JP7764025B2 - Survey response data processing device, survey response data processing method, and survey response data processing program - Google Patents

Description

The present invention relates to the technical fields of a survey response data processing device, a survey response data processing method, and a survey response data processing program. More specifically, the present invention relates to a survey response data processing device, a survey response data processing method, and a program for the survey response data processing device that processes response data indicating responses to a survey conducted over a network such as the Internet in the form of requesting answers to questions.

So-called questionnaire surveys have traditionally been conducted for purposes such as market research. Previous methods for conducting questionnaire surveys have included, for example, mailing questionnaires to randomly selected homes or visiting those homes to request responses to the questionnaire, randomly selecting passersby on the street to respond, or calling randomly selected homes to obtain responses.

In contrast, in recent years, as smartphones have become more widespread, surveys are often conducted by sending question data over the Internet to the smartphones of those expected to respond, and then retrieving the responses again via the Internet. Such surveys using smartphones and the Internet have the advantage of being able to collect responses quickly and inexpensively.

However, the above-mentioned method of conducting questionnaire surveys using smartphones and the Internet has problems, such as the small size of smartphone screens and the fact that responses are often entered while respondents are on the move, which means that if there are a large number of questions or they cover a wide range of topics, the credibility of the answers decreases (i.e., only "careless answers" are obtained). More specifically, the following problems (a) to (c) actually exist.

(a) There is a certain number of respondents who give contradictory answers to different questions; more specifically, for example, respondents who answer that they have "visited" a certain place (such as a tourist spot), but at the same time answer that they "don't know" that place (hereinafter, such respondents will be referred to simply as "contradictory respondents").

(i) For example, there are a certain number of respondents who, even when asked a question containing dozens of beverages, will answer "I have never tried any of them," i.e., respondents who are likely to find it troublesome to read all the options (hereinafter, such respondents will be referred to simply as "low responders").

(c) For example, in a questionnaire survey about a product that, according to general statistics, has almost no teenage customers, if the number of teenage respondents who answered "I use it" is disproportionately higher than respondents of other age groups, there may be doubts about the statistical distribution of those teenage respondents and whether they can be treated as legitimate respondents. Such respondents are likely to be giving false answers because they want to receive compensation for their answers, for example.

Furthermore, if respondents such as those described in paragraphs (a) to (c) above were treated the same as other respondents who provided valid responses when compiling survey results, the statistical distribution of the survey results would be adversely affected, and there would be a possibility that the intentions of the plan and decision-making associated with the survey would be misleading. Therefore, it is necessary to exclude the responses of respondents such as those described in paragraphs (a) to (c) above from the compilation of survey results. Patent Document 1 below is an example of a document disclosing prior art that has been considered with the aim of meeting this demand.

The technology described in Patent Document 1 measures the response time required by respondents to answer questions, calculates a representative response time for each question, calculates a response time index for each respondent for the questions answered, and calculates an average by dividing the sum of the response time indexes by the number of questions answered. The survey is then compiled by excluding a predetermined percentage of responses with the smallest average value.

Patent No. 4795496

However, respondents such as those described in paragraphs (a) to (c) above are thought to be "accustomed to questionnaire surveys," and may, for example, deliberately manipulate their responses to make the response time appropriate, as described in Patent Document 1, or may manipulate their responses to make them appear less appropriate by diversifying their own response trends. As a result, there is a problem in that respondents such as those described in paragraphs (a) to (c) above may appear to be legitimate respondents.

Furthermore, excluding response data obtained from respondents as described in paragraphs (a) through (c) above is possible for the survey implementer (e.g., the research company conducting the survey) who obtains the response data itself as a result of the survey. However, it is difficult for an analyst (e.g., the survey analysis company conducting the analysis) who analyzes the response data obtained from the survey implementer to distinguish between response data obtained as a result of inappropriate response manipulation such as the above and other legitimate (i.e., highly credible) response data. Therefore, there is also the problem that the analyst is forced to perform an analysis of the entire response data, including the response data that is a result of the response manipulation (in other words, an inaccurate analysis).

Furthermore, when considering using AI (Artificial Intelligence) to increase the number of response data (expand response data) resulting from a questionnaire survey, such as in the data generation method by the inventors of the present invention described in JP 2021-179865 A, if the original response data contains response data obtained from respondents such as those described in items (a) to (c) above, the AI will expand the response data based on the response data that contains this data. As a result, there is also the problem of a decrease in the quality of the expanded response data.

The present invention was made in consideration of the above-mentioned problems, and one example of its objective is to provide a survey response data processing device, a survey response data processing method, and a program for such a survey response data processing device that can effectively filter out low-quality responses that should be filtered out when compiling survey responses and when subsequently analyzing those responses.

In order to solve the above problem, the invention of claim 1 includes an answer data acquisition means such as a processing unit that acquires a plurality of answer data including content data indicating the content of the answer to a questionnaire conducted via a network in the form of requesting answers to questions and identification data for identifying the answer; an extraction means such as an inconsistent answer cleaning unit that extracts, from the acquired answer data, low-quality answer data corresponding to a low-quality answer that is an answer that falls under a low-quality standard that is preset as the answer; a deletion means such as an inconsistent answer cleaning unit that deletes the low-quality answer data from the acquired answer data based on the identification data indicating the extracted low-quality answer data; and a deletion unit that deletes the low-quality answer data from the acquired answer data after the low-quality answer data has been deleted. and a generating means such as a generating unit that generates new complementary answer data, the number of which corresponds to the number of deleted low-quality answer data, to replace the deleted low-quality answer data, based on the answer data after the deletion, wherein the low-quality criteria are configured to be any one or all of : (i) a first low-quality criterion: when the answers to the multiple-choice questions are mutually contradictory, each answer is designated as a low-quality answer; (ii) a second low-quality criterion: when the number of selections in the answer to the multiple-choice question is less than a predetermined number, the answer is designated as a low-quality answer; and (iii) a third low-quality criterion: when the answer by an answerer belonging to a group of answerers different from the group of answerers from which the answer is expected, the answer is designated as a low-quality answer.

