JP7616761B2 - QUESTION ESTIMATION DEVICE, TRAINED MODEL GENERATION DEVICE, QUESTION ESTIMATION METHOD, TRAINED MODEL PRODUCTION METHOD, PROGRAM, AND RECORDING MEDIUM - Google Patents

Description

The present invention relates to a question estimation device, a trained model generation device, a question estimation method, a trained model production method, a program, and a recording medium.

In order to support staff who respond to questions from customers, technology has been reported that automatically creates answers to the questions and allows the staff to answer them as necessary (for example, Patent Document 1, etc.).

JP 2004-171479 A

However, when one person is responsible for multiple customers, there are many cases where a response intended for one customer is mistakenly sent to another customer (also known as mis-sending).

The present invention aims to provide a question estimation device, a trained model generation device, a question estimation method, a trained model production method, a program, and a recording medium that can prevent erroneous transmissions.

In order to achieve the above object, the question estimation device of the present invention comprises:
The system includes a conversation history data acquisition unit, a sentence type identification unit, a question and answer table generation unit, an unsent sentence acquisition unit, and an estimated question sentence generation unit,
the conversation history data acquisition unit acquires conversation history data, which is character data of a completed conversation;
The conversation history data is composed of a plurality of sentence data,
The text type identification unit identifies a text type for each of the text data,
The sentence types include a question type and an answer type,
the question and answer table generating unit generates a question and answer table by linking the sentence data of the question type with the sentence data of the answer type based on a result of the identification;
The unsent text acquisition unit acquires an unsent reply text that has not yet been transmitted to another user in an ongoing conversation,
The estimated question sentence generation unit is a device that generates an estimated question sentence corresponding to the untransmitted answer sentence based on the question and answer table and the untransmitted answer sentence.

The trained model generation device of the present invention comprises:
The data acquisition unit and the trained model generation unit are included.
the data acquisition unit acquires question-type sentence data linked to answer-type sentence data as teacher data;
The trained model generation unit is a device that generates a first trained model by machine learning using the teacher data, which generates and outputs an estimated question sentence corresponding to the answer type sentence data when the answer type sentence data is input.

The question estimation method of the present invention includes the steps of:
The method includes a conversation history data acquisition step, a sentence type identification step, a question and answer table generation step, an unsent sentence acquisition step, and an estimated question sentence generation step,
The conversation history data acquisition step acquires conversation history data, which is character data of a completed conversation,
The conversation history data is composed of a plurality of sentence data,
The text type identification step includes identifying a text type for each of the text data,
The sentence types include a question type and an answer type,
the question and answer table generating step links the sentence data of the question type with the sentence data of the answer type based on a result of the identification to generate a question and answer table;
The unsent text acquisition step includes acquiring an unsent reply text that has not yet been transmitted to another user in an ongoing conversation,
The estimated question sentence generating step is a method of generating an estimated question sentence corresponding to the untransmitted answer sentence based on the question and answer table and the untransmitted answer sentence.

The method for producing a trained model of the present invention includes the steps of:
The method includes a data acquisition process and a trained model generation process.
The data acquisition step includes acquiring question-type sentence data linked to answer-type sentence data as teacher data,
The trained model generation process is a production method for generating a first trained model by machine learning using the training data, which generates and outputs an estimated question sentence corresponding to the answer type of sentence data when the answer type of sentence data is input.

The present invention makes it possible to prevent erroneous transmissions.

FIG. 1 is a block diagram illustrating a configuration of an example of a question estimation device according to the first embodiment. FIG. 2 is a block diagram illustrating an example of a hardware configuration of the question estimation device according to the first embodiment. FIG. 3 is a flowchart illustrating an example of a process in the question estimating device of the first embodiment. FIG. 4 is a diagram showing an example of the conversation history database. FIG. 5 is a diagram showing an example of a time series of each piece of transmitted document data. FIG. 6 is a diagram illustrating an example of generation of a question and answer table in the question estimation device of the first embodiment. FIG. 7(A) is a block diagram showing an example of the configuration of a trained model generation device of embodiment 2, and FIG. 7(B) is a block diagram showing an example of the hardware configuration of the trained model generation device. FIG. 8 is a flowchart showing an example of processing in the trained model generation device of the second embodiment. FIG. 9 is a diagram showing an example of creating a question sentence estimation model in the third embodiment. FIG. 10 is a diagram showing an example of analyzing the similarity between a generated estimated question sentence and a question sentence already sent in an ongoing conversation in the third embodiment.

In the question estimation device of the present invention, for example,
the sentence type identification unit includes an unnecessary sentence identification unit, an unnecessary sentence deletion unit, a morphological analysis unit, and an identification unit;
The unnecessary sentence identification unit identifies unnecessary sentences in the text data,
the unnecessary sentence deletion unit deletes the identified unnecessary sentences from the text data;
The morphological analysis unit performs a morphological analysis on the text data on a sentence-by-sentence basis;
The identification unit may identify a sentence type of the sentence data based on the morphological analysis and in accordance with a preset sentence-ending expression rule.

In the question estimation device of the present invention, for example,
The question and answer table generating unit may generate a question and answer table by linking the sentence data of the question type with the sentence data of the answer type based on at least one of the following conditions (1) to (3):
(1) Linking the answer type text data to the question type text data that was sent earlier than the answer type text data. (2) Linking the question type text data sent by a user other than the user who sent the answer type text data. (3) When there are multiple question type text data to be linked with the answer type text data, calculating the similarity between the answer type text data and the question type text data, and linking the answer type text data and the question type text data with the highest similarity.

In the question estimation device of the present invention, for example,
the estimated question sentence generation unit inputs the acquired untransmitted answer sentence into a question sentence estimation model to generate an estimated question sentence corresponding to the untransmitted answer sentence;
The question sentence estimation model may be a model trained to generate an estimated question sentence corresponding to an input untransmitted answer sentence from the input untransmitted answer sentence.

The question estimation device of the present invention, for example,
Further, a learning processing unit is included,
the learning processing unit uses the question sentence in the question and answer table as training data and uses sentence data of the answer type linked to sentence data of the question type as input data to construct the question sentence estimation model through supervised machine learning;
The estimated question sentence generation unit may input the acquired unsent answer sentence to the constructed question sentence estimation model, and generate an estimated question sentence corresponding to the unsent answer sentence.

In the question estimation device of the present invention including the learning processing unit, for example,
The learning processing unit may input each piece of text data in the question and answer list into a vector conversion trained model that has been trained in an unsupervised manner to convert input text data into a vector before performing the supervised machine learning, converting each piece of text data into a vector, and using the text data of the question type converted into the vector as teacher data and the text data of the answer type converted into the vector as input data to construct the question text estimation model by supervised machine learning.

