JP7741529B2 - Question and answer display server, question and answer display method, and question and answer display program - Google Patents

Description

This disclosure relates to a question and answer display server, a question and answer display method, and a question and answer display program that interactively answers questions from a question and answer database, and that assists in building the question and answer database by creating alternative question sentences to those entered by an administrator when registering them in the question and answer database.

In recent years, AI (Artificial Intelligence) technology has been attracting attention. In user support situations, traditionally, telephone calls were handled manually or user inquiries were answered via email, but AI is increasingly being used in an increasing number of situations. For example, in situations where AI is being used, systems that use dialogue systems to answer questions from users are being developed.

For example, Patent Document 1 proposes a configuration in which multiple question and answer patterns are registered in advance as a knowledge database, and a dialogue system is used to respond to questions from users.

Patent No. 6218057

Patent Document 1 uses an interactive interface to present the answers the user is looking for, but to do so, it is necessary to prepare answer patterns and corresponding question patterns.

Furthermore, in systems such as those described in Patent Document 1, multiple question patterns must be prepared depending on how the question is asked in response to a given answer pattern, but creating these question patterns requires a certain level of experience, posing the problem that beginners may not be able to immediately prepare useful question patterns.

The purpose of this disclosure is to automatically create a question phrase with a different expression that corresponds to the same answer pattern when a certain question pattern is input, thereby contributing to the easy creation of useful question patterns.

In one aspect of the present disclosure, the question and answer display server includes a memory unit that stores a question and answer database containing answer patterns and multiple corresponding question patterns; a question acquisition unit that acquires a question sentence input from a user terminal; an answer pattern display unit that displays the corresponding answer pattern on the user terminal based on the similarity between the acquired question sentence and the question pattern stored in the memory unit; and a question creation unit that creates a question sentence by converting the question sentence input from the administrator terminal into one or more different expressions that correspond to the same answer pattern as the input question sentence.

A question and answer display method according to one aspect of the present disclosure includes a storage step of storing a question and answer database containing answer patterns and multiple corresponding question patterns; a question acquisition step of acquiring a question sentence input from a user terminal; an answer pattern display step of displaying the corresponding answer pattern on the user terminal based on the similarity between the acquired question sentence and the question pattern stored in the storage step; and a question creation step of creating a question sentence by converting the question sentence input from the administrator terminal into one or more different expressions corresponding to the same answer pattern as the input question sentence.

In one aspect of the present disclosure, the question and answer display program is a program that causes a computer to execute the following steps: a storage step for storing a question and answer database containing answer patterns and multiple corresponding question patterns; a question acquisition step for acquiring a question entered from a user terminal; an answer pattern display step for displaying the corresponding answer pattern on the user terminal based on the similarity between the acquired question and the question pattern stored in the storage step; and a question creation step for creating a question entered from an administrator terminal by converting the question entered from the administrator terminal into one or more different expressions that correspond to the same answer pattern as the entered question.

According to the present disclosure, a question and answer display server that answers questions from users using an interactive interface automatically generates paraphrases for questions, allowing administrators to easily create a question and answer database.

FIG. 1 is a diagram illustrating an example of the configuration of a question and answer display server according to the present disclosure. FIG. 2 is a diagram illustrating an example of a data configuration of a storage unit according to the present disclosure. FIG. 10 is a diagram illustrating an example of a display of a user terminal according to the present disclosure. 10A and 10B are diagrams illustrating an example of a display of a question from a user and a response from a dialogue server on a user terminal of the present disclosure. FIG. 10 is a diagram illustrating an example of an interface for selecting a question generated in the administrator terminal of the present disclosure. 10 is a flowchart showing an example of a processing flow when a user operates the question and answer display server. 10 is a flowchart showing an example of a processing flow when the question and answer display server creates a paraphrase of a question sentence. FIG. 1 is a schematic block diagram illustrating a configuration of a computer according to an embodiment of the present disclosure.

Below, a question and answer display server, a question and answer display method, and a question and answer display program according to embodiments of the present disclosure will be described with reference to the drawings. Note that the embodiments described below do not unduly limit the content of the present disclosure as set forth in the claims. Furthermore, not all of the configurations described in the embodiments are necessarily essential components of the present disclosure. Furthermore, in all figures describing the embodiments, common components will be designated by the same reference numerals, and repeated explanations will be omitted.