In order to solve the above-mentioned problems, the invention of claim 6 is an answer data processing method executed in an answer data processing device including answer data acquisition means, extraction means, deletion means, and generation means, the answer data processing method including an answer data acquisition step of acquiring, by the answer data acquisition means, a plurality of answer data each including content data indicating the content of the answer to a questionnaire conducted via a network in a format in which an answer to a question is requested, and identification data for identifying the answer; an extraction step of extracting, by the extraction means, low-quality answer data corresponding to a low-quality answer that is an answer that falls under a low-quality standard that is preset as the answer, from the acquired answer data; and an extraction step of deleting, by the deletion means, the low-quality answer data from the acquired answer data based on the identification data indicating the extracted low-quality answer data. and a generating step of generating new complementary answer data by the generating means based on the answer data after the low-quality answer data has been deleted, the complementary answer data being in a number corresponding to the number of deleted low-quality answer data and replacing the deleted low-quality answer data, wherein the low-quality criteria are configured to be any one or all of: (i) a first low-quality criterion: when the answers to the multiple-choice questions are mutually contradictory, the answers are designated as low-quality answers; (ii) a second low-quality criterion: when the number of selections in the answer to the multiple-choice question is less than a predetermined number, the answer is designated as low-quality answers; and (iii) a third low-quality criterion: when the answer to the multiple-choice question is selected less than a predetermined number, the answer is designated as low-quality answers.

In order to solve the above problem, the invention of claim 7 provides a computer included in a response data processing device, comprising: response data acquisition means for acquiring a plurality of pieces of response data, each of which includes content data indicating the content of the response to a questionnaire conducted via a network in a format in which the response is requested to a question, and identification data for identifying the response ; extraction means for extracting, from the acquired response data, low-quality response data corresponding to a low-quality response that is an answer that falls within a low-quality standard that is preset as the answer; deletion means for deleting the low-quality response data from the acquired response data based on the identification data indicating the extracted low-quality response data; and deletion means for deleting the low-quality response data based on the answer data after the low-quality response data has been deleted. and a generating means for generating new complementary answer data corresponding to the number of deleted low-quality answer data to replace the deleted low-quality answer data, wherein the low-quality criteria are configured to be any one or all of: (i) a first low-quality criterion: when the answers to the multiple-choice questions are mutually contradictory, the answers are designated as low-quality answers; (ii) a second low-quality criterion: when the number of selections in the answer to the multiple-choice question is less than a predetermined number, the answer is designated as low-quality answers; and (iii) a third low-quality criterion: when the answer from an answerer belonging to a group of answerers different from the group of answerers from which the answer is expected, the answer is designated as low-quality answers.

According to the invention of any one of claims 1, 6 , and 7 , a plurality of pieces of answer data corresponding to a questionnaire conducted via a network in the form of requesting answers to questions are acquired, and low-quality answer data corresponding to low-quality answers are extracted and deleted from the answer data. At this time, any or all of the first to third low-quality criteria are used as low-quality criteria for extracting low-quality answers. Therefore, low-quality answers that should be excluded when collecting the answers to the questionnaire and when subsequently analyzing the answers can be effectively excluded.
Furthermore, based on the answer data after the low-quality answer data has been deleted, new complementary answer data is generated in a number corresponding to the number of deleted low-quality answer data. Therefore, by using the answer data from which the low-quality answer data has been deleted, the low-quality answer data can be deleted and a necessary and sufficient number of answer data can be obtained.

To solve the above problem, the invention described in claim 2 provides the response data processing device described in claim 1, wherein the first low-quality criterion is that the responses to each of the multiple questions included in one questionnaire are mutually contradictory, and the extraction means is configured to extract the low-quality response data corresponding to the low-quality response that falls under the contradiction from the acquired response data.

According to the invention described in claim 2, in addition to the effects of the invention described in claim 1, the first low-quality criterion is that the answers to multiple questions included in a single questionnaire are mutually contradictory, and low-quality answer data corresponding to the low-quality answers that fall under the contradiction is extracted from the answer data, so that low-quality answer data can be appropriately extracted and deleted using an appropriate first low-quality criterion.

In order to solve the above problem, the invention described in claim 3 provides a response data processing device described in claim 2, wherein the first low-quality criterion is either first low-quality criterion (a), which is the first low-quality criterion described in claim 2, or first low-quality criterion (b), which is that the questions included in the preliminary survey corresponding to the questionnaire and the questions included in the main survey corresponding to the questionnaire are similar and the answers to the questions are different from each other; the extraction means extracts low-quality answer information corresponding to the low-quality answers that fall into either of the above categories from the acquired response information; each of the first low-quality criteria is scored so that the first low-quality criterion (b) is lower in quality as an answer than the first low-quality criterion (a); the device further includes a determination means for determining the number of low-quality response data to be deleted based on the scored first low-quality criterion and a preset deletion number criterion; and the deletion means is configured to delete the determined number of low-quality answer data from the acquired response data.

According to the invention described in claim 3, in addition to the effects of the invention described in claim 2, the first low-quality criterion (a) and the first low-quality criterion (b) are each scored, and a number of low-quality answer data determined based on each scored first low-quality criterion and a predetermined deletion number criterion is deleted from the answer data, making it possible to extract and delete low-quality answer data using more objective low-quality criteria.

In order to solve the above problem, the invention described in claim 4 is a response data processing device described in any one of claims 1 to 3 , wherein the third low-quality criterion is that the difference between the ratio of the number of respondents belonging to the respondent group from which the response is expected to be given to the total number of respondents and the ratio of the number of respondents belonging to the different respondent group who have given the response to the total number of respondents is equal to or greater than a predetermined standard, and the extraction means is configured to extract from the acquired response data the low-quality response data corresponding to the low-quality responses by respondents belonging to the different respondent group that meet the third low-quality criterion.

