Interview support apparatus, interview support method, and interview support program

Abstract

To present question candidates in an interview without performing difficult preparatory work.SOLUTION: The knowledge structuring unit 114 extracts a word corresponding to a term of the target field of the interview among words included in the utterance of the respondent from the respondent utterance data indicating the utterance of the respondent to the question in the interview as an attention word. The topic generation unit 115 generates a question candidate for the answerer based on the word of interest. The information presentation unit 116 presents the generated question candidates.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present disclosure relates to a technique for supporting interviews. [Background technology]

[0002] An interview is conducted to find out information that a person has. Before an interview, preparation is done to determine what you want to find out from the respondent. In an interview, it is necessary to extract the desired information from the other person through dialogue, so the interviewer needs to have appropriate skills and experience. Therefore, various support devices have been proposed to enable people with limited interview skills and experience to conduct appropriate interviews.

[0003] Patent Document 1 discloses a device for presenting question candidates. For this device, a general set of questions is determined in advance, assuming an interview technique called a semi-structured interview. This device estimates the topics that respondents are most interested in based on the content of the interview dialogue and presents candidate questions.

[0004] In order to use the device of Patent Document 1, it is essential to carry out preliminary work to determine appropriate questions in advance. If appropriate question items cannot be determined in advance, the device of Patent Document 1 cannot present appropriate question candidates. In order to determine appropriate questions in the preliminary work, the questioner needs to have appropriate knowledge about the information (target information) that he or she wants to obtain from the respondent. For example, if the respondent is a skilled technician, the target information will be the tacit knowledge that the respondent possesses. In this case, it is difficult to determine appropriate questions about tacit knowledge in advance, even if the respondent has sufficient knowledge about tacit knowledge that has not yet become explicit knowledge. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Patent No. 5258056 Summary of the Invention [Problem to be solved by the invention]

[0006] The present disclosure aims to enable candidate questions to be presented in an interview without the need for difficult advance work. [Means for solving the problem]

[0007] The interview support device of the present disclosure includes: a knowledge structuring unit that extracts, as noteworthy words, words corresponding to terms in the subject field of the interview from among words included in the statements of the respondent from respondent statement data indicating the statements of the respondent in response to questions in the interview; a topic generation unit that generates candidate questions for the respondent based on the attention words; an information presenting unit that presents the generated question candidates; Equipped with. [Effects of the Invention]

[0008] The present disclosure aims to enable candidate questions to be presented during an interview without the need for difficult preparatory work. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a configuration diagram of an interview support system 200 according to the first embodiment. [Figure 2] FIG. 1 is a configuration diagram of an interview support device 100 according to a first embodiment. [Figure 3] FIG. 1 is a functional configuration diagram of an interview support system 200 according to the first embodiment. [Figure 4] 3 is a state transition diagram of the interview support device 100 according to the first embodiment. [Figure 5]3 is a flowchart of the interview support method according to the first embodiment. [Figure 6] 4 is a table showing an example of a result of morphological analysis in the first embodiment. [Figure 7] 4 is a table showing an example of cosine similarity between words according to the first embodiment. [Figure 8] FIG. 2 is a diagram showing an example of a dependency structure (word by word) according to the first embodiment. [Figure 9] FIG. 2 shows an example of a dependency structure (phrase unit) in the first embodiment. [Figure 10] FIG. 3 shows an example of an information presentation screen 300 according to the first embodiment. [Figure 11] FIG. 3 shows an example of an information presentation screen 300 according to the first embodiment. [Figure 12] FIG. 3 shows an example of an information presentation screen 300 according to the first embodiment. [Figure 13] FIG. 3 shows an example of an information presentation screen 300 according to the first embodiment. [Figure 14] FIG. 2 is a diagram showing an example of generating candidate questions using LLM according to the first embodiment. [Figure 15] FIG. 2 is a diagram showing an example of generating candidate questions using LLM according to the first embodiment. [Figure 16] FIG. 10 is a functional configuration diagram of an interview support system 200 according to a second embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0010] In the embodiments and drawings, the same or corresponding elements are denoted by the same reference numerals. The description of elements denoted by the same reference numerals as those already described will be omitted or simplified as appropriate. Arrows in the drawings primarily indicate the flow of data or the flow of processing.

[0011] Embodiment 1 The interview support system 200 will be described with reference to FIGS.

[0012] ***Configuration Description*** The configuration of the interview support system 200 will be described with reference to FIG. The interview support system 200 comprises an interview support device 100, a voice input device such as microphones (211, 212), and input / output devices such as a keyboard 213 and a display 214. The interview support device 100 is an information processing device that supports interviews. The interview support device 100 cooperates with microphones (211, 212) and the like to support the interview. The microphone 211 is a voice input device that collects the voice of the questioner 201 and outputs voice data. The microphone 212 is a voice input device that collects the voice of the respondent 202 and outputs voice data. The keyboard 213 is a user interface that is operated to control the interview support system 200 . The display 214 is an information presentation device that presents information to support the interview.

[0013] The configuration of the interview support device 100 will be described with reference to FIG. The interview support device 100 is a computer that includes hardware such as a processor 101, a memory 102, an auxiliary storage device 103, a communication device 104, and an input / output interface 105. These pieces of hardware are connected to one another via signal lines.

[0014] The processor 101 is an IC that performs arithmetic processing and controls other hardware. For example, the processor 101 is a CPU, a DSP, a GPU, or a combination thereof. IC is an abbreviation for Integrated Circuit. CPU is an abbreviation for Central Processing Unit. DSP is an abbreviation for Digital Signal Processor. GPU is an abbreviation for Graphics Processing Unit.

