A food flavor assessment method and system based on big data analysis

Abstract

The invention discloses a food flavor evaluation method and system based on big data analysis, which specifically relates to the technical field of food evaluation, and is used to solve the problem of evaluation result deviation caused by sequence effect in the existing traditional flavor evaluation; the food sample flavor evaluation data of multiple evaluators are collected to construct an initial data set; a sequence effect model is established through statistical analysis to quantify the influence of the sequence effect on the evaluation result; the evaluation data of each food sample is evaluated based on the sequence effect model, and if the influence of the sequence effect exceeds a preset threshold, the data is marked for correction; the marked data is adjusted through a correction algorithm to generate a corrected evaluation data set; different evaluation sequence combinations are generated through simulation to analyze whether the corrected data is still affected by the sequence effect; a potential Dirichlet allocation model is applied to identify potential factors in the evaluation data, and the sequence effect caused by the potential factors is evaluated, thereby effectively improving the accuracy and reliability of flavor evaluation.

Description

Technical Field

[0001] The present invention relates to the technical field of food evaluation, and more specifically, to a food flavor evaluation method and system based on big data analysis. Background Art

[0002] In the food industry, flavor evaluation is an important part of evaluating food quality and consumer acceptance. Traditional flavor evaluation usually relies on professional assessors' sensory evaluation of food samples, including subjective scores of multiple dimensions such as aroma, taste, and flavor. However, in the actual evaluation process, the evaluation results are often affected by many factors, among which the evaluation order is a key but often overlooked factor.

[0003] The evaluation order effect refers to the fact that when the assessor continuously evaluates multiple food samples, the previously tasted food samples will interfere with and affect the evaluation of subsequent samples. For example, if the flavor of the food sample evaluated first is strong, it may cause the assessor's taste sense to adapt or fatigue, which will make the flavor evaluation results of subsequent food samples low or inaccurate. On the contrary, if the flavor of the food evaluated first is light, the subsequent samples may be evaluated higher. This order effect will introduce systematic bias and affect the accuracy and reliability of the evaluation results.

[0004] At present, in order to solve the problem of sequence effect in flavor evaluation, traditional methods generally reduce the impact of sequence effect by randomizing the sample order or setting evaluation intervals. However, these methods have limitations in actual operation and cannot completely eliminate the bias caused by sequence effect. How to identify and correct sequence effect in the evaluation process and improve the accuracy of flavor evaluation results has become a technical problem that needs to be solved urgently.

[0005] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0006] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a food flavor assessment method and system based on big data analysis to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A food flavor assessment method based on big data analysis comprises the following steps:

[0009] S1: Collect the flavor evaluation data of food samples by the assessors, record the score, evaluation order and related background information of each assessor, and construct an initial data set including the evaluation score and evaluation order;

[0010] S2: Analyze the initial data set through statistical analysis, establish a sequential effect model for the impact of the assessment order on the assessment results, and quantify the impact of the sequential effect on the assessment accuracy based on the sequential effect model;

[0011] S3: Based on the established sequential effect model, evaluate whether the evaluation data of each food sample is affected by a significant sequential effect. If the degree of sequential effect of the evaluation data exceeds a preset threshold, mark the evaluation data for correction processing;

[0012] S4: adjusting the marked assessment data using a correction algorithm to generate a corrected assessment data set;

[0013] S5: evaluating the corrected assessment data set, including: generating different permutations and combinations of assessment sequences through simulation, analyzing the changing trend of assessment results under each permutation, comparing the simulation results with the actual corrected data, and determining whether there is an uneliminated sequence effect; identifying potential factors that are not directly observed in the assessment data through a latent Dirichlet allocation model, and evaluating the degree of sequence effect caused by the potential factors;

[0014] S6: Based on the evaluation result of step S5, determine whether the corrected evaluation data set is accurate. If it is accurate, form a comprehensive evaluation of the food flavor.

[0015] In a preferred embodiment, S1 specifically includes: S101: selecting assessors covering different ages, genders and eating habits; S102: providing assessors with clear scoring criteria; S103: during the assessment process, recording the assessment order of each assessor; S104: collecting relevant background information of the assessors; S105: constructing an initial data set including assessment scores and assessment order; the initial data set should include each assessor's score for each food sample, the order of the scores and its relevant background information.

[0016] In a preferred embodiment, S2 specifically includes:

[0017] S201: preprocessing the evaluation scores and evaluation sequence data in the initial data set;

[0018] S202: Use multiple regression analysis to analyze the correlation between the evaluation order and the evaluation score of the flavor evaluation data of food samples, and preliminarily identify the existence of the order effect;

[0019] S203: Applying a neural network model to train the preprocessed data, and establishing a sequential effect model of the impact of the evaluation sequence on the evaluation results: using a multi-layer neural network model to train the data in the preprocessed initial data set; the neural network input layer includes the evaluation score, the evaluation sequence and the background information of the assessor, and the output layer is the scoring result predicted by the model; the loss function of the neural network adopts the mean square error;

[0020] S204: Based on the sequential effect model, the sequential effect impact value of each assessment data is calculated, and the impact of the sequential effect on the assessment accuracy is quantified by the sequential effect impact value:

[0021] Calculate the order effect impact value of each evaluation data. The specific calculation formula is as follows: ;in, Indicates The order effect of samples is For the The evaluation order of samples is For the Other variables that may affect the scores of the samples, is the mean of the evaluation order of all samples, is the mean of other variables that may affect the score, and are the corresponding model coefficients.

