Multi-stratified sampling method using machine learning algorithm, and system supporting same

Abstract

A multi-stratified sampling system using a machine learning algorithm, according to one embodiment of the present invention, comprises: a sample size calculation unit for calculating a sample size on the basis of the size of a target population; a multi-stratified sampling unit for performing multi-stratified sampling on the target population on the basis of the calculated sample size; and a multi-stratified sample group calculation unit for calculating a multi-stratified sample group from the target population on the basis of the performed multi-stratified sampling.

Description

Multilayered sampling method using machine learning algorithms and a system supporting the same

[0001] The present invention relates to a multilayered sampling method using a machine learning algorithm and a system supporting the same.

[0002] The present invention is the result of the research of the Blockchain Industry Advanced Technology Development Project (Project No.: RS-2024-00397538, Project Name: Development of Web3-based Public Opinion Survey Technology Guaranteeing Fairness, Anonymity, and Transparency), which was carried out with funding from the government (Ministry of Science and ICT) in 2024 and support from the Korea Institute of Information and Communications Planning and Evaluation.

[0003] In public opinion survey systems based on conventional technology, surveys are conducted using simple random sampling methods or samples provided by telecommunications companies. However, since these methods have difficulty adequately reflecting the representativeness of specific population groups, there is a high possibility of loss of representativeness, such as under- or over-sampling of specific population groups, which consequently limits the ability to guarantee the accuracy of the survey.

[0004] Furthermore, since the sample size and margin of error of the public opinion survey were calculated manually, there is a problem in securing the sampling results necessary to guarantee the consistency and accuracy of the survey results. Along with this, because the sampling error is not properly managed, there is a problem in that the survey results do not accurately reflect the actual opinions of the population.

[0005] The present invention aims to solve the problem described above by providing a multi-stratified sampling method using a machine learning algorithm that classifies a population into several subgroups and automatically extracts representative samples from each subgroup using statistical methods and machine learning techniques, and a system that supports the same.

[0006] The objectives of the present invention are not limited to those mentioned above, and other unmentioned objectives will be clearly understood by those skilled in the art from the description below.

[0007] A multi-stratified sampling system using a machine learning algorithm according to an embodiment of the present invention for achieving the above technical problem includes a sample size calculation unit that calculates a sample size based on the size of a target population, a multi-stratified sampling unit that performs multi-stratified sampling on a target population based on the calculated sample size, and a multi-stratified sample group calculation unit that calculates a multi-stratified sample group from a target population based on the performed multi-stratified sampling.

[0008] In addition, the multi-stratified sampling unit can perform multi-stratified sampling on the target population based on a clustering algorithm and a dimensionality reduction algorithm.

[0009] In addition, the multi-stratified sampling unit can perform multi-stratified sampling on the target population based on a k-means-based clustering algorithm and a PCA-based dimensionality reduction algorithm.

[0010] In addition, a multi-layered sampling system using a machine learning algorithm according to one embodiment of the present invention may further include a blockchain record unit that records the calculated multi-layered sample set on a blockchain using a smart contract.

[0011] In addition, a multi-layered sampling system using a machine learning algorithm according to one embodiment of the present invention may further include a public opinion survey unit that investigates and analyzes public opinion regarding a multi-layered sample group calculated using a sentiment analysis algorithm based on a natural language processing algorithm.

[0012] In addition, the public opinion research department can identify trends in public opinion fluctuations by inputting the accumulated results of public opinion surveys over a specified period into an artificial intelligence model.

[0013] In addition, the multi-stratification sampling unit can perform multi-stratification based on stratification criteria determined by at least one of demographic characteristics, regional characteristics, and socioeconomic characteristics.

[0014] A multi-stratified sampling method using a machine learning algorithm according to an embodiment of the present invention for achieving the above technical problem includes the steps of: calculating a sample size based on the size of a target population; performing multi-stratified sampling on a target population based on the calculated sample size; and calculating a multi-stratified sample group from a target population based on the performed multi-stratified sampling.

[0015] Additionally, the step of performing multi-stratified sampling on the target population based on the calculated sample size may include the step of performing multi-stratified sampling on the target population based on a clustering algorithm and a dimensionality reduction algorithm.

