JPWO2019167281A1 - Response processing program, response processing method, response processing device and response processing system - Google Patents

Abstract

When registering the question data (Q), the response processing device (101) morphologically analyzes the question data (Q) and extracts words included in the question data (Q). The response processing device (101) presents a candidate word to be set as a synonym for the extracted word. When the response processing device (101) accepts the selection of one of the presented candidates as the target word to be set as the synonym of the extracted word, it is included in the question data (Q) and the question data (Q). Learn so that both the question data (Q') obtained by replacing the word with the selected word are associated with the same answer data (A).

Description

The present invention relates to a response processing program, a response processing method, a response processing device, and a response processing system.

In recent years, an increasing number of call centers have introduced FAQ (freaky Asked Question) systems that use a technology called chatbots. A chatbot is a computer system that uses artificial intelligence to interact with humans and exchange messages.

In addition, a synonym for a word included in a question registered as FAQ is registered in a dictionary, and when a question including that word (synonym) is input, the word (synonym) is replaced with the corresponding word. , May search for answers. The synonym dictionary is automatically generated, for example, by inputting the past dialogue history, the corresponding manual, etc. as learning data into the learning engine.

Prior art is to receive question data, if the answer to the question is in the database section, output the answer from the output device, if there is no answer, search the question content of the unsolved data, and multiple items contained therein Some keywords are added to a list and accumulated.

Japanese Unexamined Patent Publication No. 2001-125901

However, in the prior art, there is a problem that it is difficult to register an appropriate synonym for a word included in the question data registered as FAQ.

In one aspect, the present invention aims to assist in the registration of appropriate synonyms when registering question data.

In one embodiment, when registering question data, the question data is morphologically analyzed, words included in the question data are extracted, and candidate words to be set as synonyms for the words are presented. When the selection of any of the candidates is accepted as the target word to be set as the synonym of the word, the question data and the word included in the question data are replaced with the selected word. A response processing program is provided that learns that both the question data and the question data are related to the same answer data.

According to one aspect of the present invention, it is possible to support the registration of appropriate synonyms when registering question data.

FIG. 1 is an explanatory diagram showing an embodiment of the response processing device 101 according to the embodiment. FIG. 2 is an explanatory diagram showing a system configuration example of the response processing system 200. FIG. 3 is a block diagram showing a hardware configuration example of the response processing device 101. FIG. 4 is an explanatory diagram showing an example of the stored contents of the FAQ master 220. FIG. 5 is an explanatory diagram showing an example of the stored contents of the synonym evaluation table 230. FIG. 6 is an explanatory diagram showing an example of the stored contents of the chat log DB 240. FIG. 7 is a block diagram showing a functional configuration example of the response processing device 101. FIG. 8 is an explanatory diagram showing a screen example of the topic selection screen. FIG. 9 is an explanatory diagram showing a screen example of the phrase selection screen. FIG. 10 is an explanatory diagram showing an example of the correspondence between the topic and the phrase. FIG. 11 is an explanatory diagram (No. 1) showing another screen example of the topic selection screen. FIG. 12 is an explanatory diagram (No. 2) showing another screen example of the topic selection screen. FIG. 13 is a flowchart (No. 1) showing an example of the information processing procedure of the response processing device 101. FIG. 14 is a flowchart (No. 2) showing an example of the information processing procedure of the response processing device 101.

Hereinafter, embodiments of a response processing program, a response processing method, a response processing apparatus, and a response processing system according to the present invention will be described in detail with reference to the drawings.

(Embodiment)
FIG. 1 is an explanatory diagram showing an embodiment of the response processing device 101 according to the embodiment. In FIG. 1, the response processing device 101 is a computer that supports registration of synonyms for words included in question data. The question data is, for example, question data registered as FAQ.

The FAQ is a "frequently asked question", and is a collection of question data and answer data for a question expected in advance. The question data is for asking how to solve some problem. For example, question data is a question about a product or service. The question data may be represented by a word or a combination of words, or may be represented by one or more sentences.

Here, although the question data is registered, if the question data in which the words included in the question data are replaced with synonyms is input, if the synonyms are not registered, it cannot be judged as the same or similar question and the answer is answered. It may not be possible to present it. On the other hand, if knowledge data on synonyms is applied unconditionally, there is a problem that unexpected question data is judged to be the same or the same type of question and an incorrect answer is presented.

Therefore, it is important to confirm whether or not the synonyms registered in the dictionary are appropriate words. For example, a synonym dictionary is automatically generated by inputting a past dialogue history, a corresponding manual, etc. as learning data into a learning engine. At this time, if a large number of synonyms are generated, the load on maintenance work such as confirmation increases. When the load on maintenance work increases, synonym dictionaries containing inappropriate synonyms may be left unattended and become obsolete.

Therefore, in the present embodiment, when registering the question data, the response processing device 101 that supports the registration of appropriate synonyms for the words included in the question data will be described. Hereinafter, a processing example of the response processing device 101 will be described.

(1) When registering the question data Q, the response processing device 101 morphologically analyzes the question data Q and extracts words included in the question data Q. Here, the question data Q is, for example, question data registered together with the answer data as FAQ. Morphological analysis is a natural language processing by a computer, for example, a process of dividing a sentence into units of morphemes and discriminating part of speech.

In the example of FIG. 1, it is assumed that the question data Q is "Tell me about the computer" and "computer" is extracted as the word included in the question data Q.

(2) The response processing device 101 presents a candidate word to be set as a synonym for the extracted word. Here, the target word that is a candidate to be set as a synonym is automatically generated by inputting, for example, a past dialogue history (FAQ) in the call center, a corresponding manual, or the like as learning data into the learning engine.