The question estimation device of the present invention, for example,
Further, the apparatus includes an output unit, a question sentence acquisition unit, and a similarity analysis unit,
the question sentence acquisition unit acquires question-type sentence data that has been sent in the ongoing conversation and has been sent by another user,
the similarity analysis unit compares the transmitted sentence data of the question type with the generated estimated question sentence to analyze the similarity;
The output unit may output the degree of similarity.

In the trained model generation device of the present invention, for example,
The data acquisition unit acquires conversation history data,
The conversation history data is composed of a plurality of sentence data,
The trained model generation unit may further generate a second trained model by unsupervised machine learning using the conversation history data, which, when input, converts the sentence data into a vector and outputs the sentence data.

In the question estimation method of the present invention, for example,
The sentence type identification step includes an unnecessary sentence identification step, an unnecessary sentence deletion step, a morphological analysis step, and an identification step,
The unnecessary sentence identification step includes identifying unnecessary sentences in the text data,
the unnecessary sentence deletion step deletes the identified unnecessary sentences from the text data;
The morphological analysis step performs a morphological analysis on the text data on a sentence-by-sentence basis;
The identification step may be performed by identifying a sentence type of the sentence data based on the morphological analysis and in accordance with a preset sentence-ending expression rule.

In the question estimation method of the present invention, for example,
The question and answer table generating step may be performed in a mode in which the question type sentence data and the answer type sentence data are linked to generate a question and answer table based on at least one of the following conditions (1) to (3):
(1) Linking the answer type text data to the question type text data that was sent earlier than the answer type text data. (2) Linking the question type text data sent by a user other than the user who sent the answer type text data. (3) When there are multiple question type text data to be linked with the answer type text data, calculating the similarity between the answer type text data and the question type text data, and linking the answer type text data and the question type text data with the highest similarity.

In the question estimation method of the present invention, for example,
the estimated question sentence generation step inputs the acquired untransmitted answer sentence into a question sentence estimation model to generate an estimated question sentence corresponding to the untransmitted answer sentence;
The question sentence estimation model may be a model trained to generate an estimated question sentence corresponding to an input untransmitted answer sentence from the input untransmitted answer sentence.

The question estimation method of the present invention includes, for example,
Further, a learning process is included,
the learning process includes constructing the question sentence estimation model by supervised machine learning using the question sentence in the question and answer table as training data and sentence data of the answer type linked to sentence data of the question type as input data;
The estimated question sentence generating step may be implemented in a manner that the acquired unsent answer sentence is input to the constructed question sentence estimation model to generate an estimated question sentence corresponding to the unsent answer sentence.

In the question estimating method of the present invention including the learning process, for example,
The learning process may be such that, before performing the supervised machine learning, each piece of text data in the question and answer list is input into a vector conversion trained model that has been trained in an unsupervised manner to convert input text data into a vector, and each piece of text data is converted into a vector, and the question text estimation model is constructed by supervised machine learning using the text data of the question type converted into the vector as teacher data and the text data of the answer type converted into the vector as input data.

The question estimation method of the present invention includes, for example,
The method further includes an output step, a question sentence acquisition step, and a similarity analysis step,
The question sentence acquisition step acquires question-type sentence data that has been sent in the ongoing conversation and has been sent by another user,
the similarity analysis step compares the transmitted sentence data of the question type with the generated estimated question sentence to analyze the similarity;
The output step may include outputting the similarity.

In the method for producing a trained model of the present invention, for example,
The data acquisition step acquires conversation history data,
The conversation history data is composed of a plurality of sentence data,
The trained model generation process may further comprise generating a second trained model by unsupervised machine learning using the conversation history data, which, when input, converts the sentence data into a vector and outputs the vector.

The program of the present invention is a program for causing a computer to execute each step of the method of the present invention as a procedure.

The recording medium of the present invention is a computer-readable recording medium on which the program of the present invention is recorded.

The application field of the present invention is not particularly limited, and may be any field in which a "question" and an "answer" to the question are transmitted and received in a conversation. In the present invention, a conversation refers to a conversation between two or more speakers. The conversation may be, for example, a conversation using text data such as chat or email, or a conversation using audio data such as a telephone call. The conversation begins with one user (questioner) presenting a question and ends with the other user (answerer) answering the question, and is considered to be one set. The relationship between the questioner and the answerer is not particularly limited, and may be, for example, a relationship between a customer and a person in charge, a student and a teacher, or an athlete and a coach. In the following description, a customer may be used as the questioner and a person in charge as the answerer, but this is not limiting.

Next, an embodiment of the present invention will be described with reference to the drawings. The present invention is not limited to the following embodiment. In each of the drawings, the same parts are given the same reference numerals. Furthermore, the explanations of each embodiment can be mutually incorporated unless otherwise specified, and the configurations of each embodiment can be combined unless otherwise specified.

[Embodiment 1]
Fig. 1 is a block diagram showing an example of the configuration of a question estimation device 10 of this embodiment. As shown in Fig. 1, the device 10 includes a conversation history data acquisition unit 11, a sentence type identification unit 12, a question and answer table generation unit 13, an unsent sentence acquisition unit 14, and an estimated question sentence generation unit 15. The device 10 may further include, for example, an output unit 16, a learning processing unit 17, a question sentence acquisition unit 18, and a similarity analysis unit 19 as optional components. The sentence type identification unit 12 may include, for example, an unnecessary sentence identification unit 121, an unnecessary sentence deletion unit 122, a morphological analysis unit 123, and an identification unit 124.

The device 10 may be, for example, a single device including each of the above-mentioned parts, or a device to which each of the above-mentioned parts can be connected via a communication line network. The device 10 may also be connected to an external device described later via the communication line network. The communication line network is not particularly limited and may be a publicly known network, for example, wired or wireless. Examples of the communication line network include the Internet line, WWW (World Wide Web), telephone line, LAN (Local Area Network), SAN (Storage Area Network), DTN (Delay Tolerant Networking), LPWA (Low Power Wide Area), L5G (Local 5G), etc. Examples of wireless communication include WiFi (Wireless Fidelity), Bluetooth (registered trademark), local 5G, and LPWA. The wireless communication may be a form in which each device communicates directly (Ad Hoc communication), infrastructure communication, indirect communication via an access point, and the like. The device 10 may be incorporated into a server as a system. The device 10 may be, for example, a personal computer (PC, for example, desktop type or notebook type) in which the program of the present invention is installed, a smartphone, a tablet terminal, or the like. Furthermore, the device 10 may be in the form of cloud computing or edge computing, for example, in which at least one of the parts is on a server and the other parts are on a terminal.