[0016] In addition, the step of performing multi-stratified sampling on the target population based on clustering algorithms and dimensionality reduction algorithms can perform multi-stratified sampling on the target population based on k-means-based clustering algorithms and PCA-based dimensionality reduction algorithms.

[0017] In addition, a multi-layered sampling method using a machine learning algorithm according to one embodiment of the present invention may further include the step of recording the calculated multi-layered sample set on a blockchain using a smart contract.

[0018] In addition, a multi-stratified sampling method using a machine learning algorithm according to one embodiment of the present invention may further include a step of investigating and analyzing public opinion regarding a multi-stratified sample group calculated using a sentiment analysis algorithm based on a natural language processing algorithm.

[0019] According to the present invention, by classifying a population into several subgroups and automatically extracting representative samples from each subgroup using statistical methods and machine learning techniques through multi-stratified sampling via cluster analysis, it is possible to perform sampling so that the results of a public opinion survey properly reflect the actual opinions of the population without loss of representativeness such as under- or over-sampling problems of specific population groups, and also automatically calculate the sample size and error margin of the public opinion survey samples, thereby increasing the consistency and accuracy of the public opinion survey results.

[0020] FIG. 1 is a configuration diagram of a multilayered sampling system using a machine learning algorithm according to one embodiment of the present invention.

[0021] FIG. 2 is an example of a minimum sample table derived according to the error limit of a multilayered sampling system using a machine learning algorithm according to one embodiment of the present invention.

[0022] FIG. 3 is an example diagram of the application of the k-means algorithm to a multilayered sampling system using a machine learning algorithm according to an embodiment of the present invention.

[0023] FIG. 4 is an example diagram of the application of a PCA algorithm to a multilayered sampling system using a machine learning algorithm according to an embodiment of the present invention.

[0024] FIG. 5 is an example of a multilayer sampling result of a multilayer sampling system using a machine learning algorithm according to an embodiment of the present invention.

[0025] FIG. 6 is a flowchart of a multilayer sampling method using a machine learning algorithm according to an embodiment of the present invention.

[0026] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings. The advantages and features of the present invention and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the attached drawings. However, the present invention is not limited to the embodiments disclosed below but may be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims. In relation to the description of the drawings, the same or corresponding components may be assigned the same reference number.

[0027]

[0028] Although terms such as "first," "second," etc. are used to describe various elements, components, and / or sections, it goes without saying that these elements, components, and / or sections are not limited by these terms. These terms are used merely to distinguish one element, component, or section from another. Accordingly, it goes without saying that the first element, first component, or first section mentioned below may be a second element, second component, or second section within the technical scope of the present invention.

[0029] The terms used herein are for describing embodiments and are not intended to limit the invention. In this specification, the singular form includes the plural form unless specifically stated otherwise in the text. As used herein, "comprises" and / or "made of" do not exclude the presence or addition of one or more other components, steps, actions, and / or elements to the mentioned components, steps, actions, and / or elements.

[0030] Unless otherwise defined, all terms used in this specification (including technical and scientific terms) may be used in a meaning commonly understood by those skilled in the art to which the present invention pertains. Additionally, terms defined in commonly used dictionaries are not to be interpreted ideally or excessively unless explicitly and specifically defined otherwise.

[0031] Hereinafter, the configuration of the present invention will be described in detail with reference to the attached drawings.

[0032]

[0033] FIG. 1 is a configuration diagram of a multilayered sampling system using a machine learning algorithm according to one embodiment of the present invention.

[0034] Referring to FIG. 1, a multi-stratified sampling system (100) using a machine learning algorithm according to one embodiment of the present invention (hereinafter referred to as the multi-stratified sampling system (100)) includes a sample size calculation unit (110) that calculates a sample size based on the size of a target population, a multi-stratified sampling unit (120) that performs multi-stratified sampling on a target population based on the calculated sample size, and a multi-stratified sample group calculation unit (130) that calculates a multi-stratified sample group from a target population based on the performed multi-stratified sampling.

[0035] In addition, a multi-layered sampling system (100) using a machine learning algorithm according to one embodiment of the present invention may further include a blockchain record unit (140) that records the calculated multi-layered sample group on a blockchain using a smart contract. In addition, a multi-layered sampling system (100) using a machine learning algorithm according to one embodiment of the present invention may further include a public opinion survey unit (150) that investigates and analyzes public opinion regarding the calculated multi-layered sample group using a sentiment analysis algorithm based on a natural language processing algorithm.