In the example of FIG. 1, candidates of "personal computer", "computer" and "software" are presented on the display 110 as candidates of the target word to be set as a synonym of the word "computer". The display 110 may be possessed by the response processing device 101, or may be possessed by another computer connected to the response processing device 101.

(3) When the response processing device 101 accepts the selection of one of the presented candidates as the target word to be set as the synonym of the extracted word, both the question data Q and the question data Q'are displayed. Learn to relate to the same answer data A. The question data Q'is question data obtained by replacing the words included in the question data Q with the selected words.

Here, learning so that both the question data Q and the question data Q'are related to the same answer data A means that, for example, the selected word is used as a synonym for the word extracted from the question data Q, such as a synonym dictionary. Is to register with. As a result, when question data including this synonym is input thereafter, it is possible to search by the question data Q in which the synonym is replaced with the corresponding word, and the hit rate of the FAQ system can be increased. The hit rate is an index value indicating the rate at which the answer requested by the user can be presented in response to the question from the user input to the FAQ system.

In addition, as another example of learning, when generating a synonym dictionary using a learning model, a condition is set so that the selected word is extracted as a synonym of the word extracted from the question data Q. You may. As a result, when the next synonym dictionary is generated, it becomes possible to generate a synonym dictionary in which an appropriate synonym is associated with a word included in the question data Q registered as FAQ.

Further, as another example of learning, the question data Q'obtained by replacing the word included in the question data Q with the selected word may be registered as the FAQ together with the answer data A. As a result, when the question data Q'is input, the answer data A associated with the question data Q can be output.

In the example of FIG. 1, among the candidates "personal computer", "computer" and "software" presented on the display 110, "personal computer" is selected as the target word to be set as a synonym for the word "computer". Suppose. In this case, the response processing device 101 registers, for example, the selected word "personal computer" in the synonym dictionary 120 as a synonym for the word "computer" extracted from the question data Q. As a result, when question data including the synonym "PC" is input, it is possible to search by the question data in which the synonym "PC" is replaced with the corresponding word "computer", and the hit rate can be increased. ..

As described above, according to the response processing device 101, when registering the question data Q, it is possible to selectively present the candidate of the target word to be set as the synonym of the word included in the question data Q. Then, according to the response processing device 101, when any of the presented candidates is selected, both the question data Q and the question data Q'are learned to be associated with the same answer data A. can do.

As a result, when registering the question data Q, the synonyms of the words included in the question data Q can be registered by simply performing a simple operation of selecting one of the candidate words to be set as the synonyms. can do. Therefore, the work load required for registering synonyms can be reduced, and it is possible to support the registration of appropriate synonyms for the words included in the question data Q.

(System configuration example of response processing system 200)
Next, a system configuration example of the response processing system 200 including the response processing device 101 shown in FIG. 1 will be described. The response processing system 200 is applied to, for example, an FAQ system using a chatbot.

FIG. 2 is an explanatory diagram showing a system configuration example of the response processing system 200. In FIG. 2, the response processing system 200 includes a response processing device 101 and a worker terminal 201. In the response processing system 200, the response processing device 101 and the worker terminal 201 are connected via a wired or wireless network 210. The network 210 is, for example, a LAN (Local Area Network), a WAN (Wide Area Network), the Internet, or the like.

The response processing device 101 has an FAQ master 220, a synonym evaluation table 230, and a chat log DB (Database) 240. The stored contents of 220, 230, 240 such as various DBs will be described later with reference to FIGS. 4 to 6. The response processing device 101 is, for example, a server.

Further, the response processing device 101 may output answer candidates corresponding to the questions input to the response processing system 200 (FAQ system). Answer candidates corresponding to the question are searched using, for example, an existing search algorithm adopted in the FAQ system or the FAQ site.

Specifically, for example, the response processing device 101 analyzes the question input by the user of the response processing system 200 and decomposes it into morphemes. Next, the response processing device 101 refers to the FAQ master 220 and searches for the FAQ corresponding to the decomposed morpheme according to a predetermined search condition. Then, the response processing device 101 presents the searched search result (FAQ) as an answer candidate corresponding to the input question.

However, in the response processing system 200, the process of accepting the input of the question from the user and searching / outputting the answer candidate corresponding to the input question is executed by another computer different from the response processing device 101. It may be. Further, the FAQ master 220 and the chat log DB 240 may be owned by another computer (for example, a database server) accessible to the response processing device 101.

The worker terminal 201 is a computer used by the worker. The worker is a person who performs maintenance work on the synonym dictionary, and is, for example, an administrator of the response processing system 200. The synonym dictionary is a dictionary in which synonyms of words included in the question data Q registered as FAQ are registered. The worker terminal 201 is, for example, a PC (Personal Computer), a tablet-type PC, or the like.

(Hardware configuration example of response processing device 101)
FIG. 3 is a block diagram showing a hardware configuration example of the response processing device 101. In FIG. 3, the response processing device 101 includes a CPU (Central Processing Unit) 301, a memory 302, an I / F (Interface) 303, a disk drive 304, and a disk 305. Further, each component is connected by a bus 300.

Here, the CPU 301 controls the entire response processing device 101. The memory 302 includes, for example, a ROM (Read Only Memory), a RAM (Random Access Memory), a flash ROM, and the like. Specifically, for example, the flash ROM stores the OS (Operating System) program, the ROM stores the application program, and the RAM is used as the work area of the CPU 301. The program stored in the memory 302 is loaded into the CPU 301 to cause the CPU 301 to execute the coded process.