[0015] The memory 102 is a volatile or non-volatile storage device. The memory 102 is also called a primary storage device or a main memory. For example, the memory 102 is a RAM. Data stored in the memory 102 is saved in the secondary storage device 103 as needed. RAM is an abbreviation for Random Access Memory.

[0016] The auxiliary storage device 103 is a non-volatile storage device. For example, the auxiliary storage device 103 is a ROM, a HDD, a flash memory, or a combination thereof. Data stored in the auxiliary storage device 103 is loaded into the memory 102 as needed. ROM is an abbreviation for Read Only Memory. HDD is an abbreviation for Hard Disk Drive.

[0017] The communication device 104 is a receiver and a transmitter. For example, the communication device 104 is a communication chip or a NIC. Communication of the interview support device 100 is performed using the communication device 104. NIC is an abbreviation for Network Interface Card.

[0018] The input / output interface 105 is a port to which an input device and an output device are connected. For example, the input / output interface 105 is a USB terminal, the input devices are a keyboard 213 and a mouse, and the output device is a display 214. Input and output to and from the interview support device 100 are performed via the input / output interface 105. USB is an abbreviation for Universal Serial Bus.

[0019] The interview support device 100 comprises components such as a control unit 111, a speaker identification unit 112, a speech recognition unit 113, a knowledge structuring unit 114, a topic generation unit 115, and an information presentation unit 116. These components are realized by software.

[0020] The auxiliary storage device 103 stores an interview support program for causing the computer to function as a control unit 111, a speaker identification unit 112, a speech recognition unit 113, a knowledge structuring unit 114, a topic generation unit 115, and an information presentation unit 116. The interview support program is loaded into the memory 102 and executed by the processor 101. The auxiliary storage device 103 also stores an OS. At least a part of the OS is loaded into the memory 102 and executed by the processor 101. The processor 101 executes the interview support program while running the OS. OS is an abbreviation for Operating System.

[0021] Data (input data, output data, etc.) of the interview support program is stored in storage unit 190. The memory 102 functions as the storage unit 190. However, a storage device such as the auxiliary storage device 103, a register in the processor 101, or a cache memory in the processor 101 may function as the storage unit 190 instead of or together with the memory 102.

[0022] The interview support program can be recorded (stored) in a computer-readable manner on a non-volatile recording medium such as an optical disk or flash memory.

[0023] FIG. 3 shows the functional configuration of the interview support system 200. The input / output interface 105 functions as a voice input unit (121, 122), an information input unit 123, and an information output unit . Voice data is input from a voice input device such as a microphone (211, 212) via a voice input unit (121, 122). Various types of information data are input from an input device such as a keyboard 213 via the information input unit 123 . Various types of information data are output via the information output unit 124 using an output device such as a display 214 .

[0024] The storage unit 190 stores setting data 191, a structured model 192, a generative model 193, etc. These data will be described later.

[0025] ***Explanation of Operation*** The operation procedure of the interview support system 200 (particularly the interview support device 100) corresponds to an interview support method, and the operation procedure of the interview support device 100 corresponds to a processing procedure by the interview support program.

[0026] The state transitions of the interview support device 100 will be described with reference to FIG. The main states of the interview support device 100 are a standby state and a support state.

[0027] The standby state is a state in which the interview support device 100 waits until the interview starts. When the interview support device 100 is started up, the state of the interview support device 100 goes into a standby state.

[0028] In the standby state, various settings can be made to the interview support device 100. A user (for example, questioner 201) uses an input device such as keyboard 213 to input various setting information into interview support device 100. The control unit 111 receives the input information and performs various settings in accordance with the input information. For example, in the standby state, the following settings are made: The speaker labels are set as described below: a speaker label identifying the questioner 201 is set in association with the microphone 211, and a speaker label identifying the answerer 202 is set in association with the microphone 212. The input volume from the microphones (211, 212) to the audio input units (121, 122) is set. Information on various settings is saved as setting data 191.

[0029] In the standby state, the interview support device 100 can be instructed to start support. At the start of the interview, the questioner 201 inputs an instruction to start support into the interview support device 100 . The control unit 111 receives the instruction to start support and transitions the state of the interview support device 100 from the standby state to the support state.

[0030] The support state is a state in which interview support device 100 supports the interview for questioner 201 from the start of the interview to the end of the interview. In the assisted state, the interview support device 100 performs operations for supporting the interview. The operations of the interview support device 100 in the assisted state will be described later.

[0031] In the assistance state, it is possible to instruct the interview assistance device 100 to stop assistance. At the end of the interview, the questioner 201 inputs an instruction to stop the support into the interview support device 100. The control unit 111 receives the instruction to stop support and transitions the state of the interview support device 100 from the support state to the standby state.

[0032] In the standby state, it is possible to instruct the interview support device 100 to end support. The questioner 201 inputs an instruction to end support into the interview support device 100 in order to stop the interview support device 100 . The control unit 111 receives the instruction to end the support and stops the interview support device 100.

[0033] The interview support method will be explained based on FIG. While the interview support device 100 is in the support state, an interview is conducted between a questioner 201 and an answerer 202 . The field to which the topic that is the object of the interview belongs is called the subject field.

[0034] Steps S110 to S160 are continuously executed while the interview support device 100 is in the support state.

[0035] In step S110, a questioner 201 speaks to ask a question, or an answerer 202 speaks to answer the question. When questioner 201 speaks, microphone 211 converts the speech of questioner 201 into voice data and outputs it. The output voice data is input from microphone 211 to interview support device 100. Speaker identification unit 112 then receives the voice data. When the respondent 202 speaks, the microphone 212 converts the respondent's utterance into voice data and outputs it. The output voice data is input from the microphone 212 to the interview support device 100. The speaker identification unit 112 then receives the voice data.