[0022] In a preferred embodiment, S3 specifically includes:

[0023] S301: for the evaluation data of each food sample, use the sequential effect model to obtain its sequential effect impact value;

[0024] S302: Compare the order effect influence value of each evaluation data with a preset threshold value to determine whether the order effect influence is significant: when the order effect influence value of the evaluation data is greater than the preset threshold value, it is determined that the order effect influence is significant;

[0025] S303: If the order effect of the assessment data is significant, the assessment data is marked as needing correction processing;

[0026] S304: Summarize all the assessment data marked as needing correction processing to form a data set to be corrected.

[0027] In a preferred embodiment, S4 specifically includes:

[0028] S401: Adjust the marked assessment data according to the order effect influence value: the score after initial correction is the difference between the original assessment score and the order effect influence value;

[0029] S402: Select the Bayesian correction method to calculate the adjustment coefficient of each labeled data, and use the adjustment coefficient to correct the deviation of the order effect on the evaluation result:

[0030] Specifically, the following Bayesian formula is used to calculate the adjustment coefficient for labeled data:

[0031] ;in, For a given data The posterior probability of the post-assessment score, is the likelihood function, is the prior probability, is the marginal likelihood;

[0032] S403: Applying the Bayesian correction method to adjust all the labeled data, generating a final corrected evaluation score, and forming a corrected evaluation data set.

[0033] In a preferred embodiment, different permutations and combinations of evaluation sequences are generated by simulation, the changing trend of the evaluation results under each permutation is analyzed, and the simulation results are compared with the actual corrected data to determine whether there is an uneliminated sequence effect, which specifically includes:

[0034] Different permutations and combinations of evaluation sequences are generated through computer simulation to construct multiple simulated evaluation data sets;

[0035] For each simulated evaluation data set, the rater's rating results are:

[0036] ;in, Indicates The rater Food samples in The scoring results under the arrangement are To simulate the sorting order, is the initial rating of the assessor, It is a sequential effect model;

[0037] Calculate the mean and standard deviation for each permutation;

[0038] Compare the simulated assessment results with the corrected assessment data set and calculate the difference index:

[0039] The difference index is calculated using mean square error:

[0040] ;in, is the difference indicator; represents the total number of simulated permutations and combinations; Represents the corrected evaluation data set The final corrected rating score of each food sample is For the Food samples in The mean score under the simulated arrangement;

[0041] When the difference index is greater than the corresponding preset threshold, it is determined that there is an uneliminated sequence effect in the corrected assessment data set; when the difference index is less than or equal to the corresponding preset threshold, it is determined that there is no uneliminated sequence effect in the corrected assessment data set.

[0042] In a preferred embodiment, the latent factors that are not directly observed in the assessment data are identified by the latent Dirichlet allocation model, and the degree of sequential effect caused by the latent factors is evaluated, which specifically includes:

[0043] The latent Dirichlet allocation model was used to analyze the corrected assessment data set to extract potential factors that were not directly observed.

[0044] A model based on multiple linear regression is used to construct a relationship model between potential factors and assessment results. The multiple regression model is expressed as:

[0045] ;in, Indicates The rater The final corrected rating score of each food sample is Indicates The influence coefficient of potential factors on the evaluation score is represents the random error term in the model, It means The rater Potential factors in food samples The weight, represents the number of potential factors;

[0046] Defining a parameterizable variational distribution Used to approximate the true posterior distribution ,in is the corrected assessment data set, It is a potential factor. is the regression coefficient;

[0047] By minimizing the KL divergence Optimize the variational distribution; through the variational EM algorithm, alternately update the parameters of the posterior distribution until convergence; and Make estimates;

[0048] The potential order effect index is calculated as follows: ;in, is the potential order effect index, represents the number of assessors, Indicates the quantity of food sample;

[0049] The potential sequential effect index is compared with its corresponding preset threshold: when the potential sequential effect index is greater than its corresponding preset threshold, the influence of the potential factor on the evaluation result is significant; when the potential sequential effect index is less than or equal to its corresponding preset threshold, the influence of the potential factor on the evaluation result is not significant.

[0050] In a preferred embodiment, S6 is specifically:

[0051] When the difference index is less than or equal to the corresponding preset threshold, and the potential order effect index is less than or equal to the corresponding preset threshold, the corrected assessment data set is determined to be accurate; otherwise, the corrected assessment data set is determined to be inaccurate;

[0052] When the corrected evaluation data set is determined to be accurate, a comprehensive evaluation of the food flavor is formed.

[0053] On the other hand, the present invention provides a food flavor assessment system based on big data analysis, including a data acquisition module, a sequential effect modeling module, a data evaluation marking module, a correction processing module, a correction data evaluation module, and an evaluation result judgment module;

[0054] Data collection module: collects the flavor evaluation data of food samples by assessors, records the score, evaluation order and related background information of each assessor, and constructs an initial data set including evaluation scores and evaluation order;

[0055] Sequential effect modeling module: Analyze the initial data set through statistical analysis, establish a sequential effect model of the impact of the assessment order on the assessment results, and quantify the impact of the sequential effect on the assessment accuracy based on the sequential effect model;

[0056] Data evaluation and marking module: Based on the established sequential effect model, it evaluates whether the evaluation data of each food sample is affected by a significant sequential effect. If the degree of sequential effect of the evaluation data exceeds a preset threshold, the evaluation data is marked for correction processing;

[0057] Correction processing module: uses a correction algorithm to adjust the marked assessment data to generate a corrected assessment data set;

[0058] Correction data evaluation module: evaluates the corrected assessment data set, including: generating different permutations and combinations of assessment sequences through simulation, analyzing the changing trend of assessment results under each permutation, comparing the simulation results with the actual corrected data, and determining whether there is an uneliminated sequence effect; identifying potential factors that are not directly observed in the assessment data through the potential Dirichlet allocation model, and evaluating the degree of sequence effect caused by the potential factors;

[0059] Evaluation result judgment module: Based on the evaluation result of the correction data evaluation module, it is judged whether the corrected evaluation data set is accurate. If it is accurate, a comprehensive evaluation of the food flavor is formed.

