

  The user interface unit 14 has a function of providing the user terminal 51 with an input / output interface using a screen when monitoring the contents of the negotiation. For example, an interactive search screen, an interactive monitoring screen, etc., which will be described later, are displayed on the Web browser on the user terminal 51 by a Web server program (not shown).

<Identification of dialogue to be monitored>
In the present embodiment, the user first accesses the user interface unit 14 of the dialog analysis system 1 using the user terminal 51, searches for dialogs in the analyzed dialog data 15, and sets them as monitoring targets. Identify the conversation.

  FIG. 2 is a diagram showing an outline of an example of a dialog search screen displayed on the user terminal 51 in the present embodiment. This screen displays a list of search results of dialogs in the analyzed dialog data 15 for telephone calls between sales representatives and customers. In the upper part of the screen, there are displayed an input field, a drop-down list, and the like for accepting designation from the user about “keyword”, “sales employee name”, “period”, “time”, “confirmation status”, and the like. By specifying here, it is possible to perform keyword search for the text of the dialogue content and narrow down using various index information.

  The list of search results at the bottom of the screen lists each dialog extracted from the analyzed dialog data 15 and items indicating attribute information of each dialog. Here, as shown in the “topic tag” column on the right, as a result of analyzing the content of the dialog, for example, a tag indicating the topic or topic of the dialog such as a product targeted by the dialog is given. Since the analysis result may be inappropriate, the topic tag automatically assigned can be manually added / deleted / modified by the user. Based on the content of the topic tag, an analysis rule to be used on a monitoring screen described later is determined.

  Further, in the “risk level” column, the degree of compliance risk with respect to business activities is determined and displayed from the content of the dialogue, the attribute information of the target customer, and the like. Note that the “contents of dialogue” here indicates general contents and features that do not depend on the contents of the negotiation, such as “I say the same thing many times” and “There are few conflicts”. By making a determination based on such conversation contents and customer attribute information, for example, the risk may be increased when the other party is a dementia or elderly customer, regardless of the target or contents of the explanation. And increase the priority in identifying the dialogue to be monitored. In the example in the figure, in the “Danger Level” column, the risk level is shown in three levels of high, medium, and low by colored marks among the three marks, but other display methods may be used. Good.

  In the “call ID” column on the left, call IDs uniquely assigned to each dialog are displayed as hyperlinks, and when the user selects this link, monitoring will be described later for the selected dialog. Transition to the screen. In the “Group ID” column, for each dialogue, for example, based on the destination telephone number and the analysis result of the topic tag, etc. Are grouped as a series, and the group ID assigned to each group is displayed by a hyperlink. When the user selects this link, a transition is made to a monitoring screen, which will be described later, for a combination of dialogs in the group.

  FIG. 3 is a flowchart showing an outline of an example of the flow of processing in which the dialog attribute extracting unit 12 in the present embodiment extracts and assigns attribute information to each dialog. In batch processing at a daily timing, the dialogue attribute extraction unit 12 first acquires dialogue data as a result of analysis by the dialogue analysis unit 11 (S01).

  Then, a topic tag is assigned to each dialogue in the obtained dialogue data (S02). Here, for example, based on the keywords being interacted with and customer attribute data 42 acquired from the CRM system 4 (for example, the type of product purchased in the past, purchase time, customer age, etc.), The current financial product and customer attributes are determined on a rule basis, and a topic tag is attached from the viewpoint of monitoring performed later. A plurality of topic tags may be assigned.

  The types of topic tags that can be assigned are registered in advance in a table or file of topic tag information 121, for example. In the topic tag information 121, for example, a topic tag and information such as keywords and customer attributes associated therewith are registered. Maintenance such as registration and update of the content of the topic tag information 121 is separately performed irregularly by the user, for example (S00).

  For example, in the screen example of the dialog search result in FIG. 2 described above, the user can manually correct the topic tag given to each dialog. The corrected topic tag and the text data of the target dialog Based on the above, the contents of the topic tag information 121 can be updated. Specifically, for example, for each topic tag, a TF-IDF (Term Frequency-Inverse Document Frequency) value is calculated for a word in the text data of the conversation to which the target topic tag is assigned, and a feature score is assigned. It is possible to take a technique such as setting a high word as a keyword associated with a target topic tag.

