JP2013089109A - Device, method and program for recommending question item - Google Patents

Abstract

PROBLEM TO BE SOLVED: To elicit high-quality and useful response from a respondent.SOLUTION: A question item recommending device 1 for sequentially recommending an item to be questioned next during an interview, stores, in a question item storage unit 13, a plurality of question items that are hierarchically configured according to the depth of the contents of the question items. When there is not much time left for the interview, the question item recommending unit 12 puts a high priority on a question shallow in the hierarchy and, when there is a lot of time left for the interview, puts the high priority on a question deep in the hierarchy and sequentially recommends it as next question item.

Description

  The present invention relates to a technique for recommending an item to be questioned next during an interview.

  In many research fields such as sociology and nursing, interview surveys are conducted to extract the problems of the subjects.

  There are various methods for interview surveys. In recent years, a method called “semi-structured interview” has been carried out in the field of engineering research in addition to sociology and nursing. This is a method in which only rough question items are determined in advance, and the order of questions and the depth of questions are changed according to the flow of conversation with interview respondents (hereinafter, respondents) (Non-Patent Document 1). reference).

Uwe Flick, "Introduction to Qualitative Research", Shunjusha, p.180-201

  As described above, the interview method that changes the order of questions and the depth of the question according to the flow of conversation with the respondent is an effective method for listening to the true intention of the respondent.

  However, since an interviewer (hereinafter referred to as a questioner) needs to select a question flexibly in accordance with the flow of the talk with the respondent, a very advanced interview technique is required for the questioner.

  In particular, in interviews aimed at qualitative research, the question items prepared in advance by the questioner are not asked in equal order. While considering the balance of the remaining time of the interview, listen to the prepared questions exhaustively to some extent, and if necessary, discard some questions (do not ask), while digging into specific questions To be heard. Thereby, it is possible to collect high-quality data while maintaining a certain degree of completeness.

  “Quality data” here refers to data that enables new discoveries that were not clear until now. Specifically, data with abundant variations in experiences, abstract Instead, it means data based on specific cases, data that includes detailed explanations and ideas about experiences.

  Therefore, if the questioner is too focused on “questioning all the items without omissions” within a limited time, he / she will not be able to ask deep questions. If you dig deeper into questions and spend too much time, there is a problem that you can ask even the basic questions that you wanted to hear.

  The present invention has been made in view of the above problems, and the problem is that even if a questioner is unfamiliar with an interview, it is possible to extract good quality useful data (answers) from the respondent. Is to make it.

  The question item recommendation device according to claim 1 is a question item recommendation device that sequentially recommends items to be asked next during an interview, and a plurality of pieces of question item data hierarchically structured according to the depth of the content of the question items And storing the plurality of question item data from the storage unit, and when the remaining time of the interview is low, the question item in the shallow hierarchy is preferentially recommended and the remaining time of the interview is large In this case, there is a recommendation unit that preferentially recommends question items in a deep hierarchy.

  According to the present invention, in a question item recommendation device that sequentially recommends items to be asked next during an interview, a plurality of question item data that is hierarchically structured according to the depth of the content of the question item is stored. When the remaining time of the interview is low, the question item in the shallow hierarchy is given priority, and when the remaining time in the interview is high, the question item in the deep hierarchy is given priority and recommended. Even so, it can make it possible to extract useful answers of good quality from respondents.

  The question item recommendation device according to claim 2, wherein, in the question item recommendation device according to claim 1, when the recommendation means recommends the shallow hierarchy question item preferentially and there is a remaining time for the interview, The deep question item is recommended.

  According to the present invention, when there is remaining time for an interview even if a question item in a shallow hierarchy is preferentially recommended, a better quality and useful answer is recommended from the respondent in order to recommend the deep question item. Can be pulled out.

  The question item recommendation device according to claim 3 is the question item recommendation device according to claim 1 or 2, wherein the plurality of question item data is classified according to the type of content of the question item, When the remaining time of the interview is large, the question item is classified into the same type as the current question item, and the question item at the next deeper level than the current question item is recommended.