[0060] Technical effects and advantages of a food flavor assessment method and system based on big data analysis of the present invention:

[0061] 1. By collecting the scores, evaluation order and related background information of multiple assessors and constructing an initial data set, a sequential effect model is established using statistical analysis technology to quantify the impact of sequential effects on the accuracy of assessments; identify which assessment data exceeds a preset threshold due to the impact of sequential effects, and automatically mark these data for correction processing to ensure that the final assessment data set is more accurate and reliable. In addition, the present invention adjusts the marked data through a correction algorithm, and further evaluates the corrected data by simulating different permutations and combinations of the assessment order to ensure that the sequential effect has been effectively eliminated.

[0062] 2. By introducing the latent Dirichlet allocation model to identify potential factors that are not directly observed in the evaluation data, and evaluating the degree of sequential effects caused by potential factors through the variational inference algorithm, the implicit bias in the evaluation data can be quantified and corrected, ensuring the high accuracy and objectivity of the evaluation results. Finally, through the dual evaluation mechanism difference index and the potential sequential effect index, it is possible to reliably determine whether the corrected evaluation data is accurate, significantly improving the credibility of the flavor evaluation results, and providing a more scientific and effective basis for food quality assessment and product development. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A schematic diagram of a food flavor assessment method based on big data analysis according to the present invention;

[0064] Figure 2 The figure is a schematic diagram of the structure of a food flavor assessment system based on big data analysis according to the present invention. DETAILED DESCRIPTION

[0065] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0066] Example 1

[0067] Figure 1 The present invention provides a food flavor assessment method based on big data analysis, which comprises the following steps:

[0068] S1: Collect the flavor evaluation data of food samples by assessors, record each assessor's score, evaluation order and related background information, and construct an initial data set including evaluation scores and evaluation order.

[0069] S2: Analyze the initial data set through statistical analysis, establish a sequential effect model for the impact of the assessment order on the assessment results, and quantify the degree of influence of the sequential effect on the assessment accuracy based on the sequential effect model.

[0070] S3: Based on the established sequential effect model, evaluate whether the evaluation data of each food sample is affected by a significant sequential effect. If the degree of sequential effect of the evaluation data exceeds a preset threshold, mark the evaluation data for correction processing.

[0071] S4: The marked assessment data is adjusted using a correction algorithm to generate a corrected assessment data set.

[0072] S5: Evaluate the corrected assessment data set, including: generate different permutations and combinations of assessment orders through simulation, analyze the changing trend of assessment results under each permutation, compare the simulation results with the actual corrected data, and determine whether there is an uneliminated sequence effect; identify potential factors that are not directly observed in the assessment data through the latent Dirichlet allocation model, and evaluate the degree of sequence effect caused by the potential factors.

[0073] S6: Based on the evaluation result of step S5, determine whether the corrected evaluation data set is accurate. If it is accurate, form a comprehensive evaluation of the food flavor.

[0074] S1 specifically includes:

[0075] S101: Select a certain number of assessors to ensure sample diversity, covering different ages, genders, and dietary habits to increase the representativeness of the assessment:

[0076] Select a sufficient number of assessors to ensure that the collected data is statistically significant. Specifically, the selection of assessors should be based on the principle of sample diversity, covering different age groups, genders, professional backgrounds and eating habits. This diversity ensures that the flavor evaluation results are representative of a wide range of consumer groups. For example, 20-50 assessors can be selected, with an age range of 18 to 60 years old and a balanced ratio of men to women. When selecting assessors, they should also be ensured to be in good health when participating in the evaluation to avoid sensory abnormalities or physical conditions that affect the evaluation results.

[0077] S102: Provide clear scoring criteria for assessors, including flavor, taste, aroma and other dimensions, using a five-point or ten-point scale to quantify the assessment results:

[0078] The scoring criteria should cover multiple dimensions, including flavor, taste, aroma, appearance and texture. The scoring of each dimension can be based on a five-point system (1 to 5 points) or a ten-point system (1 to 10 points) to ensure that each dimension can be clearly quantified. For example, in the flavor dimension, 1 point means extremely poor flavor and 10 points means excellent flavor; in the taste dimension, 1 point means very bad taste and 10 points means extremely smooth taste. Each assessor needs to have a full understanding of the scoring criteria before participating in the assessment to ensure the consistency and comparability of the data.

[0079] S103: During the evaluation process, record the evaluation order of each assessor, that is, the order in which the assessors score the food samples. This order information is crucial for the subsequent analysis of the order effect:

[0080] During the evaluation process, record the order in which each rater scored the food samples. Evaluation order refers to the process in which raters taste and score multiple food samples one by one in a specific order. For example, the rater may taste sample A first, then sample B, sample C, and so on. The tasting order may be different for each rater, so the specific evaluation order of each rater must be fully recorded. This information is of great value for subsequent sequential effect analysis and helps to identify potential bias caused by the order of evaluation process.