  When a topic tag is assigned to each dialogue in step S02, it is determined whether or not a sales representative explains “necessary items” in each dialogue based on the given topic tag. Based on this, a process for extracting an utterance to be monitored is performed (S03). The details of the monitoring location extraction process will be described later. If the accuracy of the process is not high, it is not essential to perform the process at this timing, and the process may be skipped.

  Thereafter, the degree of compliance risk for each dialogue is further determined (S04). Here, for example, keywords in dialogue text data, how customers speak and respond (for example, “I say the same thing many times (possibility of dementia)”, “ Based on the customer attribute data 42 obtained from the CRM system 4 (for example, “whether it is an elderly person”, “whether the transaction history is shallow”, etc.) Determine the risk level.

  Further, when the monitoring point extraction process is performed in step S03, the risk level is corrected based on the result information, that is, the information indicating how much “necessary items” are explained in the target dialogue. You may do it.

Specifically, the calculation of the risk level is, for example,
・ If the customer's age is 70 years old or higher → the risk is +0.5
・ If the customer's transaction history is less than 3 years → the risk is +0.5
・ The sales representative has been talking for over 30 seconds in the dialogue.
If the customer has never spoken during that time → the risk is +1.0
・ (When monitoring points are being extracted)
If even one explanation of “Required Information” is missing → Increase the risk level to +1.5
The final risk level is calculated by integrating (or subtracting) the risk level values when the above rules are met.

  After that, based on the keyword in the text data of the dialogue and the information of the customer attribute data 42 acquired from the CRM system 4 (for example, the type of product purchased in the past, the purchase time, etc.) The continuity with the dialogue is determined, a series of dialogues having continuity is grouped (S05), and the process is terminated.

  For example, if the dialogue text data contains a keyword such as “I had previously considered…”, it was heard that the target product was already explained in the previous negotiation, It can be determined that there is continuity. Also, if a keyword such as “This time a new product has been recruited…” is included, it will be asked to propose a new product for the first time in this business negotiation, and it will be judged that there is no continuity with past business negotiations. be able to. In addition, even if the conversation is supposed to be continuous with the previous conversation by the same salesperson, for example, if the conversation has no history of response for more than one month, a new group is created. To associate this dialog with this group. On the other hand, if the dialogue has a history of response within the most recent month, the dialogue is associated with the group to which the previous dialogue belongs.

<Monitoring for specified dialogue>
In the present embodiment, the user selects, for example, a dialogue (or a group of dialogues) determined to have a high degree of risk from the dialogue search result screen as shown in the example of FIG. Monitor whether the “necessary items” are explained in the specific utterances of the sales representatives and whether “danger expressions” are issued.

  FIG. 4 is a diagram showing an overview of an example of a dialogue content monitoring screen displayed on the user terminal 51 in the present embodiment. In this screen, for the dialogue selected by the user from the dialogue search result screen as shown in the example of FIG. 2 (including those obtained by combining the dialogues in the same group), the contents on the left side, that is, The contents of the dialogue between the sales representative and the customer are displayed in time series for each utterance. In the right area, a summary of the selected dialogue (that is, an explanation item explained to the customer by the sales representative) is displayed for each utterance category as a response record.

  The dialogue content displayed in the left area is based on the type of each utterance set by the natural language analysis in the dialogue analysis unit 11 and, for example, unnecessary conversations such as “conversation” and “greeting” are hidden. It is displayed as “Simple”. Each utterance is provided with a reproduction icon, and by selecting this icon, the user can reproduce the contents of the conversation partially as utterances in units of utterances and perform monitoring by voice. Also, a playback icon and a stop icon are attached to the entire dialog, and the entire dialog can be played / stopped. In addition, various operations related to audio reproduction such as pause, fast forward, and rewind are possible. During playback, for example, the playback speed may be selected such as “1 × speed”, “1.5 × speed”, “double speed”. Further, highlighting such as blinking or highlighting may be performed for the corresponding part of the utterance being reproduced.

  Each utterance is given a category to be described later. In the example in the figure, categories such as “document arrival confirmation”, “subordinated special contract”, and “bond rating” are assigned. Thereby, it is possible to easily grasp the “necessary items” that are the subject of explanation in each utterance.