According to the invention of claim 4 , in addition to the effects of the invention of any one of claims 1 to 3 , when the difference between the proportion of respondents belonging to a respondent group from which an answer is expected and the proportion of respondents belonging to a respondent group different from the expected respondent group and who have given the answer is equal to or greater than a predetermined standard, low-quality answer data corresponding to low-quality answers given by respondents belonging to the different respondent group is extracted from the answer data. Thus, low-quality answer data can be properly extracted and deleted using an appropriate third low-quality criterion.

In order to solve the above problem, the invention described in claim 5 is the response data processing device described in claim 1 , wherein the generation means is configured to newly generate the complementary response data by referring to the distribution of the entire acquired response data.

According to the invention described in claim 5 , in addition to the effect of the invention described in claim 1 , new complementary answer data is generated by referring to the distribution of the entire original answer data, so that answer data that corresponds to the distribution of the original answer data and is necessary and sufficient in terms of the number of answers can be obtained.

As described above, according to the present invention, multiple pieces of response data corresponding to a survey conducted over a network in the form of requesting answers to questions are obtained, and low-quality response data corresponding to low-quality answers are extracted and deleted from the response data. At this time, any or all of the first to third low-quality criteria are used as the low-quality criteria for extracting low-quality answers.

Therefore, low-quality responses that should be excluded when collecting survey responses and when subsequently analyzing the responses can be effectively excluded.
Furthermore, based on the answer data after the low-quality answer data has been deleted, new complementary answer data is generated in a number corresponding to the number of deleted low-quality answer data. Therefore, by using the answer data from which the low-quality answer data has been deleted, the low-quality answer data can be deleted and a necessary and sufficient number of answer data can be obtained.

FIG. 10 is a diagram showing an example of a response by a contradiction respondent in the first embodiment. FIG. 10 is a diagram showing an example of a response from a low responder in the first embodiment. FIG. 10 is a diagram illustrating an example of a statistically distributed low-quality response in the first embodiment. 1 is a block diagram showing a schematic configuration of a questionnaire response processing device according to a first embodiment; 10 is a flowchart showing a questionnaire response process according to the first embodiment. FIG. 10 is a block diagram showing a schematic configuration of a questionnaire response processing device according to a second embodiment.

Next, embodiments of the present invention will be described with reference to the drawings. Note that each embodiment described below is an embodiment in which the present invention is applied to a survey response processing device that improves the quality of a questionnaire survey conducted among users of terminal devices such as smartphones by deleting responses by respondents such as those described in items (a) to (c) above from the responses that result from the survey. Note that in the following description, responses by respondents such as those described in items (a) to (c) above will be collectively referred to as "low-quality responses."

(I) First embodiment
First, a first embodiment of the present invention will be described with reference to FIGS. 1 to 5. FIG.

(A) Low-quality answers
First, before explaining the questionnaire response processing device of the first embodiment, specific examples of low-quality responses that are the target of questionnaire response processing by the questionnaire response processing device will be explained. Note that Fig. 1 is a diagram showing example responses of a contradictory respondent in the first embodiment, Fig. 2 is a diagram showing example responses of a low-responder in the first embodiment, and Fig. 3 is a diagram showing example responses of a statistically distributed low-quality response in the first embodiment.

As described above, the low-quality answers that are the target of the survey response processing in the first embodiment include answers by inconsistent respondents described in section (a) above, answers by low responders described in section (b) above, and answers with questionable statistical distribution described in section (c) above. In the following description, answers by inconsistent respondents described in section (a) above will be referred to as "inconsistent answers," answers by low responders described in section (b) above will be referred to as "low-response answers," and answers with questionable statistical distribution described in section (c) above will be referred to as "answers with questionable distribution."

The above-mentioned contradictory answers are "mutually contradictory answers to different questions" given by the contradictory respondent. An example of this is answer DX1, as shown in Figure 1, in survey result D1 of a questionnaire survey regarding awareness of a drink (beverage) and whether or not the respondent has actually consumed it, in which the respondent selected "x" for the question item "I know," but also selected "o" for the question item "I have consumed it." The contradictory respondent who gave this contradictory answer ultimately answered that "I have consumed that drink" despite "not knowing that drink," and these answers are mutually contradictory; in other words, answer DX1 shown in Figure 1 is a typical example of the above-mentioned contradictory answer.

Next, the low response response is a "simple (lazy) response indicating that none of the many options apply" given by the low responder. An example of this is response DX2, as shown in Figure 2, in which a respondent selected "none of these apply" (i.e., there is not a single beverage that they have ever drunk) in survey result D2 of a questionnaire survey asking whether or not the respondent has ever drunk a large number of beverages. It is generally difficult to imagine a low responder who gave such a low response response answering "none of these apply" when there are as many beverages as shown in Figure 2 as options. Therefore, the respondent who gave response DX2 in Figure 2 is likely to be a respondent (low responder) who found it troublesome to read through all the options, for example.

Finally, the distribution-questionable answers are answers given by respondents who are suspected of giving false answers based on the statistical distribution, and an example of this is, as shown in Figure 3, answer DX3 (shown as " 16.5 " in Figure 3) given most frequently by teenage males in survey result D3 of a questionnaire survey about a certain product or service, even though the proportion of teenage males is extremely low (shown as "1.2" in Figure 3) in the statistical distribution of all customer data for that product or service. Respondents who give such distribution-questionable answers are considered to have doubts about the original statistical distribution and are therefore considered to be respondents who are suspected of giving false answers for a specific purpose, as described above, for example.

(B) Regarding the questionnaire response processing device of the first embodiment
Next, a questionnaire response processing device according to a first embodiment, which deletes low-quality responses from responses resulting from a question-and-answer survey conducted targeting users of the terminal device, will be described in detail with reference to Figures 4 and 5. Note that Figure 4 is a block diagram showing the general configuration of the questionnaire response processing device according to the first embodiment, and Figure 5 is a flowchart showing the questionnaire response processing according to the first embodiment. In Figure 4, "database" is appropriately represented as "DB." In the flowchart shown in Figure 5, the chronological flow of processing is indicated by solid arrows, and the data flow in each process is indicated by dashed arrows.