Figure 2 shows an example block diagram of the hardware configuration of the device 10. The device 10 includes, for example, a central processing unit (CPU, GPU, etc.) 101, a memory 102, a bus 103, a storage device 104, an input device 105, a display device 106, a communication device 107, etc. Each part of the device 10 is connected to each other via the bus 103 by its respective interface (I/F).

The central processing unit 101 is responsible for the overall control of the device 10. In the device 10, the central processing unit 101 executes, for example, the program of the present invention and other programs, and also reads and writes various information. Specifically, for example, the central processing unit 101 functions as a conversation history data acquisition unit 11, a sentence type identification unit 12, a question and answer table generation unit 13, an unsent sentence acquisition unit 14, and an estimated question sentence generation unit 15. The central processing unit 101 also functions as, for example, an output unit 16, a learning processing unit 17, a question sentence acquisition unit 18, and a similarity analysis unit 19.

The bus 103 can also be connected to, for example, an external device. Examples of the external device include an external storage device (external database, etc.), a printer, an external input device, an external display device, an external imaging device, etc. The present device 10 can be connected to an external network (the above-mentioned communication line network) by, for example, a communication device 107 connected to the bus 103, and can also be connected to other devices via the external network. The present device 10 can also be connected to a conversation history database 100 via the above-mentioned communication line network by, for example, a communication device 107 connected to the bus 103, as shown in FIG. 1.

The memory 102 may be, for example, a main memory (primary storage device). When the central processing unit 101 performs processing, the memory 102 reads various operating programs, such as the program of the present invention, stored in the storage device 104 described below, and the central processing unit 101 receives data from the memory 102 and executes the program. The main memory may be, for example, a RAM (random access memory). The memory 102 may also be, for example, a ROM (read only memory).

The storage device 104 is also referred to as an auxiliary storage device, for example, in contrast to the main memory. As described above, the storage device 104 stores operating programs including the program of the present invention. The storage device 104 may be, for example, a combination of a recording medium and a drive that reads and writes from the recording medium. The recording medium is not particularly limited, and may be, for example, an internal or external type, such as a hard disk drive (HDD), CD-ROM, CD-R, CD-RW, MO, DVD, flash memory, memory card, etc. The storage device 104 may be, for example, a hard disk drive (HDD) in which the recording medium and the drive are integrated, or a solid state drive (SSD).

In the device 10, the memory 102 and storage device 104 can also store user log information as well as information obtained from an external database (not shown).

The device 10 may further include, for example, an input device 105 and a display device 106. The input device 105 is, for example, a touch panel, a keyboard, a mouse, etc. The display device 106 is, for example, an LED display, a liquid crystal display, etc.

Next, an example of the question estimation method of this embodiment will be described with reference to the flowchart in FIG. 3. The question estimation method of this embodiment is implemented as follows, for example, using the question estimation device 10 of FIG. 1. Note that the question estimation method of this embodiment is not limited to use of the question estimation device 10 of FIG. 1.

First, the conversation history data acquisition unit 11 acquires conversation history data, which is character data of a finished conversation (S11). The conversation history data is composed of a plurality of sentence data. When the conversation is in the form of talking using voice data, the conversation history data acquisition unit 11 may acquire conversation history data by converting the voice data of the finished conversation into character data, for example. The conversation history data acquisition unit 11 may acquire the conversation history data from the conversation history database 100 that stores the conversation history data, for example, via the communication line network. An example of the conversation history database 100 is shown in FIG. 4. As shown in FIG. 4, the conversation history data may be stored as table data for each conversation. In addition, the conversation history data may include, for example, transmission information regarding the transmission time and the sender for each sentence data. A "completed conversation" means a conversation in which an answerer (a person in charge, etc.) answers a question from a questioner (a customer, etc.) that triggered the start of the conversation. Hereinafter, a "completed conversation" is also referred to as a "conversation that has been handled." The conversation history database 100 may, for example, distinguish and store conversation history data of conversations that have already been handled and conversation history data of conversations that are currently being handled (also called ongoing conversations). An "ongoing conversation" refers to a conversation in which the answerer (e.g., a person in charge) has not responded to a question from a questioner (e.g., a customer) that triggered the conversation. The time series of each piece of text data shown in FIG. 4 is shown in FIG. 5. As shown in FIG. 5, one person in charge, for example, is having conversations with multiple customers in parallel.

Next, the sentence type identification unit 12 identifies the sentence type for each of the sentence data (S12).

An example of the processing by the sentence type identification unit 12 (step (S12)) will be specifically described. The sentence type identification unit 12 includes, for example, an unnecessary sentence identification unit 121, an unnecessary sentence deletion unit 122, a morphological analysis unit 123, and an identification unit 124. First, the unnecessary sentence identification unit 121 identifies unnecessary sentences in the sentence data (S121). Specifically, for example, preset sentences such as "Thank you for your continued support" and "Thank you" and sentences that are highly similar to the preset sentences are identified as unnecessary sentences. The similarity is not particularly limited, but for example, the cosine similarity described below can be used. Next, the unnecessary sentence deletion unit 122 deletes the identified unnecessary sentences from the sentence data (S122). Next, the morphological analysis unit 123 performs morphological analysis on the sentence data on a sentence-by-sentence basis (S123). Next, the identification unit 124 identifies the sentence type of the sentence data based on the morphological analysis in accordance with preset sentence-ending expression rules (S124). The sentence-ending expression rule associates sentence-ending expressions with the sentence type, and can be set arbitrarily. Specifically, examples of the sentence-ending expressions associated with the question type include "Please...", "Please do...", "What is...?", "Will... happen?", etc. Examples of the sentence-ending expressions associated with the answer type include "It will be...", "It will solve...", etc. The sentence type identification unit 12, for example, executes the processes of steps (S121) to (S124) on all sentence data.

Next, the question and answer table generating unit 13 generates a question and answer table by linking the sentence data of the question type with the sentence data of the answer type based on the result of the identification (S13). The question and answer table may be stored, for example, in the memory 102 and the storage device 104. The question and answer table generating unit 13 may generate the question and answer table based, for example, on at least one of the following conditions (1) to (3):
(1) Linking the answer type text data to the question type text data that was sent earlier than the answer type text data. (2) Linking the question type text data sent by a user other than the user who sent the answer type text data. (3) When there are multiple question type text data to be linked with the answer type text data, calculating the similarity between the answer type text data and the question type text data, and linking the answer type text data and the question type text data with the highest similarity.