[0081] S104: Collect relevant background information of the assessor, including age, gender, occupation, dietary preferences and allergies to specific foods. This background information will help analyze potential factors affecting the assessment results.

[0082] S105: Construct an initial data set containing evaluation scores and evaluation order. The initial data set should include each assessor's score for each food sample, the order of the scores, and related background information to ensure the integrity and availability of the data.

[0083] S2 specifically includes:

[0084] S201: Preprocess the evaluation scores and evaluation sequence data in the initial data set, clean up outliers, and fill in missing data:

[0085] Outlier cleaning: Use statistical analysis methods, such as box plots, to identify and remove outliers in the evaluation scores. Outliers refer to values ​​that are far away from most data points in the statistical distribution. These values ​​may be caused by misoperation or external interference.

[0086] Missing value filling: For missing data in the initial data set, interpolation or K-nearest neighbor algorithm (KNN) is used to fill in the missing data. For example, if a rater is missing the evaluation data of a sample, it can be filled in based on the rater's scoring pattern for other samples and the scoring data of similar raters.

[0087] S202: Use multiple regression analysis to analyze the correlation between the evaluation order and the evaluation score of the flavor evaluation data of food samples, and preliminarily identify the existence of the order effect:

[0088] Multiple regression analysis was used to explore the correlation between the evaluation order and the evaluation score. The regression model was constructed as follows: ;in, Indicates The evaluation scores of samples, For the The evaluation order of samples is For the Other variables that may affect the scores of the samples (such as the background information of the raters), is a constant term, and is the regression coefficient, is the random error term, is the sample number.

[0089] Through the regression coefficient The significance test can preliminarily identify whether the sequence effect exists significantly.

[0090] S203: Apply the neural network model to train the preprocessed data and establish a sequential effect model of the impact of the evaluation sequence on the evaluation results:

[0091] In order to further accurately model the impact of the evaluation order on the evaluation results, a multi-layer neural network model is used to train the data in the preprocessed initial data set. The neural network consists of an input layer, a hidden layer, and an output layer. The input layer includes the evaluation score, evaluation order, and the background information of the assessor, and the output layer is the scoring result predicted by the model.

[0092] The loss function of the neural network uses the mean square error (MSE), and the formula is as follows: ;in, is the loss function, For the real score, Predict scores for the neural network model, are model parameters, is the sample size, is the sample number.

[0093] By optimizing the loss function, the model can accurately predict the impact of the rating order on the score.

[0094] The training process uses stochastic gradient descent (SGD) to update parameters. The update formula is as follows: ;in, are the updated model parameters, is the learning rate, is the current gradient.

[0095] S204: Based on the established sequential effect model, the sequential effect impact value of each assessment data is calculated, and the impact of the sequential effect on the assessment accuracy is quantified by the sequential effect impact value:

[0096] After the sequential effect model is established, the sequential effect impact value of each assessment data is calculated. The specific calculation formula is as follows: ;in, Indicates The order effect of samples is is the mean of the evaluation order of all samples, is the mean of other variables that may affect the score, and are the corresponding model coefficients.

[0097] By calculating the impact value of the order effect, the degree of influence of the order effect on the assessment accuracy can be quantified.

[0098] The larger the order effect value, the greater the impact of the order effect on the accuracy of the assessment. That is, the larger the order effect value, the more serious the interference of the assessment order on the assessment results, which means that the accuracy of the scoring is more affected by the deviation.

[0099] S3 specifically includes:

[0100] S301: For the evaluation data of each food sample, use the sequential effect model to obtain its sequential effect impact value:

[0101] First, for each food sample's evaluation data, the order effect model established in step S2 is used to calculate the order effect impact value of each evaluation data. Specifically, the order effect impact value is calculated based on the model formula, reflecting the specific degree of influence of the evaluation order on each score.

[0102] S302: Compare the order effect impact value of each evaluation data with a preset threshold value to determine whether the order effect impact is significant:

[0103] Compare the sequential effect impact value of the assessment data with the preset threshold to determine whether its impact is significant. The preset threshold is set based on the historical performance of the data set and statistical methods. It is usually set by standard deviation calculation or empirical value. For example, the preset threshold can be set to the mean of all impact values ​​plus one standard deviation.

[0104] When the order effect impact value of the assessment data is greater than the preset threshold, the order effect is judged to be significant.

[0105] S303: If the order effect of the assessment data is significant, the assessment data is marked as needing correction processing.

[0106] S304: Summarize all the assessment data marked as needing correction processing to form a data set to be corrected.

[0107] S4 specifically includes:

[0108] S401: Adjust the marked assessment data according to the order effect influence value:

[0109] According to the sequential effect influence value, the marked evaluation data is initially adjusted. Specifically, the original evaluation score is corrected using the sequential effect influence value obtained by the sequential effect model, so that the evaluation score can more accurately reflect the true flavor characteristics of the food sample after removing the sequential effect.

[0110] The initial corrected score is the difference between the original assessment score and the influence value of the sequence effect.

[0111] S402: Select the Bayesian correction method, calculate the adjustment coefficient for each labeled data, and use the adjustment coefficient to correct the deviation of the order effect on the evaluation result:

[0112] Commonly used correction algorithms include the inverse probability weighting method and the Bayesian correction method. The Bayesian correction method can make full use of prior information and sample data, and is suitable for situations where the assessment data has complex characteristics. The Bayesian correction method corrects the deviation caused by the sequence effect by constructing the posterior distribution of the sample.