Figure 1 is a diagram showing the overall configuration of a question and answer display system 1 according to the present disclosure. Below, the configuration of the question and answer display system 1 according to the present disclosure will be explained with reference to this diagram.

<Configuration>
The question and answer display server 100 includes a memory unit 101, a question acquisition unit 102, an answer pattern display unit 103, a question creation unit 104, a masking unit 105, and a question selection unit 106. The question and answer display server 100 is connected to user terminals 200-1 to 200-N and administrator terminals 201-1 to 201-M via a network NW. The question and answer display server 100 is implemented by installing a program on a computer. Note that the components of the question and answer display server 100 may be implemented in a distributed manner on multiple computers. The memory unit 101 may also be located outside the dialogue server.

User terminals 200-1 to 200-N are terminals that accept input questions, send them to question and answer display server 100, and receive and display answer patterns generated by question and answer display server 100 in response to those questions. User terminals 200-1 to 200-N are assumed to be terminals used by general users who wish to inquire about unclear points in user support, for example.

Administrator terminals 201-1 to 201-M are terminals that access and manage a database containing questions and corresponding answer patterns stored in storage unit 101. Administrator terminals 201-1 to 201-M are assumed to be terminals used by support managers who handle inquiries in user support, for example. Administrators use administrator terminals 201-1 to 201-M to perform registration operations such as adding, changing, and deleting question and answer patterns.

The storage unit 101 stores a question and answer database that includes question patterns and answer patterns. For example, the question and answer database stores one or more question patterns for a given answer pattern. In situations such as user support, multiple question patterns are possible for a single answer, so it is desirable for the question and answer database to also register as many possible question patterns as possible for a single answer pattern.

The question and answer display server 100 uses the question and answer database in the storage unit 101 to display answer patterns corresponding to questions entered by users via user terminals 200-1 to 200-N, and provides an interface that allows users to engage in interactive exchanges such as chat.

The question acquisition unit 102 acquires questions from user terminals 200-1 to 200-N and/or administrator terminals 201-1 to 200-M via the network NW. Questions can be acquired in the form of natural language. By allowing questions to be asked in the form of natural language, users can easily make inquiries in the same way as if they were asking a person. Note that words indicating keywords, etc. may also be acquired.

The answer pattern display unit 103 compares the question sentence acquired by the question sentence acquisition unit 102 with question patterns registered in the memory unit 101 and calculates the similarity to the question pattern. Here, similarity to a question pattern includes similarity to a question pattern group, which is a collection of question patterns corresponding to the same answer pattern, and similarity to functions, categories, etc. based on question patterns. The similarity is calculated, for example, using natural language processing. Specifically, keywords from the question pattern and the question sentence may be extracted, and the degree of keyword agreement may be calculated using weights for each keyword. Alternatively, the question pattern and the question sentence may be converted into vectors, and the similarity may be calculated using cosine similarity. In this way, by using similarity to calculate which question pattern the question sentence is similar to, it is possible to select an answer pattern even if it is not the same as the question pattern stored in the memory unit.

When calculating the similarity between the question sentence and question pattern used in the answer pattern display unit 103, the parameters for calculating the similarity may be updated by performing machine learning using the question pattern entered by the user and/or administrator as a candidate for learning data (also called teacher data). By performing such machine learning, it is possible to improve the accuracy of presenting answer patterns that match the user's intentions.

The answer pattern display unit 103 further displays, on the user terminals 200-1 to 200-N, answer patterns corresponding to question patterns that have a high degree of similarity with the question sentence based on the degree of similarity.

The confidence level may also be displayed along with the answer pattern. The confidence level is a measure of the dialogue server's degree of confidence that the question sentence has the same meaning as the question pattern stored in the memory unit 101, and the similarity calculated by the answer pattern display unit 103 may be used. The confidence level may be calculated by evaluating the degree of match between the question sentence and the question pattern using another method, rather than using the similarity level as is. Other methods include, for example, calculating the confidence level using the degree of keyword match when the similarity level is calculated using cosine similarity. When displaying the confidence level, it may be displayed numerically, or the degree may be expressed using an icon or the like.

It is also possible to set a certain threshold for the confidence level and then display all answer patterns above that threshold, or the top n answer patterns. In this way, by displaying multiple answer patterns assuming a certain confidence level, it is possible to increase the likelihood that the answer pattern the user is looking for will be displayed.