6 shows an example of the generation of the question and answer table. In FIG. 6, each piece of text data in the conversation history data is shown in order of transmission time. In each piece of text data shown in FIG. 6, text data identified by the text type identification unit 12 as a question type is shown as "Qn (n is an integer of 1 or more)" in order of transmission time, and text data identified as an answer type is shown as "Am (m is an integer of 1 or more)" in order of transmission time. For example, as shown in FIG. 6, the text data A1 of the answer type is linked to the text data Q2 of the question type that was sent before the transmission time t3 of the text data A1 and was sent by the sender Y different from the sender X of the text data A1 by the question and answer table generation unit 13 in accordance with the above conditions (1) and (2). The text data A2 and A3 of the answer type are also linked to the text data Q1 and Q3 of the question type, respectively, in the same manner as the text data A1. In this way, the text data of the answer type may be linked to, for example, multiple text data of the question type. On the other hand, when there are multiple pieces of question-type text data to be linked, such as answer-type text data A2 and A3, the similarity may be calculated according to the above condition (3), and the answer-type text data and the question-type text data with the high similarity may be linked. The similarity is not particularly limited, but may be, for example, cosine similarity. Specifically, for example, the similarity is calculated by vectorizing each piece of text data and calculating the cosine similarity. The higher the similarity, the more similar the pieces of text data are. Specifically, in the case of answer-type text data A2, the similarity in the combination of text data A2 and text data Q1, and the similarity in the combination of text data A2 and text data Q3 are calculated, and the combination with the high similarity is linked.

The similarity will be described in more detail by taking cosine similarity as an example. First, the question estimation device 10 performs morphological analysis on each piece of text data and divides each piece of text data into words. Next, the question estimation device 10 calculates the number of occurrences of each word for each piece of text data, and converts the text data into a vector whose length is the number of words (also called vocabulary count, not the number of occurrences). The process up to this point is called, for example, vectorization (embedding) of text, and techniques such as TF-IDF (Term Frequency-Inverse Document Frequency) can be used. Then, the vector is used to calculate cosine similarity. The cosine similarity is an index that indicates the degree to which the vectors to be compared are oriented in the same direction, and for example, its value ranges from -1 to 1, and the closer it is to 1, the more it can be said that the vectors are oriented in the same direction. In other words, the closer the value is to 1, the higher the similarity of the text data to be compared is determined to be.

Next, the unsent text acquisition unit 14 acquires unsent answer texts in the ongoing conversation that have not yet been sent to other users (S14). The unsent text acquisition unit 14 may acquire the unsent answer texts from an external input device by the answerer (e.g., a person in charge, etc.) via the communication line network, for example.

Then, the estimated question sentence generation unit 15 generates an estimated question sentence corresponding to the unsent answer sentence based on the question and answer table and the unsent answer sentence (S15), and ends (END). The estimated question sentence generation unit 15 may, for example, input the acquired unsent answer sentence into a question sentence estimation model to generate an estimated question sentence corresponding to the unsent answer sentence. The question sentence estimation model is a model that has been trained to generate an estimated question sentence corresponding to the unsent answer sentence from the input unsent answer sentence. The question sentence estimation model may, for example, be acquired from the outside, or may be constructed by the learning processing unit 17 when the device 10 includes a learning processing unit 17. The question sentence estimation model includes, for example, an input layer that inputs the unsent answer sentence, an output layer that outputs the estimated question sentence, and at least one intermediate layer provided between the input layer and the output layer. The question sentence estimation model may, for example, be a program module that is a part of artificial intelligence software. The multi-layered network may, for example, be a neural network. The neural network may be, for example, a convolution neural network (CNN). Note that the question sentence estimation model is not limited to the CNN, and may be, for example, a neural network other than CNN, or a trained model constructed using other learning algorithms such as a support vector machine (SVM), a Bayesian network, or a regression tree.

The learning processing unit 17 will be described. After the step (S13), the learning processing unit 17 constructs the question sentence estimation model by supervised machine learning using the question sentence in the question and answer table as teacher data and the answer type sentence data linked to the question type sentence data as input data (S17). The supervised machine learning is, for example, machine learning using CNN, SVM, Bayesian network, regression tree, etc., as described above. For example, before performing the supervised machine learning, the learning processing unit 17 may input each sentence data of the question and answer table to a vector conversion trained model that has been trained in an unsupervised manner to convert the input sentence data into a vector, and convert each sentence data into a vector. Then, the learning processing unit 17 may construct the question sentence estimation model by supervised machine learning using the question type sentence data converted into the vector as teacher data and the answer type sentence data converted into the vector as input data. The vector conversion trained model may be obtained, for example, from outside, or may be constructed by the learning processing unit 17. That is, the learning processing unit 17 may be configured with two phases, for example, a pre-training phase in which the unsupervised learning is performed, and a fine tuning phase in which supervised machine learning is performed thereafter. Specifically, the pre-training is, for example, a process of masking words in a sentence from a large amount of unlabeled sentence data (sentence data of conversation history data) and learning to infer the masked words. Also, specifically, the fine tuning is, for example, a process of using the vector conversion trained model as an initial value and re-training using a small amount of labeled data (the question and answer table) to improve the accuracy of the generated estimated question sentence.

For example, after the step (S15), the output unit 16 may output the generated estimated question sentence (S16) and terminate (END). The output unit 16 may output the estimated question sentence to an external device via the communication line network, or may output the estimated question sentence to the display device 106, for example.

Furthermore, for example, the question sentence acquisition unit 18 may acquire sentence data of a question type that has already been sent in the ongoing conversation and that has been sent by another user (S18). The timing of executing the step (S18) is not particularly limited. For example, the question sentence acquisition unit 18 may acquire all sentence data that has already been sent in the ongoing conversation, identify the sentence type of the acquired sentence data by the sentence type identification unit 12 in the same manner as described above, and acquire only the sentence data identified as the question type (also referred to as the sent question sentence).

When the step (S18) is executed, after the steps (S15) and (S18), the similarity analysis unit 19 compares the transmitted question type text data with the generated estimated question text to analyze the similarity (S19). The similarity can be, for example, a cosine similarity.

Then, the output unit 16 outputs the similarity (S20) and ends the process (END). For example, if the similarity is equal to or lower than a preset value, the output may output a warning indicating that the similarity is low. Furthermore, the output unit 16 may be controllable to disable output (transmission) of the unsent response document if the similarity is equal to or lower than a preset value.

According to this embodiment, by generating an estimated question sentence, it is possible to use the generated question sentence to check the similarity between the question sentence in an ongoing conversation and the estimated question sentence, and to prevent erroneous transmission. Also, according to this embodiment, for example, the similarity can be calculated, so that it is easy to check the similarity between the question sentence in an ongoing conversation and the estimated question sentence. Furthermore, according to this embodiment, it is possible to generate a more accurate estimated question sentence by using two phases: a pre-training phase and a fine tuning phase that is performed after the pre-training phase.