  The question item recommendation device according to claim 4 is the question item recommendation device according to claim 3, wherein the recommendation means is shallowly classified into a type different from the current question item when the remaining time of the interview is small. It is characterized by recommending question items in the hierarchy.

  The question item recommendation method according to claim 5 is a question item recommendation method performed by a question item recommendation device that sequentially recommends items to be asked next during an interview. A storage step of storing a plurality of question item data hierarchically structured according to the storage means, and reading the plurality of question item data from the storage means, and a shallow hierarchy when the remaining time of the interview is small And a recommendation step that preferentially recommends question items in a deep hierarchy when the remaining time of the interview is large.

  According to the present invention, in a question item recommendation device that sequentially recommends items to be asked next during an interview, a plurality of question item data that is hierarchically structured according to the depth of the content of the question item is stored. When the remaining time of the interview is low, the question item in the shallow hierarchy is given priority, and when the remaining time in the interview is high, the question item in the deep hierarchy is given priority and recommended. Even so, it can make it possible to extract useful answers of good quality from respondents.

  The question item recommendation method according to claim 6 is the question item recommendation method according to claim 5, wherein the recommendation step has a remaining time of an interview even if the shallow question item is preferentially recommended. The deep question item is recommended.

  According to the present invention, when there is remaining time for an interview even if a question item in a shallow hierarchy is preferentially recommended, a better quality and useful answer is recommended from the respondent in order to recommend the deep question item. Can be pulled out.

  The question item recommendation method according to claim 7 is the question item recommendation method according to claim 5 or 6, wherein the plurality of question item data is classified according to the type of content of the question item, and the recommendation step includes When the remaining time of the interview is large, the question item is classified into the same type as the current question item, and the question item at the next deeper level than the current question item is recommended.

  A question item recommendation program according to claim 8 causes a computer to execute each step according to any one of claims 5 to 7.

  ADVANTAGE OF THE INVENTION According to this invention, even if it is a questioner unfamiliar with an interview, it can be made possible to draw a good-quality useful answer from an answerer.

It is a figure which shows the functional block structure of a question item recommendation apparatus. It is a figure which shows a question item list example. It is a figure which shows the hierarchical structure example of a question item. It is a figure which shows the whole operation | movement flow of a question item recommendation apparatus. It is a figure which shows the operation | movement flow of the recommendation calculation of a question item. It is a figure which shows the other example of a question item list.

  Hereinafter, an embodiment for carrying out the present invention will be described with reference to the drawings. However, the present invention can be implemented in many different modes and should not be construed as being limited to the description of the present embodiment.

  FIG. 1 is a diagram showing a functional block configuration of the question item recommendation device according to the present embodiment. The question item recommendation device 1 sequentially recommends items to be questioned next after each question is finished during an interview, and a conversation data input unit for inputting a conversation of a questioner who conducts an interview and its answerer 11, a question item recommendation unit 12 that analyzes a current question item from an inputted conversation and obtains a question item to be recommended next, and a question item storage that stores a plurality of question items to be asked in an interview The unit 13 and the question data output unit 14 for outputting the next recommended question item are mainly configured.

  As shown in the figure, the question item recommendation unit 12 analyzes the current question item and uses the question ID specifying unit 12a for specifying the ID of the question and the specified question ID to It is comprised with the recommendation question calculation part 12b which asks a question item by calculation.

  Next, the question item list stored in the question item storage unit 13 will be described. In the question item list, as illustrated in FIG. 2, the ID of the question to be asked in the interview, the classification ID to which the question belongs, the depth of the question, the question sentence as the question item, the time required for each question, Are stored in association with each other.

  The question ID is a unique identifier assigned to each question. In the present embodiment, as shown in FIG. 2, numbers in ascending order are sequentially given. Basically, all questions are sequentially asked in this order (1 → 2 → 3 → ...). However, according to the present invention, some questions are truncated (no questions are asked) according to the remaining time of the interview. ).