As shown in FIG. 4, the questionnaire response processing device S of the first embodiment is specifically realized by, for example, a server computer or a personal computer, and is composed of a processing unit 1 consisting of a CPU or the like, a recording unit 2 consisting of an HDD (Hard Disk Drive) or SSD (Solid State Drive) or the like, an operation unit 3 consisting of a keyboard, mouse, etc., and a display 4 consisting of a liquid crystal display, etc.

The processing unit 1 is also composed of a contradictory respondent cleaning unit 10, a low responder cleaning unit 11, a statistical distribution cleaning unit 12, and a generation unit 13.

In this case, the contradictory respondent cleaning unit 10, low responder cleaning unit 11, statistical distribution cleaning unit 12, and generation unit 13 may be realized by a hardware logic circuit including a CPU or the like constituting the processing unit 1, or may be realized in software by the CPU or the like reading and executing a program corresponding to the questionnaire response processing of the first embodiment described below (see FIG. 5). In this case, the program may be pre-recorded in the recording unit 2 and read by the CPU or the like, or the program may be recorded on an external server device (not shown) and retrieved and used by the CPU or the like via a network such as the Internet.

The processing unit 1 corresponds to an example of the "response data acquisition means" of the present invention, the inconsistent respondent cleaning unit 10, the low responder cleaning unit 11, and the statistical distribution cleaning unit 12 correspond to an example of the "extraction means" and an example of the "deletion means" of the present invention, respectively, and the inconsistent respondent cleaning unit 10 corresponds to an example of the "determination means" of the present invention. Furthermore, the low responder cleaning unit 11 corresponds to an example of the "classification means" and an example of the "second determination means" of the present invention, respectively, and the generation unit 13 corresponds to an example of the "generation means" of the present invention.

In the above configuration, the processing unit 1 of the questionnaire response processing device S is connectable to a survey results database 100 and a statistical database 102. The survey results database 100 is a database that stores response data, which is the results (responses) of a question-and-answer survey conducted for users of the terminal device. The response data includes content data indicating the content of the response and identification data for identifying the response or the respondent. The response data stored in the survey results database 100 includes response data for the low-quality responses (see Figures 1 to 3) and response data for responses that are not low-quality responses (i.e., normal, legitimate responses). In contrast, the statistical database 102 is a database that stores various general statistical data collected by, for example, public institutions or local governments. The data for the survey results database 100 and the statistical database 102 may be pre-recorded in the recording unit 2, or may be obtained via a network such as the Internet from an external server device (not shown) each time the questionnaire response processing of the first embodiment is executed.

Then, the survey response processing device S uses the survey results database 100 and the statistical database 102 to delete the response data of low-quality responses from the response data that constitutes the survey results database 100. In addition, the survey response processing device S generates complementary response data in a number equivalent to the total number of deleted response data (for example, the same number as the number of deleted response data) based on the response data that constitutes the survey results database 100 after the response data of low-quality responses has been deleted, while referencing the response data that constitutes the survey results database 100 before the deletion. The method for generating this complementary response data will be described in detail later.

Then, the survey response processing device S adds the generated complementary response data to the post-deletion survey results database 100 to generate a new high-quality database 103. At this time, the number of response data constituting the newly generated high-quality database 103 will be the same as the number of response data constituting the survey results database 100 before the response data of the low-quality responses was deleted.

More specifically, first, the inconsistent answerer cleaning unit 10 of the processing unit 1 of the questionnaire response processing device S uses a preset method for deleting inconsistent answers to delete the answer data of inconsistent answers from the answer data that make up the survey results database 100. This method for deleting inconsistent answers will be described in detail later.

Next, the low-response person cleaning unit 11 of the processing unit 1 uses a preset method for deleting low-response answers to delete the response data of low-response answers from the response data that makes up the survey results database 100. This method for deleting low-response answers will be described in detail later. The deletion of the response data of low-response answers by the low-response person cleaning unit 11 is carried out in parallel with the deletion of the response data of contradictory answers by the contradictory answerer cleaning unit 10.

Finally, the statistical distribution cleaning unit 12 of the processing unit 1 uses a preset method for deleting distribution-questioning answers to delete the answer data for distribution-questioning answers from the answer data constituting the survey results database 100, based on the statistical data stored in the statistical database 102. This method for deleting distribution-questioning answers will also be described in detail later. The deletion of the answer data for distribution-questioning answers by the statistical distribution cleaning unit 12 is carried out in parallel with the deletion of the answer data for contradictory answers by the contradictory answerer cleaning unit 10 and the deletion of the answer data for low-response answers by the low-response answer cleaning unit 11.

Then, after the response data constituting the survey results database 100 has been deleted for each of the contradictory responses, low-reflection responses, and distribution-questionable responses, the response data is temporarily recorded in the recording unit 2 as a cleaned database 101.

Next, the generation unit 13 of the processing unit 1 generates the above-mentioned complementary answer data in the same number as the total number of answer data deleted by the contradictory answerer cleaning unit 10, the low responder cleaning unit 11, and the statistical distribution cleaning unit 12, based on the answer data that constitutes the survey results database 100 after the answer data of the low-quality answers has been deleted, while referencing the answer data that constitutes the survey results database 100 before the deletion. The generation unit 13 then adds the generated complementary answer data to the survey results database 100 after the answer data of the low-quality answers has been deleted. In this way, the generation unit 13 generates high-quality answer data, which is answer data that should constitute the above-mentioned high-quality database 103 and is the same number as the number of answer data that constitutes the survey results database 100 before the answer data of the low-quality answers has been deleted, and stores this in the above-mentioned high-quality database 103.

The operations required to execute each of the functions described above are performed by the operation unit 3, and operation signals corresponding to these operations are output to the processing unit 1. Based on these operation signals, the processing unit 1 then executes the series of functions described above. Furthermore, information required to execute these functions is displayed, for example, on the display 4 and presented to the operator of the questionnaire response processing device S.

Next, the survey response processing of the first embodiment executed by the survey response processing device S will be explained in detail using Figures 4 and 5.

The survey response processing of the first embodiment, which is executed by the survey response processing device S having the above-described functions, begins, for example, when a power switch (not shown) of the survey response processing device S is turned on. It is assumed that the response data that is the subject of the survey response processing of the first embodiment is stored in advance in the survey results database 100.