[0113] Specifically, the following Bayesian formula is used to calculate the adjustment coefficient for labeled data: ;in, For a given data the posterior probability of the post-assessment score; is the likelihood function, which indicates the data in the model parameters The probability of the following; is the prior probability, which indicates the prior knowledge of the model parameters; is the marginal likelihood.

[0114] The adjustment coefficient of each labeled data is calculated by the Bayesian formula to correct the scoring deviation. The adjustment coefficient of the labeled data is usually calculated in the posterior analysis to adjust the evaluation results to make it more accurately reflect the actual situation, which will not be repeated here.

[0115] in, is the representation of the model parameters. Represents an observational dataset, which is based on the assessors' scoring results of food samples, as well as the relevant assessment order and background information.

[0116] The "bias" in the evaluation results refers to the inaccurate scoring caused by the order effect. The order effect refers to the fact that when the assessor evaluates samples continuously, the scoring is affected by the order of sample evaluation. Specifically, the difference in characteristics between the previous and next samples may cause the assessor to deviate from the current sample's score. For example, if the sample with a stronger taste is tasted first, the subsequent sample may be scored lower. This kind of deviation affects the evaluation results, making the score unable to truly reflect the actual characteristics of the sample.

[0117] S403: Apply the Bayesian correction method to adjust all the labeled data, generate the final corrected evaluation score, and form a corrected evaluation data set:

[0118] All assessment data marked as needing correction are subjected to Bayesian correction. The specific adjustment steps are as follows:

[0119] Using the calculated posterior probabilities, each rater's score is adjusted to more accurately reflect the true rating results.

[0120] The final corrected assessment score was calculated by applying a Bayesian correction method, which was obtained by multiplying the initial corrected score with the adjustment factor of the labeled data.

[0121] The final corrected assessment scores are integrated to form a corrected assessment data set.

[0122] By simulating different permutations and combinations of the evaluation sequence, we analyzed the changing trend of the evaluation results under each permutation, compared the simulation results with the actual corrected data, and determined whether there was an uneliminated sequence effect, including:

[0123] Different permutations and combinations of evaluation orders were generated through computer simulation to construct multiple simulated evaluation data sets:

[0124] Computer simulation is used to generate multiple permutations of the evaluation order to ensure that the evaluation order of each food sample covers as many possible situations as possible in different permutations. Due to the limitation of computing resources, multiple representative permutations are usually selected through random sampling or stratified sampling methods.

[0125] In order to ensure the coverage of the simulated data, the Monte Carlo method can be used for random sampling to generate multiple evaluation order arrangements to ensure the diversity and representativeness of the arrangement samples.

[0126] For each simulated evaluation data set, analyze the changing trend of the evaluation results under different arrangements:

[0127] After generating the simulated evaluation data set, for each simulated evaluation data set, the changing trend of the evaluation results under different evaluation orders is analyzed; each permutation combination corresponds to a simulated evaluation data set, and the scoring results under the permutation are analyzed.

[0128] For each simulated evaluation data set, the rater's rating results can be described by the following formula: ;in, Indicates The rater Food samples in The scoring results under the arrangement are To simulate the sorting order, is the initial rating of the assessor, This is a sequential effect model.

[0129] : The index of the assessor, the value range is .

[0130] : The index of the food sample, the value range is .

[0131] : The index of the permutation combination, indicating the Permutations.

[0132] represents the number of assessors, Indicates the quantity of food sample.

[0133] Compute the mean and standard deviation for each permutation:

[0134] ;in, For the Food samples in The mean score under the simulated arrangement.

[0135] ;in, To indicate the Food samples in The standard deviation of the scores under the simulated arrangement.

[0136] By analyzing the changing trends of these statistics, we can evaluate the overall impact of the evaluation order on the results.

[0137] Compare the simulated assessment results with the corrected assessment data set and calculate the difference index:

[0138] After analyzing the changing trends under different arrangements, it is necessary to compare these simulation results with the actual corrected assessment data to evaluate their differences. By defining the difference index, the distance or difference between the simulation assessment results and the corrected data can be quantified.

[0139] The difference index is calculated using the mean square error (MSE): ;in, is the difference index, which means The mean square error of the food samples; represents the total number of simulated permutations and combinations; Represents the corrected evaluation data set The final corrected rating score for each food sample.

[0140] Evaluate the difference index to determine whether there are uneliminated order effects in the corrected assessment data set:

[0141] After calculating the difference index, we can determine whether there is still an uneliminated order effect by analyzing whether the difference index exceeds the preset threshold. The specific preset threshold can be determined based on historical data or statistical analysis. For example, the preset threshold of the difference index can be set to a range where the mean square error is below a certain percentile:

[0142] When the difference index is greater than its corresponding preset threshold, it is determined that there is an uneliminated sequence effect in the corrected assessment data set, indicating that there is still a large deviation between the simulated assessment results and the corrected data, which means that the sequence effect has not been completely eliminated.

[0143] When the difference index is less than or equal to its corresponding preset threshold, it is determined that there is no uneliminated sequence effect in the corrected assessment data set.

[0144] In addition, statistical test methods (such as t-test) can be used to analyze the significant differences between the simulation results and the calibration data. If the difference is significant, the model needs to be further modified or the calibration algorithm needs to be adjusted.