[0036] A sample size calculation unit (110) of a multilayered sampling system (100) according to one embodiment of the present invention calculates a sample size based on the size of the target population. Since the reliability of the results can be guaranteed only if the sample size is set appropriately, it is very important to calculate the minimum sample size required.

[0037] Specifically, the sample size calculation unit (110) according to one embodiment can derive the minimum sample size required to estimate a specific proportion of a population at a given confidence level and error range using [Equation 1].

[0038]

[0039] Here, n represents the required sample size and may refer to the minimum sample size. z is the z-value corresponding to the target confidence level (for example, the z-value is 1.96 at a 95% confidence level). p is the probability of success or the expected rate of occurrence of the characteristic to be investigated (for example, if the expected probability of an event occurring is 50%, p=0.5). e is the margin of error, which is the sampling error (for example, if a ±3% error is allowed, e=0.03).

[0040] Meanwhile, the sample size calculation unit (110) according to one embodiment can adjust the initial sample size when the size of the entire population is relatively small using [Equation 2]. For example, if the sample size initially calculated is larger than the entire population, it can be used to reduce the actual required sample size to prevent over-sampling of the population. Through such correction, the sampling can be prevented from depleting the population.

[0041]

[0042] n represents the initial sample size calculated using the basic formula of [Equation 1], and N represents the size of the entire population. represents the adjusted sample size considering the correction for the target population size. [Equation 2] can be used to adjust the initial sample size when the total population size is relatively small. For example, if the initially calculated sample size is large compared to the total population, it can be used to reduce the actual required sample size to prevent over-sampling the population. Through this correction, it is possible to ensure that sampling does not deplete the population.

[0043] In this way, the sample size calculation unit (110) according to one embodiment can calculate a sample size that can obtain accurate and reliable data even from a small population by using [Mathematical Formula 1] and [Mathematical Formula 2].

[0044] A sample size calculation unit (110) according to one embodiment can calculate a sample size based on a predetermined minimum sample table based on the target survey results as shown in FIG. 2. That is, the sample size calculation unit (110) according to one embodiment can calculate the minimum sample size required within a predetermined allowable error range, the sample error range, and the proportion range. The proportion on the y-axis of FIG. 2 represents the expected proportion of the characteristic or phenomenon to be investigated and is used when predicting the proportion of a population having specific conditions or characteristics.

[0045] For example, as shown in FIG. 2, a sample size calculation unit (110) according to one embodiment can calculate the minimum sample size (n=1068) required when the ratio is 50% and the sample error is 3% based on a predetermined minimum sample table.

[0046] A multi-stratified sampling unit (120) of a multi-stratified sampling system (100) according to one embodiment of the present invention performs multi-stratified sampling on a target population based on a calculated sample size. A multi-stratified sampling unit (120) according to one embodiment may perform multi-stratified sampling on a target population based on a dimensionality reduction algorithm and clustering.

[0047] The dimensionality reduction algorithm performed in the multilayer sampling unit (120) according to one embodiment includes PCA (Principal Component Analysis), LDA (Linear Discriminant Analysis), t-SNE (t-Distributed Stochastic Neighbor Embedding), UMAP (Uniform Manifold Approximation and Projection), Factor Analysis, and Isomap.

[0048] LDA can reduce dimensionality by maximizing between-class variance and minimizing within-class variance. t-SNE is a non-linear dimensionality reduction method designed to visualize the structure of high-dimensional data in lower dimensions. UMAP is similar to t-SNE but provides effective dimensionality reduction for data of various dimensions in a faster manner suitable for large-scale datasets. Factor analysis identifies potential factors based on correlations between variables and explains the data through this process. Isomap is a non-linear dimensionality reduction technique that reduces dimensionality while maintaining geodesic distances between data points.

[0049] A multilayered sampling unit (120) according to one embodiment can perform PCA for dimensionality reduction. Generally, PCA is used to reduce dimensionality while preserving important information in high-dimensional datasets, and helps with data visualization, removal of noise (unnecessary variability), and improvement of data processing speed. PCA identifies new axes (principal components) in a direction that maximizes the variance of the data, thereby enabling the extraction of the most important characteristics of the data.