When the survey response processing begins, the processing unit 1 of the survey response processing device S first acquires the response data to be processed from the survey result database 100 (step S1). The response data acquired from the survey result database 100 is then output in parallel to the inconsistent respondent cleaning unit 10, the low-responder cleaning unit 11, and the statistical distribution cleaning unit 12. The inconsistent respondent cleaning unit 10 then executes the inconsistent response deletion process (inconsistent respondent cleaning process) described above for the response data acquired from the survey result database 100 (step S2), the low-responder cleaning unit 11 executes the low-responder response deletion process (low-responder cleaning process) (step S3), and the statistical distribution cleaning unit 12 executes the distribution-questionable response deletion process (statistical distribution cleaning process) (step S4). The response data from the survey result database 100 from which the inconsistent responses, low-responder responses, and distribution-questionable responses have been deleted is temporarily stored in the cleaned database 101. In this case, if the number of response data (in other words, the number of respondents; the same applies below to the number of response data in each database) stored in the original survey results database 100 is 1,000, and 50 responses are deleted by the inconsistent respondent cleaning process (step S2), 100 responses are deleted by the low responder cleaning process (step S3), and 30 responses are deleted by the statistical distribution cleaning process (step S4), the number of response data stored in the cleaned database 101 will be 820.

(a) Regarding the contradictory respondent cleaning process of the first embodiment
The inconsistent answerer cleaning process of step S2 will now be described in more detail. In the inconsistent answerer cleaning process, a determination procedure (hereinafter, referred to as the "inconsistent answer determination procedure") is set in advance to determine inconsistent answers from answers indicated by the answer data acquired from the survey result database 100 in step S1. The inconsistent answer determination procedure is implemented by the inconsistent answerer cleaning unit 10 reading and executing a program or the like for implementing the procedure, such as a program pre-recorded in the recording unit 2. The inconsistent answer determination procedure of the first embodiment is not fixed to the inconsistent answerer cleaning unit 10 but can be changed based on, for example, the attributes of respondents to be excluded as inconsistent answerers of the first embodiment. The following examples of the inconsistent answer determination procedure of the first embodiment may be used individually, or two or more of the determination procedures may be used in combination.

A first example of such a contradictory answer determination procedure is a determination procedure in which, when answers to multiple questions included in a questionnaire survey (or corresponding to a questionnaire survey) are mutually contradictory, such as answer DX1 described using Figure 1, each answer is determined to be a contradictory answer.

Next, a second example of a procedure for determining contradictory answers is a more specific example of the first example of the procedure for determining contradictory answers, in which, even though questions included in a preliminary survey (so-called screening survey) corresponding to the questionnaire survey from which the response data recorded in the survey results database 100 was obtained and questions included in a main survey corresponding to that questionnaire survey are of the same nature, the answers to those questions differ (i.e., are contradictory) from each other, and the answers in the preliminary survey and the main survey are determined to be contradictory answers.

Next, as a third example of a procedure for determining inconsistent answers, the degree of inconsistency in an inconsistent answer may be assigned a score (score; the same applies below) in advance. More specifically, a score (score; the same applies below) corresponding to a respondent whose answer was determined to be inconsistent by the first example of the procedure for determining inconsistent answers above may be set to, for example, "-100 points," and this information may be recorded in association with the respondent. A score for a respondent whose answer was determined to be inconsistent by the second example of the procedure for determining inconsistent answers above may be set to "-70 points," and this information may be recorded in association with the respondent. The scores are then added up for each respondent, and respondents with relatively higher scores (negative numbers) after the addition are ranked lower as inconsistent respondents (in other words, closer to normal respondents; the same applies below).

Finally, as a fourth example of the procedure for determining a contradictory answer, the following method can be given as an example of how to actually delete the answer data of answers determined to be contradictory in the first to third examples of the procedure for determining a contradictory answer.
(a) A method of deleting all response data of answers determined to be inconsistent answers from the survey result database 100, and excluding all respondents who gave such inconsistent answers (i.e., inconsistent respondents) from the survey results. (b) A method of excluding inconsistent respondents whose answers were determined to be inconsistent answers in the third example above from the survey results according to the score they received as an inconsistent answer. More specifically, for example, a method of excluding the top 200 inconsistent respondents in descending order of the score for their inconsistent answers from the survey results, or a method of excluding the top 10 percent of inconsistent respondents with the highest scores from the survey results.

(b) Low-responder cleaning process in the first embodiment
Next, the low-responder cleaning process of step S3 will be described in more detail. In the low-responder cleaning process, a discrimination procedure (hereinafter referred to as the "low-responder response discrimination procedure") is set in advance to discriminate low-responder responses from among the responses indicated by the response data acquired from the survey result database 100 in step S1. The low-responder response discrimination procedure is realized by the low-responder cleaning unit 11 reading and executing a program or the like for implementing the procedure, such as a program pre-recorded in the recording unit 2. The low-responder response discrimination procedure of the first embodiment is not fixed to the low-responder cleaning unit 11, but is changeable based on, for example, the attributes of respondents to be excluded as inconsistent respondents of the first embodiment. Furthermore, the multiple examples of the low-responder response discrimination procedure of the first embodiment illustrated below may be used alone, or two or more of the discrimination procedures may be used in combination.

A first example of such a procedure for identifying low-response responses is a procedure in which the response data recorded in the survey results database 100 is analyzed using the so-called cluster analysis method, and responses from respondents (low responders) who belong to the cluster with the lowest response rate (i.e., the fewest specific options selected in the question) among the analyzed (classified) clusters are identified as low-response responses.

Next, a second example of a procedure for determining low-response responses is a procedure for determining that a response from a respondent (low responder) who answered "none of the options apply" to multiple (large number of) options in a question, such as response DX2 described using Figure 2, is a low-response response.

A third example of a procedure for identifying low-response responses is a procedure for identifying responses from respondents (low responders) who answer "not applicable" to all of a number of questions as low-response responses.

