JP2007304793A - Document retrieval device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a document retrieval device for efficiently retrieving a document even when a conversation is inadequate, in a document retrieving device for extracting a keyword from a conversation between the operator and customer of a call center, and for retrieving a document relevant to the keyword from a database, and for providing content corresponding to the inquiry of the customer from the document to the operator. <P>SOLUTION: This document retrieval device for extracting a keyword based on voice recognition from a conversation between an operator and a customer, and for retrieving a document from the keyword, is configured to automatically update retrieval conditions by changing the combination of keywords even when any error occurs in voice recognition, or converting a portion of the keywords, and to efficiently retrieve the document by effectively utilizing the obtained keywords without bothering the operator. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a document search apparatus that improves the convenience of a person who is talking or who is listening to a conversation by presenting a document related to the contents of the conversation exchanged by voice.

2. Description of the Related Art Conventionally, for example, in a product manufacturing / sales company, a call center operator answers an inquiry from a customer to the call center based on knowledge and experience.
However, depending on only the operator's knowledge and experience, there is a risk that the quality of customer service will be reduced for an inexperienced operator. In other words, the quality of response varies depending on the knowledge and experience of the operator.
Training to homogenize operator knowledge and training as much as possible is very expensive. Further, when products and services provided to customers are wide, complex, and frequently updated, it is difficult for even a skilled operator to maintain knowledge necessary for business.

Therefore, a FAQ system in which typical inquiries and answers are stored in a database is used to support the operator's response.
When an operator uses the FAQ system, it is necessary to search for necessary documents by entering keywords during response, display the searched documents on the display, and read out the contents of the displayed documents. Information is communicated to customers and used as responses to customer inquiries.
To further reduce the burden on the operator, the operator's voice or the customer's voice is recognized by the voice recognition device, keywords are extracted from the obtained voice recognition results, and the database is searched using the extracted keywords. Has also been done. This can reduce the labor for the operator to think and input keywords.

For example, Patent Document 1 discloses a method of performing a search by inputting a keyword in the form of text data with respect to a database storing voice data. This document describes a retrieval device that performs speech recognition on speech data stored in a database and expands the result in consideration of recognition errors to improve retrieval performance. Since this device assumes keyword input by text, it is assumed that an appropriate search keyword is given.
Japanese Patent Application Laid-Open Publication No. 2004-246824 “Speech Document Retrieval Method and Apparatus and Speech Document Retrieval Program”

However, when a database is searched using a keyword obtained from a speech recognition device, a keyword that can search for a document that matches the content of a conversation is not always obtained from the speech.
In addition, since there is a possibility that an error is included in voice recognition, an appropriate search is not performed if an incorrect keyword is extracted.
That is, when a document search is performed based on a keyword included in the speech recognition result, the combination of keywords is not always appropriate, and a speech recognition error may be included. There was a problem that the usefulness of the operator was low.

  SUMMARY OF THE INVENTION An object of the present invention is to solve the above-described problems caused by the prior art, and suppresses a decrease in document search performance due to a combination of keywords or a voice recognition error, thereby providing a document search device with higher search performance than before. Is to provide.

  In order to achieve the above-described object, the first feature of the document search apparatus according to the present invention is in addition to the conventional processing in which document search is performed using one or more keywords extracted from the speech recognition result. Thus, a combination of a plurality of keywords is configured such that a keyword is deleted, a keyword is added, and a keyword is changed to automatically generate a new keyword combination and perform a plurality of searches.

  According to the above-described configuration, even when a combination of keywords is inappropriate or a speech recognition error of a keyword is included, many document search results are obtained by a plurality of searches with the keyword combination changed. This increases the possibility that the operator can find a document necessary for the response, and the document retrieval performance is improved.

The configuration as the second feature is that, in the configuration as the first feature described above, the combination is changed by deleting one or more keywords from a plurality of keywords.
When a search result is narrowed down because a certain keyword is included, or there are keywords mixed in due to a voice recognition error, the document search performance can be improved by deleting these keywords. .

The configuration as the third feature is that, in the configuration as the first feature described above, the combination is changed by adding one or more keywords to the combination of keywords.
If the search results cannot be narrowed down enough with just the keywords obtained from a single speech recognition result, add the keywords obtained in the previous utterance to narrow down the search results and improve the document search performance. be able to. For example, it is conceivable to add the accumulated keyword from the keyword history before the earliest accumulated keyword among the current keyword combinations.