It is also possible to display only the answer pattern with the highest confidence level. Furthermore, if there are no answer patterns that meet or exceed a set threshold, it is possible to not display any answer patterns or to display a message prompting the user to ask a different question, such as "Please change your question." This allows the user to avoid being shown an answer pattern that is different from the answer to the question they intended, and avoids unnecessary confusion.

The question creation unit 104 creates a question by converting the question entered from the administrator terminals 201-1 to 201-M into a different expression.

A question sentence converted into a different expression is a sentence whose content should display the same answer pattern as the input question sentence, and may be a sentence that is synonymous with the input question sentence. Also, it does not necessarily have to be a sentence that is synonymous with the input question sentence, but may be a question sentence that corresponds to an answer pattern, such as asking about the same thing.

The question creation unit 104 acquires the question entered from the administrator terminals 201-1 to 201-M via the question acquisition unit 102. Then, the question creation unit 104 uses the answer patterns and associated question patterns, i.e., question pattern groups, stored in the storage unit 101 to create a question converted from the question into a different expression.

The question creation unit 104 may, for example, use a synonym dictionary to perform natural language processing and convert words in the input question into different words, thereby creating a question that has been converted into a different expression.

The question creation unit 104 may also create a question that is converted from the input question into a different expression using, for example, a synonym dictionary or rule-based rules.

The question creation unit 104 may further create a question expressed differently from the input question by generating a learning model using a statistical model or machine learning.

When using a statistical model or machine learning, the question creation unit 104 may generate learning data as follows. That is, the answer patterns and question patterns stored in the memory unit 101 have a correspondence relationship in which a question pattern corresponds to a certain answer pattern. Furthermore, an answer pattern is linked to one or more question patterns, and multiple question patterns corresponding to one answer pattern can each be assumed to be paraphrases of the other question patterns.

The question creation unit 104, in a group of multiple question patterns corresponding to a certain answer pattern, uses a certain question pattern belonging to that group as correct answer data that is a paraphrase of other question patterns belonging to that group, i.e., as learning data for machine learning, and generates a learning model, etc. using this, and then creates a question that has been converted from the acquired question into a different expression.

Note that the learning data used for machine learning is not limited to the question and answer database stored in the storage unit 101; question patterns converted from sentences registered in the question and answer database using a synonym dictionary may also be added.

The learning data used by the question creation unit 104 when using statistical models and machine learning has been described above. Below, we will describe a method for creating paraphrased questions using statistical models and machine learning.

The question creation unit 104 uses, for example, machine translation technology to create a question in a different expression, which is a paraphrase expression. Furthermore, using neural translation technology in particular makes it easy to obtain multiple converted question candidates.

The question creation unit 104 may, for example, use training data to prepare a translation model and a language model based on automatic translation technology using a statistical model, and create a question that is converted from the input question into a different expression.

The question creation unit 104 may, for example, build a learning model based on machine translation technology and then create a question by converting the input question into a different expression.

In this case, it is preferable to build a learning model based on neural translation as a machine translation technology. Using a learning model based on neural translation makes it easier to output multiple candidate questions when creating a question that has been converted from the input question into a different expression.

The question creation unit 104 preferably further constructs a learning model in the same way as neural translation, for example, using an attention-enabled LSTM translation model. This makes it possible to convert the input question into one or more questions expressed differently and create them.

In the question and answer display system 1, the administrator adds answer patterns and question patterns via administrator terminals 201-1 to 201-M and stores them in the storage unit 101. However, adding question patterns requires a certain level of experience, and if question and answer display server 100 could suggest question patterns, it would be possible to add question patterns more efficiently. The question creation unit 104 achieves this, proposing question sentences that paraphrase the question entered by the administrator and assisting in the addition of question patterns to the storage unit.

The masking unit 105 uses natural language processing to mask proper nouns in the question created by the question creation unit 104.

When extracting proper nouns from a question sentence using natural language processing, the masking unit 105 performs the extraction using, for example, a word dictionary or a method such as machine learning. The following method is used: That is, a word dictionary is used to perform natural language processing on the question sentence, and words that are not in the word dictionary are extracted as proper nouns.

The masking unit 105 performs natural language processing on the question sentence using, for example, a word dictionary, and extracts words that are not in the word dictionary as proper nouns.