The I / F 303 is connected to the network 210 through a communication line, and is connected to an external computer (for example, the worker terminal 201 shown in FIG. 2) via the network 210. The I / F 303 controls the interface between the network 210 and the inside of the device, and controls the input / output of data from an external computer. For the I / F 303, for example, a modem, a LAN adapter, or the like can be adopted.

The disk drive 304 controls data read / write to the disk 305 according to the control of the CPU 301. The disk 305 stores the data written under the control of the disk drive 304. Examples of the disk 305 include a magnetic disk and an optical disk.

The response processing device 101 may include, for example, an SSD (Solid State Drive), an input device, a display, or the like, in addition to the above-mentioned components. Further, the worker terminal 201 shown in FIG. 2 can also be realized by the same hardware configuration as the response processing device 101. However, the worker terminal 201 has an input device, a display, and the like in addition to the above-mentioned components.

(Memory contents of FAQ master 220)
Next, the stored contents of the FAQ master 220 included in the response processing device 101 will be described. The FAQ master 220 is realized by, for example, a storage device such as the memory 302 and the disk 305 shown in FIG.

FIG. 4 is an explanatory diagram showing an example of the stored contents of the FAQ master 220. In FIG. 4, the FAQ master 220 has fields for FAQ-ID, title, question, and answer, and by setting information in each field, FAQ information (for example, FAQ information 400-1, 400-2) can be obtained. Store as a record.

Here, the FAQ-ID is an identifier that identifies the FAQ. The title is a representative question among multiple questions. Multiple questions that have similar meanings are given the same title. Then, the same FAQ-ID is given to the FAQs having the same title. The question is question data Q indicating the question forming the FAQ. The answer is answer data A indicating the answer to the question. That is, one FAQ is a collection of n questions and m answers (n and m are natural numbers of 1 or more).

For example, FAQ information 400-1 indicates the title of FAQ1, "Tell me about your computer", the question "Tell me about your computer", and the answer "It's a notebook-type personal computer."

(Memory contents of synonym evaluation table 230)
Next, the stored contents of the synonym evaluation table 230 included in the response processing device 101 will be described. The synonym evaluation table 230 is realized by, for example, a storage device such as the memory 302 and the disk 305 shown in FIG.

FIG. 5 is an explanatory diagram showing an example of the stored contents of the synonym evaluation table 230. In FIG. 5, the synonym evaluation table 230 stores synonym evaluation information (for example, synonym evaluation information 500-1) representing a question, a topic, a phrase, and an acceptance / rejection flag in association with each other.

Here, the question is the question data Q registered as FAQ. The topic is a word included in the question data Q. A phrase is a candidate for a word to be set as a synonym for a topic. The acceptance / rejection flag is a flag indicating whether or not to adopt a phrase as a target word to be set as a synonym of a topic. The acceptance / rejection flag "1" indicates that the phrase is adopted. The acceptance / rejection flag "0" indicates that the phrase is not adopted. The acceptance / rejection flag is "0" in the initial state.

For example, the synonym evaluation information 500-1 indicates the acceptance / rejection flag of each phrase "personal computer, computer, software" for the topic "computer" included in the question "Tell me about the computer." Here, the acceptance / rejection flag of each phrase "personal computer, computer, software" is in the initial state "0".

(Memory contents of chat log DB240)
Next, the stored contents of the chat log DB 240 included in the response processing device 101 will be described. The chat log DB 240 is realized by, for example, a storage device such as the memory 302 and the disk 305 shown in FIG.

FIG. 6 is an explanatory diagram showing an example of the stored contents of the chat log DB 240. In FIG. 6, the chat log DB 240 has fields of an input question, FAQ-ID, and FAQ-LIST, and by setting information in each field, chat log information (for example, chat log information 600-1,600-) 2) is stored as a record.

Here, the input question is question data input to the response processing system 200 (FAQ system). The FAQ-ID is the FAQ-ID of the answer candidate (answer data) selected from the answer candidates presented for the input question. If no answer candidate is presented for the input question, "-(Null)" is set in the FAQ-ID field. If none of the answer candidates are selected even though the answer candidates are presented, "Not Found" is set in the FAQ-ID field.

The FAQ-LIST is a list of candidate answers presented for input questions. If no answer candidate is presented for the input question, "Not Found" is set in the FAQ-LIST field. The FAQ-LIST "Not Found" indicates that there is no FAQ corresponding to the input question. The FAQ-ID "Not Found" indicates that the FAQ expected by the user was not among the answer candidates presented for the input question.

For example, the chat log information 600-1 indicates that no answer candidate was presented to the input question "Tell me about the computer." In addition, the chat log information 600-2 indicates that none of the answer candidates (FAQ100, FAQ101, FAQ102) presented to the input question "Please tell me the specifications" was selected.

(Example of functional configuration of response processing device 101)
FIG. 7 is a block diagram showing a functional configuration example of the response processing device 101. In FIG. 7, the response processing device 101 includes an acquisition unit 701, an extraction unit 702, a display control unit 703, a reception unit 704, and a learning unit 705. The acquisition unit 701 to the learning unit 705 are functions that serve as control units. Specifically, for example, by causing the CPU 301 to execute a program stored in a storage device such as the memory 302 and the disk 305 shown in FIG. Alternatively, the function is realized by the I / F 303. The processing result of each functional unit is stored in a storage device such as a memory 302 or a disk 305, for example.

The acquisition unit 701 acquires the question data Q to be registered. The question data Q is, for example, question data registered as FAQ in the FAQ master 220 shown in FIG. The question data Q may be newly generated question data, or may be question data registered in the FAQ master 220 in the past.