[0145] The latent Dirichlet allocation model is used to identify potential factors that are not directly observed in the assessment data and to evaluate the degree of sequential effects caused by potential factors, including:

[0146] The latent Dirichlet allocation model is used to analyze the corrected rating data set to extract potential factors that are not directly observed:

[0147] The corrected rating data set was analyzed using the Latent Dirichlet Allocation (LDA) model. LDA is a generative probabilistic model that identifies latent factors in the data that are not directly observed. These latent factors may represent the raters' sensitivity to certain specific flavor characteristics or the influence of the rating order that was not noticed during the rating process.

[0148] The latent factor distribution of each rater follows a Dirichlet distribution: ;in, Indicates The latent factor distribution vector of the raters, is a hyperparameter of the Dirichlet distribution (representing prior knowledge of the latent factors).

[0149] For each assessor and each food sample, the assessment score Generated by the following process:

[0150] Select a latent factor from the latent factor distribution ;

[0151] According to the potential factors selected , from the feature distribution Generated evaluation score: .

[0152] This model can be used to identify the weights that each rater places on different potential factors. And the characteristic distribution of each latent factor These potential factors It reflects structures or patterns in the rating data that are not directly observed and may represent some potential preferences or effects that affect flavor perception or rating order.

[0153] Indicates The feature distribution vector of latent factors, Represents the number of potential factors.

[0154] Construct a relationship model between potential factors and assessment results to quantify the impact of potential factors on assessment results:

[0155] After identifying the potential factors, a relationship model is constructed to quantify the impact of these potential factors on the assessment results. A model based on multiple linear regression is used to represent the potential factors. How to influence each rater's rating.

[0156] The multiple regression model is expressed as: ;in, Indicates The rater The final corrected rating score of each food sample is represents the first The rater Potential factors in food samples, Indicates The influence coefficient of potential factors on the evaluation score is represents the random error term in the model, It means The rater Potential factors in food samples The amount of

[0157] The multivariate regression model is used to quantify the influence of each potential factor on the final assessment score, especially the potential bias of the assessment order on the score. It can reveal the complex structure in the assessment data, not only identify the implicit influence of the assessment order on the scoring results, but also provide a basis for the subsequent correction model, thereby reducing the interference of the order effect on the final score.

[0158] Apply variational inference algorithms to assess the extent of sequential effects caused by latent factors:

[0159] Variational inference infers latent variables by approximating the posterior distribution and model parameters distribution.

[0160] The specific steps are as follows:

[0161] Defining a parameterizable variational distribution , which is used to approximate the true posterior distribution ,in is the corrected assessment data set, It is a potential factor. is the regression coefficient.

[0162] By minimizing the Kullback-Leibler (KL) divergence To optimize the variational distribution. The KL divergence is defined as follows: ; KL divergence measures the variational distribution and the true posterior distribution The difference between.

[0163] Through the variational EM algorithm, the parameters of the posterior distribution are updated alternately until convergence. Finally, the estimated and Used to assess the impact of potential factors.

[0164] The potential order effect index is calculated as follows: ;in, is the potential order effect index.

[0165] The larger the potential order effect index is, the more significant the order effect caused by the potential factors is. This means that the potential factors have a stronger impact on the assessment results, and the assessment results may be more affected by the order effect, thereby reducing the accuracy of the assessment.

[0166] Compare the potential order effect index with its corresponding preset threshold:

[0167] When the potential sequential effect index is greater than its corresponding preset threshold, it means that the potential factor has a significant impact on the assessment result, and the assessment result may be affected by a large sequential effect. Therefore, correction processing is required to improve the accuracy of the assessment.

[0168] When the potential sequential effect index is less than or equal to its corresponding preset threshold, it means that the potential factor has no significant impact on the assessment result, and the assessment result is basically not significantly affected by the sequential effect. It can usually be considered that the assessment result is reliable and does not require further correction.

[0169] S6 is specifically:

[0170] When the difference index is less than or equal to the corresponding preset threshold, and the potential sequence effect index is less than or equal to the corresponding preset threshold, the corrected evaluation data set is determined to be accurate; at this time, it means that the corrected evaluation data set performs well in terms of consistency between the evaluation results and the simulation results, and the influence of the sequence effect on the evaluation results is also within a reasonable range. Therefore, it can be considered that the corrected data set is accurate, and the comprehensive evaluation of food flavor can be directly based on this data set.

[0171] Otherwise, the corrected assessment data set is judged to be inaccurate, indicating that the corrected assessment data set still has large errors or sequence effects that have not been completely eliminated, and the assessment data may still be affected by potential factors. In this case, the correction result is judged to be inaccurate, and the assessment data needs to be further optimized or recalibrated.

[0172] When the corrected evaluation data set is determined to be accurate, a comprehensive evaluation of the food flavor is formed.

[0173] The comprehensive evaluation is based on the final corrected scores of each assessor for different food samples, combined with the scores of different flavor dimensions (such as aroma, taste, aftertaste, etc.) for weighted calculation. The final comprehensive evaluation result can reflect the overall flavor score of each food sample and can accurately reflect the real perception of consumers.

[0174] By considering both the difference index and the potential order effect index, the double standard was used to evaluate the corrected assessment data set, which significantly improved the accuracy. Compared with the single indicator judgment, this method is more comprehensive and can solve the problems of explicit and implicit errors at the same time.

[0175] The potential order effect index is calculated through the latent variable model and variational inference algorithm. It can effectively identify and quantify the order effects that are not directly observed in the evaluation data, accurately evaluate the impact of implicit factors on the evaluation results, and ensure the reliability of the evaluation results. The corrected data is used to form a comprehensive evaluation result of food flavor, which not only ensures the accuracy of the evaluation, but also can analyze different flavor dimensions to obtain a more comprehensive flavor evaluation that is more in line with consumer perception.