The masking unit 105 extracts proper nouns, for example, using a learning model for machine learning to extract proper nouns.

The masking unit 105 performs masking processing by replacing the extracted proper nouns with specific characters or symbols.

The question and answer display system 1 is envisioned as a system in which, for example, multiple businesses providing user support operate administrator terminals 201-1 to 201-M, and users of these businesses use user terminals 200-1 to 200-N. Therefore, question patterns created by other businesses may be used as training data. In this case, masking the converted question text can be performed to eliminate proper nouns and events specific to other businesses, allowing for general use of the data.

The question selection unit 106 allows the administrator to select one or more questions to be registered as answer patterns from among multiple candidate paraphrased questions created by the question creation unit 104 or masked by the masking unit 105, and accepts and acquires the questions selected by the administrator. The question selection unit 106 stores the accepted questions in the memory unit 101 as question patterns for the answer patterns.

The question selection unit 106 provides an interface to the administrator, allowing the administrator to register the question they want to add to the storage unit 101 simply by selecting it, making it easier to add question patterns.

When creating paraphrases, the question creation unit 104 may use the question entered from the administrator terminal and the question selected by the question selection unit 106 as learning data, or may use them in combination with a group of question patterns as learning data, which may be useful for building a learning model for creating paraphrases of questions.

<Basic operations of the question and answer database and the question and answer display server>
2 shows a specific example of the question and answer database stored in the storage unit 101. The question and answer database is made up of a combination of question patterns and answer patterns, and one or more question patterns are prepared for each answer pattern.

When a user makes an inquiry to the question and answer display server, the user operates the user terminal 200-1 to 200-N to input a question, and the question acquisition unit 102 acquires the question input by the user.

Figure 3 shows an example of the user interface 111 when the question acquisition unit acquires a question for the user terminals 200-1 to 200-N and/or the administrator terminals 201-1 to 201-M. The question acquisition unit acquires a question by inputting it from a keyboard, touch panel, or the like. It is also possible to configure the question to be input by voice via a voice recognition engine.

The answer pattern display unit 103 calculates the similarity between the question sentence and the question pattern stored in the memory unit 101. Here, similarity to a question pattern includes similarity to a question pattern group, which is a collection of question patterns corresponding to the same answer pattern, as well as similarity to functions, categories, etc. based on the question pattern.

For example, suppose that N answer patterns, from A1 to AN, are registered in the memory unit 101. Then, for each answer pattern, question patterns Q11 to Q1M1, Q21 to Q2M2, ... QN1 to QNMN are registered. In this case, question patterns Q11 to Q1M1 are question pattern group C1, question patterns Q21 to Q2M2 are question pattern group C2, ... question patterns QN1 to QNMN are question pattern group CN. In this case, the answer pattern display unit 103 may calculate the similarity between the question sentence and the question pattern (Q11 to QNMN). It may also calculate the similarity between the question sentence and the question pattern group (C1 to CN). Furthermore, it may calculate the similarity with functions, categories, etc. based on the question pattern.

The answer pattern display unit 103 calculates the similarity between the question sentence and the question patterns stored in the memory unit 101. For example, suppose a question sentence such as "I would like to change my address information" is acquired, and question pattern groups C1, C2, and C3 are registered in the memory unit 101. The answer pattern display unit 103 calculates the similarity between the question sentence and each question pattern group; in this case, the similarities with each question pattern group are calculated to be, for example, 0.92, 0.70, and 0.21, respectively.

The answer pattern display unit 103 calculates the confidence level using the question sentence, question pattern, and similarity. Here, for example, it calculates 98% for question pattern group C1, 78% for C2, and 20% for C3. Furthermore, if the threshold is set to 60%, the confidence levels of C1 and C2 exceed the threshold, and the corresponding answer patterns, A1 and A2, are displayed on user terminals 200-1 to 200-N along with confidence level information.

If only the top answer pattern is displayed, only the answer pattern A1, which has the highest confidence level among those whose confidence level exceeds the threshold, will be displayed.

Figure 4 shows an example of the user interface 111 that the answer pattern display unit 103 displays on the user terminals 200-1 to 200-N. In this example, the question entered by the user is shown on the right, and the answer pattern selected by the dialogue server is shown on the left.

The above is the basic operation of the question and answer display server 100.