Furthermore, as a fourth example of the procedure for determining low response responses, the degree of low response in low response responses may be assigned a score in advance. More specifically, the score corresponding to a respondent belonging to the lowest response cluster in the first example of the procedure for determining low response responses may be recorded as "-90 points," with this information associated with the respondent; the score of a respondent whose answer was determined to be a low response response in the second example of the procedure for determining low response responses may be recorded as "-30 points," with this information associated with the respondent; and the score of a respondent whose answer was determined to be a low response response in the third example of the procedure for determining low response responses (when there are three questions) may be recorded as "-150 points," with this information associated with the respondent. Then, the scores for each respondent may be added up, and respondents with relatively larger scores (negative numbers) after the addition may be ranked as respondents with lower low response ratings.

Finally, as a fifth example of the procedure for determining low response responses, the following method can be given as an example of how to actually delete the response data of each of the responses determined to be low response responses in the first to fourth examples of the procedure for determining low response responses.
(a) A method of deleting all response data for answers determined to be low response answers from the survey result database 100 and excluding all respondents who gave such low response answers (i.e., low responders) from the survey results database 100; (b) A method of excluding low responders whose answers were determined to be low response answers in the fourth example above from the survey results database 100 according to the score of the low response answer. More specifically, for example, a method of excluding the top 200 low responders in descending order of the score of the low response answer from the survey results database, or a method of excluding the top 10% of low responders with the highest score from the survey results database; (c) A method of excluding respondents who answered "not applicable" to all three preset questions among respondents belonging to the lowest response cluster in the first example of the low response answer discrimination procedure from the survey results database; (d) A method of setting a limit in advance for excluding the inconsistent respondents and the low responders together from the survey results database, and excluding respondents below that limit from the survey results database;

(c) Statistical Distribution Cleaning Process of the First Embodiment
Next, the statistical distribution cleaning process of step S4 will be described in more detail. In this statistical distribution cleaning process, a discrimination procedure (hereinafter referred to as the "distribution-questioned answer discrimination procedure") is preset for discriminating distribution-questioned answers from among the answers indicated by the answer data acquired from the survey result database 100 in step S1 based on the statistical data stored in the statistical database 102. This distribution-questioned answer discrimination procedure is implemented by the statistical distribution cleaning unit 12 reading and executing a program or the like for implementing the procedure, such as a program pre-recorded in the recording unit 2. Note that the distribution-questioned answer discrimination procedure of the first embodiment is not fixed to the statistical distribution cleaning unit 12 but can be changed based on, for example, the attributes of respondents who are to be excluded as having given the distribution-questioned answer of the first embodiment. Furthermore, the multiple examples of the distribution-questioned answer discrimination procedure of the first embodiment shown below may be used individually, or two or more of these discrimination procedures may be used in combination.

A first example of such a procedure for determining whether an answer is distributionally questionable is a procedure for determining whether an answer from a respondent who belongs to a statistically inappropriate segment (see "DX3" in Figure 3) is a distributionally questionable answer based on the statistical data stored in the statistical database 102.

More specifically, suppose the statistical data is collected by gender and age group in ten-year age segments. Then, when the respondents whose response data is stored in the survey results database 100 are divided into ten-year age segments by gender and age group in the same way as the statistical data, if the number of respondents in each segment deviates by ±10% or more from the number of people belonging to the corresponding segment of the statistical data (for example, if the statistical data indicates that only 3% of respondents are teenage males, but 18% of the respondents whose response data is stored in the survey results database 100 are teenage males), the response of the teenage male respondent whose response data is stored in the survey results database 100 is determined to be a response with questionable distribution.

Next, as a second example of a procedure for determining distribution-questionable answers, as in the statistical distribution cleaning process of the first embodiment, segments of statistical data segmented in the same manner as in the first example above (hereinafter referred to as "statistical data segments") are compared with segments of response data stored in the survey results database 100 (hereinafter referred to as "response data segments"), and the response data belonging to each response data segment, whose number of responses does not correspond to the number of responses in the corresponding statistical data segment, is reduced according to the same rules until the number of respondents belonging to that response data segment is within a ±5% deviation from the number of responses in the corresponding statistical data segment. A procedure for determining whether the responses in the response data to be reduced are distribution-questionable answers is then exemplified. In this case, respondents whose responses in the response data segment to be deleted are targeted for reduction, for example, starting with those ranked highest in terms of scoring as inconsistent respondents or low responders (i.e., respondents most likely to be inconsistent respondents or low responders).

After the low-quality answers are deleted by the inconsistent answerer cleaning process (step S2), the low responder cleaning process (step S3), and the statistical distribution cleaning process (step S4), the generation unit 13 of the processing unit 1 then generates new complementary answer data in the same number as the total number of deleted low-quality answer data, based on the answer data that constitutes the survey results database 100 after the answer data of the low-quality answers has been deleted, while referencing the answer data that constitutes the survey results database 100 before the deletion, using, for example, the data generation method developed by the inventors of the present invention described in JP 2021-179865 A (step S5). Using the AI-based data generation method described in JP 2021-179865 A to generate the complementary answer data in step S5 can improve the overall quality of the answer data complemented by the complementary answer data.

More specifically, if 50 answers are deleted by the inconsistent answerer cleaning process (step S2), 100 answers are deleted by the low responder cleaning process (step S3), and 30 answers are deleted by the statistical distribution cleaning process (step S4), the generation unit 13 generates 180 new complementary answer data using the data generation method, etc. The generation unit 13 then acquires the answer data of the survey results database 100 after the low-quality answer data has been deleted from the cleaned database 101, and adds the complementary answer data generated in step S5 to the acquired answer data (step S6). As a result, the generation unit 13 generates high-quality answer data in the same number as the number of answer data constituting the survey results database 100 before the low-quality answer data was deleted, and stores this in the high-quality database 103 (step S6).