[0036] In step S120, the speaker identification unit 112 assigns a speaker label to the received voice data.

[0037] When voice data is input from the microphone 211, the speaker identifying unit 112 assigns a speaker label indicating the questioner 201 to the voice data. When voice data is input from the microphone 212, the speaker identification unit 112 assigns a speaker label indicating the respondent 202 to the voice data.

[0038] In step S130, the speech recognition unit 113 performs speech recognition on the speech data to which the speaker labels have been assigned, and generates utterance data to which the speaker labels have been assigned. The utterance data is data indicating what the speaker has said. Specifically, the utterance data indicates what the speaker has said in text format.

[0039] The utterance data to which the speaker label indicating the questioner 201 is assigned is referred to as questioner utterance data. The utterance data to which the speaker label indicating the respondent 202 is assigned is referred to as respondent utterance data.

[0040] For example, if respondent 202 says, "It's important not to mix too much to make it fluffy," speech recognition is performed on the voice data of this utterance, and respondent utterance data is generated. This respondent utterance data shows the utterance, "It's important not to mix too much to make it fluffy," in text form.

[0041] In step S140, the knowledge structuring unit 114 extracts structured knowledge from the utterance data to which the speaker labels have been assigned.

[0042] The structured knowledge indicates important words contained in the utterance. The attention words are words contained in the utterance that correspond to terms in the target field. A word equivalent to a subject matter term is a word that is the same as the subject matter term or a word that is semantically similar to the subject matter term. Examples of subject matter terms are frequent words in the subject matter.

[0043] Subject matter frequent words are words that occur frequently in conversations about topics in the subject matter, i.e., words that are commonly used when describing the subject matter or communicating actual work in the subject matter to others. For example, if the target field is "cooking," frequently occurring words in the target field include words that describe the texture of the food being prepared, words that describe the work process, words that indicate the state of the process, and words that indicate tools or ingredients. Examples of words that describe the texture of the food being prepared include "plump," "firm," "crispy," and "crunchy." Examples of words that describe the work process include "cut," "bake," "fry," "simmer," and "steam." Examples of words that indicate the state of a process include "slowly," "firmly," "quickly," "many times," and "for a while." Examples of words that indicate tools or ingredients include "bowl," "knife," "mixer," "flour," and "sugar." As such, a wide variety of frequently occurring words can be exemplified. Note that words other than these words may be selected as frequently occurring words, and these words do not necessarily have to be selected as frequently occurring words. The frequently occurring words in the target field can be extracted from literature related to the target field, conversations that actually take place during work, etc. Note that the frequently occurring words may also be extracted from sources other than literature and conversations.

[0044] Words are extracted from text data by natural language processing, for example. Specifically, extraction of words from text is achieved by morphological analysis.

[0045] Figure 6 shows an example of the results obtained by decomposing the statement, "It is important not to mix too much to achieve a fluffy texture," using morphological analysis. In this example, the surface form, part of speech, and normalized form are obtained for each word. However, the granularity of the decomposition and the output content are not limited to this example, as long as the granularity and content required for subsequent processing are secured. For example, after decomposition into the primitive elements "funwari," "to," "shi," and "ta," the decomposed elements can be grouped together to form a compound word such as "funwari shita" (it was done) and treated as a single word. Anything is acceptable as long as the granularity and content required for subsequent processing are secured.

[0046] Semantic similarity can be determined by using a model such as Word2Vec or BERT to obtain distributed word representations and determining the similarity based on the cosine similarity of the word vectors.

[0047] FIG. 7 shows an example of the cosine similarity between words expressed as vectors. The first row shows the most frequently occurring words in the subject domain. The first column shows the words in the utterance. The values ​​in the second and subsequent columns are the cosine similarity of the words in the first column to the frequently occurring words in the first row. The cosine similarity takes a value between -1 and 1, and the closer the cosine similarity is to 1, the higher the similarity between the words.

[0048] A word whose cosine similarity with any frequently occurring word is equal to or greater than a threshold is extracted as a word of interest. For example, if the threshold is 0.6, the cosine similarity between "funwari" (soft) and "fukkura" (plump), is equal to or greater than 0.6, so "fukkura" (plump) is extracted as a word of interest. If there is a word whose cosine similarity with two or more frequently occurring words is equal to or greater than the threshold, that word is extracted as a word of interest corresponding to the frequently occurring word with the largest cosine similarity. For example, "mix" (mix) has a cosine similarity of 0.6 or greater with both "knead" (knead) and "mix." However, the cosine similarity with "mix" is greater. Therefore, "mix" is extracted as a word of interest corresponding to "mix." The threshold is set in advance, and is not limited to 0.6.

[0049] For example, respondent 202 stated, "It's important not to mix too much to get a fluffy texture." In this case, "fluffy," "texture," and "mix" are extracted as attention words based on the cosine similarity in Figure 7. The attention word "fluffy" corresponds to the frequently occurring word "plump," the attention word "texture" corresponds to the frequently occurring word "texture," and the attention word "mix" corresponds to the frequently occurring word "churned." These attention words are words related to the content that the interviewer 201 should ask in order to achieve the purpose of the interview.

[0050] Continuing with the explanation of structured knowledge. The structured knowledge indicates the dependency structure of the syntactic elements that make up an utterance. The dependency structure indicates the dependency relationship between syntax elements. Syntactic elements are the elements that make up sentences, such as words, phrases, and clauses.

[0051] The dependency structure is extracted by, for example, syntactic analysis.

[0052] Figures 8 and 9 show examples of dependency structures for the statement, "It is important not to mix too much to achieve a fluffy texture." In FIG. 8, the dependency structure shows the dependency relationship on a word-by-word basis. In FIG. 9, the dependency structure shows the dependency relationship in phrase units. The granularity of the dependency relationships may be either a word unit or a phrase unit, or may be another unit. In the following description, FIG. 9 will be used instead of FIG.