  The classification ID is an identifier of a category to which each question sentence belongs. Since the question is given according to the type of content of the question text, the question given the same ID indicates that the question is semantically similar.

  The depth of a question is the depth (degree of depth) of the content of a question sentence. As will be described later, this corresponds to the depth of the hierarchy when all questions are expressed in a hierarchical structure according to the depth of the content. For example, a question with a high level of abstraction of a question item is defined as a “shallow question”, a more specific question is defined as a “deep question”, or a question that simply asks the fact is a “shallow question”. It is possible to consider a method that makes a deep question a question that goes into your feelings and emotions. The depth to which each question item corresponds may be defined by the system provider such as the manufacturer or provider of the question item recommendation device 1, or the system user such as the questioner may store the question item storage unit 13. When registering a question, the depth may be registered subjectively.

  The question sentence is the content of the question item that the questioner plans to make during the interview. The question sentence may be in a format that can be referred to by the questioner at the time of the question, and may be registered as a sentence, a figure, a picture, or a symbol. A format in which a plurality of keywords that can recall the question content are listed may be used.

  The required time is the time required to ask each question. This is the time including the answer time by the respondent. This required time is an estimated required time, and when an interview is actually performed, the required time is not always within the required time.

  In the question item storage unit 13, in addition to the elements described above, the interviewable time is also registered. Interviewable time is the maximum time for an interview and is a value determined depending on the time of interview respondents and questioners, and is registered in advance. Although it is desirable that the total time required for each question item and the interviewable time match, it is not always necessary to match.

  The above-described registration of each element in the question item storage unit 13 may be performed in advance by the questioner before the interview, but the registrant does not necessarily have to be the questioner. For example, for the classification ID, after the questioner registers the question item, the system provider described above performs grouping according to the similarity of the contents of each question item, and automatically assigns the classification ID to each group. A method is also conceivable.

  FIG. 3 shows an example of the hierarchical structure of the question items shown in FIG. A plurality of question items (question item data) are hierarchically structured by classification ID and question depth. Two numbers in each block in FIG. 3 represent “classification ID−question depth”. For example, “1-2” means that the classification ID is 1 and the question depth is 2. Is shown.

  The depth of the hierarchy indicates how deep the question is, and in the examples of FIGS. 2 and 3, the smaller the value, the shallower the content of the question. That is, the question on the left side of FIG. 3 indicates that the depth of the question is shallower, and the question on the right side indicates that the depth of the question is deeper. The deep question of “1-1” is “1-2”, which is a further deep question. Is “1-3”.

  The depth of this question corresponds to the priority order of items to be questioned next. In this embodiment, the “shallow question” (more abstract question), which is the upper hierarchy, The priority order recommended as the question item is increased, and the priority level is lower for the “deep question” (more specific question) which is the lower hierarchy.

  In the present invention, using the question items prioritized (hierarchical structure) according to the depth of the question (hierarchy depth) as described above, the next recommended question item is determined as the remaining time of the interview. It is characterized in that it is determined according to.

  Specifically, the priority is to give priority to “shallow questions”, but if there is enough time remaining in the interview (that is, even if all of the “shallow questions” are asked) If there is room in the remaining time of the interview), the “priority question” with the lower priority is recommended in preference.

  In the present invention, since the next question item is determined according to the remaining time of the interview as described above, the above recommendation method is based on the remaining time of the interview. When the time is short, it is possible to prioritize the “shallow question”, and when the remaining time of the interview is large, the “deep question” can be prioritized and recommended as the next question item.

  Next, the operation of the question item recommendation device 1 will be described. FIG. 4 is a diagram showing an overall operation flow of the question item recommendation device 1 according to the present embodiment.

  First, in step S1, the conversation data input unit 11 acquires the conversation between the questioner and the respondent, inputs the conversation data (for example, voice data or text data) to the question item recommendation device 1, and asks the question. Output to the item recommendation unit 12. When the conversation data is input for the first time, the time when the interview is started is also output to the question item recommendation unit 12. If no conversation data is input, the conversation data input waiting state is continued (see step S0).