[0050] In the multilayered sampling unit (120) according to one embodiment, since the computational cost of dimensionality-reduced data is lowered, the clustering algorithm (e.g., k-means) to be described later can be performed more quickly and efficiently. At this time, when applying PCA, an appropriate k value is input and performed, and then hierarchical clustering is applied to derive an even more appropriate k value.

[0051] That is, the multilayered sampling unit (120) according to one embodiment can derive a k value more suitable for PCA application by using a dendrogram obtained by performing hierarchical clustering. That is, as shown in FIG. 3, the hierarchical structure of the data and the formation status of data groups can be visually confirmed from the dendrogram. That is, the dendrogram forms clusters hierarchically based on the distance or similarity between each data point or data group, and represents them in a tree form.

[0052] A multilayered sampling unit (120) according to one embodiment may input a derived dendrogram into an artificial intelligence model (including a vision recognition machine learning algorithm) that performs vision recognition to calculate an appropriate k value, or a user of a multilayered sampling system (100) according to one embodiment may derive an appropriate k value based on the dendrogram. For example, it may be determined that k=4 is appropriate in the dendrogram shown in FIG. 3. A multilayered sampling unit (120) according to one embodiment may perform optimal dimensionality reduction by repeatedly performing PCA and hierarchical clustering in this manner.

[0053] A clustering algorithm performed in a multilayered sampling unit (120) according to one embodiment can divide data into several groups or clusters so that data points within each cluster have similar characteristics.

[0054] The clustering algorithm performed in the multilayer sampling unit (120) according to one embodiment may use k-Means clustering, DBSCAN (Density-Based Spatial Clustering of Applications with Noise), Hierarchical Clustering, Mean Shift, Spectral Clustering, and OPTICS (Ordering Points To Identify the Clustering Structure).

[0055] DBSCAN is a density-based clustering method that forms clusters based on regions where data points are densely concentrated. Agglomerative Hierarchical Clustering (AHC) is a type of hierarchical clustering method that starts every data point as an individual cluster and gradually merges them. The mean shift method identifies the density center of data points and automatically determines the number of clusters without requiring a predefined number. Spectral clustering constructs a graph using the similarity between data points and forms clusters using the spectrum (eigenvalues) of this graph. OPTICS performs density-based clustering similarly to DBSCAN and operates effectively across various densities.

[0056] A multilayered sampling unit (120) according to one embodiment may apply a k-means algorithm among clustering algorithms to the result of applying the aforementioned dimensionality reduction algorithm. For example, the multilayered sampling unit (120) according to one embodiment may divide the data into four clusters when performing the k-means algorithm. This indicates that each group forms a single cluster and that data points can be classified into four major groups, and implies that each group internally shares similar characteristics and is highly likely to be differentiated from other groups.

[0057] A multilayered sampling unit (120) according to one embodiment performs k-means-based clustering using the k value obtained as a result of performing a dimensionality reduction algorithm. Accordingly, k-means-based clustering classifies data by repeatedly adjusting the cluster center point so that data points within each cluster become closest to the cluster center point.

[0058] A multilayer sampling unit (120) according to one embodiment can perform a clustering algorithm and a dimensionality reduction algorithm as described above to derive a rotated component matrix table as shown in FIG. 4.

[0059] The Rotated Component Matrix is ​​used to identify major patterns or structures among variables in a dataset and to determine how each variable contributes to the principal components. The composition of the Rotated Component Matrix in Fig. 4 is as follows.

[0060] 1) Variable: Refers to variables included in the dataset, such as BMI, WAIST, Smoking, SEX, AGE, Drink, Diabetes_Status, etc.

[0061] 2) Components 1, 2, 3, and 4: As the main axes (principal components) explaining the variability of the data, these principal components simplify the structure of the data, allowing for a clear understanding of the relationships between variables.

[0062] 3) Component Loading

[0063] Loading: A value indicating how each variable contributes to each principal component; a higher loading value means that the variable contributes significantly to that principal component. For example, Component 1, BMI, shows a very high value of 0.948, which means that BMI plays an important role in forming the first principal component. The loading value for Component 2, Smoking, is 0.834, indicating that it makes a major contribution to the second principal component.

[0064] Components 3 and 4 show that age (AGE) and diabetes status (Diabetes_Status) have high loading values ​​in the third and fourth principal components, respectively, indicating that these components act as important variables in forming the principal components.