The processing unit 1 then determines whether or not to terminate the survey response processing of the first embodiment, for example, by performing a termination operation on the operation unit 3 (step S7). If the determination in step S7 is that the survey response processing is to be terminated (step S7: YES), the processing unit 1 terminates the survey response processing. On the other hand, if the determination in step S7 is that the survey response processing is to be continued, for example, with another survey results database 100 as the target (step S7: NO), the processing unit 1 returns to step S1 and continues the processing described above with the other survey results database 100 as the target.

As described above, according to the survey response processing by the survey response processing device S of the first embodiment, multiple pieces of response data corresponding to a survey conducted over a network in the form of requesting answers to questions are acquired (see step S1 in Figure 5), and low-quality response data are extracted from the response data and deleted (steps S2 to S4 in Figure 5). At this time, response data for inconsistent responses (see step S2 in Figure 5), low-response responses (see step S3 in Figure 5), and distribution-questionable responses (see step S4 in Figure 5) are deleted as low-quality response data, thereby effectively excluding (deleting) low-quality response data that should be excluded when compiling survey responses and when subsequently analyzing those responses.

Furthermore, when discriminating answer data for contradictory answers using the first example of the contradictory answer discrimination procedure to the fourth example of the contradictory answer discrimination procedure (see step S2 in Figure 5), the answer data for the contradictory answer can be extracted and deleted appropriately and properly.

In this case, when identifying inconsistent answer data using the third or fourth example of the inconsistent answer identification procedure described above (see step S2 in Figure 5), it is possible to extract and delete low-quality answer data using objective criteria based on scoring.

Furthermore, when identifying response data for low-response responses using the first example of the low-response response discrimination procedure or the third example of the low-response response discrimination procedure (see step S3 in Figure 5), the response data for low-response responses can be appropriately extracted and deleted using appropriate criteria.

Furthermore, when using the fourth example of the low-response response determination procedure described above, response data for low-response responses can be deleted using more objective criteria.

Furthermore, when determining the distribution of questioned answers using the above-mentioned distribution of questioned answers determination procedure (see step S4 in Figure 5), the distribution of questioned answers can be appropriately extracted and deleted using appropriate criteria.

Furthermore, new complementary answer data for complementary answers in the same number as the deleted low-quality answers is generated based on the answer data after the answer data for low-quality answers has been deleted (see steps S5 and S6 in Figure 5).By using the answer data after the answer data for low-quality answers has been deleted, it is possible to construct a high-quality database 103 in which the answer data for low-quality answers has been deleted and which has a necessary and sufficient number of answer data.

In this case, if a data generation method using AI, for example, is used to generate the complementary response data, the quality of the entire response data complemented by the complementary response data (i.e., the entire response data of the complemented answers in the questionnaire survey) can be improved. Furthermore, the number of complementary response data generated in steps S5 and S6 above may be the same as the number of deleted low-quality answers, or it may be a number that is smaller than the number of deleted low-quality answers and corresponds to the number of deleted low-quality answers.

Furthermore, when generating the complementary response data, the new complementary response data is generated by referencing the distribution in the original survey results database 100, making it possible to construct a high-quality database 103 that corresponds to the distribution in the original survey results database 100 while also having a necessary and sufficient number of responses.

In the questionnaire response processing of the first embodiment shown in FIG. 5, the inconsistent respondent cleaning processing (step S2), the low respondent cleaning processing (step S3), and the statistical distribution cleaning processing (step S4) are described as being executed in parallel. However, these cleaning processes may also be executed serially, i.e., in the order of inconsistent respondent cleaning processing (step S2) → low responder cleaning processing (step S3) → statistical distribution cleaning processing (step S4).

Furthermore, in the survey response processing of the first embodiment described above, the case was explained in which the response data of low-quality answers to be deleted in the survey response processing is the above-mentioned response data of contradictory answers, response data of low-response answers, or response data of answers with questionable distribution. However, in addition to the above-mentioned response data of contradictory answers, the low-quality response data to be deleted may also include response data of low-time answers, where the response time required by the survey respondent to answer the included questions is shorter than a preset threshold. In this case, the system may be configured to delete response data that corresponds to any or all of the above-mentioned response data of contradictory answers, response data of low-response answers, response data of answers with questionable distribution, or the above-mentioned response data of low-time answers.

(II) Second embodiment
Next, a second embodiment of the present invention will be described with reference to Fig. 6. Fig. 6 is a block diagram showing the general configuration of a questionnaire response processing device according to the second embodiment.

In the survey response processing of the first embodiment described above, response data for low-quality responses is deleted from the survey results database 100, and new complementary response data is generated in the same number as the deleted response data. As a result, high-quality response data is generated in the same number as the number of response data constituting the survey results database 100 before the low-quality response data was deleted, and stored in the high-quality database 103. In contrast, in the database generation processing of the second embodiment described below, new response data is further generated by the generation unit 13, and a new large-scale database is generated, for example as a virtual market database, with an even larger number of response data than the number of response data in the high-quality database 103 of the first embodiment.

The hardware configuration of the questionnaire response processing device of the second embodiment is basically the same as the hardware configuration of the questionnaire response processing device S of the first embodiment. Therefore, in the following explanation, components that are similar to those of the questionnaire response processing device S are assigned the same component numbers, and detailed explanations will be omitted.

As shown in FIG. 6, the generation unit 13A of the processing unit 1A of the questionnaire response processing device S1 of the second embodiment, in addition to the functions of the generation unit 13 of the first embodiment, further generates a predetermined number of new response data corresponding to the number of high-quality data 103, for example, in a manner similar to that of the generation unit 13, based on the response data in the high-quality database 103, while referring to, for example, the distribution of statistical data constituting the statistical database 102.

The generation unit 13A then combines the newly generated response data with the response data already stored in the high-quality database 103 (i.e., expands the response data) to generate response data that constitutes the large-scale database 110 of the second embodiment, and stores the response data in the large-scale database 110.

The questionnaire response processing of the second embodiment described above not only provides the same effects as the questionnaire response processing of the first embodiment, but also has the effect of making it possible to easily construct a large-scale database 110.

In addition, according to the questionnaire response processing of the second embodiment described above, when a large-scale virtual market database is constructed by expanding response data using complementary response data (e.g., generated using AI) similar to that of the first embodiment, the quality of the data contained in the virtual market database can also be improved.