[Embodiment 2]
7A is a block diagram showing an example of a configuration of a trained model generation device 20 according to the present embodiment. As shown in FIG. 7A, the device 20 includes a data acquisition unit 21 and a trained model generation unit 22.

The device 20 may be, for example, a single device including the above-mentioned units, or a device to which the above-mentioned units can be connected via the communication line network. The device 20 may also be connectable to the above-mentioned external device via the communication line network. The device 20 may be, for example, incorporated into a server as a system. The device 20 may also be, for example, a personal computer (PC, for example, desktop type or notebook type) in which the program of the present invention is installed, a smartphone, a tablet terminal, or the like. Furthermore, the device 20 may be in the form of cloud computing or edge computing, for example, in which at least one of the above-mentioned units is on a server and the other units are on a terminal.

Figure 7 (B) illustrates an example block diagram of the hardware configuration of the trained model generation device 20. The trained model generation device 20 includes, for example, a central processing unit (CPU, GPU, etc.) 101, a memory 102, a bus 103, a storage device 104, an input device 105, a display device 106, a communication device 107, etc. Unless otherwise specified, the descriptions of the respective parts of the hardware configuration of the trained model generation device 20 can be used for the respective parts of the hardware configuration of the question estimation device 10 shown in Figure 3.

The central processing unit 101 functions, for example, as a data acquisition unit 21 and a trained model generation unit 22.

The memory 102 and storage device 104 may store the trained model generated by the device 20.

Next, an example of a method for producing a trained model according to this embodiment will be described with reference to the flowchart of FIG. 8. The method for producing a trained model according to this embodiment is implemented as follows, for example, using the trained model generation device 20 of FIG. 7. Note that the method for producing a trained model according to this embodiment is not limited to the use of the trained model generation device 20 of FIG. 7.

First, the data acquisition unit 21 acquires question-type sentence data linked to answer-type sentence data as teacher data (S21). Next, the trained model generation unit 22 generates a first trained model by machine learning using the teacher data, which generates and outputs an estimated question sentence corresponding to the answer-type sentence data when the answer-type sentence data is input (S22), and ends (END). The first trained model may be used as a question sentence estimation model in the question estimation device 10 described in the first embodiment, for example.

In other words, the device 20 can be said to construct (generate) a neural network that uses the answer type text data as input and outputs the estimated question text by performing machine learning. The neural network may be, for example, the same as described above. The neural network may include, for example, an input layer that accepts the answer type text data, an output layer that outputs the estimated question text, and at least one intermediate layer provided between the input layer and the output layer.

The input layer includes, for example, a plurality of neurons that accept input of sentence data of the answer type, and passes the input sentence data of the answer type to the intermediate layer. The intermediate layer includes a plurality of neurons that extract features of the sentence data of the answer type, and passes the extracted features to the output layer. The output layer includes one or more neurons that output an estimated question sentence, and generates the estimated question sentence based on the features output from the intermediate layer.

The teacher data and the input answer-type sentence data may be subjected to, for example, morphological analysis (Natural Language Processing (NLP)) before being subjected to the machine learning and first trained model.

The data acquisition unit 21 may further acquire conversation history data, for example. The conversation history data is similar to that described above and is composed of a plurality of sentence data. The trained model generation unit 22 may further generate a second trained model that converts the sentence data into a vector and outputs it when the sentence data is input, for example, by unsupervised machine learning using the conversation history data. The second trained model may be used, for example, as a vector conversion trained model in the question estimation device 10 described in the first embodiment. That is, the device 20 may generate two trained models, the question sentence estimation model and the vector conversion trained model.

The first and second trained models generated by this embodiment are used, for example, in the question estimation device 10 described in the first embodiment. This makes it possible to generate estimated question sentences and prevent erroneous transmissions.

[Embodiment 3]
An example of creating a question sentence estimation model will be specifically described with reference to FIG.

9 is a diagram showing an example of creating a question sentence estimation model. As shown in FIG. 9, the question estimation device 10 uses the acquired conversation history data by the question-answer table generation unit 13 to link the question type sentence data with the answer type sentence data to generate a question-answer table (QA table). On the other hand, the question estimation device 10 uses the acquired conversation history data by the learning processing unit 17 to perform unsupervised machine learning (pre-training) and construct the vector conversion trained model. The vector conversion trained model may be obtained from an external source, as described above, or may use a trained model generated by the trained model generation device 20. Then, the question estimation device 10 inputs each piece of sentence data in the QA table by the learning processing unit 17 and converts each piece of sentence data into a vector. The QA table may be, for example, data that links and stores each piece of sentence data with the vector, as shown in FIG. 9. Next, the question estimation device 10 performs natural language processing on the question type sentence data in the QA table using the learning processing unit 17, and performs supervised machine learning (fine-tuning) using the question type sentence data as training data and the answer type sentence data linked to the question type sentence data as input data, thereby constructing the question sentence estimation model. For example, the question sentence estimation model may use a trained model generated by the trained model generation device 20 as described above.

An example of outputting similarity will be specifically explained using Figure 10.

Figure 10 is a diagram showing an example of analyzing the similarity between the generated estimated question sentence and a sent question sentence in an ongoing conversation. The respondent terminal 30 shown in Figure 10 is a conversation support device on the respondent's side, and is, for example, a device that performs an exchange (dialogue/conversation) between the questioner and the respondent. For example, as shown in Figure 10, a plurality of sentence data in the ongoing conversation is displayed on the display (display device) of the respondent terminal 30. In addition, an input field is displayed on the display. The respondent inputs sentence data into the input field using an input device. In Figure 10, the sentence data input into the input field is answer-type sentence data, but sentence data corresponding to a question may also be input. The question estimation device 10 may, for example, acquire unsent sentence data input into the input field by the unsent sentence acquisition unit 14, and identify the sentence type of the sentence data input into the input field by the sentence type identification unit 12. Then, when the text type of the text data entered in the input field is identified as an answer type, the question estimation device 10 executes, for example, the generation of an estimated question text by the estimated question text generation unit 15, the acquisition of the transmitted answer text by the question text acquisition unit 18, and the analysis of similarity by the similarity analysis unit 19. On the other hand, the question estimation device 10 may acquire a request for the analysis of the similarity from, for example, the answerer terminal 30. Specifically, the answerer who inputs the text data of the answer type presses, for example, "verify validity" displayed on the display. Then, the request is transmitted from the answerer terminal 30 to the question estimation device 10. The question estimation device 10 that has acquired the request acquires the text data of the answer type (untransmitted answer text) entered in the input field by the untransmitted text acquisition unit 14. Next, as described above, the question estimation device 10 executes the generation of an estimated question text by the estimated question text generation unit 15, the acquisition of the transmitted answer text by the question text acquisition unit 18, and the analysis of similarity by the similarity analysis unit 19. Then, the output unit 16 outputs the similarity to the respondent terminal 30 as "legitimacy" indicating the degree of likelihood that the unsent answer sentence is a transmission target for the conversation. The output similarity (legitimacy) is displayed, for example, on the display of the respondent terminal 30. This allows the respondent to check the degree of legitimacy displayed on the display, which helps prevent erroneous transmission.