  In the response record displayed in the area on the right side, for each category, the part that is determined to explain the “necessary items”, that is, the utterance that the user should monitor is extracted from the dialog and displayed. Yes. Here, for example, when the user selects and designates an utterance to be confirmed in detail, the user interface unit 14 automatically scrolls the conversation content displayed in the left area to the position where the utterance is displayed, and You may make it start reproduction | regeneration of an audio | voice from speech.

  Further, the user operates the slider bar at the top of the screen to determine the granularity (extraction amount) of the extracted part (utterance) according to the specific accuracy of the description part of the “necessary items” in the monitoring part extracting unit 13. You may enable it to change dynamically. For example, if the granularity is increased, many utterances are extracted covering a wide range considered to include “necessary items”, and if the granularity is decreased, utterances (or a part thereof) are extracted by narrowing down to a narrower range. With these mechanisms, the user can efficiently identify the utterance to be monitored, which seems to explain “necessary items”, and eliminate unnecessary monitoring.

  FIG. 5 is a flowchart showing an outline of an example of a flow of processing in which the monitoring location extraction unit 13 in the present embodiment extracts a location to be monitored from the target dialogue. When a dialog is selected by the user via the dialog search result screen as shown in the example of FIG. 2, the monitoring location extraction unit 13 first acquires dialog data related to the target dialog from the analyzed dialog data 15. (S11).

  Then, a category is assigned to each utterance in the acquired dialogue data (S12). Here, for example, utterances are made by rule-based matching based on information such as speakers and words (or their utilization), specific expressions, utterance intentions, and the like extracted by natural language analysis in the dialog analysis unit 11. Is given a category. Information on the rule for assigning the category is registered in advance in, for example, a table or file of the category assignment information 131. FIG. 6 is a diagram showing an outline of an example of the matching rule of the category assignment information 131 in the present embodiment. It is determined whether or not the content of each utterance matches each rule shown in FIG. 6 (in the example in the figure, “rule a”, “rule b”, “rule c”,...).

  If the target utterance does not match any of the category rules, for example, the category assigned to the immediately preceding utterance may be assigned as it is. On the other hand, when the target utterance matches a plurality of categories, for example, one of the categories is assigned to each category based on a preset priority and rank. FIG. 7 is a diagram showing an outline of an example of setting the priority and appearance order of the utterance category in the present embodiment. When a plurality of categories are matched for a certain utterance, first, a category with a high priority is assigned (rule 1). Therefore, for example, when the first utterance in the dialogue matches the category of “subordinated special contract” and the category of “principal interest exemption risk”, the category of “subordinated special contract” with high priority is given by rule 1. .

  On the other hand, if the priorities of multiple matched categories are the same, information on the category assigned to the utterance immediately before the utterance is acquired, and the category with the latest rank is assigned after the category. (Rule 2). Thus, for example, if the second utterance matches the category of “document arrival confirmation (rank 1)” and the “substantial failure exemption special contract (rank 3)” category, the first utterance immediately before according to rule 2 The category of “exemption at the time of substantial bankruptcy” is assigned to the subordinate category (and the latest ranking) from the category of “subordinated special agreement (ranking 2)”.

  If there is no rear-ranked category, the category with the rank closest to the rank of the category assigned to the immediately preceding utterance is assigned (rule 3). Thus, for example, with the category of “credit risk (rank 5)” assigned to the third utterance, the fourth utterance is classified as “document arrival confirmation (rank 1)” and “exempt at the time of substantial failure”. If it matches the category of “Special Contract (Rank 3)”, the “Real” of the rank (Rank 3) closest to the rank of the “Credit Risk” category (Rank 5) assigned to the third utterance immediately before by the rule 3 The category of “exemption on bankruptcy” is granted. By assigning a category based on the above rules, it is possible to automatically assign a category in a natural manner considering the flow of utterances during conversation.

  Returning to FIG. 5, when a category is assigned to each utterance, next, an utterance (one or more utterances, or a part of the utterance) that is considered to explain “necessary items” during the dialogue is selected. Extract (S13), and the process ends. Here, for example, for each utterance in the target dialogue recorded in the analyzed dialogue data 15, the feature is extracted for each utterance, and the extraction result is evaluated by the learning model (explained part extraction information 132). The evaluation result is indicated by, for example, a score normalized to a value range of −1 to 1, and an utterance whose score is larger than a predetermined threshold is determined as an explanation part of “necessary items” and extracted.