The fourth feature is that the second and third features described above are combined.
It is possible to obtain better search results by deleting inappropriate keywords and adding new keywords.

The configuration as the fifth feature is that the combination is changed by replacing the expression of the keyword with the synonym / synonym dictionary in the configuration as the first feature described above.
Search performance can be improved by replacing keywords with words that have the same or similar meaning and are easy to match.

The configuration as the sixth feature is that, in the configuration as the first feature described above, the combination is changed by adding a keyword using a synonym / synonym dictionary.
Search performance can be improved by performing a document search using a plurality of words having the same meaning or similar meaning as the keyword.

The seventh characteristic configuration is the above-described first to sixth characteristic configuration. When the keyword combination is changed, the keyword to be deleted, added, or replaced is a keyword obtained in the course of the speech recognition process. The point is that the selection is made based on the reliability score.
By deleting a keyword with a low reliability score or adding or replacing a keyword with a high reliability score, the reliability of the combination of keywords is increased, and an improvement in search performance can be expected.

The configuration with the eighth feature is the above-described configuration with the first to seventh features. The search results obtained by performing a plurality of document searches by changing the combination of keywords are spaced at time intervals. The point is that it is configured to present.
Rather than simultaneously displaying a plurality of search results obtained by changing keyword combinations, the operator can easily confirm the displayed results by presenting them at intervals.

  According to the present invention, in a document search using a keyword extracted from a speech recognition result of conversational speech, even if an inappropriate keyword or a keyword due to a speech recognition error is included, the combination of keywords is changed, and the condition is automatically set. By performing a search after updating, a search result that is not affected by inappropriate keywords can be obtained. In other words, since the search is performed under more conditions, there is a high possibility that the searched document matches the response content.

  When implementing the present invention, the entire document search apparatus can be configured by hardware. However, the simplest implementation is to install a document search program in a computer and allow the computer to function as the document search apparatus according to the present invention. The form is the best embodiment.

  In order for a computer to function as a document search apparatus according to the present invention, a speech recognition unit that converts speech input to a computer into text data by a document search program installed in the computer, and a document from text data obtained by the speech recognition unit A keyword extraction / accumulation unit that extracts keywords used for search and accumulates them as a keyword history, a keyword combination generation unit that generates keyword combinations from the accumulated keyword histories, and a document database using the generated keyword combinations The document database search unit for searching the document, the document title obtained by the search result of the document database search unit and displayed on the display unit, and the document title displayed on the display unit In a document Obtained from the speech recognition result of the speech recognition unit, the input means for designating the display on the display means, the document presentation means for reading the content of the document title designated by the input means from the document database and displaying it on the display means A keyword extraction detection unit that detects situations that do not include keywords in the text data, and this keyword extraction detection unit is activated by detecting a situation that does not include keywords, and changes the combination of keywords that are currently configured. In accordance with the combination of keywords, a keyword combination changing unit that causes the document search unit to perform a document search is constructed and functions as a document search apparatus.

  Hereinafter, an embodiment in which the present invention is applied to an operator support system of a call center will be described in detail. In the operator support system described in this embodiment, an operator answers an inquiry from a customer with reference to an FAQ system and a past response history.

  An outline of the operation will be described in advance. Each time an operator and a customer exchange a conversation via the telephone, the speech recognition unit outputs a character sequence obtained by connecting word sequences as a speech recognition result. The keyword extraction / accumulation unit compares the character sequence with a keyword group prepared in the keyword dictionary, and if a keyword exists in the character sequence, extracts the keyword and stores it as a keyword history. The keyword combination generation unit generates a combination of keywords from a plurality of keywords accumulated in the keyword history, for example, according to the extraction order. The document database retrieval unit retrieves a document group stored in the document database according to the generated keyword combination, and retrieves document data including the keyword. The document title is extracted from the retrieved document data, and the document title is displayed on the display means. The display of the document title is continued as long as the conversation between the operator and the customer continues, the keyword combination is changed each time a new keyword is extracted during the conversation, the search is executed according to the changed search condition, and the document title is changed. It is additionally displayed.

  Here, the feature of the present invention is that the search for document data is continued by changing the combination of the keywords even in a situation where the conversation between the operator and the customer does not include the keywords. That is, while the utterances such as “Utto” and “That is it” are not related to the inquiry, keyword extraction stops and no new search information is input. Under such circumstances, the present invention is characterized in that the keyword combination is changed and the search is continued according to the new search condition.