[0176] Example 2

[0177] The difference between Example 2 of the present invention and Example 1 is that this example introduces a food flavor assessment system based on big data analysis.

[0178] Figure 2 A structural schematic diagram of a food flavor evaluation system based on big data analysis of the present invention is given, a food flavor evaluation system based on big data analysis, including a data acquisition module, a sequential effect modeling module, a data evaluation marking module, a correction processing module, a correction data evaluation module and an evaluation result judgment module.

[0179] Data collection module: collects the flavor evaluation data of food samples by assessors, records the score, evaluation order and related background information of each assessor, and constructs an initial data set including the evaluation score and evaluation order.

[0180] Sequential effect modeling module: Analyze the initial data set through statistical analysis, establish a sequential effect model of the impact of the assessment order on the assessment results, and quantify the impact of the sequential effect on the assessment accuracy based on the sequential effect model.

[0181] Data evaluation and marking module: Based on the established sequential effect model, it evaluates whether the assessment data of each food sample is affected by a significant sequential effect. If the degree of sequential effect of the assessment data exceeds a preset threshold, the assessment data is marked for correction processing.

[0182] Correction processing module: The marked assessment data is adjusted using a correction algorithm to generate a corrected assessment data set.

[0183] Correction data evaluation module: evaluates the corrected assessment data set, including: generating different permutations and combinations of assessment sequences through simulation, analyzing the changing trend of assessment results under each permutation, comparing the simulation results with the actual corrected data, and determining whether there are uneliminated sequence effects; identifying potential factors that are not directly observed in the assessment data through the latent Dirichlet allocation model, and evaluating the degree of sequence effects caused by the potential factors.

[0184] Evaluation result judgment module: Based on the evaluation result of the correction data evaluation module, it is judged whether the corrected evaluation data set is accurate. If it is accurate, a comprehensive evaluation of the food flavor is formed.

[0185] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0186] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or may be transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium may be a solid-state hard disk.

[0187] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0188] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0189] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0190] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0191] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0192] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0193] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0194] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A food flavor assessment method based on big data analysis, characterized in that: The steps include: S1: Collect the flavor evaluation data of food samples by the assessors, record the score, evaluation order and related background information of each assessor, and construct an initial data set including the evaluation score and evaluation order; S2: Analyze the initial data set through statistical analysis, establish a sequential effect model for the impact of the assessment order on the assessment results, and quantify the impact of the sequential effect on the assessment accuracy based on the sequential effect model; S3: Based on the established sequential effect model, evaluate whether the evaluation data of each food sample is affected by a significant sequential effect. If the degree of sequential effect of the evaluation data exceeds a preset threshold, mark the evaluation data for correction processing; S4: adjusting the marked assessment data using a correction algorithm to generate a corrected assessment data set; S5: evaluating the corrected assessment data set, including: generating different permutations and combinations of assessment sequences through simulation, analyzing the changing trend of assessment results under each permutation, comparing the simulation results with the actual corrected data, and determining whether there is an uneliminated sequence effect; identifying potential factors that are not directly observed in the assessment data through a latent Dirichlet allocation model, and evaluating the degree of sequence effect caused by the potential factors; S6: Based on the evaluation result of step S5, determine whether the corrected evaluation data set is accurate, and if accurate, form a comprehensive evaluation of the food flavor; A sequential effect model was established to measure the impact of the evaluation sequence on the evaluation results. The impact of the sequential effect on the evaluation accuracy was quantified based on the sequential effect model. Specifically, A multi-layer neural network model is used to train the data in the preprocessed initial data set; the neural network input layer includes the evaluation score, evaluation order and the background information of the assessor, and the output layer is the scoring result predicted by the model; the loss function of the neural network adopts the mean square error; Calculate the order effect impact value of each evaluation data. The specific calculation formula is as follows: ;in, Indicates The order effect of samples is For the The evaluation order of samples is For the Other variables that may affect the scores of the samples, is the mean of the evaluation order of all samples, is the mean of other variables that may affect the score, and is the corresponding model coefficient; The marked assessment data is adjusted using a correction algorithm to generate a corrected assessment data set, specifically including: S401: Adjust the marked assessment data according to the order effect influence value: the score after initial correction is the difference between the original assessment score and the order effect influence value; S402: Selecting the Bayesian correction method to calculate the adjustment coefficient of each labeled data, and using the adjustment coefficient to correct the deviation of the order effect on the assessment result; S403: Applying the Bayesian correction method to adjust all the labeled data, generating a final corrected evaluation score, and forming a corrected evaluation data set.

2. The food flavor assessment method based on big data analysis according to claim 1, characterized in that: S1 specifically includes: S101: Select assessors, covering different ages, genders and eating habits; S102: Provide assessors with clear scoring criteria; S103: During the assessment process, record the assessment order of each assessor; S104: Collect relevant background information of assessors; S105: Construct an initial data set containing assessment scores and assessment order; the initial data set should include each assessor's score for each food sample, the scoring order and its relevant background information.

3. A food flavor assessment method based on big data analysis according to claim 2, characterized in that: Apply the neural network model to train the preprocessed data, including: S201: preprocessing the evaluation scores and evaluation sequence data in the initial data set; S202: Use multiple regression analysis to analyze the correlation between the evaluation order and the evaluation score of the flavor evaluation data of the food samples, and preliminarily identify the existence of the order effect.