[Embodiment 4]
The program of the present embodiment is a program for causing a computer to execute each step of the method of the present invention as a procedure. In the present invention, the "procedure" may be read as "processing." The program of the present embodiment may be recorded, for example, in a computer-readable recording medium. The recording medium is not particularly limited, and examples thereof include a read-only memory (ROM), a hard disk (HD), and an optical disk.

The present invention has been described above with reference to the embodiments, but the present invention is not limited to the above embodiments. Various modifications that can be understood by a person skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

<Additional Notes>
Some or all of the above embodiments may be described as follows, but are not limited to the following:
(Appendix 1)
The system includes a conversation history data acquisition unit, a sentence type identification unit, a question and answer table generation unit, an unsent sentence acquisition unit, and an estimated question sentence generation unit,
the conversation history data acquisition unit acquires conversation history data, which is character data of a completed conversation;
The conversation history data is composed of a plurality of sentence data,
The text type identification unit identifies a text type for each of the text data,
The sentence types include a question type and an answer type,
the question and answer table generating unit generates a question and answer table by linking the sentence data of the question type with the sentence data of the answer type based on a result of the identification;
The unsent text acquisition unit acquires an unsent reply text that has not yet been transmitted to another user in an ongoing conversation,
The question estimation device, wherein the estimated question sentence generation unit generates an estimated question sentence corresponding to the untransmitted answer sentence based on the question and answer table and the untransmitted answer sentence.
(Appendix 2)
the sentence type identification unit includes an unnecessary sentence identification unit, an unnecessary sentence deletion unit, a morphological analysis unit, and an identification unit;
The unnecessary sentence identification unit identifies unnecessary sentences in the text data,
the unnecessary sentence deletion unit deletes the identified unnecessary sentences from the text data;
The morphological analysis unit performs a morphological analysis on the text data on a sentence-by-sentence basis;
The question prediction device according to claim 1, wherein the identification unit identifies a sentence type of the sentence data based on the morphological analysis and in accordance with a preset sentence-ending expression rule.
(Appendix 3)
3. The question estimation device according to claim 1, wherein the question and answer table generation unit generates a question and answer table by linking the sentence data of the question type with the sentence data of the answer type based on at least one of the following conditions (1) to (3):
(1) Linking the answer type text data to the question type text data that was sent earlier than the answer type text data. (2) Linking the question type text data sent by a user other than the user who sent the answer type text data. (3) When there are multiple pieces of question type text data to be linked with the answer type text data, calculating the similarity between the answer type text data and the question type text data, and linking the answer type text data and the question type text data with the highest similarity (Appendix 4).
the estimated question sentence generation unit inputs the acquired untransmitted answer sentence into a question sentence estimation model to generate an estimated question sentence corresponding to the untransmitted answer sentence;
4. The question estimation device according to claim 1, wherein the question sentence estimation model is a model trained to generate an estimated question sentence corresponding to an input untransmitted answer sentence from the input untransmitted answer sentence.
(Appendix 5)
Further, a learning processing unit is included,
the learning processing unit uses the question sentence in the question and answer table as teacher data and uses sentence data of the answer type linked to sentence data of the question type as input data to construct the question sentence estimation model through supervised machine learning;
5. The question estimation device according to claim 1, wherein the estimated question sentence generation unit inputs the acquired unsent answer sentence into the constructed question sentence estimation model to generate an estimated question sentence corresponding to the unsent answer sentence.
(Appendix 6)
the learning processing unit inputs each piece of text data in the question and answer list into a vector conversion trained model that has been trained in an unsupervised manner to convert input text data into a vector before performing the supervised machine learning, converts each piece of text data into a vector, and uses the text data of the question type converted into the vector as supervised data and the text data of the answer type converted into the vector as input data to construct the question text estimation model by supervised machine learning.
(Appendix 7)
Further, the apparatus includes an output unit, a question sentence acquisition unit, and a similarity analysis unit,
the question sentence acquisition unit acquires question-type sentence data that has been sent in the ongoing conversation and has been sent by another user,
the similarity analysis unit compares the transmitted sentence data of the question type with the generated estimated question sentence to analyze the similarity;
7. The question estimation device according to claim 1, wherein the output unit outputs the similarity.
(Appendix 8)
The data acquisition unit and the trained model generation unit are included.
the data acquisition unit acquires question-type sentence data linked to answer-type sentence data as teacher data;
The trained model generation unit is a trained model generation device that generates a first trained model by machine learning using the teacher data, which generates and outputs an estimated question sentence corresponding to the answer type sentence data when the answer type sentence data is input.
(Appendix 9)
The data acquisition unit acquires conversation history data,
The conversation history data is composed of a plurality of sentence data,
The trained model generation device of claim 8, wherein the trained model generation unit further generates a second trained model by unsupervised machine learning using the conversation history data, which, when input, converts the sentence data into a vector and outputs the sentence data.
(Appendix 10)
The method includes a conversation history data acquisition step, a sentence type identification step, a question and answer table generation step, an unsent sentence acquisition step, and an estimated question sentence generation step,
The conversation history data acquisition step acquires conversation history data, which is character data of a completed conversation,
The conversation history data is composed of a plurality of sentence data,
The text type identification step includes identifying a text type for each of the text data,
The sentence types include a question type and an answer type,
the question and answer table generating step links the sentence data of the question type with the sentence data of the answer type based on a result of the identification to generate a question and answer table;
The unsent text acquisition step includes acquiring an unsent reply text that has not yet been transmitted to another user in an ongoing conversation,
The question estimating method includes generating an estimated question sentence corresponding to the untransmitted answer sentence based on the question and answer table and the untransmitted answer sentence, in the estimated question sentence generating step.
(Appendix 11)
The sentence type identification step includes an unnecessary sentence identification step, an unnecessary sentence deletion step, a morphological analysis step, and an identification step,
The unnecessary sentence identification step includes identifying unnecessary sentences in the text data,
the unnecessary sentence deletion step deletes the identified unnecessary sentences from the text data;
The morphological analysis step performs a morphological analysis on the text data on a sentence-by-sentence basis;
The question estimating method according to claim 10, wherein the identifying step identifies a sentence type of the sentence data based on the morphological analysis and in accordance with a preset sentence-ending expression rule.
(Appendix 12)
12. The question estimating method according to claim 10, wherein the question and answer table generating step generates a question and answer table by linking the sentence data of the question type with the sentence data of the answer type based on at least one of the following conditions (1) to (3):
(1) Linking the answer type text data to the question type text data that was sent earlier than the answer type text data. (2) Linking the question type text data sent by a user other than the user who sent the answer type text data. (3) When there are multiple pieces of question type text data to be linked with the answer type text data, calculating the similarity between the answer type text data and the question type text data, and linking the answer type text data and the question type text data that have the highest similarity (Appendix 13).