  By changing the keyword combination in this way, even if a keyword recognized in speech recognition error is included in the keyword adopted in the combination, the keyword is deleted, etc. The probability of reaching a document with appropriate contents can be increased.

Details will be described below.
FIG. 1 shows a configuration example of the entire operator support system.
The operator support system 101 includes a plurality of operator seats 104-1 to 104-m in which telephones 102-1 to 102-m and terminals 103-1 to 103-m are installed, and document search devices 105-1 to 105-m. It is comprised by.
The telephones 102-1 to 102-m are connected to the telephone line 106, and conversation voices exchanged with the telephones 102-1 to 102-m via the telephone line 106 are sent to the document search devices 105-1 to 105-m. Entered.

Terminals 103-1 to 103-m are general-purpose information processing apparatuses (personal computers) having a communication function and an information browsing function, and an operator operates the terminals 103-1 to 103-m to generate documents. A list of document titles and the contents of a document, which are search results output from the search devices 105-1 to 105-m, can be browsed.
Terminals 103-1 to 103-m and document search apparatuses 105-1 to 105-m are connected via a network.

The document search devices 105-1 to 105-m receive the conversation voice exchanged via the telephone line 106, and display the titles of one or a plurality of documents searched based on the result of recognizing the voice. Further, when the operator selects a document title by an operation such as a keyboard or mouse click, the contents of the corresponding document are displayed.
The document search device may be configured on an operator's terminal, or may be configured on a network so that a single document search device is used by a plurality of terminals.
In the case of an IP phone, the phone is configured on the terminals 103-1 to 103-m.

  FIG. 2 shows a configuration example of the document search devices 105-1 to 105-m in the operator support system 101. The document search devices 105-1 to 105-m include a voice recognition unit 201, a keyword extraction / accumulation unit 202, a keyword extraction detection unit 203, a keyword combination generation unit 204, a keyword combination change unit 205, a document database search unit 206, a document title. It consists of a presentation unit 207 and a document presentation unit 208.

The speech recognition unit 201 obtains a speech recognition result for the input speech using the acoustic model 211, the recognition dictionary 212, and the language model 213.
The keyword extraction / accumulation unit 202 refers to the keyword dictionary 214 and extracts keywords included in the speech recognition result.
The document database search unit 206 searches for a document matching the keyword from the document database 215 using one or a plurality of keywords extracted by the keyword extraction / accumulation unit 202.
The document title presentation unit 207 displays the titles of the documents searched by the document database search unit 215 on the operator terminals 103-1 to 103-m.

  When the operator designates the document title displayed by the document title presentation unit 207, the document presentation unit 208 displays the designated document on the display unit 210 provided in each operator terminal 103-1 to 103-m. To do.

Next, the operation procedure of the document search devices 105-1 to 105-m will be described with reference to FIG.
Upon receiving the keyword extraction result from the keyword extraction / accumulation unit 202, the keyword extraction detection unit 203 determines whether or not a keyword is included in the utterance (step S301).
If a keyword is included, the keyword combination generation unit 204 generates a keyword combination using the extracted keyword (step S302).
As a combination of keywords, a set of N keywords from the beginning or end in the order of keyword extraction may be set, or a set of N keywords from the top in the order of the number of hits of the same keyword may be considered.

The document database search unit 206 searches the document database 215 using the generated keyword combination (step S303).
The document title presenting unit 207 displays, for example, the top k document titles of the search result on the display unit 210 (S304). The top k search results can be, for example, the order in which the number of hits of the corresponding keyword is large or the search order.

An input for designating a document title is performed from the input means 209 such as a mouse or a keyword provided in the terminals 103-1 to 103-m (step S305).
When the document title is designated, the document presentation unit 208 reads the document data subordinate to the designated document title from the document database 215 from the document database 215, and displays the document data on the display unit 210 (step S306).

When the display of the document data is completed, the document search devices 105-1 to 105-m wait for the keyword input from the keyword extraction / accumulation unit 202 (step S307).
In the keyword input waiting state, a keyword extraction result is generated. If the keyword extraction result includes a keyword, the above-described steps S302, S303, S304, S305, and S306 are executed, and the input waiting state shown in step S307 is again performed. stand by.