<Specific steps for creating a paraphrase of a question>
The question creation unit 104 creates a question with a different expression from the input question. For example, in a user support situation, "I forgot my password,""I forgot my password," and "I lost my password" all have the same meaning, and a corresponding answer pattern could be, for example, "Please reset your password from this page."

In this case, by performing natural language processing on the question "I forgot my password" and converting "forgot" using a synonym dictionary, it is possible to create a question with a different or paraphrased expression, such as "I forgot my password" or "I lost my password."

The question creation unit 104 may use a question created using a synonym dictionary in this way as a candidate for a question with a different expression, or may add it as a learning pattern and then create another question using a statistical model or machine learning method.

The following describes in detail the method in which the question creation unit 104 creates questions, particularly using machine learning.

The question creation unit 104 uses the question patterns in the question and answer database stored in the storage unit 101 as training data. A synonym dictionary may also be prepared and used as training data. In order to obtain questions converted into paraphrased expressions, it is particularly effective to use machine learning techniques, and more specifically, an attention-enabled LSTM translation model, to build a training model similar to neural translation.

In normal machine translation, the relationship between a natural sentence and its translation is one-to-one. However, in order for the question creation unit 104 to create paraphrased expressions, it is necessary to provide flexibility to the output sentences. Therefore, it is desirable to use training data in which there are multiple question patterns and paraphrased expressions have a one-to-N relationship, such as a group of question patterns that serve as training data in which paraphrased expressions are related.

Here, let the set of answer patterns be A = (a1, a2, ... aN). ai is associated with mi question patterns Qi = (qi1, qi2, ... qimi). Furthermore, all question patterns can be expressed as Q = (Q1, Q2, ... QN).

The training dataset consists of combinations of (qij, qik), where i is a value between 1 and N, j and k are values between 1 and mi, and j and k cannot equal k.

When creating a learning model, if all combinations of data are used as training data, the amount of data will be enormous, so it is also possible to randomly select a certain amount and downsample it.

After determining the training dataset, the question creation unit 104 uses the training dataset to construct a training model for the attention-enabled LSTM translation model.

The question creation unit 104 inputs the acquired question into an attention-enabled LSTM translation model to create a differently expressed question. Here, by using attention-enabled LSTM translation, a neural translation model, likelihood can be calculated simultaneously with output. Therefore, the calculated likelihood can be used as a confidence level, which is an indicator of the reliability of the paraphrased expression for the question.

Furthermore, the question creation unit 104 sets a certain threshold and performs a beam search using a greedy method while pruning the calculated confidence level, thereby extracting question sentences with paraphrased expressions that have a confidence level above the threshold, making it possible to create multiple candidate paraphrased expressions.

In this case, if the question creation unit 104 simply performs a beam search, the influence of the confidence level at the beginning of the sentence will be large, so it may keep the top few words at the beginning of the sentence and create a paraphrased question using these words as a starting point.

The above describes a method that uses machine learning technology and an attention-assisted LSTM translation model, but even when using other machine learning methods, it is possible to similarly prepare training data, output multiple paraphrased questions, and then use a threshold to create the top N questions.

Furthermore, when using a statistical model, it is possible to similarly prepare training data, output multiple paraphrased question sentences along with their likelihoods, and then use a threshold to create the top N questions.

The masking unit 105 masks proper nouns in the question created by the question creation unit 104. For example, the masking unit 105 uses a word dictionary and natural language processing to extract words that are not in the word dictionary as proper nouns.

The masking unit 105 may also use machine learning to extract proper nouns and then apply masking.

The masking unit 105 masks the question sentences, which are paraphrases, to eliminate the influence of proper nouns and to create questions by using many question patterns as training data.

The question selection unit 106 allows the administrator to select a question pattern to be registered as an answer pattern for a question created by the question creation unit 104 that is a paraphrase of the question entered by the administrator, or for a question masked by the masking unit 105. The selected question is then added to the storage unit 101 as a question pattern.

Figure 5 shows a specific example of a question created by the question creation unit 104, and a specific example of an interface in the question selection unit 106 that displays questions to the administrator and allows them to select one. In Figure 5, the administrator selects a question and then presses the registration button, whereby a question phrased in a paraphrase of the same answer pattern as the question entered by the administrator is registered as a question pattern and stored in the storage unit 101.