[0053] Returning to FIG. 5, the description continues from step S150. In step S150, the topic generation unit 115 uses the utterance data to generate topic information based on structured knowledge.

[0054] The topic information is data indicating possible questions to be asked to the respondent 202, etc.

[0055] The question indicates what the questioner 201 should ask the answerer 202 .

[0056] A candidate question sentence is called a candidate question.

[0057] The question candidates are generated using respondent utterance data as follows. First, the topic generation unit 115 determines a noteworthy element in the reply from the reply person 202 based on the noteworthy words and dependency structure in the reply from the reply person 202 . The attention element is a syntactic element that includes the attention word among the syntactic elements that make up the utterance.

[0058] The attention element indicates the content that the questioner 201 should pay attention to. The focus element may be only the focus word, or may be a phrase, sentence, or other syntactic element that includes the focus word and words related to the focus word.

[0059] In FIG. 9, when the target words are "fluffy," "texture," and "mix," the target elements are "fluffy texture" and "not mixing too much."

[0060] The generation of question candidates will now be described. Next, the topic generation unit 115 generates sentences that confirm each of a plurality of confirmation items in the interview regarding the element of interest in the statement of the respondent 202. The plurality of confirmation items are defined in advance. For example, six points to check are defined from the perspective of 5W1H, which means six perspectives: when, where, who, what, why, and how. Each of the generated sentences is called a tentative candidate. A tentative candidate is a tentative question candidate.

[0061] In Figure 9, the elements of interest are "soft and fluffy texture" and "not mixing too much." Note that in the dependency structure, "soft and fluffy texture" depends on "not mixing too much." When applying the 5W1H perspective to "soft and fluffy texture," the following tentative candidates are generated: “When is it going to have a soft and fluffy texture?” "Which fluffy texture is it?" “Whose fluffy texture is it?” “What fluffy texture is that?” "Why does it have such a fluffy texture?" "How fluffy is the texture?"

[0062] Similarly, tentative candidates are generated for the focus element "not mixing too much."

[0063] The generation of question candidates will now be described. Then, the topic generation unit 115 excludes tentative candidates that correspond to confirmed matters from the generated tentative candidates, and each of the remaining tentative candidates becomes a question candidate. The confirmed matters are, among the multiple confirmed matters, matters that have been explained in the respondent's 202 previous statements.

[0064] In Figure 9, the key elements are "soft texture" and "not mixing too much." The statement "soft and fluffy" already explains the degree of texture. Therefore, the tentative candidate "How soft and fluffy is the texture?" is a tentative candidate that corresponds to the confirmed matter and is therefore excluded.

[0065] Previous statements may be referenced along with or instead of the current statement. For example, let's assume that respondent 202 had previously said, "We're proud of the fact that the texture can be maintained even if you leave it for a few days instead of eating it right away." This statement explains the "texture" period as "even if you leave it for a few days." Therefore, the candidate sentence "When is the fluffy texture?" is a provisional candidate that corresponds to the confirmed matter and is therefore excluded.

[0066] Similarly, for the focus element of "not mixing too much," tentative candidates corresponding to confirmed items are excluded.

[0067] As a result, question candidates are generated for confirmation items other than the confirmed items among the plurality of confirmation items.

[0068] Returning to FIG. 5, the topic information generated in step S150 will be further explained. The topic information includes candidate questions for the answerer 202, each statement by the questioner 201 and the answerer 202, notable words in each statement by the questioner 201 and the answerer 202, and notable elements in each statement by the questioner 201 and the answerer 202. The words of interest in the statements of the questioner 201 are extracted in the same way as the words of interest in the statements of the answerer 202, using the questioner statement data. The notable elements of the statements of the questioner 201 are determined in the same way as the notable elements of the statements of the answerer 202, using the questioner statement data.

[0069] In step S160, the information presenter 116 presents the topic information.

[0070] Topical information is presented as follows: The information presenting unit 116 generates data for the information presenting screen 300 and displays the information presenting screen 300 on the display 214, thereby presenting the topic information. The information presentation screen 300 is a screen for presenting topic information to the questioner 201.

[0071] 10 to 13 show examples of the information presentation screen 300. FIG. The information presentation screen 300 has a comment area 310 and a structured knowledge area 320 . The comment area 310 is an area where comments (311 to 314) from the questioner 201 and the answerer 202 are displayed. The structured knowledge area 320 is an area where structured knowledge is displayed.

[0072] In the comment area 310, each comment (311 to 314) from the questioner 201 and the answerer 202 is written in text in a speech bubble. The statements of respondent 202 (311, 314) are written in speech bubbles from the left. The statements of questioner 201 (312, 313) are written in speech bubbles from the right. In this way, the statements (311 to 314) of the questioner 201 and the answerer 202 are displayed so as to be easily distinguishable and visible.

[0073] The important words in each statement (311-314) are highlighted. Different attention words are highlighted in different colors, and the same attention words are highlighted in the same color. In Figure 10, "fluffy," "texture," and "mixed" are highlighted in different colors in comment 311. In Figures 11 to 13, "mixed" in comments (312 to 314) is highlighted in the same color as "mixed" in comment 311. In this way, the words of interest in each utterance are displayed so that they can be easily distinguished and visually recognized.