  Next, in step S2, the question ID specifying unit 12a converts the conversation data input from the conversation data input unit 11 into a format that can be interpreted, and collates it with the question item read from the question item storage unit 13, so that the current The question ID is specified and output to the recommended question calculation unit 12b. At that time, the interview start time is also output.

  Specifically, first, voice recognition of voice data input from the conversation data input unit 11 is performed to determine a speaker (questioner or answerer) and to determine a conversation break. Hereinafter, when the speaker of the divided conversation data is a questioner, the conversation data is referred to as “question data”.

  Thereafter, in order to determine which question in the question item list the questioner has made in the conversation data during the interview, the question item list stored in the question item storage unit 13 is referred to, and the question data, the question item, To identify the question ID and output the ID of the question.

  For example, with all the question items and question data stored in the question item storage unit 13, each sentence is morphologically analyzed, the similarity between both sentences is calculated, and the question item with the highest similarity is obtained. The question item indicated by the question data is determined.

  Note that the question item and the question data need not be automatically collated. For example, it is possible to adopt a method in which the questioner selects on the GUI the content that he / she asked during the interview. The morphological analysis is a known natural language processing technique performed using a computer, and any method can be adopted as long as the sentence and words used for the similarity calculation can be analyzed from the sentence.

  Next, in step S3, the recommended question calculation unit 12b receives the current question ID and the interview start time output from the question ID specifying unit 12a as input, and from the question item list read from the question item storage unit 13, The next recommended question item is calculated according to the remaining time. Hereinafter, a specific example of the recommendation calculation will be described in detail with reference to FIG.

  First, in step S3-0, the current question ID and the interview start time specified in step S2 are input to the recommended question calculation unit 12b. Hereinafter, the question ID number is assumed to be s.

  Next, in step S3-1, the interview remaining time LeftTime is calculated by subtracting the interview start time from the current time and further subtracting it from the available interview time in the question item list.

  Next, in step S3-2, a question from the next question ID (s + 1) to the last question of the input current question ID (s), that is, a total value L (n of required times of all remaining questions. ) For each question depth n (= 1, 2, 3,...).

  For example, in the example of FIG. 2, when the ID of the input question is 2, that is, ID (2), the total value L (1) of the time required for the question depth 1 is calculated from the question ID (3) to the question ID. Of (7), L (1) = 12 (minutes) from the total value of the required time of the question item of depth 1 (question ID = 4, 6). Similarly, the total value of the required time of the question depth 2 is L (2) = 10 (minutes), and the total value of the required time of the question depth 3 is L (3) = 10 (minutes).

Next, in step S3-3, the depth n of the question that can be completed within the remaining time of the interview is calculated by calculating n satisfying the equation (1).

For example, in the example of FIG. 2, when the remaining time Left of the interview is 20, Σ _{k = 1 to 1} L (k) = 12, and Σ _{k = 1 to 2} L (k) = 22. N satisfying the formula (1) is 1. That is, in this example, if the depth of the question is up to 1, it indicates that it is possible to finish all the questions by asking questions within the remaining time of the interview.

  Next, in step S3-4, the magnitude relationship between the depth d of the question ID (s + 1) next to the input current question ID (s) and the depth n of the question calculated in step S3-3. The candidate question to be recommended next is determined.

  Specifically, it is determined whether d is n or less, n + 1, or n + 2 or more. If n is n or less, the process proceeds to step S3-5. If d is n + 1, step S3-6 is performed. If n + 2 or more, the process proceeds to step S3-7.

  For example, in the example of FIG. 2, when the current question ID input is 2 and the interview remaining time LeftTime is 20, the depth n of the question calculated in step S3-3 is 1, Since the depth d of the question with question ID (3 (= 2 + 1)) is 3, which corresponds to n + 2 or more, the process proceeds to step S3-7.

  Next, when the determination result of step S3-4 shows that d is n or less, the process proceeds to step S3-8 with the question ID (s + 1) as the next candidate question in step S3-5.