Specifically, for example, the acquisition unit 701 acquires the question data Q to be registered by receiving one or more question data Q from the worker terminal 201. Further, the acquisition unit 701 may acquire the question data Q to be registered by the operation input of the user using an input device (not shown). The acquired question data Q is registered in the FAQ master 220 together with the answer data A as FAQ, for example.

The extraction unit 702 morphologically analyzes the acquired question data Q and extracts words (topics) included in the question data Q. Specifically, for example, the extraction unit 702 divides the acquired question data Q into units of morphemes to determine part of speech, and extracts words included in the question data Q as topics.

In addition, the extraction unit 702 specifies a candidate word (phrase) to be set as a synonym for the extracted word (topic). Specifically, for example, the extraction unit 702 refers to a synonym dictionary and identifies a phrase that is a candidate for a target word to be set as a synonym for the extracted topic. A synonym dictionary is information that indicates the correspondence between a topic and a phrase.

However, the synonym dictionary referred to here is a synonym dictionary to be maintained, and is, for example, a temporary synonym dictionary generated by existing technology. The synonym dictionary may be generated in the response processing device 101, or may be generated in another computer different from the response processing device 101.

For example, a synonym dictionary is generated by inputting a past dialogue history (FAQ) in a call center, a corresponding manual, or the like as learning data into a learning engine. Specifically, for example, the response processing device 101 performs natural language processing (morphological analysis, parsing, semantic analysis, etc.) on the training data and extracts the correspondence between the topic and the phrase to obtain a synonym dictionary. It may be generated.

Synonyms may also be presented based on search results within a given site for a topic. The predetermined site is, for example, a predetermined search site or an encyclopedia site. Specifically, for example, the response processing device 101 extracts a word by performing natural language processing using HTML (Hyper Text Markup Language) data of a web page obtained by using a topic as a search word in a predetermined site as learning data. To do.

At this time, the response processing device 101 may use all the data (for example, HTML data of all pages) registered in a predetermined site (for example, an encyclopedia site) as learning data. Then, the response processing device 101 may use an existing classifier to classify the extracted words to extract the correspondence between the topic and the phrase and generate a synonym dictionary.

The word (topic) and the candidate word (phrase) to be set as a synonym of the word (topic) are, for example, associated with the question data Q including the word (topic), and the synonym evaluation table shown in FIG. It is stored in 230. However, at this point, the acceptance / rejection flag is "0".

If there is no candidate word (phrase) to be set as a synonym for a word (topic), the word (topic) does not have to be registered in the synonym evaluation table 230. When a word (topic) for which no candidate (phrase) does not exist is registered in the synonym evaluation table 230, each field of the phrase and the acceptance / rejection flag of the word (topic) becomes "-".

The display control unit 703 presents a candidate word (phrase) to be set as a synonym for the extracted word (topic). Specifically, for example, first, the display control unit 703 displays the topic selection screen with reference to the synonym evaluation table 230. The topic selection screen is an operation screen for displaying a topic extracted from one or a plurality of question data Qs by the extraction unit 702 so as to be selectable. A screen example of the topic selection screen will be described later with reference to FIG.

Next, when the display control unit 703 accepts the selection of any topic on the topic selection screen, the display control unit 703 displays the phrase selection screen for the selected topic with reference to the synonym evaluation table 230. The phrase selection screen is an operation screen that is specified by the extraction unit 702 and displays a phrase that is a candidate for a target word to be set as a synonym of the topic so as to be selectable. A screen example of the phrase selection screen will be described later with reference to FIG.

As a result, it is possible to selectively present a phrase that is a candidate for a target word to be set as a synonym for the topic included in the question data Q. The display of the display destination of the various screens may be, for example, the display of the worker terminal 201 (not shown) or the display of the own device (not shown).

Further, the display control unit 703 may display the similarity of each phrase when displaying a phrase that is a candidate of a target word to be set as a synonym of the topic in a selectable manner. Here, the similarity of the phrase is an index value indicating the similarity with the topic. The similarity of each phrase can be obtained, for example, by using an existing technique when generating a synonym dictionary using a learning engine. Further, when a plurality of topic phrases exist, the display control unit 703 may display the plurality of phrases in descending order of similarity. Further, the display control unit 703 may display only phrases having a similarity equal to or higher than a preset threshold value among a plurality of phrases.

Further, the display control unit 703 sets each of the plurality of words (topics) as a synonym when the question data Q contains a plurality of words (topics) as a result of morphological analysis of the question data Q by the extraction unit 702. It may be presented so that the existence or number of candidates (phrases) of the target word can be known.

Specifically, for example, the display control unit 703 specifies the presence / absence or number of phrases with reference to the synonym evaluation table 230 for the topics included in each of the one or a plurality of question data Qs. Then, the display control unit 703 displays the topics extracted from one or a plurality of question data Qs in a selectable manner on the topic selection screen, and displays the presence / absence or the number of phrases specified for each topic. A screen example of the topic selection screen for displaying the presence / absence or number of phrases for each topic will be described later with reference to FIG.

Further, the display control unit 703 may control the display mode according to the priority of the topic when displaying the topic extracted by the extraction unit 702. The priority of a topic can be set by any method, and the topic is set so that it is desirable to perform maintenance of synonyms preferentially.

Specifically, for example, the display control unit 703 may highlight a high-priority topic on the topic selection screen. A high-priority topic is, for example, a topic whose priority is equal to or higher than a preset threshold value. Highlighting a topic means, for example, displaying it in a color, background color, icon, font, etc. that is different from other topics.