On the other hand, when the keyword is not included in the utterance (when the content of the utterance is, for example, “Et” or “That is it”), the keyword combination changing unit 205 is activated. The keyword combination changing unit 205 determines whether or not there is a previous keyword combination (step S308).
If no keyword combination exists, the process proceeds to a standby state shown in step S307. If a keyword combination already exists, the keyword combination changing unit 205 changes the previously used keyword combination (step S309).
In the first embodiment, the combination is changed by deleting a part of the keyword. If the previous keyword combination includes, for example, three keywords “Internet”, “modem”, and “connection”, for example, the combination is changed to “Internet” and “modem” (step S309).

The document database search unit 206 searches the document database 215 using the changed keyword combination (step S310).
The document title presentation unit 207 displays the title of the document data obtained by the search on the display unit 210 (step S304).
When the document title is displayed, if the input means 209 inputs which document title content is to be displayed, the desired document title content (document data) is displayed on the display means 210 (step S306).

  When the document data is displayed on the display means 210, it again enters an input waiting state (step S307). Steps S308, S309, S310, S304, S305, and S306 are repeatedly executed as long as the utterance between the operator and the customer does not include the keyword. Each time the keyword combination changing unit 205 operates, the keyword combination is changed to, for example, “modem ”,“ Connection ”, etc., and search is executed under new conditions.

  Therefore, even if a speech recognition error is included in the combination by changing the keyword combination in this way, the keyword recognized by the speech recognition error is deleted because the keyword is deleted at a high rate. The search is executed under the search conditions and the probability is high, and the probability that the operator finds the necessary document can be increased.

In this embodiment, as proposed in claim 3, the keyword combination changing unit 205 changes the keyword combination to be executed, so that a new keyword is added to the already existing keyword combination. It is what.
For example, if the existing keyword combination is “Internet”, “modem”, “connection”, and “mail” is already extracted as a keyword, for example, the new keyword combination is “mail”. ”,“ Internet ”,“ modem ”, and“ connection ”, and the document database 215 is searched by a combination of these keywords.
According to this embodiment, since the number of keywords increases, the search condition becomes stricter and the narrowing down to the target document can be achieved.

In this embodiment, the keyword combination change unit 205 executes the change of the keyword combination as “a keyword by deleting a part of a keyword from an already existing keyword combination and adding a new keyword to the deleted keyword combination”. Change the combination of ".
Specifically, when the existing combination is “Internet”, “modem”, “connection”, for example, “Internet” is deleted and “mail” is added, and “mail”, “modem”, “ Search is executed with this combination as a new combination.
Accordingly, in this case as well, the rate of removing keywords obtained by voice recognition errors is increased, and the probability that correct search results are efficiently obtained is increased.

FIG. 4 shows the configuration of the document search devices 105-1 to 105-m proposed in the fourth embodiment of the present invention. In this embodiment, a synonym / synonym dictionary 216 is added to the keyword combination changing unit 205, and, for example, “mail” is replaced with “mail” as a synonym or “mail” using the synonym / synonym dictionary 216. , “Mail”, “Mail”, “MAIL”, and the like.
By replacing synonyms as in this embodiment or listing synonyms and synonyms, highly versatile and robust search performance can be obtained.

In the fifth embodiment of the present invention, when the keyword combination changing unit 205 changes the keyword combination, the selection of the keyword is selected based on the reliability score obtained by the speech recognition unit 201. That is, when the speech recognition unit 201 recognizes the input speech and converts it into a word in the form of text data, a large number of recognition candidate models are generated as speech recognition results. Each recognition candidate model is given a reliability score (likelihood), and the recognition candidate model with the largest sum of reliability scores is set as the most promising candidate, and the recognition candidate model is set as the most reliable speech recognition result. A word is determined by a recognition candidate model. Therefore, a reliability score is always given to a word obtained as a speech recognition result.
Therefore, by using a speech recognition result having a high reliability score as a keyword, a word with a low speech recognition error is used as a keyword, and the keyword used by the keyword combination changing unit 205 is selected according to the value of the reliability score. By selecting, the reliability of the combination of keywords can be improved.

That is, for example, when deleting a part of a keyword from a keyword combination, the reliability of the keyword combination can be improved by deleting the keyword having the lowest reliability score in the current combination. .
Also, even if you delete a part of a keyword from a keyword combination and then add a new keyword, the deleted keyword will have the lowest confidence score in the combination, As a keyword to be added, a keyword having the highest reliability score may be selected from the candidate addition keywords.