[0065] At this point, Varimax rotation with Kaiser Normalization can be applied as a rotation method. This method performs rotation in a way that increases the independence between variables (Varimax), allowing each principal component to distinguish between variables more clearly and making the interpretation of each principal component clearer, while applying a normalization method (Kaiser Normalization) during the rotation process minimizes the impact of differences in variable scales on the results.

[0066] As such, the rotational component matrix provides insights into how variables interact within the data and which variables best explain the data's variability, which can enhance data understanding and be used for further analysis or clustering.

[0067] A multi-stratified sampling unit (120) according to one embodiment can perform multi-stratified sampling by performing a clustering algorithm and a dimensionality reduction algorithm as described above to divide the population into several layers and then randomly extracting samples from each layer (thereby enabling the characteristics of each layer to be evenly represented).

[0068] A multi-layered sampling unit (120) according to one embodiment can perform multi-layered sampling to derive a multi-layered sampling table as shown in FIG. 5. That is, the multi-layered sampling table derived according to one embodiment as shown in FIG. 5 may include the frequency and ratio of each cluster for specific variables, and the results of a Chi-Square test to evaluate the statistical significance of the cluster distribution.

[0069] In the multi-stratified sampling table according to one embodiment shown in FIG. 5, information regarding each variable (e.g., gender, presence or absence of diabetes, drinking status, etc.) is recorded in the variable column, and how the data is divided into four different clusters is recorded in the cluster number (cluster 1, cluster 2, cluster 3, cluster 4) row. At this time, the clusters may be formed according to specific criteria (e.g., region, age group, health status, etc.).

[0070] The frequencies and proportions listed in the multistratified sampling table according to one embodiment represent the frequency and proportion of the corresponding variable within each cluster. The chi-square value and the p-value are used to evaluate the statistical significance of the cluster distribution for each variable within the multistratified sampling table. Generally, a large chi-square value and a very low p-value indicate that there is a statistically significant difference between clusters.

[0071] The process of the multilayered sampling unit (120) according to one embodiment performing multilayered sampling to derive a multilayered sampling table is as follows.

[0072] 1) Setting Stratification Criteria: Establish criteria for stratification according to the purpose of the study. For example, stratification criteria may include region, gender, age, etc.

[0073] 2) Sampling from each stratum: Random samples are extracted from each stratum by performing the clustering algorithm and dimensionality reduction algorithm described above according to the set stratification criteria.

[0074] 3) Data Collection and Cluster Analysis: Collect data from the extracted samples and analyze them to form clusters.

[0075] 4) Perform statistical analysis: Use the chi-square test to statistically evaluate how well the distribution of variables within each cluster represents the entire population.

[0076] 5) Summary of results and formation of a table: Based on the analysis results, create a table to summarize the frequency and proportion of variables and the chi-square test results for each cluster.

[0077] In this way, it is possible to verify how evenly various characteristics of the population are included through the multilayered sampling table derived from the multilayered sampling unit (120) according to one embodiment.

[0078] A multi-stratified sample group calculation unit (130) of a multi-stratified sampling system (100) according to one embodiment of the present invention calculates a multi-stratified sample group from a target population based on multi-stratified sampling performed in a multi-stratified sampling unit (120).

[0079] That is, the multi-stratified sample group calculation unit (130) according to one embodiment extracts samples from each stratum of the stratified population. By extracting samples by stratum in this way, the characteristics of each stratum can be reflected in the samples. At this time, the sampling can be performed randomly, and the sample size can be determined in proportion to the size of each stratum using a proportional stratification method.

[0080] In addition, a multi-stratified sample group calculation unit (130) according to one embodiment can form a single sample group by collecting samples extracted from each stratum. This sample group can be designed to comprehensively reflect various characteristics of the entire population. Through this process, information can be obtained from various segments of the population to improve the accuracy and reliability of the analysis.

[0081] Through such multi-stratified sampling, samples can be made to better reflect the characteristics of the population; by drawing samples equally from each stratum, selection bias is minimized, and more accurate data can be collected with fewer resources. This multi-stratified sampling process can be utilized as a data collection technique in various fields, such as statistical research, market research, and public policy design.