Further, the display control unit 703 may give a message on the topic selection screen prompting the topic with high priority to be preferentially maintained for synonyms. Further, the display control unit 703 may display a topic having a high priority at a higher level on the topic selection screen.

An example of setting the priority of the topic will be described later with reference to FIG.

The reception unit 704 accepts the selection of one of the presented candidates (phrases) as a target word to be set as a synonym for the word (topic). The phrase is selected, for example, on the phrase selection screen (for example, the phrase selection screen 900 shown in FIG. 9 described later).

Specifically, for example, the reception unit 704 receives a phrase selection from the worker terminal 201 by receiving a selection result indicating the phrase selected on the phrase selection screen. This makes it possible to specify the target word (phrase) to be set as a synonym for the topic.

When the learning unit 705 accepts the selection of a word (phrase), both the question data Q and the question data Q'obtained by replacing the word (topic) included in the question data Q with the selected word (phrase). However, it learns to be related to the same answer data A.

Specifically, for example, the learning unit 705 refers to the synonym evaluation table 230 and combines the question data Q (question data Q including the extracted topic), the extracted topic, and the selected phrase. Identify the synonym evaluation information corresponding to. Next, the learning unit 705 sets "1" to the acceptance / rejection flag of the phrase of the specified synonym evaluation information.

Then, the learning unit 705 refers to the synonym evaluation table 230 and generates a synonym dictionary showing the correspondence between the topic of the acceptance / rejection flag “1” and the phrase. At this time, the learning unit 705 may regenerate a new synonym dictionary, or may modify the synonym dictionary (provisional) referred to when specifying the phrase of each topic.

As a result, after that, when the question data Q'including the phrase is input to the response processing system 200, the generated synonym dictionary can be used to search by the question data Q in which the phrase is replaced with a topic. , The hit rate can be increased.

In addition, for example, when the learning unit 705 inputs the learning data into the learning model and generates a synonym dictionary, the phrase of the acceptance / rejection flag “1”, that is, the selected phrase is extracted as a synonym of the topic. The condition may be set to. As a result, when the next synonym dictionary is generated, it is possible to generate a synonym dictionary in which an appropriate phrase is associated with the topic included in the question data Q registered as FAQ.

Further, when the learning unit 705 detects the input of the question data Q'' in which the topic is replaced with a specific phrase in the question data Q after learning, both the question data Q and the question data Q'' have the same answer. It may be learned so as to be related to the data A. Here, the specific phrase is a phrase that is not selected as a target word to be set as a synonym of the topic among the candidate words to be set as a synonym of the topic.

Specifically, for example, when the learning unit 705 detects the input of the question data Q'', it refers to the synonym evaluation table 230 and replaces a specific phrase in the question data Q'' with a topic. And identify synonym evaluation information corresponding to the combination of the topic and a specific phrase. Next, the learning unit 705 sets "1" to the acceptance / rejection flag of the specific phrase of the specified synonym evaluation information. Then, the learning unit 705 modifies the synonym dictionary with reference to the synonym evaluation table 230.

As a result, after that, when the question data Q'' including the phrase is input to the response processing system 200, the search by the question data Q in which the specific phrase is replaced with the topic is performed by using the modified synonym dictionary. It becomes possible and the hit rate can be increased.

The input of the question data Q'' is, for example, whether or not the input question stored in the chat log DB 240 includes the question data Q'' in which the topic is replaced with a specific phrase in the question data Q. It can be detected by making a judgment.

Further, when the learning unit 705 detects the input of the question data Q'' in which the topic is replaced with a specific phrase in the question data Q after learning, it suggests the registration of the topic of the specific phrase as a synonym. May be good. Specifically, for example, when the learning unit 705 detects the input of the question data Q'', the learning unit 705 may send a message recommending registration as a synonym of the topic of the specific phrase to the worker terminal 201. .. Thereby, the administrator of the response processing system 200 can notice the existence of a specific phrase that should be registered as a synonym for the topic.

It should be noted that each functional unit of the response processing device 101 may be realized by another computer in the response processing system 200, for example, the worker terminal 201. Further, each functional unit of the response processing device 101 may be realized by a plurality of computers in the response processing system 200.

(Screen example of topic selection screen)
Next, a screen example of the topic selection screen will be described with reference to FIG. The topic selection screen is displayed on the display (not shown) of the worker terminal 201 based on the stored contents of the synonym evaluation table 230 by, for example, the display control unit 703 of the response processing device 101.

FIG. 8 is an explanatory diagram showing a screen example of the topic selection screen. In FIG. 8, the topic selection screen 800 is an operation screen for selectively displaying topics (for example, topics tp1 to tp10) extracted from the question data Q registered as FAQ. However, in FIG. 8, a part of the topics extracted from the question data Q is extracted and displayed.

On the topic selection screen 800, when any topic is selected by the user's operation input using the input device (not shown) of the worker terminal 201, the phrase selection screen for the selected topic can be displayed. According to the topic selection screen 800, it is possible to select a topic for which maintenance work of synonyms is performed.

Further, on the topic selection screen 800, when the end button 801 is selected by the user's operation input, the topic selection can be completed. Specifically, for example, when the end button 801 is selected, a topic selection end notification is transmitted from the worker terminal 201 to the response processing device 101, and the display of the topic selection screen 800 ends.

In the example of FIG. 8, it is assumed that the topic tp1 is selected. The topic tp1 is the topic "computer" of the synonym evaluation information 500-1 (see FIG. 5) in the synonym evaluation table 230. In this case, the selection result indicating the topic tp1 is transmitted from the worker terminal 201 to the response processing device 101. As a result, the display control unit 703 of the response processing device 101 displays the phrase selection screen 900 for the topic "computer" on the worker terminal 201 as shown in FIG.