<Processing flow>
FIG. 6 shows the flow of processing when a user uses the question and answer display server according to the present disclosure.

The question acquisition unit 102 acquires questions from user terminals 200-1 to 200-N via the network NW. The questions can be acquired in the form of natural language (step S1).

The answer pattern display unit 103 compares the question sentence acquired by the question sentence acquisition unit 102 with question patterns registered in the storage unit 101 and calculates the degree of similarity with the question pattern. The degree of similarity can be calculated, for example, using natural language processing.

Furthermore, the answer pattern display unit 103 displays answer patterns for question patterns that have a high degree of similarity with the question sentence on the user terminals 200-1 to 200-N based on the similarity (step S2).

Figure 7 shows the processing flow when a question and answer display server according to the present disclosure converts a question input from an administrator terminal 201-1 to 201-M, creates a paraphrased question, and then allows the administrator to select a question.

The question creation unit 104 converts questions input from the administrator terminals 201-1 to 201-M to create paraphrased questions. Specifically, it uses the question patterns stored in the storage unit 101 as training data to build an attention-enabled LSTM translation model and creates questions that are paraphrased versions of the input questions (step S3).

The masking unit 105 masks proper nouns in the question created by the question creation unit 104. At this time, natural language processing is used to determine that words not in the dictionary are proper nouns (step S4).

The question selection unit 106 displays the rephrased questions created by the question creation unit 104 on the administrator terminals 201-1 to 201-M to allow the administrator to select one, and stores the selected question in the memory unit 101 as a question pattern (step S5).

<Effects>
According to the present disclosure, when an administrator creates a question pattern for a certain answer pattern in a question and answer display server, the question creation unit 104 can present a paraphrased question sentence on the system side, making it possible to easily create a question pattern.

This allows even novice administrators who are unfamiliar with creating question patterns to easily create them, and also helps experienced administrators register appropriate question patterns.

Furthermore, by controlling a threshold for the parameters output from the translation model and selecting only those exceeding the threshold as question candidates, it is possible to present multiple paraphrases while still providing highly accurate paraphrases.

In addition, by masking the converted questions, it is possible to use a large amount of data as training data, and by providing a user interface for registering selected questions as question patterns, it is possible to easily register question patterns.

<Program>
8 is a schematic block diagram showing the configuration of a computer 801. The computer 801 includes a CPU 802, a main storage device 803, an auxiliary storage device 804, and an interface 805.

Here, we will explain in detail the programs that realize each function that makes up the question and answer display server 100.

The question and answer display server 100 is implemented in a computer 801. The operation of each component of the evaluation device 1 is stored in the form of a program in the auxiliary storage device 804. The CPU 802 reads the program from the auxiliary storage device 804, loads it into the main storage device 803, and executes the above-mentioned processing in accordance with the program. The CPU 802 also allocates storage areas in the main storage device 803 corresponding to the above-mentioned storage units in accordance with the program.

Specifically, this program causes the computer 801 to execute the following steps: a storage step for storing a question and answer database containing answer patterns and multiple corresponding question patterns; a question acquisition step for providing an interactive interface to the user terminal and acquiring a question entered from the user terminal; an answer pattern display step for displaying a corresponding answer pattern on the user terminal based on the similarity between the acquired question and the question pattern stored in the storage step; and a question creation step for converting a question entered from the administrator terminal into one or more question sentences expressed differently that correspond to the same answer pattern as the entered question, based on the question and answer database.

Note that auxiliary storage device 804 is an example of a non-transitory tangible medium. Other examples of non-transitory tangible media include magnetic disks, optical magnetic disks, CD-ROMs (Compact Disc Read-Only Memory), DVD-ROMs (Digital Versatile Disc Read-Only Memory), and semiconductor memories connected via interface 805. Furthermore, if this program is distributed to computer 801 via a network, computer 801 that receives the program may load the program into main storage device 803 and execute the above-described processing.

The program may also be one that realizes some of the functions described above. Furthermore, the program may be a so-called differential file (differential program) that realizes the functions described above in combination with other programs already stored in the auxiliary storage device 804.

The above describes the embodiments of the present disclosure, but various modifications and combinations that may be necessary due to design considerations or other factors are intended to fall within the scope of the invention described in the claims and the inventions corresponding to the specific examples described in the embodiments of the invention.