On the other hand, if d is n + 1 as a result of the determination in step S3-4, all the requirements for the question that can be completed within the remaining time of the interview by calculating equation (2) in step S3-6. The remaining time S after the time subtraction is calculated. If the remaining time S is equal to or longer than the required time T of the question ID (s + 1), the question ID (s + 1) is set as the next candidate question and the process proceeds to step S3-8. If S <T, the process proceeds to step S3-7.

  However, n is the value of n calculated in step S3-3.

  If this procedure is followed, a question with a younger current question ID will be recommended with priority given to a deeper question. For example, in step S3-3, when it is calculated that all questions of depth 1 can be answered within the remaining time of the interview, for questions of depth 2, some questions can be asked, but the remaining The question will be truncated. In this procedure, no matter how many questions have a depth of 2, if the depth of the question with question ID (s + 1) is 2 and S ≧ T is satisfied, the question is preferentially recommended. Since the question with question ID (s + 1) is selected, time is used for the question with question ID (s + 1), and even if there is a question of depth 2 among the remaining questions, the question must be discarded. This is because the situation can be considered.

  Instead of recommending a question with a younger question ID, that is, a question deeper in the first part of the interview, in order to be able to ask a question that is equally deep regardless of the size of the question ID, for example, the following A method is conceivable.

  For all questions that satisfy the relationship of question ID = s> s + 1, select a question item with a required time S ≧ T in a question deeper than the depth n calculated in step S3-3 (question with depth n + 1). As a result, the question items are sequentially selected randomly from the selected questions, and the question items are continuously selected until the total time (ΣT) of the time required for the selected questions becomes S or less. If the selected question includes a question whose question ID is s + 1, the question proceeds to step S3-8 as the next candidate question, and if not included, Proceed to step S3-7. By adopting such a method, even deep questions can be made regardless of the size of the question ID.

  Next, when the determination result of step S3-4 is that d is n + 2 or more, or when S <T in step S3-6, the question ID is s + 2, s + 3, s + 4,. In this way, the question ID is larger than s + 1 and the question depth is n + 1 or less is searched. Then, the question ID of n + 1 or less searched first is u and rewritten as s = u−1. Thereafter, it is determined whether the depth of the question with question ID (s + 1 = u) is n + 1 or less than n. If n + 1, the process proceeds to step S3-6. If n + 1 or less, the process proceeds to step S3-5.

  In step S3-5 and step S3-6, when the remaining time of the interview is large, it belongs to the same category as the current question, and the “deep question” next to the current question is recommended. Further, by passing through step S3-7, when the remaining time of the interview is small, a “shallow question” of a classification different from the current question item is recommended.

  Finally, in step S3-8, the next candidate question calculated in step S3-5 or step S3-6 is output to the question data output unit 14. Thereafter, the question data output unit 14 presents the candidate question to the questioner. The presentation method may be any form such as voice, image, text, and the like.

  The operation described with reference to FIGS. 4 and 5 is performed every time the questioner asks a question, and the next recommended question item is presented as needed.

  As described above, according to the present invention, in the question item recommendation device 1 that sequentially recommends items to be asked next during an interview, a plurality of question items that are hierarchically structured according to the depth of the contents of the question items are stored. If the remaining time of the interview is low, give priority to the shallow question (“Shallow Question”), and give priority to the deep question (“Deep Question”) if the remaining time is long. Since it is recommended as the next question item, even a questioner who is unfamiliar with the interview can make it possible to draw out a good and useful answer from the respondent.

  In addition, according to the present embodiment, priority is given to recommending a “shallow question” having a high priority, but when there is sufficient time remaining in the interview (that is, all “shallow questions” are questions). If there is still enough time remaining in the interview), it is recommended to prioritize low-priority “deep questions” so that better quality useful answers can be drawn from respondents. can do.