[0082] A blockchain record unit (140) of a multi-layered sampling system (100) according to one embodiment records the calculated multi-layered sample group on a blockchain using a smart contract. The smart contract is executed automatically according to predefined conditions, specific rules, and logic, thereby minimizing human intervention related to data processing and reducing the possibility of errors. For example, when specific demographic data is recorded on the blockchain, the smart contract automatically verifies and processes whether the data meets the criteria.

[0083] By using smart contracts to record sampled data on a blockchain, the data becomes immutable and difficult to forge, thereby enhancing the reliability of poll results. Furthermore, blockchain technology provides high transparency by making all transactions and data changes publicly verifiable. Data recorded on the blockchain is accessible across various applications and services, facilitating easy scalability when additional analysis or utilization is required.

[0084] A blockchain record unit (140) according to one embodiment can be configured so that a smart contract is executed and recorded on the blockchain only when the sampled data satisfies a specific criterion (e.g., when it satisfies a specific demographic criterion). To this end, preprocessing can be performed to refine and format the data obtained through multi-layered sampling before it is processed by the smart contract. At this time, the data obtained through multi-layered sampling can be encrypted to enhance personal information protection and data security.

[0085] The public opinion survey unit (150) of the multi-stratified sampling system (100) according to one embodiment can investigate and analyze public opinion regarding a multi-stratified sample group calculated using a sentiment analysis algorithm based on a natural language processing algorithm. Data is collected from various sources (online surveys, social media, news articles, etc.) written by public opinion survey subjects belonging to the sample group calculated by the multi-stratified sample group calculation unit (130) according to one embodiment. This data mainly consists of text data and may include the opinions, emotions, attitudes, etc. of the public opinion survey subjects.

[0086] The collected data can be preprocessed. Specifically, preprocessing steps such as tokenization, stop-word removal, stemming, or lemmatization may be performed. Additionally, the preprocessed data may undergo vectorization to convert it into a numerical form (e.g., TF-IDF (term frequency-inverse document frequency), word embeddings).

[0087] The data processed in this way can be used to perform sentiment analysis on the responses of poll participants using machine learning (e.g., SVM, Random Forest) or deep learning models (e.g., LSTM, CNN) (e.g., positive, negative, neutral).

[0088] Hereinafter, based on the above description, a multilayered sampling method using a machine learning algorithm according to an embodiment of the present invention will be described.

[0089]

[0090] FIG. 6 is a multilayer sampling method using a machine learning algorithm according to an embodiment of the present invention, and is a flowchart of a multilayer sampling method using a machine learning algorithm using a multilayer sampling system using a machine learning algorithm according to FIG. 1 of the present invention.

[0091] Referring to FIG. 6, a multi-stratified sampling method using a machine learning algorithm according to one embodiment of the present invention calculates a sample size based on the size of a target population by a sample size calculation unit (110) (S610), and a multi-stratified sampling unit (120) performs multi-stratified sampling on the target population based on the calculated sample size (S620).

[0092] At this time, the multi-stratified sampling unit (120) can perform multi-stratified sampling on the target population based on a clustering algorithm and a dimensionality reduction algorithm. When the multi-stratified sampling unit (120) performs multi-stratified sampling on the target population based on a clustering algorithm and a dimensionality reduction algorithm, it can perform multi-stratified sampling on the target population based on a k-means clustering algorithm and a PCA (Principal Component Analysis) dimensionality reduction algorithm.

[0093] The sample group calculation unit (130) calculates a multi-stratified sample group from the target population based on the multi-stratified sampling performed (S630). The blockchain record unit (140) can record the calculated multi-stratified sample group on the blockchain using a smart contract (S640). The public opinion survey unit (150) can investigate and analyze public opinion regarding the calculated multi-stratified sample group using a sentiment analysis algorithm based on a natural language processing algorithm (S650).

[0094]

[0095] As explained above, the present invention classifies a population into several subgroups and automatically extracts representative samples from each subgroup using artificial intelligence technology including statistical methods and machine learning algorithms through multi-stratified sampling via cluster analysis. This allows for sampling that properly reflects the actual opinions of the population without loss of representativeness, such as under- or over-sampling problems of specific population groups, and also enables the automatic calculation of the sample size and error margin of the public opinion survey samples, thereby increasing the consistency and accuracy of the public opinion survey results.