  Maintenance such as updating of the explanation location extraction information 132, which is a learning model, is separately performed irregularly, for example, in accordance with a user instruction (S10). It may be performed periodically at a predetermined timing such as monthly. Here, for example, as learning data when performing machine learning, a place where a user has confirmed by monitoring through a screen as shown in FIG. 4 for a past dialogue (or a place designated to be performed) ) Information. In order to enable extraction, a flag may be set in the analyzed dialogue data 15 for the utterance (or the utterance designated to be performed) monitored by the user.

  In the maintenance process, for each utterance, information on the elements constituting the utterance and its features are further extracted. The features may include, for example, information on the words used and their parts of speech, utterance intentions, utterance positions in dialogue, and the like.

  The information of the utterance (the utterance to be performed) and the information of the utterance characteristics that have been monitored are converted into a format that can be learned by a machine learning engine (not shown), and then learned by the machine learning engine. The description location extraction information 132, which is a learning model, is output. The machine learning engine to be used is not particularly limited. For example, an open source machine learning library such as LIBLINEAR can be used as appropriate.

FIG. 8 is a diagram showing an outline of an example of learning data imported into the machine learning engine according to the present embodiment. The upper data is, for example, information on whether or not to monitor each utterance (“Stmt”) in the dialogue (“Dlg”) extracted from the analyzed dialogue data 15 (“OK” if necessary), If not required, “NG”) is set. On the other hand, the lower data is a format in which the upper data can be interpreted by, for example, LIBLINEAR.
[Class number] [Feature number 1]: [frequency] [Feature number 2]: [frequency]
It is converted to the format. Here, a score indicating whether or not to monitor (1 for “OK” and −1 for “NG”) is set as [class number]. The explanation location extraction information 132 (learning model) is created and updated by taking this file into LIBREAR and learning.

  In the above-mentioned example, the part explaining “necessary items” is extracted and displayed as a response record on the monitoring screen as shown in FIG. 4, but a “prohibited expression” is issued. It is also possible to extract and display the same location.

  As described above, according to the dialog analysis system 1 according to an embodiment of the present invention, the content of the dialog related to the explanation given to the customer is recorded as voice data, and this is converted into text. Thus, natural language analysis is performed to automatically determine whether “necessary items” are explained and whether “prohibited expressions” are not issued, and the result can be displayed. As a result, the sales staff and their managers can monitor the status of compliance in sales activities.

  At this time, the time required for the monitoring work can be reduced by providing a user interface that can easily identify the utterances that have issued “necessary items” and “prohibited expressions” during the conversation. Furthermore, by automatically assigning topic tags to each dialogue based on the analyzed dialogue content, the time required for searching for dialogues that actually need to be monitored from a large number of dialogues is reduced. Shortening is possible.

  As mentioned above, the invention made by the present inventor has been specifically described based on the embodiments. However, the present invention is not limited to the above-described embodiments, and various modifications can be made without departing from the scope of the invention. Needless to say. For example, the above-described embodiment has been described in detail for easy understanding of the present invention, and is not necessarily limited to the one having all the configurations described. In addition, it is possible to add, delete, and replace other configurations for a part of the configuration of the above-described embodiment.

  Each of the above-described configurations, functions, processing units, processing means, and the like may be realized by hardware by designing a part or all of them with, for example, an integrated circuit. Each of the above-described configurations, functions, and the like may be realized by software by interpreting and executing a program that realizes each function by the processor. Information such as programs, tables, and files for realizing each function can be stored in a recording device such as a memory, a hard disk, or an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD.

  Moreover, in each said figure, the control line and the information line have shown what is considered necessary for description, and do not necessarily show all the control lines and information lines on mounting. Actually, it may be considered that almost all the components are connected to each other.

  The present invention can be used for a dialog analysis system and a dialog analysis program for explaining matters to be explained by a sales person in a business talk with a customer and checking that matters that should not be stated are not stated.