  2 and 3 are simple examples in which there is only one question in the same hierarchy for a question item having a question depth of 2 or more. For example, as shown in FIG. This is also the case. This figure shows an example in which there are two question items with a question depth 2 and three question items with a question depth 3 when the classification ID is 1. Even in such a case, the question item to be recommended next can be determined by calculating based on the same procedure as described above.

  Note that the rules for assigning the question IDs do not have to be duplicated as described above, but when following the procedure shown in FIG. 5, the priority is given to the depth priority in the hierarchy direction, not the breadth priority in the classification direction. .

  Specifically, when the classification ID is 1, the question with the depth 1 of the question is ID (1), the first question with the depth 2 of the question is ID (2), and one of the questions with the depth 3 The eye question is ID (3). In the example of FIG. 6, since there is no question of depth 4, a question whose upper hierarchy question belongs to the same category in the second question of question depth 3 is set as ID (4). Next, since there are no more questions in which the upper-level question is the same question, the second question item of depth 2 is set to ID (5).

  Thus, by assigning an ID to a question sentence having a hierarchical structure with priority on the depth of the question, it becomes possible to correctly select the next candidate question according to the procedure shown in the flowchart of FIG.

  In addition, by assigning question IDs in this way, it is possible to proceed from a shallow question to a deep question little by little. If there is no deeper question, a similar question (a question in the upper hierarchy to which the question belongs) It is possible to recommend a question item in a natural manner, such as recommending a question that is the same as the previous question, and moving to the next question if there is no more similar question.

  Finally, the question item recommendation device 1 described in the present embodiment can be realized by a control unit such as a CPU or a storage unit such as a memory. The processing of each functional unit can be realized by a program that can execute each processing.

DESCRIPTION OF SYMBOLS 1 ... Question item recommendation apparatus 11 ... Conversation data input part 12 ... Question item recommendation part 12a ... Question ID specific | specification part 12b ... Recommended question calculation part 13 ... Question item memory | storage part 14 ... Question data output part S0-S3, S3-0 S3-8 ... Step (Processing Procedure)

Claims (8)

In the question item recommendation device that sequentially recommends the next item to be asked during the interview,
Storage means for storing a plurality of question item data hierarchically structured according to the depth of the content of the question item;
The plurality of question item data is read from the storage means, and when the remaining time of the interview is small, the question item of the shallow hierarchy is preferentially recommended, and when the remaining time of the interview is large, the deep question item is selected. Recommending means to give priority,
A question item recommendation device characterized by comprising: The recommendation means is:
2. The question item recommendation device according to claim 1, wherein when there is a remaining interview time even if the shallow question item is preferentially recommended, the deep question item is recommended. The plurality of question item data is classified according to the type of content of the question item,
The recommendation means is:
The question item according to claim 1 or 2, wherein when the remaining time of the interview is large, the question item is classified into the same type as the current question item, and a question item at a deeper level next to the current question item is recommended. Recommendation device. The recommendation means is:
4. The question item recommendation device according to claim 3, wherein when the remaining time of the interview is short, a question item in a shallow hierarchy classified into a type different from the current question item is recommended. In the question item recommendation method that is performed by the question item recommendation device that sequentially recommends items to be asked next during the interview,
By the question item recommendation device,
A storage step of storing a plurality of question item data hierarchically structured according to the depth of the content of the question item in a storage means;
The plurality of question item data is read from the storage means, and when the remaining time of the interview is small, the question item of the shallow hierarchy is preferentially recommended, and when the remaining time of the interview is large, the deep question item is selected. A recommendation step to preferentially recommend,
A question item recommendation method characterized by comprising: The recommendation step includes
6. The question item recommendation method according to claim 5, wherein when there is a remaining interview time even if the shallow question item is preferentially recommended, the deep question item is recommended. The plurality of question item data is classified according to the type of content of the question item,
The recommendation step includes
The question item according to claim 5 or 6, wherein when the remaining time of the interview is large, the question item is classified into the same type as the current question item, and a question item at a deeper level next to the current question item is recommended. Recommendation method.   A question item recommendation program that causes a computer to execute each step according to claim 5.

Images