the estimated question sentence generation step inputs the acquired untransmitted answer sentence into a question sentence estimation model to generate an estimated question sentence corresponding to the untransmitted answer sentence;
The question estimation method according to any one of appendices 10 to 12, wherein the question sentence estimation model is a model trained to generate an estimated question sentence corresponding to an input unsent answer sentence from the input unsent answer sentence.
(Appendix 14)
Further, a learning process is included,
the learning process includes constructing the question sentence estimation model by supervised machine learning using the question sentence in the question and answer table as training data and sentence data of the answer type linked to sentence data of the question type as input data;
The question estimating method according to any one of Appendices 10 to 13, wherein the estimated question sentence generating step inputs the acquired unsent answer sentence into the constructed question sentence estimation model to generate an estimated question sentence corresponding to the unsent answer sentence.
(Appendix 15)
The question estimation method according to claim 14, wherein the learning process step inputs each piece of text data in the question and answer list into a vector conversion trained model that has been trained in an unsupervised manner to convert input text data into a vector before performing the supervised machine learning, converts each piece of text data into a vector, and uses the text data of the question type converted into the vector as teacher data and the text data of the answer type converted into the vector as input data to construct the question text estimation model by supervised machine learning.
(Appendix 16)
The method further includes an output step, a question sentence acquisition step, and a similarity analysis step,
The question sentence acquisition step acquires question-type sentence data that has been sent in the ongoing conversation and has been sent by another user,
the similarity analysis step compares the transmitted sentence data of the question type with the generated estimated question sentence to analyze the similarity;
16. The question estimating method according to any one of appendices 10 to 15, wherein the outputting step outputs the similarity.
(Appendix 17)
The method includes a data acquisition process and a trained model generation process.
The data acquisition step includes acquiring question-type sentence data linked to answer-type sentence data as teacher data,
The trained model generation process is a trained model production method that generates a first trained model by machine learning using the training data, which generates and outputs an estimated question sentence corresponding to the answer type sentence data when the answer type sentence data is input.
(Appendix 18)
The data acquisition step acquires conversation history data,
The conversation history data is composed of a plurality of sentence data,
The trained model generation process further generates a second trained model by unsupervised machine learning using the conversation history data, which converts the sentence data into a vector and outputs the sentence data when the sentence data is input.
(Appendix 19)
A program for causing a computer to execute steps including a conversation history data acquisition step, a sentence type identification step, a question and answer table generation step, an unsent sentence acquisition step, and an estimated question sentence generation step:
The conversation history data acquisition step acquires conversation history data, which is character data of a completed conversation;
The conversation history data is composed of a plurality of sentence data,
The text type identification step includes identifying a text type for each of the text data;
The sentence types include a question type and an answer type,
the step of generating a question and answer table includes linking the sentence data of the question type with the sentence data of the answer type based on a result of the identification to generate a question and answer table;
The unsent text acquisition step includes acquiring an unsent reply text that has not yet been transmitted to another user in an ongoing conversation;
The estimated question sentence generating step generates an estimated question sentence corresponding to the untransmitted answer sentence based on the question and answer table and the untransmitted answer sentence.
(Appendix 20)
The sentence type identification step includes an unnecessary sentence identification step, an unnecessary sentence deletion step, a morphological analysis step, and a classification step;
The unnecessary sentence identification step includes identifying unnecessary sentences in the text data;
the unnecessary sentence deletion step deletes the identified unnecessary sentences from the sentence data;
The morphological analysis step performs a morphological analysis on the text data on a sentence-by-sentence basis;
20. The program according to claim 19, wherein the identification step identifies a sentence type of the sentence data based on the morphological analysis and in accordance with a preset sentence-ending expression rule.
(Appendix 21)
The program according to claim 19 or 20, wherein the question and answer table generating step generates a question and answer table by linking the sentence data of the question type with the sentence data of the answer type based on at least one of the following conditions (1) to (3):
(1) Linking the answer type text data to the question type text data that was sent earlier than the answer type text data. (2) Linking the question type text data sent by a user other than the user who sent the answer type text data. (3) When there are multiple pieces of question type text data to be linked with the answer type text data, calculating the similarity between the answer type text data and the question type text data, and linking the answer type text data and the question type text data with the highest similarity (Appendix 22).
the estimated question sentence generation step inputs the acquired untransmitted answer sentence into a question sentence estimation model to generate an estimated question sentence corresponding to the untransmitted answer sentence;
A program described in any one of appendices 19 to 21, wherein the question sentence estimation model is a model trained to generate an estimated question sentence corresponding to an input unsent answer sentence from the input unsent answer sentence.
(Appendix 23)
Further, a learning procedure is included,
the learning process step includes constructing the question sentence estimation model by supervised machine learning using the question sentence in the question and answer table as training data and sentence data of the answer type linked to sentence data of the question type as input data;
The program according to any one of Appendices 19 to 22, wherein the estimated question sentence generation step inputs the acquired unsent answer sentence into the constructed question sentence estimation model to generate an estimated question sentence corresponding to the unsent answer sentence.
(Appendix 24)
The program described in Appendix 23, wherein the learning process procedure includes inputting each piece of sentence data in the question and answer list into a vector conversion trained model that has been trained in an unsupervised manner to convert input sentence data into a vector before performing the supervised machine learning, converting each piece of sentence data into a vector, and using the sentence data of the question type converted into the vector as teacher data and the sentence data of the answer type converted into the vector as input data to construct the question sentence estimation model by supervised machine learning.
(Appendix 25)
Further, the method includes an output step, a question sentence acquisition step, and a similarity analysis step,
The question sentence acquisition step acquires question-type sentence data that has been sent in the ongoing conversation and has been sent by another user,
the similarity analysis step compares the transmitted question type sentence data with the generated estimated question sentence to analyze the similarity;
25. The program according to any one of appendices 19 to 24, wherein the output step outputs the similarity.
(Appendix 26)
A program for causing a computer to execute a procedure including a data acquisition procedure and a trained model generation procedure:
The data acquisition step includes acquiring question-type sentence data linked to answer-type sentence data as training data;
The trained model generation procedure generates, through machine learning using the training data, a first trained model that, when inputted with the answer type of sentence data, generates and outputs an estimated question sentence corresponding to the answer type of sentence data.
(Appendix 27)
The data acquisition step acquires conversation history data;
The conversation history data is composed of a plurality of sentence data,
The program described in Appendix 26, wherein the trained model generation procedure further generates a second trained model by unsupervised machine learning using the conversation history data, which, when the sentence data is input, converts the sentence data into a vector and outputs it.
(Appendix 28)
A computer-readable recording medium having a program according to any one of appendices 19 to 27 recorded thereon.