[0096]

[0097] The above embodiments may be implemented using various forms of computing means including one or more processors, memory, and storage means. Additionally, a network interface connected to a wired or wireless network may be included. The processor may be a central processing unit or a semiconductor device that executes processing instructions stored in memory and / or storage units. The memory and storage units may include volatile storage media or non-volatile storage media. For example, the memory may include ROM and RAM. Accordingly, embodiments of the present invention may be implemented as a method implemented by a computer or as a non-transient computer-readable medium having computer-executable instructions stored on said computer. In one embodiment of the present invention, when executed by a processor, the computer-readable instructions may perform a method according to at least one aspect of the present invention.

[0098] Although the present invention has been described with reference to illustrated embodiments as above, these are merely exemplary, and it will be evident to those skilled in the art that various modifications, changes, and equivalent alternative embodiments are possible without departing from the gist and scope of the present invention. For example, the multilayered sampling unit (120) and the sample group calculation unit (130) may be implemented as a single integrated module or divided into two or more devices. Accordingly, the true technical scope of protection of the present invention should be determined by the technical concept of the appended claims.

[0099] The present invention can be used in opinion poll systems that conduct opinion polls by reflecting the actual opinions of a population.

Claims

1. A sample size calculation unit that calculates a sample size based on the size of the target population; A multi-stratification sampling unit that performs multi-stratification sampling on the target population based on the calculated sample size; and A multi-stratification sampling system using a machine learning algorithm comprising a multi-stratification sample group calculation unit that calculates a multi-stratification sample group from the target population based on the multi-stratification sampling performed above.

2. In Claim 1, The above multilayer sampling unit is a multilayer sampling system using a machine learning algorithm that performs multilayer sampling on the target population based on a clustering algorithm and a dimensionality reduction algorithm.

3. In Claim 1, The above-described multi-stratified sampling unit is a multi-stratified sampling system using a machine learning algorithm that performs multi-stratified sampling on the target population based on a k-means-based clustering algorithm and a PCA (Principal Component Analysis)-based dimensionality reduction algorithm.

4. In Claim 1, A multi-layered sampling system using a machine learning algorithm that further includes a blockchain record that records the calculated multi-layered sample group on a blockchain using a smart contract.

5. In Claim 1, A multi-stratified sampling system using a machine learning algorithm, further comprising a public opinion survey unit that investigates and analyzes public opinion regarding the multi-stratified sample group calculated using a sentiment analysis algorithm based on a natural language processing algorithm.

6. In Claim 1, The above-mentioned public opinion survey department is a multi-layered sampling system using a machine learning algorithm that inputs the accumulated results of public opinion surveys over a predetermined period into an artificial intelligence model to identify trends in public opinion fluctuations.

7. In Claim 1, The above-mentioned multi-stratification sampling unit is a multi-stratification sampling system using a machine learning algorithm that performs multi-stratification based on stratification criteria determined by at least one of demographic characteristics, regional characteristics, and socioeconomic characteristics.

8. A step of calculating the sample size based on the size of the target population; A step of performing multistratified sampling on the target population based on the calculated sample size; and A multi-stratification sampling method using a machine learning algorithm comprising the step of calculating a multi-stratification sample group from the target population based on the multi-stratification sampling performed above.

9. In Claim 8, The step of performing multi-stratified sampling on the target population based on the sample size calculated above A multi-stratified sampling method using a machine learning algorithm, comprising the step of performing multi-stratified sampling on the target population based on a clustering algorithm and a dimensionality reduction algorithm.

10. In Claim 9, The step of performing multi-stratified sampling on the target population based on the clustering algorithm and dimensionality reduction algorithm is A multi-stratified sampling method using a machine learning algorithm, comprising the step of performing multi-stratified sampling on the target population based on a k-means-based clustering algorithm and a Principal Component Analysis (PCA)-based dimensionality reduction algorithm.

11. In Claim 8, A multi-stratified sampling method using a machine learning algorithm that further includes the step of recording the calculated multi-stratified sample group on a blockchain using a smart contract.

12. In claim 8, A multi-stratified sampling method using a machine learning algorithm, further comprising the step of investigating and analyzing public opinion regarding the multi-stratified sample group calculated using a sentiment analysis algorithm based on a natural language processing algorithm.

Images