1 ... Dialog analysis system, 2 ... Dialog recording system, 3 ... Speech recognition system, 4 ... CRM system,
DESCRIPTION OF SYMBOLS 11 ... Dialog analysis part, 12 ... Dialog attribute extraction part, 13 ... Monitoring location extraction part, 14 ... User interface part, 15 ... Analyzed dialog data,
21 ... Dialogue recording data,
31 ... voice recognition unit, 32 ... voice recognition text,
41 ... CRM data, 42 ... Customer attribute data 51 ... User terminal,
121: Topic tag information, 131: Category assignment information, 132: Explanation location extraction information

Claims (9)

A dialog analysis system for determining whether or not there is an explanation of a predetermined matter during a dialog between an explainer and the other party,
A dialogue analysis unit that performs natural language analysis processing including morphological analysis on speech recognition text obtained by converting the content of speech data in which dialogue content is recorded into text using speech recognition technology, and outputs an analysis result;
A dialog attribute extraction unit that adds a tag related to a topic to each dialog included in the analysis result output from the dialog analysis unit based on data including the other party's attribute data, and records the data as analyzed dialog data When,
Based on conditions specified by the user via the information processing terminal, a user interface unit that extracts dialogs that match the conditions from the analyzed dialog data and displays the list on the information processing terminal;
For dialogs designated by the user from the list display via the information processing terminal, a first area for displaying the utterances of the presenter and the other party during the conversation in time series, and the predetermined by the presenter A second area for extracting and displaying an utterance determined to have been explained, and a monitoring location extracting section for displaying the utterance on the information processing terminal via the user interface section. Analysis system. The dialog analysis system according to claim 1,
The dialog attribute extraction unit determines the degree of compliance risk of each dialog based on the content of the dialog that does not depend on the attribute data of the other party and the explanation target for each dialog included in the analysis result. Dialogue analysis system. The dialog analysis system according to claim 1,
The dialogue attribute extraction unit, for each dialogue related to the presenter included in the analysis result, based on information including a keyword during dialogue, the tag provided, and the opponent, a plurality of Dialog analysis system that groups dialogs. The dialog analysis system according to claim 1,
The dialog analysis system, wherein the user interface unit scrolls the first area so that an utterance designated by the user in the second area displayed on the information processing terminal is displayed in the first area. In the dialog analysis system according to claim 4,
The dialog analysis system, wherein the user interface unit reproduces voice data related to an utterance designated by a user in the second area on the information processing terminal. The dialog analysis system according to claim 1,
The monitoring location extraction unit is determined to explain the predetermined matter by the instructor according to a predetermined operation amount of the user in the second area displayed on the information processing terminal. A dialog analysis system that changes the granularity when extracting utterances. The dialog analysis system according to claim 1,
The monitoring location extraction unit assigns a category corresponding to the predetermined item by matching with a predetermined rule for each utterance in the conversation designated by the user from the list display, and the matching category is Dialog analysis system that assigns the category assigned to the previous utterance if there is not. The dialog analysis system according to claim 1,
The monitoring location extraction unit assigns a category corresponding to the predetermined item by matching with a predetermined rule for each utterance in the conversation designated by the user from the list display, and the matching category is If there are multiple categories, the category with the highest priority is selected and assigned, and if there are multiple categories with the highest priority, any one category is selected and assigned based on the category assigned to the immediately preceding utterance. , Dialogue analysis system. A dialogue analysis program for causing a computer to execute a process so as to function as a dialogue analysis system for determining whether or not there is an explanation of a predetermined item during a dialogue between an explainer and a partner,
Dialogue analysis processing that performs natural language analysis processing including morphological analysis and outputs the analysis results to the speech recognition text that has been converted into text by speech recognition technology, the content of the speech data recorded the dialogue content,
Dialog attribute extraction processing for adding a tag related to a topic based on data including attribute data of the other party to each dialogue included in the analysis result output by the dialogue analysis processing, and recording it as analyzed dialogue data When,
Based on conditions specified by the user via the information processing terminal, a dialog that matches the conditions from the analyzed dialog data is extracted and displayed in a list on the information processing terminal;
For dialogs designated by the user from the list display via the information processing terminal, a first area for displaying the utterances of the presenter and the other party during the conversation in time series, and the predetermined by the presenter The interactive analysis program which makes the said computer perform the monitoring location extraction process which makes the said information processing terminal display the 2nd area | region which extracts and displays the speech determined that description of the matter is performed.