1...Question and answer display system, 100...Question and answer display server, 101...Memory unit, 102...Question acquisition unit, 103...Answer pattern display unit, 104...Masking unit, 105...Question creation unit, 106...Question selection unit, 111...Interface presented to user or administrator, 200-1 to 200-N...User terminal, 201-1 to 201-M...Administrator terminal, 801...Computer, 802...CPU, 803...Main memory, 804...Auxiliary memory, 805...Interface, NW...Network

Claims (6)

A server that answers questions based on pre-registered answer patterns and corresponding question patterns,
a storage unit that stores a question and answer database including answer patterns and a plurality of question patterns corresponding to the answer patterns;
a question acquisition unit that acquires a question input by a user terminal;
an answer pattern display unit that displays a corresponding answer pattern on a user terminal based on the similarity between the acquired question sentence and the question pattern stored in the storage unit;
a question creation unit that generates a learning model using machine translation technology by inputting the question pattern into the learning model , and creates a question that converts a question input from an administrator terminal into one or more different expressions that correspond to the same answer pattern as the input question ;
a question selection unit that displays one or more of the question sentences in the alternative phrases converted by the question creation unit as candidates, and acquires one or more of the question sentences in the alternative phrases selected by an administrator and stores them in a question and answer database ;
The learning model is a question and answer display server that is trained using learning data in which there are multiple question patterns, and paraphrases are expressed such that the answer pattern is 1 and there are N question patterns corresponding to this answer pattern , and converts the question text input from the administrator terminal into one or more other expressions corresponding to the same answer pattern. The question and answer display server described in claim 1, wherein the question creation unit generates the learning model using a group of multiple question patterns corresponding to answer patterns in the question and answer database as learning data for paraphrased expressions, and uses the learning model to create questions converted into different expressions. 3. The question and answer display server according to claim 1, further comprising a masking unit that extracts and masks proper nouns included in one or more of the question sentences expressed in different ways converted by the question sentence creation unit. The question and answer display server of claim 1, wherein the question creation unit further performs machine learning using the question input from the administrator terminal and the question of a different expression selected by the administrator in the question selection unit as learning data. A method for answering questions based on pre-registered answer patterns and corresponding question patterns,
a storage step of storing a question and answer database including answer patterns and a plurality of question patterns corresponding to the answer patterns;
a question acquisition step of acquiring a question inputted by a user terminal;
an answer pattern display step of displaying a corresponding answer pattern on a user terminal based on the similarity between the acquired question sentence and the question pattern stored in the storage step;
a question generation step of generating a learning model using machine translation technology based on the question and answer database by inputting the question pattern into the learning model , and generating a question sentence in which the question sentence input from the administrator terminal is converted into one or more different expressions corresponding to the same answer pattern as the input question sentence ;
a question selection step of displaying one or more of the question sentences in the alternative phrases converted by the question creation step as candidates, and acquiring one or more of the question sentences in the alternative phrases selected by an administrator and storing them in a question and answer database ;
A question and answer display method in which the learning model is trained using learning data in which there are multiple question patterns, and paraphrases are expressed such that the answer pattern is 1 and there are N question patterns corresponding to this answer pattern , and the learning model converts the question text input from the administrator terminal into one or more other expressions corresponding to the same answer pattern. A program that answers questions based on pre-registered answer patterns and corresponding question patterns,
a storage step of storing a question and answer database including answer patterns and a plurality of question patterns corresponding to the answer patterns;
a question acquisition step of acquiring a question inputted by a user terminal;
an answer pattern display step of displaying a corresponding answer pattern on a user terminal based on the similarity between the acquired question sentence and the question pattern stored in the storage step;
a question generation step of generating a learning model using machine translation technology based on the question and answer database by inputting the question pattern into the learning model , and generating a question sentence in which the question sentence input from the administrator terminal is converted into one or more different expressions corresponding to the same answer pattern as the input question sentence ;
a question selection step of displaying one or more of the question sentences in the alternative phrases converted by the question creation step as candidates, and acquiring one or more of the question sentences in the alternative phrases selected by an administrator and storing them in a question and answer database ;
The learning model is a question and answer display program that is a learning model that learns using learning data in which there are multiple question patterns, and paraphrases are expressed such that the answer pattern is 1 and there are N question patterns corresponding to this answer pattern , and converts the question text input from the administrator terminal into one or more other expressions that correspond to the same answer pattern.