(Screen example of phrase selection screen)
Here, a screen example of the phrase selection screen will be described with reference to FIG.

FIG. 9 is an explanatory diagram showing a screen example of the phrase selection screen. In FIG. 9, the phrase selection screen 900 is an operation screen for selectively displaying the adoption / rejection of the phrases “personal computer”, “computer”, and “software” which are candidates for the target word to be set as a synonym for the topic “computer”. ..

On the phrase selection screen 900, the adoption / rejection of the phrase "personal computer" can be selected by selecting the adoption button b1 or the rejection button b2 by the user's operation input. Further, by selecting the adopt button b3 or the non-adopt button b4, it is possible to select whether or not to adopt the phrase "computer". Further, by selecting the adopt button b5 or the non-adopt button b6, it is possible to select whether or not to adopt the phrase "software".

In the initial state, the rejected buttons b2, b4, and b6 are selected for the phrases "personal computer", "computer", and "software".

Further, on the phrase selection screen 900, when the enter button 901 is selected by the user's operation input, the selection of acceptance / rejection of the phrase can be completed. Specifically, for example, when the decision button 901 is selected, the worker terminal 201 transmits a selection result indicating the phrase adopted in the response processing device 101, and the display of the phrase selection screen 900 ends.

For example, when the enter button 901 is selected while the adopt button b1 and the non-adopt buttons b4 and b6 are selected, the selection result indicating the adopted phrase "personal computer" is transmitted from the worker terminal 201 to the response processing device 101. Will be sent. As a result, the reception unit 704 of the response processing device 101 accepts the selection of the phrase "personal computer" to be set as a synonym for the topic "computer".

In this case, the learning unit 705 of the response processing device 101 refers to the synonym evaluation table 230 and sets "1" in the acceptance / rejection flag of the phrase "personal computer" of the synonym evaluation information 500-1. As a result, the phrase "personal computer" is set as a synonym for the topic "computer", and a synonym dictionary showing the correspondence between the topic "computer" and the phrase "personal computer" is generated.

That is, the learning department 705 replaces the question data Q "Tell me about the computer" and the topic "Computer" included in the question data Q with the phrase "PC", and the question data Q'"Tell me about the personal computer". Both of them are learned to be related to the same answer data A.

After learning, the question data Q'' "Tell me about the computer." Is entered by replacing the topic "Computer" with the specific phrase "Calculator" in the question data Q "Tell me about the computer." Suppose it is detected. The specific phrase "calculator" is a phrase that was not selected as a target word to be set as a synonym for the topic "computer".

In this case, the learning unit 705 sets "1" in the acceptance / rejection flag of the phrase "computer" of the synonym evaluation information 500-1 with reference to the synonym evaluation table 230. As a result, the phrase "computer" is set as a synonym for the topic "computer", and the synonym dictionary is modified to show the correspondence between the topic "computer" and the phrase "computer".

That is, the learning unit 705 learns so that both the question data Q "Tell me about the computer" and the question data Q "" Tell me about the computer "are related to the same answer data A.

(Example of setting topic priority)
Next, an example of setting the priority of the topic will be described with reference to FIG.

FIG. 10 is an explanatory diagram showing an example of the correspondence between the topic and the phrase. For example, the display control unit 703 sets a higher priority than other topics for a plurality of topics sharing the same phrase with reference to the synonym evaluation table 230. At this time, the display control unit 703 may set a higher priority as the topic has more phrases shared with other topics.

For example, as shown in (10-1) of FIG. 10, it is assumed that the phrase "computer" has "personal computer" and "computer", and the phrase "terminal" has "personal computer" and "computer". .. In this case, the topic "computer" and the topic "terminal" share the phrases "computer" and "computer". Therefore, the display control unit 703 may raise the priority of the topic "computer" and the topic "terminal" by two levels.

Further, the display control unit 703 refers to the synonym evaluation table 230, and when the relationship that one topic is the same as the word used in the phrase of the other topic is established between the topics, both sides You may want to set a high priority for the topic.

For example, as shown in (10-2) of FIG. 10, it is assumed that the phrase "computer" has "computer" and "computer", and the phrase "computer" has "computer" and "computer". .. In this case, the topic "computer" is the same as the word "computer" used in the phrases of other topics "computers". In addition, the topic "personal computer" is the same as the word "personal computer" used in the phrases of other topics "computer". Therefore, the display control unit 703 sets a high priority for both the topic "computer" and the topic "personal computer".

Further, as a method of setting the priority of the topic based on the information obtained during the operation of the response processing system 200, for example, there are the following.

After learning, the display control unit 703 refers to the chat log DB 240 for each phrase of the unselected topic that is not selected on the topic selection screen, and includes it in the question data (input question) input to the response processing system 200. It may be decided to count the number of times. Then, the display control unit 703 may consider a topic having a phrase having a certain number of times counted (for example, 5 or more) as a high-priority topic. As a result, it is possible to set a high priority for a topic having a phrase having a certain number of times or more appearing in the question data input during the operation of the response processing system 200.

(Examples of other screens on the topic selection screen)
Next, a screen example of a topic selection screen for displaying the presence / absence or number of phrases for each topic will be described with reference to FIG.

FIG. 11 is an explanatory diagram (No. 1) showing another screen example of the topic selection screen. In FIG. 11, the topic selection screen 1100 is an operation screen for selectively displaying topics (for example, topics tp1 to tp10) extracted from the question data Q registered as FAQ. However, in FIG. 10, a part of the topics extracted from the question data Q is extracted and displayed.