As explained above, the present invention can be used in the field of processing response data that makes up a database, and can produce particularly significant effects when applied to the field of improving the quality of that response data.

1, 1A Processing unit 2 Recording unit 3 Operation unit 4 Display 10 Inconsistent respondent cleaning unit 11 Low responder cleaning unit 12 Statistical distribution cleaning unit 13, 13A Generation unit 100 Survey result database 101 Cleaned database 102 Statistical database 103 High-quality database 110 Large-scale database S, S1 Questionnaire response processing device D1, D2, D3 Survey results DX1, DX2, DX3 Responses

Claims (7)

an answer data acquisition means for acquiring a plurality of answer data including content data indicating the content of the answers to a questionnaire conducted via a network in a format requesting answers to questions, and identification data for identifying the answers;
An extraction means for extracting low-quality answer data corresponding to a low-quality answer that is an answer that falls under a low-quality standard that is preset as the answer from the acquired answer data;
a deletion means for deleting the extracted low-quality response data from the acquired response data based on the identification data indicating the extracted low-quality response data;
a generating means for generating new complementary answer data in a number corresponding to the number of deleted low-quality answer data, based on the answer data after the low-quality answer data has been deleted, to replace the deleted low-quality answer data;
Equipped with
The low quality standard is:
(i) a first low-quality criterion: when the answers to a plurality of the questions are mutually contradictory, the answers are determined to be low-quality answers;
(ii) a second low-quality criterion: determining an answer to the multiple-choice question in which the number of selections is less than a predetermined number as the low-quality answer; and
(iii) Third low-quality criterion: determining that the answer given by a respondent belonging to a different respondent group from the respondent group from which the answer is expected is a low-quality answer;
Response data processing device characterized in that it is any one or all of the above. 2. The response data processing device according to claim 1,
the first low-quality criterion is that the answers to the questions included in one of the questionnaires are mutually contradictory;
The answer data processing device is characterized in that the extraction means extracts the low-quality answer data corresponding to the low-quality answer corresponding to the contradiction from the acquired answer data. 3. The response data processing device according to claim 2,
The first low quality criterion is
A first low-quality criterion (a) according to claim 2, or
A first low-quality criterion (b) is that the questions included in the preliminary survey corresponding to the questionnaire and the questions included in the main survey corresponding to the questionnaire are of the same meaning, and the answers to the questions are different from each other;
Either
The extraction means extracts low-quality answer information corresponding to any of the low- quality answers from the acquired answer information,
Each of the first low quality standards is scored so that the first low quality standard (b) is lower in quality as the answer than the first low quality standard (a),
The method further includes determining means for determining the number of the low-quality response data to be deleted based on each of the first low-quality criteria scored and a predetermined deletion number criterion;
The answer data processing device, wherein the deletion means deletes the determined number of low-quality answer data from the acquired answer data. 4. The response data processing device according to claim 1,
the third low-quality criterion is that the difference between the ratio of the number of respondents belonging to the group of respondents from which the answer is expected to be given to the total number of respondents and the ratio of the number of respondents belonging to the different group of respondents who have given the answer to the total number of respondents is equal to or greater than a predetermined standard;
The answer data processing device is characterized in that the extraction means extracts the low-quality answer data corresponding to the low-quality answers by respondents belonging to the different answer group that fall under the third low-quality criterion from the acquired answer data . 2. The response data processing device according to claim 1 ,
The answer data processing device, wherein the generating means generates the complementary answer data by referring to the distribution of the entire acquired answer data . A response data processing method executed in a response data processing device including a response data acquisition means, an extraction means, a deletion means, and a generation means,
a response data acquisition step of acquiring, by the response data acquisition means, a plurality of response data pieces, each of which includes content data indicating the content of the response to a questionnaire conducted via a network in a format requesting an answer to a question, and identification data for identifying the response;
an extraction step of extracting, from the acquired answer data, low-quality answer data corresponding to a low-quality answer that is an answer that falls under a low-quality standard that is preset as the answer by the extraction means;
a deleting step of deleting the extracted low-quality response data from the acquired response data by the deleting means based on the identification data indicating the extracted low-quality response data;
a generating step of generating, by the generating means, new complementary answer data in a number corresponding to the number of deleted low-quality answer data, based on the answer data after the low-quality answer data has been deleted, to replace the deleted low-quality answer data;
Including,
The low quality standard is:
(i) a first low-quality criterion: when the answers to a plurality of the questions are mutually contradictory, the answers are determined to be low-quality answers;
(ii) a second low-quality criterion: determining an answer to the multiple-choice question in which the number of selections is less than a predetermined number as the low-quality answer; and
(iii) Third low-quality criterion: determining that the answer given by a respondent belonging to a different respondent group from the respondent group from which the answer is expected is a low-quality answer;
The answer data processing method is characterized in that any one or all of the above is performed . The computer included in the response data processing device is
a response data acquisition means for acquiring a plurality of response data including content data indicating the content of the response to a questionnaire conducted via a network in a format requesting an answer to a question, and identification data for identifying the response;
an extraction means for extracting, from the acquired answer data, low-quality answer data corresponding to a low-quality answer that is an answer that falls under a low-quality standard that is preset as the answer;
a deletion means for deleting the extracted low-quality response data from the acquired response data based on the identification data indicating the extracted low-quality response data; and
a generating means for generating new complementary answer data in a number corresponding to the number of deleted low-quality answer data, based on the answer data after the low-quality answer data has been deleted, to replace the deleted low-quality answer data;
A response data processing program that functions as
The low quality standard is:
(i) a first low-quality criterion: when the answers to a plurality of the questions are mutually contradictory, the answers are determined to be low-quality answers;
(ii) a second low-quality criterion: determining an answer to the multiple-choice question in which the number of selections is less than a predetermined number as the low-quality answer; and
(iii) Third low quality criterion: Answers that belong to a different group of respondents than the group of respondents from whom the answer is expected.
determining the answer by the user as the low quality answer;
A program for processing answer data, characterized in that it is any one or all of the above .