[0074] In the structured knowledge area 320, the structured knowledge is displayed in a tabular format. The first column, "Process," shows attention words that correspond to processes in the target field. For example, attention words that are verbs are selected as attention words that correspond to processes. The second column, "Content," shows the attention words related to the processes in the target field and the candidate questions related to the processes in the target field. For example, attention words related to the processes are selected based on the dependency structure, which have a dependency relationship with the attention word corresponding to the process. For example, question candidates related to the processes are selected based on the statements containing the attention word corresponding to the process. In Figure 10, the attention word corresponding to the process is the verb "mix," which corresponds to the frequently occurring word "stir." The attention words related to the process are "fluffy" and "texture." The attention element "fluffy texture" containing the attention word related to the process is selected based on the dependency structure as the purpose of the attention element "do not mix too much" which contains the attention word corresponding to the process.

[0075] The explanation of step S160 continues. The information presenting unit 116 determines related words in the statement of the questioner 201 based on the dependency structure of the statement of the questioner 201. The related words are words included in the statement of the questioner 201 that have a modification relationship with the target word in the statement of the answerer 202. The information presenting unit 116 displays the related words on the information presenting screen 300 .

[0076] The information presenting unit 116 determines whether the new comment from the answerer 202 is a correction comment based on the dependency structure of the new comment from the answerer 202. A correction statement is a statement that corrects the content expressed by the related words. If the new comment from the answerer 202 is a correction comment, the information presenter 116 determines the correction word in the new comment from the answerer 202 based on the dependency structure of the new comment from the answerer 202. The correction words are words included in the new statement of the respondent 202 that correct the content expressed by the related words. Then, the information presenting unit 116 corrects the related words displayed on the information presenting screen 300 to the corrected words.

[0077] Specific examples of related words and correction words will be explained. The comment area 310 and the structured knowledge area 320 are updated in response to new comments (312 to 314). In Figure 12, a new utterance 313 from questioner 201 has been added to utterance area 310. New utterance 313 is "You're mixing it gently." From new utterance 313, new information "gently" indicating the state of "mixing" corresponding to the process is obtained based on the dependency structure. Therefore, new information "gently" indicating the state has been added to structured knowledge area 320. "Gently" is a related word that has a dependency relationship with "mixing," which is the focus word in utterance 311. In Figure 13, a new statement 314 from respondent 202 has been added to statement area 310. New statement 314 is "No, it looks like I'm mixing it gently, but in reality I'm using a lot of force," which negates the state of "gently." From new statement 314, new information indicating the correct state, "use force," is obtained based on the dependency structure. Therefore, in structured knowledge area 320, "gently," which indicates the state, has been crossed out and corrected to "use force." "Use force" is a correction word that corrects "gently."

[0078] ***Additional explanation*** By identifying the microphones (211, 212) from which the voice data is input, it is possible to distinguish between the questioner 201 and the answerer 202. It is also possible to distinguish between the questioner 201 and the answerer 202 by performing voice authentication on the voice data. In this case, it is not necessary to distinguish the microphones (211, 212) that input the voice data.

[0079] Other modal output devices may be used in addition to or in place of the visual output device, display 214. For example, an auditory output device, a speaker, may be used.

[0080] A situation is assumed in which a questioner 201 conducts an interview with a respondent 202 at close range. However, the interview support method may also be implemented in a situation where the questioner 201 remotely interviews the respondent 202. In this case, the interview support system 200 comprises an interview support device 100 for the questioner 201 and an information processing device for the respondent 202. The interview support device 100 for the questioner 201 and the information processing device for the respondent 202 operate in cooperation with each other. A microphone 211 and a speaker are connected to the interview support device 100 for the questioner 201, and a microphone 212 and a speaker are connected to the information processing device for the respondent 202. Audio data and the like are communicated between the interview support device 100 for the questioner 201 and the information processing device for the respondent 202.

[0081] Common formats for audio data are WAVE format or MP3 format. However, the format of the audio data is not particularly limited, and an easy-to-process format is selected.

[0082] Data relating to the various analyses in step S140 is stored in the structured model 192. Before step S140, the structured model 192 is read, and step S140 is executed using the structured model 192.

[0083] The structured knowledge is a data structure that contains at least the words that make up the text of the spoken text data (the words in the utterance). The structured knowledge has correspondence data between words in the utterance data and frequently occurring words in the field of the interview subject that match or are semantically similar to the words in the utterance data. The structured knowledge contains dependency relationship data between words in the utterance data.

[0084] The frequently occurring words may be extracted by a method using machine learning or by other methods.

[0085] A measure other than the cosine similarity may be used. For example, the semantic distance between words may be calculated and used.

[0086] Step S150 is a process of generating, from the structured knowledge, what the questioner 201 should ask the respondent 202 in order to achieve the purpose of the interview, and what the questioner 201 should pay attention to from the utterances of the respondent 202.

[0087] Data related to the various analyses in step S150 is stored in the generative model 193. Before step S150, the generative model 193 is read, and step S150 is executed using the generative model 193.

[0088] Candidate questions may be generated using large-scale language models (LLMs). Figures 14 and 15 show examples of question candidates generated by LLM. The "Myself" column shows the input to the LLM. The question sentence to generate question candidates is the input to the LLM. The input to the LLM is generated using utterance data. In Figures 14 and 15, "It's important not to mix too much to create a fluffy texture" is the utterance of respondent 202. The "Copilot" column shows the output from the LLM. The output from the LLM is question candidates. Question candidates are generated by the LLM and extracted from the output sentences of the LLM. In Figure 14, question candidates 1 to 3 are generated by the LLM. In Figure 15, question candidates 1 and 2 are generated by the LLM.

[0089] Questions that are expressed in unnatural ways may be excluded. For example, suppose "texture" is the focus word and "soft texture" is the focus element. "Texture" is generally a word for ingredients or dishes, not a word for people. Therefore, tentative candidates that confirm the confirmation items from the perspective of "who" are excluded from the question candidates. In other words, the question candidate "Whose soft texture is this?" is excluded.