On the topic selection screen 1100, for each topic (for example, topics tp1 to tp10), the presence or absence or number of phrases is displayed together with each topic. For example, for topic tp1, the number of phrases "3" is displayed. Further, regarding the topic tp8, "0" indicating that there is no phrase is displayed.

According to the topic selection screen 1100, when selecting a topic for which maintenance work of synonyms is performed, it is possible to determine a topic for which maintenance of synonyms should be performed preferentially based on the presence or absence of phrases and the number of phrases for each topic. it can.

Next, a screen example of the topic selection screen for highlighting high-priority topics will be described with reference to FIG.

FIG. 12 is an explanatory diagram (No. 2) showing another screen example of the topic selection screen. In FIG. 12, the topic selection screen 1200 is an operation screen for selectively displaying topics (for example, topics tp1 to tp10) extracted from the question data Q registered as FAQ. However, in FIG. 12, a part of the topics extracted from the question data Q is extracted and displayed.

On the topic selection screen 1200, high-priority topics tp1 and tp9 are highlighted. Specifically, the background color of the buttons representing the topics tp1 and tp9 is changed to a background color different from the buttons representing other topics.

According to the topic selection screen 1200, when selecting a topic for which maintenance work of synonyms is performed, it is possible to determine a topic for which maintenance of synonyms should be performed preferentially from the difference in the background color of the buttons representing each topic. ..

(Information processing procedure of response processing device 101)
Next, the information processing procedure of the response processing device 101 will be described with reference to FIGS. 13 and 14.

13 and 14 are flowcharts showing an example of the information processing procedure of the response processing device 101. In the flowchart of FIG. 13, first, the response processing device 101 acquires the learning data (step S1301). The learning data is, for example, a past dialogue history (FAQ) in a call center, a corresponding manual, or the like.

Then, the response processing device 101 inputs the acquired learning data into the existing learning engine to generate a synonym dictionary (provisional) indicating the correspondence between the topic and the phrase (step S1302). Next, the response processing device 101 acquires the question data Q to be registered (step S1303).

Then, the response processing device 101 morphologically analyzes each of the acquired question data Q and extracts a topic included in each question data Q (step S1304). Next, the response processing device 101 refers to the generated synonym dictionary (provisional) and identifies a phrase that is a candidate for the target word to be set as the synonym of each extracted topic (step S1305).

Then, the response processing device 101 registers the extracted topic and the specified phrase in the synonym evaluation table 230 (step S1306) in association with each question data Q, and proceeds to step S1401 shown in FIG. However, at this point, the acceptance / rejection flag is "0".

In the flowchart of FIG. 14, first, the response processing device 101 refers to the synonym evaluation table 230 and displays the topic selection screen (step S1401). Next, the response processing device 101 determines whether or not the selection of the topic has been accepted (step S1402).

Here, if the topic selection is not accepted (step S1402: No), the response processing device 101 proceeds to step S1407. On the other hand, when the topic selection is accepted (step S1402: Yes), the response processing device 101 refers to the synonym evaluation table 230 and displays the phrase selection screen for the selected topic (step S1403). However, if there is no phrase about the topic, the response processing device 101 proceeds to step S1407.

Next, the response processing device 101 determines whether or not the phrase selection has been accepted (step S1404). Here, the response processing device 101 waits for the phrase selection to be accepted (step S1404: No). However, if no phrase is selected, that is, if no phrase is adopted as a synonym, the response processing device 101 proceeds to step S1407.

Then, when the response processing device 101 accepts the selection of the phrase (step S1404: Yes), the response processing device 101 refers to the synonym evaluation table 230, and the synonym evaluation information corresponding to the combination of the selected topic and the selected phrase. Is specified (step S1405).

Next, the response processing device 101 sets “1” in the acceptance / rejection flag of the specified synonym evaluation information (step S1406). Then, the response processing device 101 determines whether or not to finish selecting the topic (step S1407). The response processing device 101 determines that the topic selection is completed when, for example, the topic selection end notification is received from the worker terminal 201.

Here, if the topic selection is not completed (step S1407: No), the response processing device 101 returns to step S1402. On the other hand, when the topic selection is completed (step S1407: Yes), the response processing device 101 refers to the synonym evaluation table 230 and generates a synonym dictionary showing the correspondence between the topic of the acceptance / rejection flag “1” and the phrase. (Step S1408).

Then, the response processing device 101 outputs the generated synonym dictionary (step S1409), and ends a series of processes according to this flowchart. As a result, it is possible to support the registration of appropriate synonyms for the topic included in the question data Q registered in the FAQ master 220.

The processing of steps S1301 and S1302 shown in FIG. 13 may be executed by another computer different from the response processing device 101. In this case, the response processing device 101 acquires a synonym dictionary (provisional) from another computer. Further, if the response processing device 101 has a synonym dictionary generated in the past, the synonym dictionary may be acquired as a synonym dictionary (provisional).

Further, for the unselected topic not selected in step S1402, the response processing device 101 refers to, for example, the synonym evaluation table 230, and sets the acceptance / rejection flag of the phrase having the maximum similarity to the topic to “1”. It may be set automatically to.

As described above, according to the response processing device 101 according to the embodiment, the question data Q is morphologically analyzed, the topic included in the question data Q is extracted, and the target word to be set as a synonym for the extracted topic. Can present phrases that are candidates for. Then, according to the response processing device 101, when the selection of any of the presented phrases is accepted as the target word to be set as the synonym of the topic, both the question data Q and the question data Q'are the same. It can be learned to be related to the answer data A. The question data Q'is question data obtained by replacing the topic included in the question data Q with a selected phrase.