The present invention can prevent erroneous sending of answers. Therefore, the present invention is particularly useful in cases where one person answers questions from multiple customers.

10 Question estimation device 11 Conversation history data acquisition unit 12 Sentence type identification unit 13 Question and answer table generation unit 14 Untransmitted sentence acquisition unit 15 Estimated question sentence generation unit 16 Output unit 17 Learning processing unit 18 Question sentence acquisition unit 19 Similarity analysis unit 20 Trained model generation device 21 Data acquisition unit 22 Trained model generation unit 100 Conversation history database 101 Central processing unit 102 Memory 103 Bus 104 Storage device 105 Input device 106 Display device 107 Communication device

Claims (10)

The system includes a conversation history data acquisition unit, a sentence type identification unit, a question and answer table generation unit, an unsent sentence acquisition unit, and an estimated question sentence generation unit,
the conversation history data acquisition unit acquires conversation history data, which is character data of a completed conversation;
The conversation history data is composed of a plurality of sentence data,
The text type identification unit identifies a text type for each of the text data,
The sentence types include a question type and an answer type,
the question and answer table generating unit generates a question and answer table by linking the sentence data of the question type with the sentence data of the answer type based on a result of the identification;
The unsent text acquisition unit acquires an unsent reply text that has not yet been transmitted to another user in an ongoing conversation,
The question estimation device, wherein the estimated question sentence generation unit generates an estimated question sentence corresponding to the untransmitted answer sentence based on the question and answer table and the untransmitted answer sentence. the sentence type identification unit includes an unnecessary sentence identification unit, an unnecessary sentence deletion unit, a morphological analysis unit, and an identification unit;
The unnecessary sentence identification unit identifies unnecessary sentences in the text data,
the unnecessary sentence deletion unit deletes the identified unnecessary sentences from the text data;
The morphological analysis unit performs a morphological analysis on the text data on a sentence-by-sentence basis;
The question prediction device according to claim 1 , wherein the identification unit identifies a sentence type of the sentence data based on the morphological analysis and in accordance with a preset sentence-ending expression rule. the estimated question sentence generation unit inputs the acquired untransmitted answer sentence into a question sentence estimation model to generate an estimated question sentence corresponding to the untransmitted answer sentence;
The question estimation device according to claim 1 , wherein the question sentence estimation model is a model trained to generate an estimated question sentence corresponding to an input untransmitted answer sentence from the input untransmitted answer sentence. Further, a learning processing unit is included,
the learning processing unit uses sentence data of the question type in the question and answer table as teacher data and sentence data of the answer type linked to the sentence data of the question type as input data to construct the question sentence estimation model through supervised machine learning;
The question estimation device according to claim 3 , wherein the estimated question sentence generation unit inputs the acquired untransmitted answer sentence to the constructed question sentence estimation model to generate an estimated question sentence corresponding to the untransmitted answer sentence. Further, the apparatus includes an output unit, a question sentence acquisition unit, and a similarity analysis unit,
the question sentence acquisition unit acquires question-type sentence data that has been sent in the ongoing conversation and has been sent by another user,
the similarity analysis unit compares the transmitted sentence data of the question type with the generated estimated question sentence to analyze the similarity;
The question estimation device according to claim 1 , wherein the output unit outputs the similarity. The data acquisition unit and the trained model generation unit are included.
the data acquisition unit acquires question-type sentence data linked to answer-type sentence data as teacher data;
The trained model generation unit is a trained model generation device that generates a first trained model by machine learning using the teacher data, which generates and outputs an estimated question sentence corresponding to the answer type sentence data when the answer type sentence data is input. The method includes a conversation history data acquisition step, a sentence type identification step, a question and answer table generation step, an unsent sentence acquisition step, and an estimated question sentence generation step,
The conversation history data acquisition step acquires conversation history data, which is character data of a completed conversation,
The conversation history data is composed of a plurality of sentence data,
The text type identification step includes identifying a text type for each of the text data,
The sentence types include a question type and an answer type,
the question and answer table generating step links the sentence data of the question type with the sentence data of the answer type based on a result of the identification to generate a question and answer table;
The unsent text acquisition step includes acquiring an unsent reply text that has not yet been transmitted to another user in an ongoing conversation,
the estimated question sentence generating step generates an estimated question sentence corresponding to the untransmitted answer sentence based on the question and answer table and the untransmitted answer sentence.
A method of query estimation in which each step is carried out by a computer . The method includes a data acquisition process and a trained model generation process.
The data acquisition step includes acquiring question-type sentence data linked to answer-type sentence data as teacher data,
The trained model generation process is a trained model production method that generates a first trained model by machine learning using the training data, which generates and outputs an estimated question sentence corresponding to the answer type sentence data when the answer type sentence data is input. A program for causing a computer to execute steps including a conversation history data acquisition step, a sentence type identification step, a question and answer table generation step, an unsent sentence acquisition step, and an estimated question sentence generation step:
The conversation history data acquisition step acquires conversation history data, which is character data of a completed conversation;
The conversation history data is composed of a plurality of sentence data,
The text type identification step includes identifying a text type for each of the text data;
The sentence types include a question type and an answer type,
the step of generating a question and answer table includes linking the sentence data of the question type with the sentence data of the answer type based on a result of the identification to generate a question and answer table;
The unsent text acquisition step includes acquiring an unsent reply text that has not yet been transmitted to another user in an ongoing conversation;
The estimated question sentence generating step generates an estimated question sentence corresponding to the untransmitted answer sentence based on the question and answer table and the untransmitted answer sentence. A program for causing a computer to execute a procedure including a data acquisition procedure and a trained model generation procedure:
The data acquisition step includes acquiring question-type sentence data linked to answer-type sentence data as teacher data;
The trained model generation procedure generates a first trained model by machine learning using the training data, which, when inputted, generates and outputs an estimated question sentence corresponding to the answer type sentence data.

Images