[0090] Any method may be used to generate candidate questions, as long as it uses at least one of utterance data and structured knowledge.

[0091] The question candidates may be generated using questioner utterance data.

[0092] The data format of the presented topic information may be any format that can be output to an information presentation device such as the display 214. For example, the data of the presented topic information may be image data or data that defines the layout of information, such as HTML data.

[0093] Step S160 is a process of presenting to the questioner 201 the speaker labels, speech text data, structured knowledge, questions that the questioner 201 should ask the answerer 202, and questions that the questioner 201 should pay attention to from the answerer 202's speech obtained by the processing up to step S150.

[0094] ***Effects of the First Embodiment*** The first embodiment aims to realize a system that supports questioners who are unfamiliar with interviews to ask appropriate questions, even if appropriate questions are not manually determined in advance. The interview support device 100 dynamically generates questions based on the content of the dialogue during the interview.

[0095] The first embodiment provides the following effects. The interview support device 100 extracts noteworthy words from the statements of the questioner and the answerer and presents the information. This eliminates the need for the questioner to determine which words to pay attention to, reducing the workload of the questioner. The interview support device 100 extracts information such as process-related content and questions from the utterance content and presents it on the screen, thereby reducing the workload of the questioner who extracts information from the utterance content. The questioner does not need to prepare questions in advance, and the interview support device 100 dynamically generates questions from the content of the utterances, thereby reducing the workload of the questioner in preparing questions. The questioner does not need to prepare a specific question for the question target in advance. According to the first embodiment, general-purpose automatic question generation becomes possible, in which the question items are not limited to a specific question target.

[0096] Embodiment 2 The embodiment in which question candidates are generated in consideration of the behavior of the respondent 202 will be described below with reference to FIG. 16, mainly with respect to the differences from the first embodiment.

[0097] ***Configuration Description*** The configuration of the interview support system 200 will be described with reference to FIG. The interview support system 200 includes a photographing device such as a camera 215 . The camera 215 captures the respondent 202 during the interview and outputs the video data. The input / output interface 105 functions as an image input unit 125 . Video data is input from a photographing device such as a camera 215 via the image input unit 125. The video data shows a respondent 202 being interviewed.

[0098] ***Explanation of Operation*** The procedure of the interview support method is the same as that in the first embodiment. However, the difference from the first embodiment is that video data is used in step S150.

[0099] In step S150, the topic generation unit 115 uses the utterance data and the video data to generate topic information based on structured knowledge.

[0100] The video data is used as follows. The topic generation unit 115 recognizes the behavior of the respondent 202 using the video data. The topic generation unit 115 generates question candidates in consideration of the behavior of the answerer 202. The question candidates are generated mainly in consideration of the behavior of the answerer 202 when the answerer 202 makes a statement. The behavior of the respondent 202 includes the facial expressions of the respondent 202 and the gestures of the respondent 202 .

[0101] For example, suppose respondent 202 crosses his arms in a gesture of deep thought and then says, "No, it may look like I'm mixing it gently, but I'm actually using a lot of force." In this case, the gesture of deep thought by the respondent 202 is recognized, and the statement made after the gesture is determined to be a statement made after careful consideration. Then, question candidates such as "I think that statement was made after some thought, but was there anything that bothered you?" and "Is what you just said an important point?" are generated.

[0102] ***Effects of the Second Embodiment*** In the second embodiment, the behavior of the respondent is taken into consideration along with the respondent's statements, which makes it possible to estimate the importance that is not expressed in words and automatically generate questions.

[0103] Embodiment 3 The following mainly describes the differences from the first and second embodiments in the form of generating question candidates in consideration of the confidence level of the statements of the answerer 202.

[0104] ***Configuration Description*** The configuration of the interview support system 200 is the same as that of the first or second embodiment.

[0105] ***Explanation of Operation*** The procedure of the interview support method is the same as that in the first and second embodiments. However, the difference from the first and second embodiments is that the confidence level of the reply from the reply person 202 is taken into consideration in step S150.

[0106] In step S150, the topic generation unit 115 uses the utterance data to generate topic information based on structured knowledge.

[0107] The topic information is generated as follows. The topic generation unit 115 calculates the degree of certainty of the answerer 202 with respect to the answerer's 202 statement by using the answerer's statement data. The topic generation unit 115 generates topic information taking into consideration the degree of certainty of the answerer 202 .

[0108] The confidence level is calculated as follows: Expressions indicating the degree of certainty are extracted from the statements of respondent 202. Specifically, expressions indicating obligation, necessity, possibility, or negation are extracted. An example of an expression indicating obligation or necessity is "must be...". Examples of expressions indicating possibility are "may be..." and "almost impossible...". An example of an expression indicating negation is "absolutely not...". The confidence level is calculated based on the frequency (number) of the extracted expressions. For example, expressions indicating obligation or imperative are converted to "1", expressions indicating possibility are converted to a value between 0 and 1, and expressions indicating negation are converted to "0". Then, the confidence level is calculated using the value of each extracted expression. For example, statistical values ​​(average, maximum, minimum, etc.) of the extracted expressions are calculated, and the calculated statistical values ​​become the confidence level. For example, the confidence level takes a value between 0 and 1, and the closer the value is to 1, the higher the confidence level, and the closer the value is to 0, the lower the confidence level. The confidence may be calculated as follows: a conversion value is set for each word category, each word in the utterance is categorized, each word is converted to a conversion value corresponding to the category, and the confidence is calculated using the conversion value of each word. For example, the words are classified using a machine learning model. The confidence may be calculated in other ways.