4. The method for evaluating food flavor based on big data analysis according to claim 3, characterized in that: S3 specifically includes: S301: for the evaluation data of each food sample, use the sequential effect model to obtain its sequential effect impact value; S302: Compare the order effect influence value of each evaluation data with a preset threshold value to determine whether the order effect influence is significant: when the order effect influence value of the evaluation data is greater than the preset threshold value, it is determined that the order effect influence is significant; S303: If the order effect of the assessment data is significant, the assessment data is marked as needing correction processing; S304: Summarize all the assessment data marked as needing correction processing to form a data set to be corrected.

5. The method for evaluating food flavor based on big data analysis according to claim 4, characterized in that: The following Bayesian formula is used to calculate the adjustment coefficient for the labeled data: ;in, For a given data The posterior probability of the post-assessment score, is the likelihood function, is the prior probability, is the marginal likelihood.

6. A food flavor assessment method based on big data analysis according to claim 5, characterized in that: By simulating different permutations and combinations of the evaluation sequence, we analyzed the changing trend of the evaluation results under each permutation, compared the simulation results with the actual corrected data, and determined whether there was an uneliminated sequence effect, including: Different permutations and combinations of evaluation sequences are generated through computer simulation to construct multiple simulated evaluation data sets; For each simulated evaluation data set, the rater's rating results are: ;in, Indicates The rater Food samples in The scoring results under the arrangement are To simulate the sorting order, is the initial rating value of the assessor, It is a sequential effect model; Calculate the mean and standard deviation for each permutation; Compare the simulated assessment results with the corrected assessment data set and calculate the difference index: The difference index is calculated using mean square error: ;in, is the difference indicator; represents the total number of simulated permutations and combinations; Represents the corrected evaluation data set The final corrected rating score of each food sample is For the Food samples in The mean score under the simulated arrangement; When the difference index is greater than the corresponding preset threshold, it is determined that there is an uneliminated sequence effect in the corrected assessment data set; when the difference index is less than or equal to the corresponding preset threshold, it is determined that there is no uneliminated sequence effect in the corrected assessment data set.

7. The method for evaluating food flavor based on big data analysis according to claim 6, characterized in that: The latent Dirichlet allocation model is used to identify potential factors that are not directly observed in the assessment data and to evaluate the degree of sequential effects caused by potential factors, including: The latent Dirichlet allocation model was used to analyze the corrected assessment data set to extract potential factors that were not directly observed. A model based on multiple linear regression is used to construct a relationship model between potential factors and assessment results. The multiple regression model is expressed as: ;in, Indicates The rater The final corrected rating score of each food sample is Indicates The influence coefficient of potential factors on the evaluation score is represents the random error term in the model, It means The rater Potential factors in food samples The weight, represents the number of potential factors; Defining a parameterizable variational distribution Used to approximate the true posterior distribution ,in is the corrected assessment data set, It is a potential factor. is the regression coefficient; By minimizing the KL divergence Optimize the variational distribution; through the variational EM algorithm, alternately update the parameters of the posterior distribution until convergence; and Make estimates; The potential order effect index is calculated as follows: ;in, is the potential order effect index, represents the number of assessors, Indicates the quantity of food sample; The potential sequential effect index is compared with its corresponding preset threshold: when the potential sequential effect index is greater than its corresponding preset threshold, the influence of the potential factor on the evaluation result is significant; when the potential sequential effect index is less than or equal to its corresponding preset threshold, the influence of the potential factor on the evaluation result is not significant.

8. The method for evaluating food flavor based on big data analysis according to claim 7, characterized in that: S6 is specifically: When the difference index is less than or equal to the corresponding preset threshold, and the potential order effect index is less than or equal to the corresponding preset threshold, the corrected assessment data set is determined to be accurate; otherwise, the corrected assessment data set is determined to be inaccurate; When the corrected evaluation data set is determined to be accurate, a comprehensive evaluation of the food flavor is formed.

9. A food flavor assessment system based on big data analysis, used to implement a food flavor assessment method based on big data analysis as claimed in any one of claims 1 to 8, characterized in that: It includes a data acquisition module, a sequential effect modeling module, a data evaluation marking module, a correction processing module, a correction data evaluation module and an evaluation result judgment module; Data collection module: collects the flavor evaluation data of food samples by assessors, records the score, evaluation order and related background information of each assessor, and constructs an initial data set including evaluation scores and evaluation order; Sequential effect modeling module: Analyze the initial data set through statistical analysis, establish a sequential effect model of the impact of the assessment order on the assessment results, and quantify the impact of the sequential effect on the assessment accuracy based on the sequential effect model; Data evaluation and marking module: Based on the established sequential effect model, it evaluates whether the evaluation data of each food sample is affected by a significant sequential effect. If the degree of sequential effect of the evaluation data exceeds a preset threshold, the evaluation data is marked for correction processing; Correction processing module: uses a correction algorithm to adjust the marked assessment data to generate a corrected assessment data set; Correction data evaluation module: evaluates the corrected assessment data set, including: generating different permutations and combinations of assessment sequences through simulation, analyzing the changing trend of assessment results under each permutation, comparing the simulation results with the actual corrected data, and determining whether there is an uneliminated sequence effect; The latent Dirichlet allocation model is used to identify the latent factors in the assessment data that are not directly observed, and to evaluate the degree of sequential effects caused by the latent factors. Evaluation result judgment module: Based on the evaluation result of the correction data evaluation module, it is judged whether the corrected evaluation data set is accurate. If it is accurate, a comprehensive evaluation of the food flavor is formed.

Images