FIG. 5 shows a configuration example of the document search apparatuses 105-1 to 105-m proposed in the sixth embodiment.
In this embodiment, a timer 217 is added to the document title presentation unit 207 so that the search results for a plurality of times searched by the document database search unit 206 are sequentially displayed on the display unit 210 with a time interval set in 217 in the timer. Shown when configured.
By displaying the search results for a plurality of times with a certain amount of time as in this embodiment, the operator is provided with only the search results for each number of searches in the field of view, so that the search results can be easily viewed.

  When the above-described document retrieval apparatus according to the present invention is implemented by a computer, a document retrieval program that is described in a computer-readable program language and causes the computer to function as a document retrieval apparatus according to the operation procedure shown in FIG. Just install it. The document search program is stored in a recording medium such as a magnetic disk, a CD-ROM, or a semiconductor memory, and installed in the computer from the recording medium or through a communication line.

  The document retrieval apparatus according to the present invention is utilized as an operator support system such as a call center.

The block diagram for demonstrating the outline | summary of the suitable operator assistance system to which the document search apparatus by this invention is applied. The block diagram for demonstrating one Example of the document search device by this invention. 3 is a flowchart for explaining an operation procedure of the document search apparatus shown in FIG. The block diagram for demonstrating the other Example of the document search device by this invention. The block diagram for demonstrating the further another Example of the document search device by this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 101 Operator support system 209 Input means 102-1 to 102-m Telephone 210 Display means 103-1 to 103-m Terminal 211 Acoustic model 104-1 to 104-m Operator seat 212 Recognition dictionary 105-1 to 105-m Document search Device 213 Language model
106 Telephone line 214 Keyword dictionary
201 Speech recognition unit 215 Document database
202 Keyword Extraction / Storage Unit 216 Synonym / Synonym Dictionary
203 keyword extraction detection unit 217 timer
204 Keyword combination generator
205 Keyword combination change section
206 Document database search unit
207 Document title presentation section
208 Document presentation part

Claims (8)

A speech recognition unit that converts input speech into text data;
A keyword extraction / accumulation unit that extracts keywords used for document search from the text data obtained by the voice recognition unit and accumulates a keyword history;
A keyword combination generation unit for generating a combination of keywords from the keyword history;
A document database search unit that performs a document search in the document database using the generated keyword combination;
Presenting the document title obtained from the search result of the document database search unit, and displaying the document title on the display means;
Input means for designating that the document content represented by the document title displayed on the display means is displayed on the display means;
A document presentation unit that reads the content of the document title designated by the input unit from the document database and displays the content on the display unit;
A keyword extraction detection unit for detecting a situation in which no text is included in the text data obtained in the voice recognition result of the voice recognition unit;
The keyword extraction detection unit is activated by detecting a situation that does not include a keyword, changes the keyword combination generated by the keyword combination generation unit, and performs a document search to the document database search unit according to the changed keyword combination A keyword combination change part to be
A document search apparatus characterized by comprising:   2. The document search apparatus according to claim 1, wherein the keyword combination changing unit changes the keyword combination by deleting a part of the keyword from the keyword combination.   2. The document search apparatus according to claim 1, wherein the keyword combination changing unit changes the keyword combination by adding a new keyword from the keyword history to the keyword combination.   2. The document search apparatus according to claim 1, wherein the keyword combination changing unit deletes a part of the keywords from the keyword combination, and adds a new keyword from the keyword history to the deleted keyword combination. A document retrieval apparatus characterized by   The document search apparatus according to claim 1, wherein the keyword combination changing unit changes a keyword combination by converting a keyword in the keyword combination into a synonym or a synonym.   2. The document search apparatus according to claim 1, wherein the keyword combination changing unit changes a keyword combination by adding a keyword represented by a keyword synonym or synonym in the keyword combination.   7. The document search apparatus according to claim 1, wherein selection of a keyword to be changed by the keyword combination changing unit is selected based on a reliability score obtained for each keyword in the voice recognition process. Document retrieval device.   8. The document search apparatus according to claim 1, wherein a search result obtained by performing a plurality of document searches in addition to the combination change is displayed at intervals. apparatus.

Images