[0109] For example, the following question candidates are generated according to the confidence level: If the confidence level is close to 1 or 0, candidate questions to confirm the necessity and candidate questions to confirm the reason are generated. If the confidence level is close to 0.5, candidate questions are generated to confirm the criteria for judgment, such as "Under what conditions does this occur?" and "Is it okay for it to occur?"

[0110] ***Effects of the Third Embodiment*** In the third embodiment, a question is generated according to the confidence level of the statement of the answerer 202. This has the effect of automatically generating a more detailed question.

[0111] ***Supplementary explanation of implementation form*** In the first embodiment, the following interview support device 100 has been described. The interview is conducted between the questioner and the respondent. The interview is conducted between at least one person and aims to elicit the skills and knowledge of the respondent. The interview support device 100 has the following features to facilitate smooth interviews. The interview support device 100 receives as input the utterances of each person participating in the interview and identifies the speaker (questioner, respondent) from the input utterances. The interview support device 100 recognizes and analyzes the content of each speaker's utterance, extracts structured knowledge from the content of the utterance, and extracts from the structured knowledge references to events and procedures related to the skills or knowledge to be elicited. The interview support device 100 generates the content (question sentence) that the questioner should ask the respondent in order to achieve the purpose of the interview, or the content (words and sentences) that the questioner should pay attention to from the respondent's utterance, and displays all or part of the processing results by linking them to the part of the utterance from which the processing results were derived.

[0112] In the second embodiment, the following interview support device 100 has been described. The interview support device 100 receives as input an image or video of a respondent during an interview, and recognizes the respondent's facial expressions and gestures from the input. The interview support device 100 generates, depending on the respondent's facial expressions and gestures, the content (question sentence) that the questioner should ask the respondent in order to achieve the purpose of the interview, or the content (words and sentences) that the questioner should pay attention to from the respondent's speech.

[0113] In the third embodiment, the following interview support device 100 has been described. The interview support device 100 extracts a confidence level from the content of the respondent's utterance during the interview. Depending on the confidence level and the content of the utterance, the interview support device 100 generates the content (question sentence) that the questioner should ask the respondent in order to achieve the purpose of the interview, or the content (words and sentences) that the questioner should pay attention to from the respondent's utterance.

[0114] The functions of the interview support device 100 can be realized by hardware, software, firmware, or a combination of these.

[0115] The "part" of each element of the interview support device 100 may be read as "processing," "step," "circuit," or "circuitry."

[0116] Each embodiment is an example of a preferred embodiment and is not intended to limit the technical scope of the present disclosure. Each embodiment may be implemented in part or in combination with other embodiments. Procedures described using flowcharts, etc. may be modified as appropriate.

[0117] Various aspects of the present disclosure are described below as appendices. (Appendix 1) a knowledge structuring unit that extracts, as noteworthy words, words corresponding to terms in the subject field of the interview from among words included in the statements of the respondent from respondent statement data indicating the statements of the respondent in response to questions in the interview; a topic generation unit that generates candidate questions for the respondent based on the attention words; an information presenting unit that presents the generated question candidates; An interview support device comprising:

[0118] (Appendix 2) The topic generation unit generates, as the question candidate, a question sentence for confirming a confirmation item regarding a syntactic element including the word of interest among syntactic elements constituting the utterance of the respondent. 10. An interview support device as described in appendix 1.

[0119] (Appendix 3) the knowledge structuring unit extracts a dependency structure indicating a dependency relationship between syntactic elements of the respondent's utterances from the respondent's utterance data; The topic generation unit determines the element of interest based on the word of interest and the dependency structure. 10. An interview support device as described in Appendix 2.

[0120] (Appendix 4) The topic generation unit generates the question candidates for confirmation matters other than confirmed matters that are confirmation matters explained in a statement made by the respondent before the statement made by the respondent, among the plurality of confirmation matters. 4. An interview support device according to claim 2 or 3.

[0121] (Appendix 5) The topic generation unit generates question sentences that confirm each of the confirmation items as tentative candidates, and obtains each of the remaining tentative candidates as the question candidate by excluding tentative candidates that correspond to the confirmed items. 10. An interview support device as described in appendix 4.

[0122] (Appendix 6) The information presenting unit presents the question candidates by displaying an information presenting screen showing the utterances of the respondents, the attention words, and the question candidates. 6. An interview support device according to any one of appendices 1 to 5.

[0123] (Appendix 7) the knowledge structuring unit extracts a word of interest from questioner utterance data indicating a utterance of a questioner; The information presenting unit displays the information presenting screen showing the statements of the questioner and the answerer, the attention word, and the question candidates. 10. An interview support device as described in Appendix 6.

[0124] (Appendix 8) the knowledge structuring unit extracts a dependency structure from questioner utterance data indicating utterances by a questioner; The information presenting unit determines, based on the dependency structure of the statement by the questioner, words that have a dependency relationship with the focused word of the statement by the answerer among words included in the statement by the questioner as related words, and displays the related words on the information presenting screen. 8. An interview support device as described in appendix 7.

[0125] (Appendix 9) The information presentation unit determines whether the new statement of the respondent is a correction statement that corrects the content expressed by the related word based on the dependency structure of the new statement of the respondent, and if the new statement of the respondent is the correction statement, determines a word that corrects the content expressed by the related word from among the words included in the new statement of the respondent based on the dependency structure of the new statement of the respondent as a correction word, and corrects the related word displayed on the information presentation screen to the correction word. 10. An interview support device as described in Appendix 8.

[0126] (Appendix 10) The information presentation unit highlights the same attention word in the utterances displayed on the information presentation screen in the same color, and highlights different attention words in the utterances displayed on the information presentation screen in different colors. 10. An interview support device according to any one of appendices 6 to 9.