As a result, it is possible to support the easy registration of appropriate synonyms for the topic included in the question data Q registered as FAQ. For example, when registering question data Q, a synonym for a topic included in question data Q can be obtained by simply selecting one of the phrases that can be candidates for the target word to be set as a synonym. Can be registered.

Further, according to the response processing device 101, it is possible to present a phrase that is a candidate of a target word to be set as a synonym of the topic based on the search result in a predetermined site about the topic. As a result, for example, even if the past dialogue history in the call center is not sufficiently accumulated, the candidate words to be set as synonyms of the topic are extracted and presented by using the information of the published web page. be able to.

Further, according to the response processing device 101, when the question data Q contains a plurality of topics as a result of morphological analysis, the phrase which is a candidate word to be set as a synonym for each of the plurality of topics. It can be presented so that the presence or absence or number of As a result, when selecting a topic for which maintenance of synonyms is to be performed, it is possible to determine a topic for which maintenance of synonyms should be given priority based on the presence or absence of phrases and the number of phrases for each topic.

Further, according to the response processing device 101, when the input of the question data Q'' in which the topic is replaced with a specific phrase in the question data Q is detected after learning, both the question data Q and the question data Q'' are detected. , You may learn to be related to the same answer data A. A specific phrase is a phrase that is not selected as a target word to be set as a synonym of the topic among the candidate words to be set as a synonym of the topic.

As a result, when the question data Q'' including the phrase is input to the response processing system 200, it is possible to search by the question data Q in which the specific phrase is replaced with the topic, and the hit rate can be increased. ..

Further, according to the response processing device 101, when the input of the question data Q'' in which the topic is replaced with a specific phrase in the question data Q after learning is detected, the registration of the topic of the specific phrase as a synonym is suggested. be able to. As a result, it is possible to notify the administrator of the response processing system 200 and the like of the existence of a specific phrase that is desirable to be registered as a synonym of the topic, and it is possible to prevent omission of registration of an appropriate synonym.

From these facts, according to the response processing device 101 and the response processing system 200 according to the embodiment, when registering the question data Q, it is possible to easily register appropriate synonyms for the topic included in the question data Q. By providing support, it is possible to reduce the workload required for maintenance of synonyms and prevent the synonym dictionary from becoming obsolete.

The response processing method described in the present embodiment can be realized by executing a program prepared in advance on a computer such as a personal computer or a workstation. This response processing program is a computer-readable recording medium such as a hard disk, a flexible disk, a CD (Compact Disk) -ROM, an MO (Magnet-Optical disk), a DVD (Digital Versaille Disk), or a USB (Universal Serial Bus) memory. It is recorded on a DVD and executed by being read from a recording medium by a computer. Further, the response processing program may be distributed via a network such as the Internet.

101 Response processing device 110 Display 120 Synonym dictionary 200 Response processing system 201 Worker terminal 210 Network 220 FAQ master 230 Synonym evaluation table 240 Chat log DB
300 bus 301 CPU
302 Memory 303 I / F
304 Disk drive 305 Disk 701 Acquisition unit 702 Extraction unit 703 Display control unit 704 Reception unit 705 Learning unit 800, 1100, 1200 Topic selection screen 900 Phrase selection screen

Claims (8)

When registering the question data, the question data is morphologically analyzed, and the words contained in the question data are extracted.
Present the candidate words to be set as synonyms of the above words,
When the selection of any of the candidates is accepted as the target word to be set as the synonym of the word, the question data and the word included in the question data are replaced with the selected word. Learn so that both with the question data are related to the same answer data,
A response processing program characterized by having a computer perform processing. The response processing program according to claim 1, wherein the synonym is presented based on a search result in a predetermined site for the word. As a result of morphological analysis of the question data, when the question data contains a plurality of words, the presence or absence or number of candidates of the target words to be set as synonyms for each of the plurality of words should be presented so as to be known. The response processing program according to claim 1. After learning, when the input of question data in which the word is replaced with a specific word that is not selected as a target word to be set as a synonym for the word among the candidates is detected, the question data and the question data are displayed. Learning so that both the word included in the question data and the question data obtained by replacing the word with the specific word are related to the same answer data.
The response processing program according to claim 1, wherein the processing is executed by the computer. After learning, when the input of question data in which the word is replaced with a specific word that is not selected as a target word to be set as a synonym of the word among the candidates is detected, the input of the specific word is detected. Suggests registration of the word as a synonym,
The response processing program according to claim 1, wherein the processing is executed by the computer. When registering the question data, the question data is morphologically analyzed, and the words contained in the question data are extracted.
Present the candidate words to be set as synonyms of the above words,
When the selection of any of the candidates is accepted as the target word to be set as the synonym of the word, the question data and the word included in the question data are replaced with the selected word. Learn so that both with the question data are related to the same answer data,
A response processing method characterized in that processing is performed by a computer. When registering the question data, the question data is morphologically analyzed, and the words contained in the question data are extracted.
Present the candidate words to be set as synonyms of the above words,
When the selection of any of the candidates is accepted as the target word to be set as the synonym of the word, the question data and the word included in the question data are replaced with the selected word. Learn so that both with the question data are related to the same answer data,
A response processing device having a control unit. When registering question data, an extraction unit that morphologically analyzes the question data and extracts words contained in the question data,
A display control unit that presents candidates for the target word to be set as a synonym for the word,
When the selection of any of the candidates is accepted as the target word to be set as the synonym of the word, the question data and the word included in the question data are replaced with the selected word. A learning department that learns so that both the question data and the question data are related to the same answer data,
A response processing system characterized by having.

Images