[0127] (Appendix 11) The topic generation unit recognizes the behavior of the respondent using video data obtained by filming the respondent during an interview, and generates the question candidates in consideration of the behavior of the respondent. 11. An interview support device according to any one of claims 1 to 10.

[0128] (Appendix 12) The topic generation unit calculates a degree of certainty of the respondent with respect to the statement by the respondent using the respondent utterance data, and generates the question candidates in consideration of the degree of certainty of the respondent. 12. An interview support device according to any one of claims 1 to 11.

[0129] (Appendix 13) extracting, as noteworthy words, words included in the statements of the respondents from respondent statement data indicating the statements of the respondents in response to questions in the interview, the words corresponding to terms in the subject field of the interview; generating candidate questions for the respondent based on the attention words; Present generated question candidates Interview support methods.

[0130] (Appendix 14) a knowledge structuring process for extracting, as noteworthy words, words corresponding to terms in the subject field of the interview from among words included in the respondent's statements from respondent statement data indicating the respondent's statements in response to questions in the interview; a topic generation process for generating candidate questions for the respondent based on the attention words; an information presentation process for presenting the generated question candidates; An interview support program that allows a computer to execute the above. [Explanation of symbols]

[0131] 100 Interview support device, 101 Processor, 102 Memory, 103 Auxiliary storage device, 104 Communication device, 105 Input / output interface, 111 Control unit, 112 Speaker identification unit, 113 Speech recognition unit, 114 Knowledge structuring unit, 115 Topic generation unit, 116 Information presentation unit, 121 Speech input unit, 122 Speech input unit, 123 Information input unit, 124 Information output unit, 125 Image input unit, 190 Memory unit, 191 Setting data, 192 Structured model, 193 Generative model, 200 Interview support system, 201 Questioner, 202 Respondent, 211 Microphone, 212 Microphone, 213 Keyboard, 214 Display, 215 Camera, 300 Information presentation screen, 310 Speech area, 311 Speech, 312 Speech, 313 remarks, 314 remarks, 320 Structured knowledge areas.

Claims

1. a knowledge structuring unit that extracts, as noteworthy words, words corresponding to terms in the subject field of the interview from among words included in the statements of the respondent from respondent statement data indicating the statements of the respondent in response to questions in the interview; a topic generation unit that generates candidate questions for the respondent based on the attention words; an information presenting unit that presents the generated question candidates; An interview support device comprising:

2. The topic generation unit generates, as the question candidate, a question sentence for confirming a confirmation item regarding a syntactic element including the word of interest among syntactic elements constituting the utterance of the respondent. The interview support device according to claim 1 .

3. the knowledge structuring unit extracts a dependency structure indicating a dependency relationship between syntactic elements of the respondent's utterances from the respondent's utterance data; The topic generation unit determines the element of interest based on the word of interest and the dependency structure. The interview support device according to claim 2 .

4. The topic generation unit generates the question candidates for confirmation matters other than confirmed matters that are confirmation matters explained in a statement made by the respondent before the statement made by the respondent, among the plurality of confirmation matters. The interview support device according to claim 2 .

5. The topic generation unit generates question sentences that confirm each of the confirmation items as tentative candidates, and obtains each of the remaining tentative candidates as the question candidate by excluding tentative candidates that correspond to the confirmed items. The interview support device according to claim 4.

6. The information presenting unit presents the question candidates by displaying an information presenting screen showing the utterances of the respondents, the attention words, and the question candidates. The interview support device according to claim 1 .

7. the knowledge structuring unit extracts a word of interest from questioner utterance data indicating a utterance of a questioner; The information presenting unit displays the information presenting screen showing the statements of the questioner and the answerer, the attention word, and the question candidates. The interview support device according to claim 6.

8. the knowledge structuring unit extracts a dependency structure from questioner utterance data indicating utterances by a questioner; The information presenting unit determines, based on the dependency structure of the statement by the questioner, words that have a dependency relationship with the focused word of the statement by the answerer among words included in the statement by the questioner as related words, and displays the related words on the information presenting screen. The interview support device according to claim 7.

9. The information presentation unit determines whether the new statement of the respondent is a correction statement that corrects the content expressed by the related words based on a dependency structure of the new statement of the respondent, and if the new statement of the respondent is the correction statement, determines a word that corrects the content expressed by the related words from among the words included in the new statement of the respondent based on the dependency structure of the new statement of the respondent as a correction word, and corrects the related words displayed on the information presentation screen to the correction word. The interview support device according to claim 8.

10. The information presentation unit highlights the same attention word in the utterances displayed on the information presentation screen in the same color, and highlights different attention words in the utterances displayed on the information presentation screen in different colors. The interview support device according to claim 6.

11. The topic generation unit recognizes the behavior of the respondent using video data obtained by filming the respondent during an interview, and generates the question candidates in consideration of the behavior of the respondent. The interview support device according to any one of claims 1 to 10.

12. The topic generation unit calculates a degree of certainty of the respondent with respect to the statement by the respondent using the respondent utterance data, and generates the question candidates in consideration of the degree of certainty of the respondent. The interview support device according to any one of claims 1 to 10.

13. extracting, as noteworthy words, words included in the statements of the respondents from respondent statement data indicating the statements of the respondents in response to questions in the interview, the words corresponding to terms in the subject field of the interview; generating candidate questions for the respondent based on the attention words; Present generated question candidates Interview support methods.

14. a knowledge structuring process for extracting, as noteworthy words, words corresponding to terms in the subject field of the interview from among words included in the respondent's statements from respondent statement data indicating the respondent's statements in response to questions in the interview; a topic generation process for generating candidate questions for the respondent based on the attention words; an information presentation process for presenting the generated question candidates; An interview support program that allows a computer to execute the above.

Images